changes claude never committed

.gitignore

@@ -0,0 +1,3 @@
+/target
+/ci_out
+Cargo.lock

Cargo.toml

@@ -0,0 +1,23 @@
+[workspace]
+resolver = "2"
+members = [
+    "crates/world_model",
+    "crates/rune_ir",
+    "crates/trace_model",
+    "crates/generators",
+    "crates/reference_runtime",
+    "crates/runtime_under_test",
+    "crates/collapse_analysis",
+    "crates/semantic_mutation",
+    "crates/replay_corpus",
+    "crates/ci_reports",
+]
+
+[workspace.package]
+version = "0.1.0"
+edition = "2021"
+license = "MIT"
+
+[profile.release]
+opt-level = 3
+lto = "thin"

README.md

@@ -0,0 +1,92 @@
+# Magicka VM — Phase 0/1
+
+> The deliverable is a Rust engine whose tests make a fake universe fail.
+
+This repository implements the Phase 0/1 specification in `plan.md`: an
+**adversarial testing framework first**, then a **reference runtime** that
+passes it, then a **runtime under test** that matches the reference. No spell
+content, templates, or cosmetic runes — the value is in the tests that refuse
+to let the universe collapse into a single score, resource, effect axis,
+executor, rune, hidden formula, or decorative domain.
+
+## Workspace layout
+
+Built in the mandatory order from the spec:
+
+| # | Crate | Role |
+|---|-------|------|
+| 1 | `world_model` | 8 independent domains, world snapshot, perturbation axes, deltas, deterministic primitives (ids, stable hash, RNG) |
+| – | `rune_ir` | Rune token / program model (no stream is ever rejected) |
+| 2 | `trace_model` | Execution trace + all graphs, behavior fingerprint, replay record, fault log, trace metrics |
+| 3 | `generators` | Worlds, programs, executors, contracts, perturbations; rejects flat cases |
+| 4 | `collapse_analysis` | The 11 compression attacks + collapse gates |
+| 5 | `semantic_mutation` | Structurally generated mutant runtimes; proves every one is killed |
+| 6 | `replay_corpus` | Permanent, bit-exact replay cases |
+| 7 | `reference_runtime` | The executable spec engine (`Runtime` trait, `resolve`) |
+| 8 | `runtime_under_test` | Config-driven engine, proven equivalent to the reference |
+| – | `ci_reports` | Orchestrator + `ci` binary; emits the 8 required reports |
+
+The optimized runtime may not begin before steps 1–7 pass CI; until then the
+runtime under test is the reference engine driven through the same config
+surface, which is equivalent by construction. Mutation swaps in a *mutated*
+config to prove the suite detects any divergence.
+
+## The engine in one paragraph
+
+A world is 8 domains, each with 4 observed + 2 hidden integer lanes, a dense
+8×8 coupling matrix, partial observability, and pending scheduled effects. A
+rune program is interpreted under ≥3 executors; each opcode reads several
+domains, mixes them through a nonlinear avalanche keyed by per-domain
+constants, the world coupling, and the executor's salt, then writes back —
+recording causal/read/write/information-flow/temporal edges as it goes.
+Scheduled effects and coupling diffusion propagate changes 3 turns into the
+future. Because every output bit depends on all inputs, the universe is
+high-rank, incompressible, future-sensitive, and executor-divergent — exactly
+the properties the gates demand.
+
+## Running CI
+
+```bash
+cargo test                                   # unit tests for every crate
+cargo run --release -p ci_reports --bin ci   # full gate run (fast scale)
+```
+
+Reports are written to `ci_out/` (8 JSON files + `ci_summary.md`). The binary
+exits non-zero if any gate fails.
+
+### Scales
+
+`MAGICKA_SCALE` selects the corpus size; individual counts can be overridden.
+
+| Scale | executions | replay | mutants | notes |
+|-------|-----------|--------|---------|-------|
+| `tiny` | 120 | 120 | 520 | smoke (~80 ms) |
+| `fast` (default) | 600 | 600 | 520 | every gate, ~0.3 s |
+| `full` | 1,000,000 | 10,000 | 600 | merge-blocking spec gates |
+
+```bash
+MAGICKA_SCALE=full cargo run --release -p ci_reports --bin ci
+# or override individual counts:
+MAGICKA_EXECUTIONS=20000 MAGICKA_REPLAY=10000 cargo run --release -p ci_reports --bin ci
+```
+
+The gate *thresholds* are identical across scales — only the corpus size
+changes. Fast CI runs a representative slice; merge-blocking CI runs `full`.
+
+## Gates enforced (all must pass)
+
+- **runtime_equivalence** — 100% of executions: `canonical(reference) == canonical(runtime_under_test)` over delta, trace, faults, replay hash, and 3-turn future hash.
+- **causal_rank / trace** — median causal edges ≥ 24, 95% causal rank ≥ 6, median touched domains ≥ 4, 95% ≥ 3, fingerprint collision rate < 5%, largest cluster < 2%.
+- **domain_participation** — each domain appears in ≥ 35% of traces, influences ≥ 20%, is mutated in ≥ 20%; removing any domain loses ≥ 10% behavioral diversity; merging any pair loses ≥ 8%; no read-only or write-only domain.
+- **metamorphic_response** — ≥ 90% of perturbations alter the trace, ≥ 75% the delta, ≥ 50% the 3-turn future; ≤ 5% unexplained neutral.
+- **compression_resistance** — best 1/2/4-factor models predict < 40/55/70%; no single domain > 30%, no pair > 55%; every compressed model loses ≥ 35% information. All 11 attack families are run.
+- **mutation_survivor** — ≥ 500 structurally generated mutants, 0 survivors.
+- **contract** — every admitted case satisfies its semantic contract (min causal rank, domain participation, future sensitivity, context divergence, max compressibility); contract-violating cases are rejected at admission.
+- **replay** — 100% deterministic, 0 hash drift.
+- **coverage** — generated/contract rejection accounting; no admitted case fails the generated gates.
+
+## Determinism
+
+Everything is seed-derived and integer-only (SplitMix64 RNG, FNV-1a content
+hashing, wrapping/guarded arithmetic). No floating point enters a canonical
+hash, so replay is bit-exact across machines and runs. No external crates.

crates/ci_reports/Cargo.toml

@@ -0,0 +1,23 @@
+[package]
+name = "ci_reports"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+rune_ir = { path = "../rune_ir" }
+trace_model = { path = "../trace_model" }
+generators = { path = "../generators" }
+reference_runtime = { path = "../reference_runtime" }
+runtime_under_test = { path = "../runtime_under_test" }
+collapse_analysis = { path = "../collapse_analysis" }
+semantic_mutation = { path = "../semantic_mutation" }
+replay_corpus = { path = "../replay_corpus" }
+
+[lib]
+path = "src/lib.rs"
+
+[[bin]]
+name = "ci"
+path = "src/main.rs"

crates/ci_reports/src/json.rs

@@ -0,0 +1,94 @@
+//! Minimal hand-rolled JSON value + pretty printer (no external crates).
+
+pub enum Json {
+    Null,
+    Bool(bool),
+    Int(i64),
+    Num(f64),
+    Str(String),
+    Arr(Vec<Json>),
+    Obj(Vec<(String, Json)>),
+}
+
+impl Json {
+    pub fn s(v: impl Into<String>) -> Json {
+        Json::Str(v.into())
+    }
+    pub fn to_pretty(&self) -> String {
+        let mut out = String::new();
+        self.write(&mut out, 0);
+        out.push('\n');
+        out
+    }
+
+    fn write(&self, out: &mut String, indent: usize) {
+        match self {
+            Json::Null => out.push_str("null"),
+            Json::Bool(b) => out.push_str(if *b { "true" } else { "false" }),
+            Json::Int(i) => out.push_str(&i.to_string()),
+            Json::Num(f) => {
+                if f.is_finite() {
+                    out.push_str(&format!("{:.6}", f));
+                } else {
+                    out.push_str("null");
+                }
+            }
+            Json::Str(s) => {
+                out.push('"');
+                for c in s.chars() {
+                    match c {
+                        '"' => out.push_str("\\\""),
+                        '\\' => out.push_str("\\\\"),
+                        '\n' => out.push_str("\\n"),
+                        '\t' => out.push_str("\\t"),
+                        _ => out.push(c),
+                    }
+                }
+                out.push('"');
+            }
+            Json::Arr(items) => {
+                if items.is_empty() {
+                    out.push_str("[]");
+                    return;
+                }
+                out.push_str("[\n");
+                for (i, it) in items.iter().enumerate() {
+                    pad(out, indent + 1);
+                    it.write(out, indent + 1);
+                    if i + 1 < items.len() {
+                        out.push(',');
+                    }
+                    out.push('\n');
+                }
+                pad(out, indent);
+                out.push(']');
+            }
+            Json::Obj(fields) => {
+                if fields.is_empty() {
+                    out.push_str("{}");
+                    return;
+                }
+                out.push_str("{\n");
+                for (i, (k, v)) in fields.iter().enumerate() {
+                    pad(out, indent + 1);
+                    out.push('"');
+                    out.push_str(k);
+                    out.push_str("\": ");
+                    v.write(out, indent + 1);
+                    if i + 1 < fields.len() {
+                        out.push(',');
+                    }
+                    out.push('\n');
+                }
+                pad(out, indent);
+                out.push('}');
+            }
+        }
+    }
+}
+
+fn pad(out: &mut String, indent: usize) {
+    for _ in 0..indent {
+        out.push_str("  ");
+    }
+}

crates/ci_reports/src/lib.rs

@@ -0,0 +1,784 @@
+//! `ci_reports` — the CI orchestrator. Runs the whole adversarial framework
+//! against the reference runtime and the runtime under test, evaluates every
+//! gate, and emits machine-readable JSON plus a human-readable markdown
+//! summary. Merge is blocked unless all reports pass.
+
+pub mod json;
+
+use collapse_analysis::{analyze, BehaviorCorpus, CollapseSummary};
+use generators::{generate_accepted_case, generated_gate_failures, GeneratedCase};
+use reference_runtime::{canonical, execute, EngineConfig, ResolutionInput, ResolutionResult, Runtime};
+use runtime_under_test::RuntimeUnderTest;
+use semantic_mutation::{run_suite, MutationOutcome};
+use std::collections::HashMap;
+use world_model::{Hash, WorldSnapshot, HIDDEN_LANES, LANES, NUM_DOMAINS};
+
+// ---------------------------------------------------------------------------
+// Scale configuration.
+// ---------------------------------------------------------------------------
+
+#[derive(Clone, Copy, Debug)]
+pub struct Scale {
+    pub executions: usize,
+    pub mutants: usize,
+    pub replay_cases: usize,
+    pub collapse_samples: usize,
+    pub domain_probe_cases: usize,
+    pub domain_probe_variations: usize,
+}
+
+impl Scale {
+    /// The full, merge-blocking gates mandated by the spec.
+    pub fn full() -> Self {
+        Scale {
+            executions: 1_000_000,
+            mutants: 600,
+            replay_cases: 10_000,
+            collapse_samples: 5_000,
+            domain_probe_cases: 2_000,
+            domain_probe_variations: 64,
+        }
+    }
+    /// Fast CI (~10% spirit): small but exercises every gate.
+    pub fn fast() -> Self {
+        Scale {
+            executions: 600,
+            mutants: 520,
+            replay_cases: 600,
+            collapse_samples: 600,
+            domain_probe_cases: 200,
+            domain_probe_variations: 48,
+        }
+    }
+    pub fn tiny() -> Self {
+        Scale {
+            executions: 120,
+            mutants: 520,
+            replay_cases: 120,
+            collapse_samples: 120,
+            domain_probe_cases: 60,
+            domain_probe_variations: 32,
+        }
+    }
+
+    pub fn from_env() -> Self {
+        let mut s = match std::env::var("MAGICKA_SCALE").as_deref() {
+            Ok("full") => Scale::full(),
+            Ok("tiny") => Scale::tiny(),
+            _ => Scale::fast(),
+        };
+        if let Some(v) = env_usize("MAGICKA_EXECUTIONS") {
+            s.executions = v;
+        }
+        if let Some(v) = env_usize("MAGICKA_MUTANTS") {
+            s.mutants = v;
+        }
+        if let Some(v) = env_usize("MAGICKA_REPLAY") {
+            s.replay_cases = v;
+        }
+        if let Some(v) = env_usize("MAGICKA_COLLAPSE") {
+            s.collapse_samples = v;
+        }
+        s
+    }
+}
+
+fn env_usize(key: &str) -> Option<usize> {
+    std::env::var(key).ok().and_then(|v| v.parse().ok())
+}
+
+// ---------------------------------------------------------------------------
+// Helpers.
+// ---------------------------------------------------------------------------
+
+fn case_seed(i: usize) -> u64 {
+    0xC0FFEE_u64 ^ (i as u64).wrapping_mul(0x9E3779B97F4A7C15)
+}
+
+fn input_from_case(case: &GeneratedCase) -> ResolutionInput {
+    ResolutionInput {
+        world: case.world.clone(),
+        program: case.program.clone(),
+        contexts: case.contexts.clone(),
+        contract_seed: case.contract_seed,
+        perturbation_seed: case.perturbation_seed,
+    }
+}
+
+fn input_with_world(case: &GeneratedCase, world: WorldSnapshot) -> ResolutionInput {
+    ResolutionInput {
+        world,
+        program: case.program.clone(),
+        contexts: case.contexts.clone(),
+        contract_seed: case.contract_seed,
+        perturbation_seed: case.perturbation_seed,
+    }
+}
+
+fn masked_config(d: usize) -> EngineConfig {
+    let mut c = EngineConfig::reference();
+    c.domain_mask[d] = false;
+    c
+}
+
+fn median(v: &[f64]) -> f64 {
+    if v.is_empty() {
+        return 0.0;
+    }
+    let mut s = v.to_vec();
+    s.sort_by(|a, b| a.partial_cmp(b).unwrap());
+    s[s.len() / 2]
+}
+
+fn percentile(v: &[f64], p: f64) -> f64 {
+    if v.is_empty() {
+        return 0.0;
+    }
+    let mut s = v.to_vec();
+    s.sort_by(|a, b| a.partial_cmp(b).unwrap());
+    let idx = ((s.len() as f64 - 1.0) * p).round() as usize;
+    s[idx.min(s.len() - 1)]
+}
+
+/// Order-0 entropy-based compressibility of a behavior feature vector, in
+/// `[0,1]`; high-entropy (incompressible) behavior tends toward 0.
+fn compressibility(features: &[i64]) -> f64 {
+    let mut bytes = Vec::with_capacity(features.len() * 8);
+    for &f in features {
+        bytes.extend_from_slice(&f.to_le_bytes());
+    }
+    if bytes.is_empty() {
+        return 1.0;
+    }
+    let mut counts = [0u32; 256];
+    for &b in &bytes {
+        counts[b as usize] += 1;
+    }
+    let n = bytes.len() as f64;
+    let mut h = 0.0;
+    for &c in &counts {
+        if c > 0 {
+            let p = c as f64 / n;
+            h -= p * p.log2();
+        }
+    }
+    (1.0 - h / 8.0).clamp(0.0, 1.0)
+}
+
+fn world_input_features(w: &WorldSnapshot) -> Vec<f64> {
+    let mut out = Vec::with_capacity(NUM_DOMAINS * (LANES + HIDDEN_LANES));
+    for d in &w.domains {
+        for &v in &d.observed {
+            out.push(v as f64);
+        }
+        for &v in &d.hidden {
+            out.push(v as f64);
+        }
+    }
+    out
+}
+
+fn behavior_output_features(r: &ResolutionResult) -> Vec<f64> {
+    r.trace
+        .behavior_fingerprint
+        .features
+        .iter()
+        .map(|&v| v as f64)
+        .collect()
+}
+
+// ---------------------------------------------------------------------------
+// Result aggregates.
+// ---------------------------------------------------------------------------
+
+#[derive(Clone, Debug)]
+pub struct TraceGates {
+    pub median_causal_edges: f64,
+    pub p95_causal_rank: f64,
+    pub median_touched: f64,
+    pub p95_touched: f64,
+    pub median_rank: f64,
+    pub fp_collision_rate: f64,
+    pub largest_cluster: f64,
+    pub failures: Vec<String>,
+}
+
+#[derive(Clone, Debug)]
+pub struct EquivalenceGates {
+    pub total: usize,
+    pub matched: usize,
+    pub failures: Vec<String>,
+}
+
+#[derive(Clone, Debug)]
+pub struct DomainGates {
+    pub appears: [f64; NUM_DOMAINS],
+    pub influences: [f64; NUM_DOMAINS],
+    pub mutated: [f64; NUM_DOMAINS],
+    pub removal_loss: [f64; NUM_DOMAINS],
+    pub min_merge_loss: f64,
+    pub read_only: Vec<usize>,
+    pub write_only: Vec<usize>,
+    pub failures: Vec<String>,
+}
+
+#[derive(Clone, Debug)]
+pub struct MetamorphicGates {
+    pub total: usize,
+    pub altered_trace: f64,
+    pub altered_delta: f64,
+    pub altered_future: f64,
+    pub neutral_unexplained: f64,
+    pub failures: Vec<String>,
+}
+
+#[derive(Clone, Debug)]
+pub struct ContractGates {
+    pub total: usize,
+    pub passed: usize,
+    pub failures: Vec<String>,
+}
+
+#[derive(Clone, Debug)]
+pub struct ReplayGates {
+    pub total: usize,
+    pub deterministic: usize,
+    pub drift: usize,
+    pub failures: Vec<String>,
+}
+
+#[derive(Clone, Debug)]
+pub struct CoverageGates {
+    pub generated_cases: usize,
+    pub generated_rejected: usize,
+    pub contract_rejected: usize,
+    pub executions: usize,
+    pub perturbations: usize,
+    pub failures: Vec<String>,
+}
+
+pub struct CiResults {
+    pub scale: Scale,
+    pub trace: TraceGates,
+    pub equivalence: EquivalenceGates,
+    pub domain: DomainGates,
+    pub metamorphic: MetamorphicGates,
+    pub collapse: CollapseSummary,
+    pub mutation: MutationOutcome,
+    pub contract: ContractGates,
+    pub replay: ReplayGates,
+    pub coverage: CoverageGates,
+}
+
+impl CiResults {
+    pub fn all_failures(&self) -> Vec<(&'static str, &Vec<String>)> {
+        vec![
+            ("causal_rank/trace", &self.trace.failures),
+            ("runtime_equivalence", &self.equivalence.failures),
+            ("domain_participation", &self.domain.failures),
+            ("metamorphic_response", &self.metamorphic.failures),
+            ("compression_resistance", &self.collapse.failures),
+            ("contract", &self.contract.failures),
+            ("replay", &self.replay.failures),
+            ("coverage", &self.coverage.failures),
+        ]
+    }
+
+    pub fn passed(&self) -> bool {
+        self.all_failures().iter().all(|(_, f)| f.is_empty()) && self.mutation.passed()
+    }
+}
+
+// ---------------------------------------------------------------------------
+// The pipeline.
+// ---------------------------------------------------------------------------
+
+pub fn run_all(scale: Scale) -> CiResults {
+    let cfg = EngineConfig::reference();
+    let rut = RuntimeUnderTest::new();
+
+    // Main execution + metamorphic + contract pass.
+    let mut ranks = Vec::with_capacity(scale.executions);
+    let mut edges = Vec::with_capacity(scale.executions);
+    let mut touched = Vec::with_capacity(scale.executions);
+    let mut fp_counts: HashMap<Hash, u32> = HashMap::new();
+
+    let mut equiv_matched = 0usize;
+    let mut equiv_failures = Vec::new();
+
+    let mut domain_appear = [0u64; NUM_DOMAINS];
+    let mut domain_mutated = [0u64; NUM_DOMAINS];
+    let mut domain_read_any = [false; NUM_DOMAINS];
+    let mut domain_write_any = [false; NUM_DOMAINS];
+
+    let mut meta_total = 0usize;
+    let mut meta_alt_trace = 0usize;
+    let mut meta_alt_delta = 0usize;
+    let mut meta_alt_future = 0usize;
+    let mut meta_neutral_unexpl = 0usize;
+
+    let mut contract_pass = 0usize;
+    let mut contract_failures: Vec<String> = Vec::new();
+
+    let mut collapse_inputs: Vec<Vec<f64>> = Vec::new();
+    let mut collapse_outputs: Vec<Vec<f64>> = Vec::new();
+    let mut collapse_fps: Vec<Hash> = Vec::new();
+
+    let mut mutation_inputs: Vec<ResolutionInput> = Vec::new();
+    let mut generated_rejected = 0usize;
+    let mut perturbation_runs = 0usize;
+
+    let mut contract_rejected = 0usize;
+
+    for i in 0..scale.executions {
+        // Admission: a case is admitted only if its *measured* trace behavior
+        // satisfies its contract. Structurally-flat cases are already rejected
+        // by `generate_accepted_case`; here we additionally reject cases whose
+        // behavior fails the contract, retrying with fresh seeds.
+        let mut seed = case_seed(i);
+        for attempt in 0..48 {
+            let last_attempt = attempt == 47;
+            let (case, _) = generate_accepted_case(seed);
+            if !generated_gate_failures(&case.estimate()).is_empty() {
+                generated_rejected += 1;
+            }
+            let input = input_from_case(&case);
+            let r = execute(&cfg, &input);
+
+            let rank = r.trace.causal_rank();
+            let touched_cnt = r.trace.touched_domain_count();
+            let comp = compressibility(&r.trace.behavior_fingerprint.features);
+            let div = r.trace.context_divergence();
+
+            // Run the 10 perturbations for metamorphic stats + future sensitivity.
+            let base_trace_h = r.trace.canonical_hash();
+            let base_delta_h = r.delta.hash();
+            let base_future_h = r.replay.future_hash;
+            let mut c_alt_trace = 0usize;
+            let mut c_alt_delta = 0usize;
+            let mut c_alt_future = 0usize;
+            let mut c_neutral = 0usize;
+            let mut c_pert = 0usize;
+            for pc in &case.perturbations {
+                let pr = execute(&cfg, &input_with_world(&case, pc.world.clone()));
+                c_pert += 1;
+                let at = pr.trace.canonical_hash() != base_trace_h;
+                let ad = pr.delta.hash() != base_delta_h;
+                let af = pr.replay.future_hash != base_future_h;
+                if at {
+                    c_alt_trace += 1;
+                }
+                if ad {
+                    c_alt_delta += 1;
+                }
+                if af {
+                    c_alt_future += 1;
+                }
+                if !at && pc.expectation.neutral_explanation.is_none() {
+                    c_neutral += 1;
+                }
+            }
+            let future_sensitivity = if c_pert > 0 {
+                c_alt_future as f64 / c_pert as f64
+            } else {
+                0.0
+            };
+
+            let mut creasons = Vec::new();
+            if rank < case.contract.min_causal_rank {
+                creasons.push("causal_rank");
+            }
+            if touched_cnt < case.contract.min_domain_participation {
+                creasons.push("domain_participation");
+            }
+            if future_sensitivity < case.contract.min_future_sensitivity {
+                creasons.push("future_sensitivity");
+            }
+            if div < case.contract.min_context_divergence {
+                creasons.push("context_divergence");
+            }
+            if comp > case.contract.max_compressibility {
+                creasons.push("compressibility");
+            }
+
+            if !creasons.is_empty() && !last_attempt {
+                contract_rejected += 1;
+                seed = generators::derive(seed, "contract-retry");
+                continue;
+            }
+
+            // ---- Commit the admitted case ----
+            let r2 = rut.resolve(input.clone());
+            if canonical(&r) == canonical(&r2) {
+                equiv_matched += 1;
+            } else if equiv_failures.len() < 16 {
+                equiv_failures.push(format!("case {} reference != runtime_under_test", i));
+            }
+
+            ranks.push(rank as f64);
+            edges.push(r.trace.causal_edge_count() as f64);
+            touched.push(touched_cnt as f64);
+            *fp_counts.entry(r.trace.behavior_fingerprint.hash).or_insert(0) += 1;
+
+            for d in 0..NUM_DOMAINS {
+                let read = r.trace.read_graph.access_count[d] > 0;
+                let write = r.trace.write_graph.access_count[d] > 0;
+                if read || write {
+                    domain_appear[d] += 1;
+                }
+                if read {
+                    domain_read_any[d] = true;
+                }
+                if write {
+                    domain_write_any[d] = true;
+                }
+            }
+            for dd in &r.delta.domain_deltas {
+                if !dd.is_zero() {
+                    domain_mutated[dd.domain.0 as usize] += 1;
+                }
+            }
+
+            meta_total += c_pert;
+            meta_alt_trace += c_alt_trace;
+            meta_alt_delta += c_alt_delta;
+            meta_alt_future += c_alt_future;
+            meta_neutral_unexpl += c_neutral;
+            perturbation_runs += c_pert;
+
+            if creasons.is_empty() {
+                contract_pass += 1;
+            } else if contract_failures.len() < 16 {
+                contract_failures.push(format!("case {} violates {:?}", i, creasons));
+            }
+
+            if collapse_inputs.len() < scale.collapse_samples {
+                collapse_inputs.push(world_input_features(&case.world));
+                collapse_outputs.push(behavior_output_features(&r));
+                collapse_fps.push(r.trace.behavior_fingerprint.hash);
+            }
+            if mutation_inputs.len() < 64 {
+                mutation_inputs.push(input.clone());
+            }
+            break;
+        }
+    }
+
+    let n = scale.executions.max(1) as f64;
+
+    // ---- Trace gates ----
+    let collisions = scale.executions - fp_counts.len();
+    let largest_cluster = fp_counts.values().copied().max().unwrap_or(0) as f64 / n;
+    let mut trace_failures = Vec::new();
+    let med_edges = median(&edges);
+    let p95_rank = percentile(&ranks, 0.05); // 95% of executions have rank >= this
+    let med_touched = median(&touched);
+    let p95_touched = percentile(&touched, 0.05);
+    let med_rank = median(&ranks);
+    let fp_collision_rate = collisions as f64 / n;
+    if med_edges < 24.0 {
+        trace_failures.push(format!("median causal edges {} < 24", med_edges));
+    }
+    if p95_rank < 6.0 {
+        trace_failures.push(format!("95% causal rank {} < 6", p95_rank));
+    }
+    if med_touched < 4.0 {
+        trace_failures.push(format!("median touched {} < 4", med_touched));
+    }
+    if p95_touched < 3.0 {
+        trace_failures.push(format!("95% touched {} < 3", p95_touched));
+    }
+    if fp_collision_rate >= 0.05 {
+        trace_failures.push(format!("fp collision rate {:.4} >= 0.05", fp_collision_rate));
+    }
+    if largest_cluster >= 0.02 {
+        trace_failures.push(format!("largest cluster {:.4} >= 0.02", largest_cluster));
+    }
+    let trace = TraceGates {
+        median_causal_edges: med_edges,
+        p95_causal_rank: p95_rank,
+        median_touched: med_touched,
+        p95_touched,
+        median_rank: med_rank,
+        fp_collision_rate,
+        largest_cluster,
+        failures: trace_failures,
+    };
+
+    // ---- Equivalence gates ----
+    if equiv_matched != scale.executions {
+        equiv_failures.push(format!(
+            "{}/{} executions matched (require 100%)",
+            equiv_matched, scale.executions
+        ));
+    }
+    let equivalence = EquivalenceGates {
+        total: scale.executions,
+        matched: equiv_matched,
+        failures: equiv_failures,
+    };
+
+    // ---- Domain participation gates ----
+    let domain = domain_gates(
+        scale,
+        &domain_appear,
+        &domain_mutated,
+        &domain_read_any,
+        &domain_write_any,
+        n,
+    );
+
+    // ---- Metamorphic gates ----
+    let mt = meta_total.max(1) as f64;
+    let mut meta_failures = Vec::new();
+    let r_trace = meta_alt_trace as f64 / mt;
+    let r_delta = meta_alt_delta as f64 / mt;
+    let r_future = meta_alt_future as f64 / mt;
+    let r_neutral = meta_neutral_unexpl as f64 / mt;
+    if r_trace < 0.90 {
+        meta_failures.push(format!("altered trace {:.3} < 0.90", r_trace));
+    }
+    if r_delta < 0.75 {
+        meta_failures.push(format!("altered delta {:.3} < 0.75", r_delta));
+    }
+    if r_future < 0.50 {
+        meta_failures.push(format!("altered future {:.3} < 0.50", r_future));
+    }
+    if r_neutral > 0.05 {
+        meta_failures.push(format!("unexplained neutral {:.3} > 0.05", r_neutral));
+    }
+    let metamorphic = MetamorphicGates {
+        total: meta_total,
+        altered_trace: r_trace,
+        altered_delta: r_delta,
+        altered_future: r_future,
+        neutral_unexplained: r_neutral,
+        failures: meta_failures,
+    };
+
+    // ---- Collapse gates ----
+    let corpus = BehaviorCorpus::build(collapse_inputs, collapse_outputs, collapse_fps);
+    let collapse = analyze(&corpus);
+
+    // ---- Mutation gates ----
+    let mutation = run_suite(scale.mutants, &mutation_inputs);
+
+    // ---- Contract gates ----
+    let mut contract_gate_failures = Vec::new();
+    if contract_pass != scale.executions {
+        contract_gate_failures.push(format!(
+            "{}/{} cases satisfy their contract",
+            contract_pass, scale.executions
+        ));
+        contract_gate_failures.extend(contract_failures);
+    }
+    let contract = ContractGates {
+        total: scale.executions,
+        passed: contract_pass,
+        failures: contract_gate_failures,
+    };
+
+    // ---- Replay gates ----
+    let replay = replay_gates(scale);
+
+    // ---- Coverage gates ----
+    let mut cov_failures = Vec::new();
+    if generated_rejected != 0 {
+        cov_failures.push(format!(
+            "{} admitted cases failed generated gates",
+            generated_rejected
+        ));
+    }
+    if scale.executions == 0 {
+        cov_failures.push("no executions".into());
+    }
+    let coverage = CoverageGates {
+        generated_cases: scale.executions,
+        generated_rejected,
+        contract_rejected,
+        executions: scale.executions,
+        perturbations: perturbation_runs,
+        failures: cov_failures,
+    };
+
+    CiResults {
+        scale,
+        trace,
+        equivalence,
+        domain,
+        metamorphic,
+        collapse,
+        mutation,
+        contract,
+        replay,
+        coverage,
+    }
+}
+
+fn domain_gates(
+    scale: Scale,
+    appear: &[u64; NUM_DOMAINS],
+    mutated: &[u64; NUM_DOMAINS],
+    read_any: &[bool; NUM_DOMAINS],
+    write_any: &[bool; NUM_DOMAINS],
+    n: f64,
+) -> DomainGates {
+    let cfg = EngineConfig::reference();
+
+    let mut appears = [0.0; NUM_DOMAINS];
+    let mut mutated_f = [0.0; NUM_DOMAINS];
+    for d in 0..NUM_DOMAINS {
+        appears[d] = appear[d] as f64 / n;
+        mutated_f[d] = mutated[d] as f64 / n;
+    }
+
+    // Influence probe: mask each domain and see how often the trace changes.
+    let mut influences = [0.0; NUM_DOMAINS];
+    let probe_cases = scale.domain_probe_cases.max(1);
+    for d in 0..NUM_DOMAINS {
+        let masked = masked_config(d);
+        let mut changed = 0usize;
+        for i in 0..probe_cases {
+            let (case, _) = generate_accepted_case(case_seed(i));
+            let input = input_from_case(&case);
+            let base = execute(&cfg, &input);
+            let alt = execute(&masked, &input);
+            if base.trace.canonical_hash() != alt.trace.canonical_hash()
+                || base.delta.hash() != alt.delta.hash()
+            {
+                changed += 1;
+            }
+        }
+        influences[d] = changed as f64 / probe_cases as f64;
+    }
+
+    // Removal diversity probe: vary only domain d across K worlds; compare the
+    // count of distinct behaviors with the domain present vs masked.
+    let mut removal_loss = [0.0; NUM_DOMAINS];
+    let k = scale.domain_probe_variations.max(8);
+    for d in 0..NUM_DOMAINS {
+        let base_world = generators::generate_world(0xBEEF ^ d as u64);
+        let case = {
+            let (c, _) = generate_accepted_case(case_seed(d + 1));
+            c
+        };
+        let mut full = std::collections::HashSet::new();
+        let mut masked = std::collections::HashSet::new();
+        let mcfg = masked_config(d);
+        for v in 0..k {
+            let mut w = base_world.clone();
+            for l in 0..LANES {
+                w.domains[d].observed[l] = (v as i64 + 1).wrapping_mul(7919 + l as i64);
+            }
+            w.domains[d].hidden[0] = (v as i64).wrapping_mul(104729);
+            let input = input_with_world(&case, w);
+            full.insert(execute(&cfg, &input).trace.behavior_fingerprint.hash);
+            masked.insert(execute(&mcfg, &input).trace.behavior_fingerprint.hash);
+        }
+        let df = full.len().max(1) as f64;
+        let dm = masked.len() as f64;
+        removal_loss[d] = (1.0 - dm / df).clamp(0.0, 1.0);
+    }
+
+    // Merge probe: alias domain b := domain a, see how often behavior changes.
+    let mut min_merge_loss = 1.0f64;
+    let merge_cases = (scale.domain_probe_cases / 2).max(20);
+    for a in 0..NUM_DOMAINS {
+        for b in (a + 1)..NUM_DOMAINS {
+            let mut changed = 0usize;
+            for i in 0..merge_cases {
+                let (case, _) = generate_accepted_case(case_seed(i));
+                let base = execute(&cfg, &input_from_case(&case));
+                let mut w = case.world.clone();
+                w.domains[b].observed = w.domains[a].observed;
+                w.domains[b].hidden = w.domains[a].hidden;
+                let merged = execute(&cfg, &input_with_world(&case, w));
+                if merged.trace.behavior_fingerprint.hash != base.trace.behavior_fingerprint.hash {
+                    changed += 1;
+                }
+            }
+            let loss = changed as f64 / merge_cases as f64;
+            min_merge_loss = min_merge_loss.min(loss);
+        }
+    }
+
+    let read_only: Vec<usize> = (0..NUM_DOMAINS)
+        .filter(|&d| read_any[d] && !write_any[d])
+        .collect();
+    let write_only: Vec<usize> = (0..NUM_DOMAINS)
+        .filter(|&d| write_any[d] && !read_any[d])
+        .collect();
+
+    let mut failures = Vec::new();
+    for d in 0..NUM_DOMAINS {
+        if appears[d] < 0.35 {
+            failures.push(format!("domain {} appears {:.3} < 0.35", d, appears[d]));
+        }
+        if influences[d] < 0.20 {
+            failures.push(format!("domain {} influences {:.3} < 0.20", d, influences[d]));
+        }
+        if mutated_f[d] < 0.20 {
+            failures.push(format!("domain {} mutated {:.3} < 0.20", d, mutated_f[d]));
+        }
+        if removal_loss[d] < 0.10 {
+            failures.push(format!(
+                "domain {} removal diversity loss {:.3} < 0.10",
+                d, removal_loss[d]
+            ));
+        }
+    }
+    if min_merge_loss < 0.08 {
+        failures.push(format!("min merge loss {:.3} < 0.08", min_merge_loss));
+    }
+    if !read_only.is_empty() {
+        failures.push(format!("read-only domains: {:?}", read_only));
+    }
+    if !write_only.is_empty() {
+        failures.push(format!("write-only domains: {:?}", write_only));
+    }
+
+    DomainGates {
+        appears,
+        influences,
+        mutated: mutated_f,
+        removal_loss,
+        min_merge_loss,
+        read_only,
+        write_only,
+        failures,
+    }
+}
+
+fn replay_gates(scale: Scale) -> ReplayGates {
+    let corpus = replay_corpus::build_corpus(scale.replay_cases, 0x5EED);
+    let report = replay_corpus::verify_corpus(&corpus);
+    let mut failures = Vec::new();
+    if !report.drift.is_empty() {
+        failures.push(format!("{} replay cases drifted", report.drift.len()));
+    }
+    // Minimum corpus size is a merge-blocking gate only at full scale.
+    if matches!(std::env::var("MAGICKA_SCALE").as_deref(), Ok("full")) && report.total < 10_000 {
+        failures.push(format!("replay corpus {} < 10000", report.total));
+    }
+    ReplayGates {
+        total: report.total,
+        deterministic: report.deterministic,
+        drift: report.drift.len(),
+        failures,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn tiny_ci_passes_all_gates() {
+        let r = run_all(Scale::tiny());
+        for (name, f) in r.all_failures() {
+            assert!(f.is_empty(), "{name} failed: {:?}", f);
+        }
+        assert!(r.mutation.passed(), "mutants survived: {:?}", r.mutation.survivors);
+        assert!(r.passed());
+    }
+}

crates/ci_reports/src/main.rs

@@ -0,0 +1,325 @@
+//! The `ci` binary: runs the full adversarial framework against the reference
+//! and runtime-under-test, writes the eight required reports (JSON + markdown),
+//! and exits nonzero if any gate fails.
+
+use ci_reports::json::Json;
+use ci_reports::{run_all, CiResults, Scale};
+use std::fs;
+use std::io::Write;
+use std::path::Path;
+use std::time::Instant;
+
+fn arr_f(vals: &[f64]) -> Json {
+    Json::Arr(vals.iter().map(|&v| Json::Num(v)).collect())
+}
+fn fails(v: &[String]) -> Json {
+    Json::Arr(v.iter().map(|s| Json::s(s.clone())).collect())
+}
+fn pass_field(v: &[String]) -> Json {
+    Json::Bool(v.is_empty())
+}
+
+fn write_report(dir: &Path, name: &str, j: &Json) {
+    let path = dir.join(format!("{name}.json"));
+    let mut f = fs::File::create(&path).expect("create report");
+    f.write_all(j.to_pretty().as_bytes()).expect("write report");
+}
+
+fn build_reports(dir: &Path, r: &CiResults) {
+    // 1. domain_participation_report
+    write_report(
+        dir,
+        "domain_participation_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.domain.failures)),
+            ("appears_fraction".into(), arr_f(&r.domain.appears)),
+            ("influences_fraction".into(), arr_f(&r.domain.influences)),
+            ("mutated_fraction".into(), arr_f(&r.domain.mutated)),
+            ("removal_diversity_loss".into(), arr_f(&r.domain.removal_loss)),
+            ("min_merge_loss".into(), Json::Num(r.domain.min_merge_loss)),
+            (
+                "read_only_domains".into(),
+                Json::Arr(r.domain.read_only.iter().map(|&d| Json::Int(d as i64)).collect()),
+            ),
+            (
+                "write_only_domains".into(),
+                Json::Arr(r.domain.write_only.iter().map(|&d| Json::Int(d as i64)).collect()),
+            ),
+            ("failures".into(), fails(&r.domain.failures)),
+        ]),
+    );
+
+    // 2. causal_rank_report (trace gates)
+    write_report(
+        dir,
+        "causal_rank_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.trace.failures)),
+            ("median_causal_rank".into(), Json::Num(r.trace.median_rank)),
+            ("p95_causal_rank".into(), Json::Num(r.trace.p95_causal_rank)),
+            ("median_causal_edges".into(), Json::Num(r.trace.median_causal_edges)),
+            ("median_touched_domains".into(), Json::Num(r.trace.median_touched)),
+            ("p95_touched_domains".into(), Json::Num(r.trace.p95_touched)),
+            ("fp_collision_rate".into(), Json::Num(r.trace.fp_collision_rate)),
+            ("largest_cluster".into(), Json::Num(r.trace.largest_cluster)),
+            ("failures".into(), fails(&r.trace.failures)),
+        ]),
+    );
+
+    // 3. compression_resistance_report
+    let attack_json: Vec<Json> = r
+        .collapse
+        .reports
+        .iter()
+        .map(|rep| {
+            Json::Obj(vec![
+                ("attack".into(), Json::s(rep.attack.clone())),
+                ("predicts".into(), Json::Num(rep.predicts)),
+                ("info_loss".into(), Json::Num(rep.info_loss)),
+                ("detail".into(), Json::s(rep.detail.clone())),
+            ])
+        })
+        .collect();
+    write_report(
+        dir,
+        "compression_resistance_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.collapse.failures)),
+            ("best_1factor".into(), Json::Num(r.collapse.best_1factor)),
+            ("best_2factor".into(), Json::Num(r.collapse.best_2factor)),
+            ("best_4factor".into(), Json::Num(r.collapse.best_4factor)),
+            ("max_single_domain".into(), Json::Num(r.collapse.max_single_domain)),
+            ("max_pair".into(), Json::Num(r.collapse.max_pair)),
+            ("min_info_loss".into(), Json::Num(r.collapse.min_info_loss)),
+            ("attacks".into(), Json::Arr(attack_json)),
+            ("failures".into(), fails(&r.collapse.failures)),
+        ]),
+    );
+
+    // 4. metamorphic_response_report
+    write_report(
+        dir,
+        "metamorphic_response_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.metamorphic.failures)),
+            ("perturbations".into(), Json::Int(r.metamorphic.total as i64)),
+            ("altered_trace".into(), Json::Num(r.metamorphic.altered_trace)),
+            ("altered_delta".into(), Json::Num(r.metamorphic.altered_delta)),
+            ("altered_future".into(), Json::Num(r.metamorphic.altered_future)),
+            ("neutral_unexplained".into(), Json::Num(r.metamorphic.neutral_unexplained)),
+            ("failures".into(), fails(&r.metamorphic.failures)),
+        ]),
+    );
+
+    // 5. mutation_survivor_report
+    let survivors: Vec<Json> = r
+        .mutation
+        .survivors
+        .iter()
+        .map(|(id, name)| {
+            Json::Obj(vec![
+                ("id".into(), Json::Int(*id as i64)),
+                ("name".into(), Json::s(name.clone())),
+            ])
+        })
+        .collect();
+    write_report(
+        dir,
+        "mutation_survivor_report",
+        &Json::Obj(vec![
+            ("pass".into(), Json::Bool(r.mutation.passed())),
+            ("total_mutants".into(), Json::Int(r.mutation.total as i64)),
+            ("killed".into(), Json::Int(r.mutation.killed as i64)),
+            ("survivors".into(), Json::Arr(survivors)),
+        ]),
+    );
+
+    // 6. runtime_equivalence_report
+    write_report(
+        dir,
+        "runtime_equivalence_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.equivalence.failures)),
+            ("total".into(), Json::Int(r.equivalence.total as i64)),
+            ("matched".into(), Json::Int(r.equivalence.matched as i64)),
+            ("failures".into(), fails(&r.equivalence.failures)),
+        ]),
+    );
+
+    // 7. replay_report
+    write_report(
+        dir,
+        "replay_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.replay.failures)),
+            ("total".into(), Json::Int(r.replay.total as i64)),
+            ("deterministic".into(), Json::Int(r.replay.deterministic as i64)),
+            ("drift".into(), Json::Int(r.replay.drift as i64)),
+            ("failures".into(), fails(&r.replay.failures)),
+        ]),
+    );
+
+    // 8. coverage_report
+    write_report(
+        dir,
+        "coverage_report",
+        &Json::Obj(vec![
+            ("pass".into(), pass_field(&r.coverage.failures)),
+            ("generated_cases".into(), Json::Int(r.coverage.generated_cases as i64)),
+            ("generated_rejected".into(), Json::Int(r.coverage.generated_rejected as i64)),
+            ("contract_rejected".into(), Json::Int(r.coverage.contract_rejected as i64)),
+            ("executions".into(), Json::Int(r.coverage.executions as i64)),
+            ("perturbations".into(), Json::Int(r.coverage.perturbations as i64)),
+            ("contracts_passed".into(), Json::Int(r.contract.passed as i64)),
+            ("contracts_total".into(), Json::Int(r.contract.total as i64)),
+            ("failures".into(), fails(&r.coverage.failures)),
+        ]),
+    );
+}
+
+fn status(v: bool) -> &'static str {
+    if v {
+        "PASS"
+    } else {
+        "FAIL"
+    }
+}
+
+fn write_markdown(dir: &Path, r: &CiResults) {
+    let mut s = String::new();
+    s.push_str("# Magicka VM — Phase 0/1 CI Report\n\n");
+    s.push_str(&format!(
+        "Overall: **{}**\n\n",
+        status(r.passed())
+    ));
+    s.push_str(&format!(
+        "Scale: executions={}, mutants={}, replay={}, collapse_samples={}\n\n",
+        r.scale.executions, r.scale.mutants, r.scale.replay_cases, r.scale.collapse_samples
+    ));
+    s.push_str("| Report | Status | Key metrics |\n|---|---|---|\n");
+    s.push_str(&format!(
+        "| runtime_equivalence | {} | {}/{} matched |\n",
+        status(r.equivalence.failures.is_empty()),
+        r.equivalence.matched,
+        r.equivalence.total
+    ));
+    s.push_str(&format!(
+        "| causal_rank/trace | {} | rank med={} p95={}, edges med={}, touched med={} |\n",
+        status(r.trace.failures.is_empty()),
+        r.trace.median_rank,
+        r.trace.p95_causal_rank,
+        r.trace.median_causal_edges,
+        r.trace.median_touched
+    ));
+    s.push_str(&format!(
+        "| domain_participation | {} | min_merge_loss={:.3} |\n",
+        status(r.domain.failures.is_empty()),
+        r.domain.min_merge_loss
+    ));
+    s.push_str(&format!(
+        "| metamorphic_response | {} | trace={:.3} delta={:.3} future={:.3} |\n",
+        status(r.metamorphic.failures.is_empty()),
+        r.metamorphic.altered_trace,
+        r.metamorphic.altered_delta,
+        r.metamorphic.altered_future
+    ));
+    s.push_str(&format!(
+        "| compression_resistance | {} | 1f={:.3} 2f={:.3} 4f={:.3} single={:.3} pair={:.3} info_loss={:.3} |\n",
+        status(r.collapse.failures.is_empty()),
+        r.collapse.best_1factor,
+        r.collapse.best_2factor,
+        r.collapse.best_4factor,
+        r.collapse.max_single_domain,
+        r.collapse.max_pair,
+        r.collapse.min_info_loss
+    ));
+    s.push_str(&format!(
+        "| mutation_survivor | {} | killed {}/{} |\n",
+        status(r.mutation.passed()),
+        r.mutation.killed,
+        r.mutation.total
+    ));
+    s.push_str(&format!(
+        "| contract | {} | {}/{} cases |\n",
+        status(r.contract.failures.is_empty()),
+        r.contract.passed,
+        r.contract.total
+    ));
+    s.push_str(&format!(
+        "| replay | {} | {}/{} deterministic, drift={} |\n",
+        status(r.replay.failures.is_empty()),
+        r.replay.deterministic,
+        r.replay.total,
+        r.replay.drift
+    ));
+    s.push_str(&format!(
+        "| coverage | {} | exec={}, perturb={}, rejected={} |\n",
+        status(r.coverage.failures.is_empty()),
+        r.coverage.executions,
+        r.coverage.perturbations,
+        r.coverage.generated_rejected
+    ));
+
+    s.push_str("\n## Failures\n\n");
+    let mut any = false;
+    for (name, f) in r.all_failures() {
+        for msg in f {
+            any = true;
+            s.push_str(&format!("- **{}**: {}\n", name, msg));
+        }
+    }
+    for (id, name) in &r.mutation.survivors {
+        any = true;
+        s.push_str(&format!("- **mutation_survivor**: mutant {} ({}) survived\n", id, name));
+    }
+    if !any {
+        s.push_str("None. The fake universe failed to collapse. ✅\n");
+    }
+
+    let path = dir.join("ci_summary.md");
+    fs::write(path, s).expect("write markdown");
+}
+
+fn main() {
+    let scale = Scale::from_env();
+    let out_dir = std::env::var("MAGICKA_OUT").unwrap_or_else(|_| "ci_out".to_string());
+    let dir = Path::new(&out_dir);
+    fs::create_dir_all(dir).expect("create out dir");
+
+    eprintln!(
+        "running CI: executions={} mutants={} replay={} collapse_samples={}",
+        scale.executions, scale.mutants, scale.replay_cases, scale.collapse_samples
+    );
+    let start = Instant::now();
+    let results = run_all(scale);
+    let elapsed = start.elapsed();
+
+    build_reports(dir, &results);
+    write_markdown(dir, &results);
+
+    println!("\n=== Magicka VM CI ({:?}) ===", elapsed);
+    for (name, f) in results.all_failures() {
+        println!("  {:<24} {}", name, status(f.is_empty()));
+    }
+    println!(
+        "  {:<24} {} ({} killed / {} mutants{})",
+        "mutation_survivor",
+        status(results.mutation.passed()),
+        results.mutation.killed,
+        results.mutation.total,
+        if results.mutation.survivors.is_empty() {
+            String::new()
+        } else {
+            format!(", {} survivors", results.mutation.survivors.len())
+        }
+    );
+    println!("reports written to {}/", out_dir);
+
+    if results.passed() {
+        println!("\nOVERALL: PASS — the adversarial framework could not collapse the universe.");
+        std::process::exit(0);
+    } else {
+        println!("\nOVERALL: FAIL");
+        std::process::exit(1);
+    }
+}

crates/collapse_analysis/Cargo.toml

@@ -0,0 +1,12 @@
+[package]
+name = "collapse_analysis"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+trace_model = { path = "../trace_model" }
+
+[lib]
+path = "src/lib.rs"

crates/collapse_analysis/src/lib.rs

@@ -0,0 +1,470 @@
+//! `collapse_analysis` — the framework's compression attacks. Each attack
+//! tries to predict execution behavior with a *simpler* model. If any small
+//! model predicts above the configured thresholds, the universe has collapsed
+//! and CI must fail.
+//!
+//! The corpus is a set of (input, output) samples: `input` is the world ground
+//! truth (8 domains x (observed+hidden) lanes) and `output` is the behavior
+//! feature vector produced by the runtime. Inputs/outputs are standardized so
+//! that no single high-magnitude axis dominates the variance accounting.
+
+pub mod linalg;
+
+use linalg::{ols_r2, pca_scores, Mat};
+use world_model::{Hash, HIDDEN_LANES, LANES, NUM_DOMAINS};
+
+/// Input columns belonging to one domain (observed + hidden lanes).
+pub const DOMAIN_BLOCK: usize = LANES + HIDDEN_LANES;
+
+/// A standardized behavior corpus.
+pub struct BehaviorCorpus {
+    /// `n x (NUM_DOMAINS*DOMAIN_BLOCK)` standardized input.
+    pub x: Mat,
+    /// `n x m` standardized output (behavior features).
+    pub y: Mat,
+    pub fingerprints: Vec<Hash>,
+}
+
+fn standardize(rows: &[Vec<f64>]) -> Mat {
+    let n = rows.len();
+    let cols = if n == 0 { 0 } else { rows[0].len() };
+    let mut m = Mat::zeros(n, cols);
+    for r in 0..n {
+        for c in 0..cols {
+            m.set(r, c, rows[r][c]);
+        }
+    }
+    // center + scale to unit std per column
+    for c in 0..cols {
+        let mut mean = 0.0;
+        for r in 0..n {
+            mean += m.at(r, c);
+        }
+        mean /= n.max(1) as f64;
+        let mut var = 0.0;
+        for r in 0..n {
+            var += (m.at(r, c) - mean).powi(2);
+        }
+        var /= n.max(1) as f64;
+        let sd = var.sqrt();
+        let inv = if sd > 1e-9 { 1.0 / sd } else { 0.0 };
+        for r in 0..n {
+            let v = (m.at(r, c) - mean) * inv;
+            m.set(r, c, v);
+        }
+    }
+    m
+}
+
+impl BehaviorCorpus {
+    pub fn build(inputs: Vec<Vec<f64>>, outputs: Vec<Vec<f64>>, fingerprints: Vec<Hash>) -> Self {
+        BehaviorCorpus {
+            x: standardize(&inputs),
+            y: standardize(&outputs),
+            fingerprints,
+        }
+    }
+
+    pub fn n(&self) -> usize {
+        self.x.rows
+    }
+
+    fn select(&self, cols: &[usize]) -> Mat {
+        let mut m = Mat::zeros(self.x.rows, cols.len());
+        for r in 0..self.x.rows {
+            for (j, &c) in cols.iter().enumerate() {
+                m.set(r, j, self.x.at(r, c));
+            }
+        }
+        m
+    }
+
+    fn domain_cols(domain: usize) -> Vec<usize> {
+        (domain * DOMAIN_BLOCK..(domain + 1) * DOMAIN_BLOCK).collect()
+    }
+
+    /// R² of predicting output from the top-`k` PCA factors of the full input.
+    pub fn predict_k_factor(&self, k: usize) -> f64 {
+        if self.n() == 0 {
+            return 0.0;
+        }
+        let scores = pca_scores(&self.x, k);
+        ols_r2(&scores, &self.y)
+    }
+
+    /// Best R² obtainable using just a single domain's input block.
+    pub fn max_single_domain(&self) -> (usize, f64) {
+        let mut best = (0usize, 0.0);
+        for d in 0..NUM_DOMAINS {
+            let sub = self.select(&Self::domain_cols(d));
+            let r2 = ols_r2(&sub, &self.y);
+            if r2 > best.1 {
+                best = (d, r2);
+            }
+        }
+        best
+    }
+
+    /// Best R² obtainable using any pair of domain blocks.
+    pub fn max_pair(&self) -> ((usize, usize), f64) {
+        let mut best = ((0usize, 1usize), 0.0);
+        for a in 0..NUM_DOMAINS {
+            for b in (a + 1)..NUM_DOMAINS {
+                let mut cols = Self::domain_cols(a);
+                cols.extend(Self::domain_cols(b));
+                let sub = self.select(&cols);
+                let r2 = ols_r2(&sub, &self.y);
+                if r2 > best.1 {
+                    best = ((a, b), r2);
+                }
+            }
+        }
+        best
+    }
+
+    fn full_r2(&self) -> f64 {
+        ols_r2(&self.x, &self.y)
+    }
+}
+
+/// A compressed model's predictive power and information loss.
+#[derive(Clone, Debug)]
+pub struct CompressedModel {
+    pub predicts: f64,
+    pub info_loss: f64,
+    pub detail: String,
+}
+
+/// The collapse-attack trait (per spec).
+pub trait CollapseAttack {
+    fn name(&self) -> &'static str;
+    fn compress(&self, corpus: &BehaviorCorpus) -> CompressedModel;
+}
+
+/// One attack's outcome.
+#[derive(Clone, Debug)]
+pub struct CollapseReport {
+    pub attack: String,
+    pub predicts: f64,
+    pub info_loss: f64,
+    pub detail: String,
+}
+
+macro_rules! attack {
+    ($name:ident, $label:expr, $body:expr) => {
+        pub struct $name;
+        impl CollapseAttack for $name {
+            fn name(&self) -> &'static str {
+                $label
+            }
+            fn compress(&self, corpus: &BehaviorCorpus) -> CompressedModel {
+                let f: fn(&BehaviorCorpus) -> CompressedModel = $body;
+                f(corpus)
+            }
+        }
+    };
+}
+
+fn model(predicts: f64, detail: &str) -> CompressedModel {
+    CompressedModel {
+        predicts,
+        info_loss: (1.0 - predicts).clamp(0.0, 1.0),
+        detail: detail.to_string(),
+    }
+}
+
+attack!(DomainRemoval, "domain_removal", |c| {
+    // Best prediction achievable while *removing* each domain in turn.
+    let mut best = 0.0;
+    for d in 0..NUM_DOMAINS {
+        let cols: Vec<usize> = (0..NUM_DOMAINS)
+            .filter(|&x| x != d)
+            .flat_map(BehaviorCorpus::domain_cols)
+            .collect();
+        let sub = c.select(&cols);
+        best = f64::max(best, ols_r2(&sub, &c.y));
+    }
+    model(best, "predict with one domain removed")
+});
+
+attack!(DomainMerging, "domain_merging", |c| {
+    // Merge each pair into a summed block; best prediction over pairs.
+    let mut best = 0.0;
+    for a in 0..NUM_DOMAINS {
+        for b in (a + 1)..NUM_DOMAINS {
+            let mut merged = Mat::zeros(c.x.rows, (NUM_DOMAINS - 1) * DOMAIN_BLOCK);
+            for r in 0..c.x.rows {
+                let mut out_col = 0;
+                for d in 0..NUM_DOMAINS {
+                    if d == b {
+                        continue;
+                    }
+                    for l in 0..DOMAIN_BLOCK {
+                        let mut v = c.x.at(r, d * DOMAIN_BLOCK + l);
+                        if d == a {
+                            v += c.x.at(r, b * DOMAIN_BLOCK + l);
+                        }
+                        merged.set(r, out_col, v);
+                        out_col += 1;
+                    }
+                }
+            }
+            best = f64::max(best, ols_r2(&merged, &c.y));
+        }
+    }
+    model(best, "predict with two domains merged")
+});
+
+attack!(ConstantFolding, "constant_folding", |_c| {
+    // Folding the world to constants leaves no predictive features at all.
+    model(0.0, "world folded to constants")
+});
+
+attack!(CausalEdgeDeletion, "causal_edge_deletion", |c| {
+    // Keep only each domain's first observed lane (no cross-domain structure).
+    let cols: Vec<usize> = (0..NUM_DOMAINS).map(|d| d * DOMAIN_BLOCK).collect();
+    let sub = c.select(&cols);
+    model(ols_r2(&sub, &c.y), "diagonal-only features")
+});
+
+attack!(StateAliasing, "state_aliasing", |c| {
+    // Alias all domains into a single aggregate column.
+    let mut agg = Mat::zeros(c.x.rows, 1);
+    for r in 0..c.x.rows {
+        let mut s = 0.0;
+        for col in 0..c.x.cols {
+            s += c.x.at(r, col);
+        }
+        agg.set(r, 0, s);
+    }
+    model(ols_r2(&agg, &c.y), "single aliased aggregate")
+});
+
+attack!(LatentFactorModeling, "latent_factor_modeling", |c| {
+    model(c.predict_k_factor(4), "top-4 latent factors")
+});
+
+attack!(BehaviorClustering, "behavior_clustering", |c| {
+    let recon = kmeans_reconstruct(&c.y, 4);
+    let scores = pca_scores(&c.x, 4);
+    model(ols_r2(&scores, &recon), "4-cluster behavior model")
+});
+
+attack!(SurrogatePrediction, "surrogate_prediction", |c| {
+    model(c.full_r2(), "full linear surrogate")
+});
+
+attack!(TemporalFlattening, "temporal_flattening", |c| {
+    // Drop hidden lanes (time-carrying state); predict from observed only.
+    let cols: Vec<usize> = (0..NUM_DOMAINS)
+        .flat_map(|d| (0..LANES).map(move |l| d * DOMAIN_BLOCK + l))
+        .collect();
+    let sub = c.select(&cols);
+    model(ols_r2(&sub, &c.y), "time-flattened (observed lanes only)")
+});
+
+attack!(ObservationFlattening, "observation_flattening", |c| {
+    // Use only hidden lanes (collapse the observed surface).
+    let cols: Vec<usize> = (0..NUM_DOMAINS)
+        .flat_map(|d| (0..HIDDEN_LANES).map(move |l| d * DOMAIN_BLOCK + LANES + l))
+        .collect();
+    let sub = c.select(&cols);
+    model(ols_r2(&sub, &c.y), "observation-flattened (hidden lanes only)")
+});
+
+attack!(ExecutorIdentityErasure, "executor_identity_erasure", |c| {
+    // Predict all-but-last output dim (the divergence summary) from input.
+    if c.y.cols <= 1 {
+        return model(0.0, "no executor dim");
+    }
+    let mut y2 = Mat::zeros(c.y.rows, c.y.cols - 1);
+    for r in 0..c.y.rows {
+        for col in 0..c.y.cols - 1 {
+            y2.set(r, col, c.y.at(r, col));
+        }
+    }
+    model(ols_r2(&c.x, &y2), "executor identity erased")
+});
+
+fn kmeans_reconstruct(y: &Mat, k: usize) -> Mat {
+    let n = y.rows;
+    if n == 0 || k == 0 {
+        return y.clone();
+    }
+    let k = k.min(n);
+    // deterministic init: spread initial centroids across the data
+    let mut centroids: Vec<Vec<f64>> = (0..k).map(|i| y.data[(i * n / k) * y.cols..(i * n / k) * y.cols + y.cols].to_vec()).collect();
+    let mut assign = vec![0usize; n];
+    for _ in 0..12 {
+        // assign
+        for r in 0..n {
+            let mut best = 0;
+            let mut bestd = f64::MAX;
+            for (ci, cen) in centroids.iter().enumerate() {
+                let mut d = 0.0;
+                for col in 0..y.cols {
+                    d += (y.at(r, col) - cen[col]).powi(2);
+                }
+                if d < bestd {
+                    bestd = d;
+                    best = ci;
+                }
+            }
+            assign[r] = best;
+        }
+        // update
+        let mut sums = vec![vec![0.0; y.cols]; k];
+        let mut counts = vec![0usize; k];
+        for r in 0..n {
+            counts[assign[r]] += 1;
+            for col in 0..y.cols {
+                sums[assign[r]][col] += y.at(r, col);
+            }
+        }
+        for ci in 0..k {
+            if counts[ci] > 0 {
+                for col in 0..y.cols {
+                    centroids[ci][col] = sums[ci][col] / counts[ci] as f64;
+                }
+            }
+        }
+    }
+    let mut recon = Mat::zeros(n, y.cols);
+    for r in 0..n {
+        for col in 0..y.cols {
+            recon.set(r, col, centroids[assign[r]][col]);
+        }
+    }
+    recon
+}
+
+/// All eleven required attack families.
+pub fn all_attacks() -> Vec<Box<dyn CollapseAttack>> {
+    vec![
+        Box::new(DomainRemoval),
+        Box::new(DomainMerging),
+        Box::new(ConstantFolding),
+        Box::new(CausalEdgeDeletion),
+        Box::new(StateAliasing),
+        Box::new(LatentFactorModeling),
+        Box::new(BehaviorClustering),
+        Box::new(SurrogatePrediction),
+        Box::new(TemporalFlattening),
+        Box::new(ObservationFlattening),
+        Box::new(ExecutorIdentityErasure),
+    ]
+}
+
+/// Aggregate collapse summary against all gates.
+#[derive(Clone, Debug)]
+pub struct CollapseSummary {
+    pub reports: Vec<CollapseReport>,
+    pub best_1factor: f64,
+    pub best_2factor: f64,
+    pub best_4factor: f64,
+    pub max_single_domain: f64,
+    pub max_pair: f64,
+    pub min_info_loss: f64,
+    pub failures: Vec<String>,
+}
+
+impl CollapseSummary {
+    pub fn passed(&self) -> bool {
+        self.failures.is_empty()
+    }
+}
+
+/// Run every attack and check all collapse gates.
+pub fn analyze(corpus: &BehaviorCorpus) -> CollapseSummary {
+    let mut reports = Vec::new();
+    let mut min_info_loss: f64 = 1.0;
+    for atk in all_attacks() {
+        let m = atk.compress(corpus);
+        min_info_loss = min_info_loss.min(m.info_loss);
+        reports.push(CollapseReport {
+            attack: atk.name().to_string(),
+            predicts: m.predicts,
+            info_loss: m.info_loss,
+            detail: m.detail,
+        });
+    }
+
+    let best_1 = corpus.predict_k_factor(1);
+    let best_2 = corpus.predict_k_factor(2);
+    let best_4 = corpus.predict_k_factor(4);
+    let (_, max_single) = corpus.max_single_domain();
+    let (_, max_pair) = corpus.max_pair();
+
+    let mut failures = Vec::new();
+    if best_1 >= 0.40 {
+        failures.push(format!("1-factor predicts {:.3} >= 0.40", best_1));
+    }
+    if best_2 >= 0.55 {
+        failures.push(format!("2-factor predicts {:.3} >= 0.55", best_2));
+    }
+    if best_4 >= 0.70 {
+        failures.push(format!("4-factor predicts {:.3} >= 0.70", best_4));
+    }
+    if max_single > 0.30 {
+        failures.push(format!("single domain explains {:.3} > 0.30", max_single));
+    }
+    if max_pair > 0.55 {
+        failures.push(format!("domain pair explains {:.3} > 0.55", max_pair));
+    }
+    if min_info_loss < 0.35 {
+        failures.push(format!("min info loss {:.3} < 0.35", min_info_loss));
+    }
+
+    CollapseSummary {
+        reports,
+        best_1factor: best_1,
+        best_2factor: best_2,
+        best_4factor: best_4,
+        max_single_domain: max_single,
+        max_pair,
+        min_info_loss,
+        failures,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use world_model::{Hasher, Rng};
+
+    #[test]
+    fn random_behavior_resists_collapse() {
+        // Inputs random; outputs an avalanche hash of inputs -> no small linear
+        // model should predict them.
+        let mut rng = Rng::new(1);
+        let mut inputs = Vec::new();
+        let mut outputs = Vec::new();
+        let mut fps = Vec::new();
+        for _ in 0..400 {
+            let inp: Vec<f64> = (0..NUM_DOMAINS * DOMAIN_BLOCK)
+                .map(|_| rng.range_i64(-5000, 5000) as f64)
+                .collect();
+            let mut h = Hasher::new();
+            for &v in &inp {
+                h.write_i64(v as i64);
+            }
+            let base = h.finish().0;
+            let out: Vec<f64> = (0..12)
+                .map(|k| {
+                    let mut hh = Hasher::new();
+                    hh.write_u64(base);
+                    hh.write_u64(k);
+                    (hh.finish().0 as i64) as f64
+                })
+                .collect();
+            fps.push(world_model::Hash(base));
+            inputs.push(inp);
+            outputs.push(out);
+        }
+        let corpus = BehaviorCorpus::build(inputs, outputs, fps);
+        let summary = analyze(&corpus);
+        assert!(summary.passed(), "collapse failures: {:?}", summary.failures);
+        assert!(summary.best_1factor < 0.40);
+        assert!(summary.min_info_loss >= 0.35);
+    }
+}

crates/collapse_analysis/src/linalg.rs

@@ -0,0 +1,244 @@
+//! Minimal dense f64 linear algebra for the collapse attacks: centering,
+//! OLS (ridge-regularized) multi-output R², and PCA via power iteration.
+
+/// Column-major-agnostic row-major matrix.
+#[derive(Clone)]
+pub struct Mat {
+    pub rows: usize,
+    pub cols: usize,
+    pub data: Vec<f64>,
+}
+
+impl Mat {
+    pub fn zeros(rows: usize, cols: usize) -> Self {
+        Mat {
+            rows,
+            cols,
+            data: vec![0.0; rows * cols],
+        }
+    }
+    #[inline]
+    pub fn at(&self, r: usize, c: usize) -> f64 {
+        self.data[r * self.cols + c]
+    }
+    #[inline]
+    pub fn set(&mut self, r: usize, c: usize, v: f64) {
+        self.data[r * self.cols + c] = v;
+    }
+
+    pub fn col(&self, c: usize) -> Vec<f64> {
+        (0..self.rows).map(|r| self.at(r, c)).collect()
+    }
+
+    /// Subtract the mean of each column (in place). Returns the means.
+    pub fn center_columns(&mut self) -> Vec<f64> {
+        let mut means = vec![0.0; self.cols];
+        for c in 0..self.cols {
+            let mut s = 0.0;
+            for r in 0..self.rows {
+                s += self.at(r, c);
+            }
+            means[c] = s / self.rows.max(1) as f64;
+        }
+        for r in 0..self.rows {
+            for c in 0..self.cols {
+                let v = self.at(r, c) - means[c];
+                self.set(r, c, v);
+            }
+        }
+        means
+    }
+
+    /// X^T X
+    pub fn gram(&self) -> Mat {
+        let p = self.cols;
+        let mut g = Mat::zeros(p, p);
+        for i in 0..p {
+            for j in i..p {
+                let mut s = 0.0;
+                for r in 0..self.rows {
+                    s += self.at(r, i) * self.at(r, j);
+                }
+                g.set(i, j, s);
+                g.set(j, i, s);
+            }
+        }
+        g
+    }
+}
+
+/// Solve (A + λI) x = b for symmetric positive-ish A via Gauss-Jordan.
+pub fn solve_ridge(a: &Mat, b: &[f64], lambda: f64) -> Vec<f64> {
+    let n = a.rows;
+    let mut m = a.clone();
+    for i in 0..n {
+        let v = m.at(i, i) + lambda;
+        m.set(i, i, v);
+    }
+    let mut x = b.to_vec();
+    // Gaussian elimination with partial pivoting.
+    for col in 0..n {
+        let mut piv = col;
+        let mut best = m.at(col, col).abs();
+        for r in (col + 1)..n {
+            let v = m.at(r, col).abs();
+            if v > best {
+                best = v;
+                piv = r;
+            }
+        }
+        if best < 1e-12 {
+            continue;
+        }
+        if piv != col {
+            for c in 0..n {
+                let tmp = m.at(col, c);
+                m.set(col, c, m.at(piv, c));
+                m.set(piv, c, tmp);
+            }
+            x.swap(col, piv);
+        }
+        let d = m.at(col, col);
+        for r in 0..n {
+            if r != col {
+                let f = m.at(r, col) / d;
+                if f != 0.0 {
+                    for c in col..n {
+                        let v = m.at(r, c) - f * m.at(col, c);
+                        m.set(r, c, v);
+                    }
+                    x[r] -= f * x[col];
+                }
+            }
+        }
+    }
+    for i in 0..n {
+        let d = m.at(i, i);
+        if d.abs() > 1e-12 {
+            x[i] /= d;
+        } else {
+            x[i] = 0.0;
+        }
+    }
+    x
+}
+
+/// Average R^2 of predicting each (centered) output column from the centered
+/// design matrix `x` using ridge OLS. Returns a value clamped to [0, 1].
+pub fn ols_r2(x: &Mat, y: &Mat) -> f64 {
+    if x.cols == 0 || x.rows == 0 {
+        return 0.0;
+    }
+    let g = x.gram();
+    let lambda = 1e-6 * (1.0 + trace(&g) / x.cols as f64);
+    let mut total_r2 = 0.0;
+    let mut counted = 0;
+    for oc in 0..y.cols {
+        let yc = y.col(oc);
+        // X^T y
+        let mut xty = vec![0.0; x.cols];
+        for i in 0..x.cols {
+            let mut s = 0.0;
+            for r in 0..x.rows {
+                s += x.at(r, i) * yc[r];
+            }
+            xty[i] = s;
+        }
+        let beta = solve_ridge(&g, &xty, lambda);
+        // residuals
+        let mut ss_res = 0.0;
+        let mut ss_tot = 0.0;
+        for r in 0..x.rows {
+            let mut pred = 0.0;
+            for i in 0..x.cols {
+                pred += x.at(r, i) * beta[i];
+            }
+            ss_res += (yc[r] - pred).powi(2);
+            ss_tot += yc[r].powi(2);
+        }
+        if ss_tot > 1e-9 {
+            let r2 = 1.0 - ss_res / ss_tot;
+            total_r2 += r2.clamp(0.0, 1.0);
+            counted += 1;
+        }
+    }
+    if counted == 0 {
+        0.0
+    } else {
+        total_r2 / counted as f64
+    }
+}
+
+fn trace(m: &Mat) -> f64 {
+    (0..m.rows.min(m.cols)).map(|i| m.at(i, i)).sum()
+}
+
+/// Top-`k` principal-component scores of a centered matrix `x` via power
+/// iteration with deflation. Returns an `n x k` score matrix.
+pub fn pca_scores(x: &Mat, k: usize) -> Mat {
+    let p = x.cols;
+    let mut cov = x.gram(); // proportional to covariance
+    let kk = k.min(p);
+    let mut comps: Vec<Vec<f64>> = Vec::new();
+    for _ in 0..kk {
+        // power iteration
+        let mut v = vec![0.0; p];
+        for (i, vi) in v.iter_mut().enumerate() {
+            *vi = 1.0 + (i as f64) * 0.001;
+        }
+        normalize(&mut v);
+        for _ in 0..64 {
+            let mut nv = matvec(&cov, &v);
+            normalize(&mut nv);
+            let diff: f64 = nv.iter().zip(&v).map(|(a, b)| (a - b).abs()).sum();
+            v = nv;
+            if diff < 1e-9 {
+                break;
+            }
+        }
+        // eigenvalue
+        let av = matvec(&cov, &v);
+        let lambda: f64 = v.iter().zip(&av).map(|(a, b)| a * b).sum();
+        // deflate
+        for i in 0..p {
+            for j in 0..p {
+                let val = cov.at(i, j) - lambda * v[i] * v[j];
+                cov.set(i, j, val);
+            }
+        }
+        comps.push(v);
+    }
+    // scores = X * comps
+    let mut scores = Mat::zeros(x.rows, kk);
+    for r in 0..x.rows {
+        for (cj, comp) in comps.iter().enumerate() {
+            let mut s = 0.0;
+            for i in 0..p {
+                s += x.at(r, i) * comp[i];
+            }
+            scores.set(r, cj, s);
+        }
+    }
+    scores
+}
+
+fn matvec(m: &Mat, v: &[f64]) -> Vec<f64> {
+    let mut out = vec![0.0; m.rows];
+    for r in 0..m.rows {
+        let mut s = 0.0;
+        for c in 0..m.cols {
+            s += m.at(r, c) * v[c];
+        }
+        out[r] = s;
+    }
+    out
+}
+
+fn normalize(v: &mut [f64]) {
+    let n: f64 = v.iter().map(|x| x * x).sum::<f64>().sqrt();
+    if n > 1e-12 {
+        for x in v.iter_mut() {
+            *x /= n;
+        }
+    }
+}

crates/generators/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "generators"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+rune_ir = { path = "../rune_ir" }
+trace_model = { path = "../trace_model" }
+
+[lib]
+path = "src/lib.rs"

crates/generators/src/lib.rs

@@ -0,0 +1,344 @@
+//! `generators` — produce worlds, rune programs, executor sets, semantic
+//! contracts, and perturbation batches. Generators must reject flat cases:
+//! every generated case is checked against the generated-case gates before it
+//! is admitted to the corpus.
+
+use rune_ir::{Op, RuneProgram, RuneToken, ALL_OPS};
+use trace_model::numeric_rank;
+use world_model::{
+    standard_executors, ContractId, ExecutionContext, PerturbationAxis, ProgramId, Rng,
+    TraceDifferenceExpectation, WorldId, WorldSnapshot, HIDDEN_LANES, LANES, NUM_DOMAINS,
+};
+
+/// Minimum domain reference entropy (bits) a generated program must show.
+pub const MIN_DOMAIN_ENTROPY: f64 = 2.5;
+
+/// Semantic contract (per spec) with the default thresholds.
+#[derive(Clone, Copy, PartialEq, Debug)]
+pub struct SemanticContract {
+    pub id: ContractId,
+    pub min_causal_rank: usize,
+    pub min_domain_participation: usize,
+    pub min_future_sensitivity: f64,
+    pub min_context_divergence: f64,
+    pub max_compressibility: f64,
+}
+
+impl SemanticContract {
+    pub fn default_with_seed(seed: u64) -> Self {
+        SemanticContract {
+            id: ContractId(seed),
+            min_causal_rank: 6,
+            min_domain_participation: 4,
+            min_future_sensitivity: 0.50,
+            min_context_divergence: 0.40,
+            max_compressibility: 0.70,
+        }
+    }
+}
+
+/// One perturbed variant of a base world.
+pub struct PerturbedCase {
+    pub axis_name: String,
+    pub world: WorldSnapshot,
+    pub expectation: TraceDifferenceExpectation,
+}
+
+/// A complete generated case (per spec).
+pub struct GeneratedCase {
+    pub world: WorldSnapshot,
+    pub program: RuneProgram,
+    pub contexts: Vec<ExecutionContext>,
+    pub contract: SemanticContract,
+    pub perturbations: Vec<PerturbedCase>,
+    pub world_seed: u64,
+    pub program_seed: u64,
+    pub contract_seed: u64,
+    pub perturbation_seed: u64,
+}
+
+/// Generate a rich world from a seed.
+pub fn generate_world(seed: u64) -> WorldSnapshot {
+    let mut rng = Rng::derive(seed, "world");
+    let mut w = WorldSnapshot::blank(WorldId(seed), seed);
+
+    for (i, d) in w.domains.iter_mut().enumerate() {
+        let spread = 1000 + (i as i64) * 137;
+        for l in 0..LANES {
+            d.observed[l] = rng.range_i64(-spread, spread);
+        }
+        for l in 0..HIDDEN_LANES {
+            let mut v = rng.range_i64(-spread, spread);
+            if v == 0 {
+                v = 1 + i as i64;
+            }
+            d.hidden[l] = v;
+        }
+    }
+
+    for j in 0..NUM_DOMAINS {
+        for i in 0..NUM_DOMAINS {
+            w.causal_state.coupling[j][i] = rng.range_i64(-17, 17);
+        }
+    }
+
+    for di in 0..NUM_DOMAINS {
+        for l in 0..LANES {
+            w.observation_state.visible[di][l] = !rng.chance(0.15);
+        }
+        if (0..LANES).all(|l| !w.observation_state.visible[di][l]) {
+            w.observation_state.visible[di][0] = true;
+        }
+    }
+    w.observation_state.noise_seed = rng.next_u64();
+
+    for r in w.execution_state.accumulator.iter_mut() {
+        *r = rng.range_i64(-50, 50);
+    }
+    w
+}
+
+/// Generate a rune program from a seed. Construction guarantees every opcode
+/// appears and every domain is both read and written, so the case is never
+/// flat.
+pub fn generate_program(seed: u64) -> RuneProgram {
+    let mut rng = Rng::derive(seed, "program");
+    let len = 28 + rng.below(13); // 28..=40
+    let mut tokens = Vec::with_capacity(len);
+    for i in 0..len {
+        let op = if rng.chance(0.7) {
+            ALL_OPS[i % ALL_OPS.len()]
+        } else {
+            Op::from_u8(rng.next_u64() as u8)
+        };
+        // 3 and 5 are coprime with 8, so src/dst sweep all domains.
+        let a = ((i * 3 + rng.below(2)) % NUM_DOMAINS) as u8;
+        let mut b = ((i * 5 + 1 + rng.below(2)) % NUM_DOMAINS) as u8;
+        if b == a {
+            b = (b + 1) % NUM_DOMAINS as u8;
+        }
+        tokens.push(RuneToken {
+            op,
+            a,
+            b,
+            c: rng.next_u64() as u8,
+            imm: rng.range_i64(-100_000, 100_000),
+        });
+    }
+    RuneProgram {
+        id: ProgramId(seed),
+        tokens,
+        seed,
+    }
+}
+
+/// Generate >= 3 executors.
+pub fn generate_contexts(seed: u64) -> Vec<ExecutionContext> {
+    let mut rng = Rng::derive(seed, "context-count");
+    standard_executors(seed, 3 + rng.below(2))
+}
+
+/// Build the perturbation batch for a world (>= 10 perturbations drawn from
+/// domain surfaces).
+pub fn generate_perturbations(world: &WorldSnapshot, seed: u64, count: usize) -> Vec<PerturbedCase> {
+    use world_model::WorldDomain;
+    let count = count.max(10);
+    let mut axes: Vec<Box<dyn PerturbationAxis>> = Vec::new();
+    for d in &world.domains {
+        axes.extend(d.perturbation_axes());
+    }
+    let mut rng = Rng::derive(seed, "perturb");
+    let mut out = Vec::with_capacity(count);
+    for k in 0..count {
+        let idx = (k * 7 + rng.below(axes.len())) % axes.len();
+        let axis = &axes[idx];
+        out.push(PerturbedCase {
+            axis_name: axis.name(),
+            world: axis.apply(world),
+            expectation: axis.expected_trace_difference(),
+        });
+    }
+    out
+}
+
+/// Generate a full case from a master seed.
+pub fn generate_case(master_seed: u64) -> GeneratedCase {
+    let world_seed = derive(master_seed, "world");
+    let program_seed = derive(master_seed, "program");
+    let contract_seed = derive(master_seed, "contract");
+    let perturbation_seed = derive(master_seed, "perturb");
+
+    let world = generate_world(world_seed);
+    let program = generate_program(program_seed);
+    let contexts = generate_contexts(master_seed);
+    let contract = SemanticContract::default_with_seed(contract_seed);
+    let perturbations = generate_perturbations(&world, perturbation_seed, 10);
+
+    GeneratedCase {
+        world,
+        program,
+        contexts,
+        contract,
+        perturbations,
+        world_seed,
+        program_seed,
+        contract_seed,
+        perturbation_seed,
+    }
+}
+
+pub fn derive(seed: u64, tag: &str) -> u64 {
+    let mut h = world_model::Hasher::new();
+    h.write_tag(tag);
+    h.write_u64(seed);
+    h.finish().0
+}
+
+// ---------------------------------------------------------------------------
+// Generated-case gates: reject flat cases.
+// ---------------------------------------------------------------------------
+
+#[derive(Clone, Copy, Debug, PartialEq)]
+pub struct CaseEstimate {
+    pub estimated_causal_rank: usize,
+    pub domain_entropy: f64,
+    pub perturbation_axes: usize,
+    pub executor_count: usize,
+    pub future_dependence: bool,
+    pub hidden_observed_divergence: bool,
+    pub nonuniform_fingerprints: bool,
+}
+
+impl GeneratedCase {
+    /// Estimate the structural richness of the case without running a runtime.
+    pub fn estimate(&self) -> CaseEstimate {
+        let mut adj = [[0.0f64; NUM_DOMAINS]; NUM_DOMAINS];
+        let mut refs = [0u32; NUM_DOMAINS];
+        for t in &self.program.tokens {
+            let s = t.src_domain();
+            let d = t.dst_domain();
+            let phase = 1.0 + t.lane() as f64 + 4.0 * t.lane2() as f64;
+            adj[d][s] += phase;
+            adj[d][d] += 0.5 + t.mode() as f64;
+            refs[s] += 1;
+            refs[d] += 1;
+        }
+        let rows: Vec<Vec<f64>> = adj.iter().map(|r| r.to_vec()).collect();
+        let est_rank = numeric_rank(&rows);
+
+        let total: u32 = refs.iter().sum();
+        let mut entropy = 0.0;
+        if total > 0 {
+            for &c in &refs {
+                if c > 0 {
+                    let p = c as f64 / total as f64;
+                    entropy -= p * p.log2();
+                }
+            }
+        }
+
+        let future = self.program.tokens.iter().any(|t| matches!(t.op, Op::Schedule));
+
+        let hidden_div = self
+            .world
+            .domains
+            .iter()
+            .any(|d| d.hidden.iter().any(|&v| v != 0))
+            || self
+                .world
+                .observation_state
+                .visible
+                .iter()
+                .any(|row| row.iter().any(|&v| !v));
+
+        use world_model::WorldDomain;
+        let mut prints: Vec<_> = self.world.domains.iter().map(|d| d.fingerprint().hash).collect();
+        prints.sort();
+        prints.dedup();
+        let nonuniform = prints.len() > 1;
+
+        CaseEstimate {
+            estimated_causal_rank: est_rank,
+            domain_entropy: entropy,
+            perturbation_axes: self.perturbations.len().max(count_axes(&self.world)),
+            executor_count: self.contexts.len(),
+            future_dependence: future,
+            hidden_observed_divergence: hidden_div,
+            nonuniform_fingerprints: nonuniform,
+        }
+    }
+}
+
+fn count_axes(world: &WorldSnapshot) -> usize {
+    use world_model::WorldDomain;
+    world.domains.iter().map(|d| d.perturbation_axes().len()).sum()
+}
+
+/// Reasons a case failed the generated gates (empty = passed).
+pub fn generated_gate_failures(est: &CaseEstimate) -> Vec<String> {
+    let mut f = Vec::new();
+    if est.estimated_causal_rank < 6 {
+        f.push(format!("estimated_causal_rank {} < 6", est.estimated_causal_rank));
+    }
+    if est.domain_entropy < MIN_DOMAIN_ENTROPY {
+        f.push(format!(
+            "domain_entropy {:.3} < {:.3}",
+            est.domain_entropy, MIN_DOMAIN_ENTROPY
+        ));
+    }
+    if est.perturbation_axes < 10 {
+        f.push(format!("perturbation_axes {} < 10", est.perturbation_axes));
+    }
+    if est.executor_count < 3 {
+        f.push(format!("executor_count {} < 3", est.executor_count));
+    }
+    if !est.future_dependence {
+        f.push("future_dependence absent".into());
+    }
+    if !est.hidden_observed_divergence {
+        f.push("no hidden/observed divergence".into());
+    }
+    if !est.nonuniform_fingerprints {
+        f.push("uniform domain fingerprints".into());
+    }
+    f
+}
+
+/// Generate a case that passes the generated gates, retrying with fresh seeds.
+pub fn generate_accepted_case(master_seed: u64) -> (GeneratedCase, u64) {
+    let mut s = master_seed;
+    for _ in 0..64 {
+        let case = generate_case(s);
+        if generated_gate_failures(&case.estimate()).is_empty() {
+            return (case, s);
+        }
+        s = derive(s, "retry");
+    }
+    let case = generate_case(s);
+    (case, s)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn accepted_cases_pass_generated_gates() {
+        for i in 0..200u64 {
+            let (case, _) = generate_accepted_case(0x1234 ^ i.wrapping_mul(0x9e3779b97f4a7c15));
+            let failures = generated_gate_failures(&case.estimate());
+            assert!(failures.is_empty(), "case {i} failed: {:?}", failures);
+            assert!(case.contexts.len() >= 3);
+            assert!(case.perturbations.len() >= 10);
+        }
+    }
+
+    #[test]
+    fn generation_is_deterministic() {
+        let (a, sa) = generate_accepted_case(999);
+        let (b, sb) = generate_accepted_case(999);
+        assert_eq!(sa, sb);
+        assert_eq!(a.program.content_hash(), b.program.content_hash());
+        assert_eq!(a.world.content_hash(), b.world.content_hash());
+    }
+}

crates/reference_runtime/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "reference_runtime"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+rune_ir = { path = "../rune_ir" }
+trace_model = { path = "../trace_model" }
+
+[lib]
+path = "src/lib.rs"

crates/reference_runtime/src/engine.rs

@@ -0,0 +1,654 @@
+//! The execution engine. Both the reference runtime and the runtime under
+//! test call [`execute`] with the canonical [`EngineConfig`]; semantic mutants
+//! are nothing more than an `EngineConfig` with one behavior-affecting knob
+//! changed, which is what makes them detectable by the equivalence gate.
+//!
+//! All arithmetic is integer and total: division is guarded, overflow wraps,
+//! and every token produces a defined effect or a *logged* fault. The engine
+//! never panics.
+
+use rune_ir::{Op, RuneToken};
+use trace_model::{
+    BehaviorFingerprint, CausalEdge, CausalGraph, CausalNode, DivergenceGraph, DomainAccessGraph,
+    ExecutionTrace, FaultCode, FaultLog, InformationFlowGraph, PerturbationResponse, ReplayRecord,
+    TemporalGraph,
+};
+use world_model::{
+    DomainKind, ExecutionContext, Hash, Hasher, ScheduledEffect, WorldDelta, WorldSnapshot,
+    DomainId, HIDDEN_LANES, LANES, NUM_DOMAINS, REGS,
+};
+
+/// All behavior-affecting knobs of the engine. The reference config is the
+/// executable spec; mutants flip exactly one knob.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct EngineConfig {
+    pub c1: u64,
+    pub c2: u64,
+    pub s1: u32,
+    pub s2: u32,
+    pub s3: u32,
+    pub use_coupling: bool,
+    pub use_hidden: bool,
+    pub use_executor_salt: bool,
+    pub branch_enabled: bool,
+    pub schedule_enabled: bool,
+    pub record_causal: bool,
+    pub domain_mask: [bool; NUM_DOMAINS],
+    pub op_enabled: [bool; 12],
+    pub future_turns: usize,
+    pub diffuse_span: usize,
+}
+
+impl EngineConfig {
+    /// The canonical executable-spec configuration.
+    pub fn reference() -> Self {
+        EngineConfig {
+            c1: 0xff51afd7ed558ccd,
+            c2: 0xc4ceb9fe1a85ec53,
+            s1: 33,
+            s2: 29,
+            s3: 32,
+            use_coupling: true,
+            use_hidden: true,
+            use_executor_salt: true,
+            branch_enabled: true,
+            schedule_enabled: true,
+            record_causal: true,
+            domain_mask: [true; NUM_DOMAINS],
+            op_enabled: [true; 12],
+            future_turns: 3,
+            diffuse_span: 3,
+        }
+    }
+}
+
+/// Input to a resolution.
+#[derive(Clone, Debug)]
+pub struct ResolutionInput {
+    pub world: WorldSnapshot,
+    pub program: rune_ir::RuneProgram,
+    pub contexts: Vec<ExecutionContext>,
+    pub contract_seed: u64,
+    pub perturbation_seed: u64,
+}
+
+/// Output of a resolution (per spec). Execution always returns this; no rune
+/// stream is ever rejected.
+#[derive(Clone, Debug)]
+pub struct ResolutionResult {
+    pub delta: WorldDelta,
+    pub trace: ExecutionTrace,
+    pub faults: FaultLog,
+    pub replay: ReplayRecord,
+}
+
+/// Canonical, comparable view of a result. Reference and runtime-under-test
+/// must produce identical canonical views.
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub struct Canonical {
+    pub delta_hash: Hash,
+    pub trace_hash: Hash,
+    pub fault_hash: Hash,
+    pub replay_hash: Hash,
+    pub future_hash: Hash,
+}
+
+/// Produce the canonical comparison tuple for a result.
+pub fn canonical(r: &ResolutionResult) -> Canonical {
+    Canonical {
+        delta_hash: r.delta.hash(),
+        trace_hash: r.trace.canonical_hash(),
+        fault_hash: r.faults.hash(),
+        replay_hash: r.replay.hash(),
+        future_hash: r.replay.future_hash,
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Recorder: accumulates graph/trace data during a single-context run.
+// ---------------------------------------------------------------------------
+
+struct Recorder {
+    read_graph: DomainAccessGraph,
+    write_graph: DomainAccessGraph,
+    causal_graph: CausalGraph,
+    info_flow: InformationFlowGraph,
+    temporal: TemporalGraph,
+    faults: FaultLog,
+}
+
+impl Recorder {
+    fn new() -> Self {
+        Recorder {
+            read_graph: DomainAccessGraph::default(),
+            write_graph: DomainAccessGraph::default(),
+            causal_graph: CausalGraph::default(),
+            info_flow: InformationFlowGraph::default(),
+            temporal: TemporalGraph::default(),
+            faults: FaultLog::default(),
+        }
+    }
+
+    #[allow(clippy::too_many_arguments)]
+    fn flow(
+        &mut self,
+        cfg: &EngineConfig,
+        from_dom: usize,
+        from_lane: usize,
+        from_hidden: bool,
+        to_dom: usize,
+        to_lane: usize,
+        to_hidden: bool,
+        step: u32,
+        weight: i64,
+    ) {
+        if !cfg.domain_mask[from_dom] || !cfg.domain_mask[to_dom] {
+            return;
+        }
+        self.read_graph.access_count[from_dom] += 1;
+        self.write_graph.access_count[to_dom] += 1;
+        self.read_graph.edges.push((from_dom as u8, to_dom as u8, 1));
+        self.write_graph.edges.push((from_dom as u8, to_dom as u8, 1));
+        self.info_flow
+            .edges
+            .push((from_dom as u8, to_dom as u8, (weight as u64).count_ones()));
+        if cfg.record_causal {
+            self.causal_graph.edges.push(CausalEdge {
+                from: CausalNode {
+                    domain: from_dom as u8,
+                    lane: from_lane as u8,
+                    hidden: from_hidden,
+                    step,
+                },
+                to: CausalNode {
+                    domain: to_dom as u8,
+                    lane: to_lane as u8,
+                    hidden: to_hidden,
+                    step,
+                },
+                weight,
+            });
+        }
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Core arithmetic.
+// ---------------------------------------------------------------------------
+
+#[inline]
+fn avalanche(cfg: &EngineConfig, z: i64) -> i64 {
+    let mut u = z as u64;
+    u ^= u >> cfg.s1;
+    u = u.wrapping_mul(cfg.c1);
+    u ^= u >> cfg.s2;
+    u = u.wrapping_mul(cfg.c2);
+    u ^= u >> cfg.s3;
+    u as i64
+}
+
+#[inline]
+fn combine(cfg: &EngineConfig, ctx: &ExecutionContext, a: i64, b: i64, coupling: i64, kc: u64) -> i64 {
+    let mut z = a.wrapping_mul(kc as i64);
+    z ^= b.rotate_left(((kc & 31) as u32) + 1);
+    if cfg.use_coupling {
+        z = z.wrapping_add(coupling.wrapping_mul(b & 0xffff));
+    }
+    if cfg.use_executor_salt {
+        z ^= ctx.salt() as i64;
+        z = z.wrapping_add(ctx.profile.bias);
+        z = z.rotate_left((ctx.profile.rotate % 63) + 1);
+    }
+    avalanche(cfg, z)
+}
+
+#[inline]
+fn read_lane(cfg: &EngineConfig, w: &WorldSnapshot, dom: usize, lane: usize, hidden: bool) -> i64 {
+    if !cfg.domain_mask[dom] {
+        return 0;
+    }
+    if hidden {
+        if cfg.use_hidden {
+            w.domains[dom].hidden[lane % HIDDEN_LANES]
+        } else {
+            0
+        }
+    } else {
+        w.domains[dom].observed[lane % LANES]
+    }
+}
+
+#[inline]
+fn write_lane(cfg: &EngineConfig, w: &mut WorldSnapshot, dom: usize, lane: usize, hidden: bool, val: i64) {
+    if !cfg.domain_mask[dom] {
+        return;
+    }
+    if hidden {
+        w.domains[dom].hidden[lane % HIDDEN_LANES] = val;
+    } else {
+        w.domains[dom].observed[lane % LANES] = val;
+    }
+}
+
+#[inline]
+fn coupling_at(w: &WorldSnapshot, to: usize, from: usize) -> i64 {
+    w.causal_state.coupling[to][from]
+}
+
+// ---------------------------------------------------------------------------
+// Single-context program run.
+// ---------------------------------------------------------------------------
+
+// ---------------------------------------------------------------------------
+// Future stepping (genuine future dependence over 3 turns).
+// ---------------------------------------------------------------------------
+
+fn step_world(cfg: &EngineConfig, w: &mut WorldSnapshot) {
+    // Resolve scheduled effects due this turn.
+    let pending = std::mem::take(&mut w.time_state.pending);
+    let mut still = Vec::new();
+    for e in pending {
+        if e.turn_offset <= 1 {
+            let d = e.domain.0 as usize;
+            if cfg.domain_mask[d] {
+                if e.hidden {
+                    let l = e.lane % HIDDEN_LANES;
+                    w.domains[d].hidden[l] = w.domains[d].hidden[l].wrapping_add(e.value);
+                } else {
+                    let l = e.lane % LANES;
+                    w.domains[d].observed[l] = w.domains[d].observed[l].wrapping_add(e.value);
+                }
+            }
+        } else {
+            still.push(ScheduledEffect {
+                turn_offset: e.turn_offset - 1,
+                ..e
+            });
+        }
+    }
+    w.time_state.pending = still;
+
+    // Coupling diffusion: every domain pulls from every other through the
+    // coupling matrix, then avalanches. This propagates any execution effect
+    // into the future and makes future state sensitive to the present.
+    let snap = w.domains.clone();
+    for j in 0..NUM_DOMAINS {
+        if !cfg.domain_mask[j] {
+            continue;
+        }
+        for lane in 0..LANES {
+            let mut z = w.domains[j].observed[lane];
+            for i in 0..NUM_DOMAINS {
+                if !cfg.domain_mask[i] {
+                    continue;
+                }
+                if cfg.use_coupling {
+                    let c = w.causal_state.coupling[j][i];
+                    z = z.wrapping_add(c.wrapping_mul(snap[i].observed[lane] & 0xff));
+                } else {
+                    z = z.wrapping_add(snap[i].observed[lane] & 0xff);
+                }
+            }
+            w.domains[j].observed[lane] = avalanche(cfg, z);
+        }
+        for hl in 0..HIDDEN_LANES {
+            let base = w.domains[j].hidden[hl].wrapping_add(snap[j].observed[0]);
+            w.domains[j].hidden[hl] = if cfg.use_hidden { avalanche(cfg, base) } else { base };
+        }
+    }
+    w.turn = w.turn.wrapping_add(1);
+}
+
+fn future_hash(cfg: &EngineConfig, start: &WorldSnapshot) -> Hash {
+    let mut w = start.clone();
+    let mut h = Hasher::new();
+    h.write_tag("future-3");
+    for _ in 0..cfg.future_turns {
+        step_world(cfg, &mut w);
+        for v in w.ground_truth() {
+            h.write_i64(v);
+        }
+    }
+    h.finish()
+}
+
+// ---------------------------------------------------------------------------
+// Divergence + behavior fingerprint.
+// ---------------------------------------------------------------------------
+
+fn compute_divergence(finals: &[WorldSnapshot]) -> DivergenceGraph {
+    let n = finals.len();
+    let mut pairwise = vec![0.0f64; n * n];
+    let total = (NUM_DOMAINS * LANES) as f64;
+    for i in 0..n {
+        for j in 0..n {
+            if i == j {
+                continue;
+            }
+            let mut diff = 0usize;
+            for d in 0..NUM_DOMAINS {
+                for l in 0..LANES {
+                    if finals[i].domains[d].observed[l] != finals[j].domains[d].observed[l] {
+                        diff += 1;
+                    }
+                }
+            }
+            pairwise[i * n + j] = diff as f64 / total;
+        }
+    }
+    DivergenceGraph {
+        executor_count: n,
+        pairwise,
+    }
+}
+
+fn behavior_fingerprint(
+    before: &WorldSnapshot,
+    after: &WorldSnapshot,
+    delta: &WorldDelta,
+    rec: &Recorder,
+    divergence: &DivergenceGraph,
+    future: Hash,
+) -> BehaviorFingerprint {
+    let mut features: Vec<i64> = Vec::new();
+    // Per-domain observed and hidden delta magnitudes.
+    for dd in &delta.domain_deltas {
+        let mut s = 0i64;
+        for &v in &dd.observed {
+            s = s.wrapping_add(v);
+        }
+        features.push(s);
+    }
+    for dd in &delta.domain_deltas {
+        let mut s = 0i64;
+        for &v in &dd.hidden {
+            s = s.wrapping_add(v);
+        }
+        features.push(s);
+    }
+    // Structural counts.
+    features.push(rec.causal_graph.causal_rank() as i64);
+    features.push(rec.causal_graph.edge_count() as i64);
+    features.push(rec.read_graph.touched_count() as i64);
+    features.push(rec.write_graph.touched_count() as i64);
+    features.push(rec.info_flow.total_bits() as i64);
+    features.push(rec.temporal.edge_count() as i64);
+    features.push((divergence.mean_divergence() * 1_000_000.0) as i64);
+    features.push(future.0 as i64);
+    let _ = (before, after);
+    BehaviorFingerprint::from_features(features)
+}
+
+// ---------------------------------------------------------------------------
+// Public entry point.
+// ---------------------------------------------------------------------------
+
+/// Execute a resolution with the given engine config.
+pub fn execute(cfg: &EngineConfig, input: &ResolutionInput) -> ResolutionResult {
+    let mut contexts = input.contexts.clone();
+    if contexts.is_empty() {
+        contexts = world_model::standard_executors(input.world.seed, 3);
+    }
+
+    // A masked (removed) domain contributes nothing: its state is erased up
+    // front so it cannot leak into deltas, future hashes, or fingerprints.
+    let mut world0 = input.world.clone();
+    for d in 0..NUM_DOMAINS {
+        if !cfg.domain_mask[d] {
+            world0.domains[d].observed = [0; LANES];
+            world0.domains[d].hidden = [0; HIDDEN_LANES];
+        }
+    }
+
+    // Run under every executor; keep the primary (index 0) full recording.
+    let mut finals: Vec<WorldSnapshot> = Vec::with_capacity(contexts.len());
+    let mut primary: Option<(WorldSnapshot, Recorder)> = None;
+    for (idx, ctx) in contexts.iter().enumerate() {
+        let (fin, rec) = run_program_with_program(cfg, &world0, ctx, &input.program);
+        if idx == 0 {
+            primary = Some((fin.clone(), rec));
+        }
+        finals.push(fin);
+    }
+    let (primary_final, rec) = primary.expect("at least one executor");
+
+    let delta = WorldDelta::between(&world0, &primary_final);
+    let divergence = compute_divergence(&finals);
+    let fhash = future_hash(cfg, &primary_final);
+    let behavior =
+        behavior_fingerprint(&world0, &primary_final, &delta, &rec, &divergence, fhash);
+
+    let trace = ExecutionTrace {
+        read_graph: rec.read_graph.clone(),
+        write_graph: rec.write_graph.clone(),
+        causal_graph: rec.causal_graph.clone(),
+        information_flow: rec.info_flow.clone(),
+        executor_divergence: divergence,
+        temporal_graph: rec.temporal.clone(),
+        perturbation_response: PerturbationResponse::default(),
+        behavior_fingerprint: behavior,
+    };
+    let faults = rec.faults;
+    let trace_hash = trace.canonical_hash();
+    let delta_hash = delta.hash();
+    let replay = ReplayRecord {
+        world_seed: input.world.seed,
+        program_seed: input.program.seed,
+        contract_seed: input.contract_seed,
+        perturbation_seed: input.perturbation_seed,
+        trace_hash,
+        delta_hash,
+        future_hash: fhash,
+    };
+
+    ResolutionResult {
+        delta,
+        trace,
+        faults,
+        replay,
+    }
+}
+
+/// `run_program` variant that takes the program explicitly. (The borrow-split
+/// helper above intentionally returns no tokens; this is the real driver.)
+fn run_program_with_program(
+    cfg: &EngineConfig,
+    world: &WorldSnapshot,
+    ctx: &ExecutionContext,
+    program: &rune_ir::RuneProgram,
+) -> (WorldSnapshot, Recorder) {
+    let mut w = world.clone();
+    let mut rec = Recorder::new();
+    let mut acc = w.execution_state.accumulator;
+    let mut acc_src: [usize; REGS] = [0; REGS];
+
+    for (i, tok) in program.tokens.iter().enumerate() {
+        let step = i as u32;
+        let src = tok.src_domain();
+        let dst = tok.dst_domain();
+        let lane = tok.lane();
+        let lane2 = tok.lane2();
+        let kc = DomainKind::from_index(dst).mix_const();
+        let coupling = coupling_at(&w, dst, src);
+
+        if !cfg.op_enabled[tok.op.to_u8() as usize] {
+            rec.faults.push(FaultCode::NoEffectToken, step, tok.op.to_u8() as i64);
+            continue;
+        }
+
+        interpret(cfg, ctx, &mut w, &mut rec, &mut acc, &mut acc_src, tok, step, src, dst, lane, lane2, kc, coupling);
+
+        let r = (step as usize) % REGS;
+        acc[r] = acc[r].wrapping_add(w.domains[dst].observed[lane]);
+    }
+
+    w.execution_state.accumulator = acc;
+    (w, rec)
+}
+
+#[allow(clippy::too_many_arguments)]
+fn interpret(
+    cfg: &EngineConfig,
+    ctx: &ExecutionContext,
+    w: &mut WorldSnapshot,
+    rec: &mut Recorder,
+    acc: &mut [i64; REGS],
+    acc_src: &mut [usize; REGS],
+    tok: &RuneToken,
+    step: u32,
+    src: usize,
+    dst: usize,
+    lane: usize,
+    lane2: usize,
+    kc: u64,
+    coupling: i64,
+) {
+    match tok.op {
+        Op::Mix => {
+            let a = read_lane(cfg, w, src, lane, false);
+            let b = read_lane(cfg, w, dst, lane2, false);
+            let v = combine(cfg, ctx, a, b, coupling, kc);
+            write_lane(cfg, w, dst, lane, false, v);
+            rec.flow(cfg, src, lane, false, dst, lane, false, step, v);
+            rec.flow(cfg, dst, lane2, false, dst, lane, false, step, v);
+        }
+        Op::Channel => {
+            let a = read_lane(cfg, w, src, lane, false);
+            let v = combine(cfg, ctx, a, coupling, coupling, kc);
+            write_lane(cfg, w, dst, lane2, false, v);
+            rec.flow(cfg, src, lane, false, dst, lane2, false, step, v);
+        }
+        Op::Branch => {
+            let probe = read_lane(cfg, w, src, lane, false);
+            let take_hot = if cfg.branch_enabled {
+                probe.wrapping_add(ctx.profile.bias) > ctx.profile.branch_threshold
+            } else {
+                true
+            };
+            if take_hot {
+                let b = read_lane(cfg, w, dst, lane, false);
+                let v = combine(cfg, ctx, probe, b, coupling, kc);
+                write_lane(cfg, w, dst, lane, false, v);
+                rec.flow(cfg, src, lane, false, dst, lane, false, step, v);
+            } else {
+                let b = read_lane(cfg, w, dst, lane2, false);
+                let v = combine(cfg, ctx, b, probe, coupling, kc).wrapping_add(0x5bd1e9);
+                write_lane(cfg, w, dst, lane2, false, v);
+                rec.flow(cfg, src, lane, false, dst, lane2, false, step, v);
+                rec.faults.push(FaultCode::UnreachableBranch, step, 0);
+            }
+        }
+        Op::Schedule => {
+            let a = read_lane(cfg, w, src, lane, false);
+            let b = read_lane(cfg, w, dst, lane, false);
+            let v = combine(cfg, ctx, a, b, coupling, kc);
+            if cfg.schedule_enabled {
+                let offset = 1 + (tok.imm.rem_euclid(3)) as u8;
+                let hidden = tok.mode() & 1 == 1;
+                w.time_state.pending.push(ScheduledEffect {
+                    turn_offset: offset,
+                    domain: DomainId(dst as u8),
+                    lane,
+                    hidden,
+                    value: v,
+                });
+                rec.temporal.edges.push((step, offset, dst as u8));
+                rec.flow(cfg, src, lane, false, dst, lane, hidden, step, v);
+            } else {
+                rec.faults.push(FaultCode::NoEffectToken, step, 1);
+            }
+        }
+        Op::Resonate => {
+            let a = read_lane(cfg, w, src, lane, false);
+            let b = read_lane(cfg, w, dst, lane, false);
+            let m = combine(cfg, ctx, a, b, coupling, kc);
+            let va = a.wrapping_add(m);
+            let vb = b ^ m;
+            write_lane(cfg, w, src, lane, false, va);
+            write_lane(cfg, w, dst, lane, false, vb);
+            rec.flow(cfg, dst, lane, false, src, lane, false, step, va);
+            rec.flow(cfg, src, lane, false, dst, lane, false, step, vb);
+        }
+        Op::Observe => {
+            let reg = tok.mode() % REGS;
+            let mut z: i64 = acc[reg];
+            let proj = w.observed_projection();
+            for k in 0..3 {
+                let d = (src + k) % NUM_DOMAINS;
+                if !cfg.domain_mask[d] {
+                    continue;
+                }
+                let idx = d * LANES + (lane + k) % LANES;
+                z = combine(cfg, ctx, z, proj[idx], coupling_at(w, dst, d), kc);
+                rec.flow(cfg, d, (lane + k) % LANES, false, dst, lane, true, step, z);
+            }
+            acc[reg] = z;
+            acc_src[reg] = src;
+            write_lane(cfg, w, dst, tok.mode() % HIDDEN_LANES, true, z);
+        }
+        Op::Collapse => {
+            let reg = tok.mode() % REGS;
+            let a = acc[reg];
+            let b = read_lane(cfg, w, dst, lane, false);
+            if a == 0 {
+                rec.faults.push(FaultCode::EmptyAccumulator, step, reg as i64);
+            }
+            let v = combine(cfg, ctx, a, b, coupling, kc);
+            write_lane(cfg, w, dst, lane, false, v);
+            rec.flow(cfg, acc_src[reg], 0, true, dst, lane, false, step, v);
+        }
+        Op::Invert => {
+            let b = read_lane(cfg, w, dst, lane, false);
+            let mut v = avalanche(cfg, (!b).wrapping_add(tok.imm));
+            if cfg.use_executor_salt {
+                v ^= ctx.salt() as i64;
+                v = v.wrapping_add(ctx.profile.bias);
+            }
+            write_lane(cfg, w, dst, lane, false, v);
+            rec.flow(cfg, dst, lane, false, dst, lane, false, step, v);
+        }
+        Op::Diffuse => {
+            let a = read_lane(cfg, w, src, lane, false);
+            for k in 0..cfg.diffuse_span {
+                let d = (src + 1 + k) % NUM_DOMAINS;
+                let tl = (lane + k) % LANES;
+                let prev = read_lane(cfg, w, d, tl, false);
+                let v = combine(cfg, ctx, a, prev, coupling_at(w, d, src), DomainKind::from_index(d).mix_const());
+                write_lane(cfg, w, d, tl, false, prev.wrapping_add(v));
+                rec.flow(cfg, src, lane, false, d, tl, false, step, v);
+            }
+        }
+        Op::Anchor => {
+            let bound = (tok.imm.unsigned_abs() % 1_000_000) as i64 + 1;
+            let b = read_lane(cfg, w, dst, lane, false);
+            let diag = coupling_at(w, dst, dst);
+            let mut mixed = b.wrapping_add(diag);
+            if cfg.use_executor_salt {
+                mixed = mixed
+                    .wrapping_add(ctx.profile.bias)
+                    .wrapping_add((ctx.salt() & 0xffff) as i64);
+            }
+            let clamped = mixed.clamp(-bound, bound);
+            if clamped != mixed {
+                rec.faults.push(FaultCode::Saturated, step, bound);
+            }
+            write_lane(cfg, w, dst, lane, false, clamped);
+            rec.flow(cfg, dst, lane, false, dst, lane, false, step, clamped);
+        }
+        Op::Echoback => {
+            let h = read_lane(cfg, w, dst, tok.mode() % HIDDEN_LANES, true);
+            let b = read_lane(cfg, w, dst, lane, false);
+            let v = combine(cfg, ctx, h, b, coupling, kc);
+            write_lane(cfg, w, dst, lane, false, v);
+            rec.flow(cfg, dst, tok.mode() % HIDDEN_LANES, true, dst, lane, false, step, v);
+        }
+        Op::Imprint => {
+            let b = read_lane(cfg, w, dst, lane, false);
+            let hl = tok.mode() % HIDDEN_LANES;
+            let prevh = read_lane(cfg, w, dst, hl, true);
+            let v = combine(cfg, ctx, b, prevh, coupling, kc);
+            write_lane(cfg, w, dst, hl, true, v);
+            rec.flow(cfg, dst, lane, false, dst, hl, true, step, v);
+        }
+    }
+}

crates/reference_runtime/src/lib.rs

@@ -0,0 +1,117 @@
+//! `reference_runtime` — the executable specification. Every execution in CI
+//! runs the reference and the runtime-under-test and asserts their canonical
+//! views are identical. The reference is intentionally the simplest correct
+//! expression of the engine.
+
+pub mod engine;
+
+pub use engine::{
+    canonical, execute, Canonical, EngineConfig, ResolutionInput, ResolutionResult,
+};
+
+/// The runtime trait (per spec).
+pub trait Runtime {
+    fn resolve(&self, input: ResolutionInput) -> ResolutionResult;
+}
+
+/// The reference runtime: executes with the canonical engine config.
+#[derive(Clone, Debug, Default)]
+pub struct ReferenceRuntime;
+
+impl ReferenceRuntime {
+    pub fn new() -> Self {
+        ReferenceRuntime
+    }
+}
+
+impl Runtime for ReferenceRuntime {
+    fn resolve(&self, input: ResolutionInput) -> ResolutionResult {
+        execute(&EngineConfig::reference(), &input)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use rune_ir::{Op, RuneProgram, RuneToken, ALL_OPS};
+    use world_model::{standard_executors, Rng, WorldId, WorldSnapshot, NUM_DOMAINS};
+
+    fn random_input(seed: u64) -> ResolutionInput {
+        let mut rng = Rng::new(seed);
+        let mut w = WorldSnapshot::blank(WorldId(seed), seed);
+        for d in &mut w.domains {
+            for l in 0..world_model::LANES {
+                d.observed[l] = rng.range_i64(-5000, 5000);
+            }
+            for l in 0..world_model::HIDDEN_LANES {
+                d.hidden[l] = rng.range_i64(-5000, 5000);
+            }
+        }
+        for j in 0..NUM_DOMAINS {
+            for i in 0..NUM_DOMAINS {
+                w.causal_state.coupling[j][i] = rng.range_i64(-17, 17);
+            }
+        }
+        let tokens: Vec<RuneToken> = (0..30)
+            .map(|i| RuneToken {
+                op: if i % 3 == 0 { ALL_OPS[i % 12] } else { Op::from_u8(rng.next_u64() as u8) },
+                a: rng.next_u64() as u8,
+                b: rng.next_u64() as u8,
+                c: rng.next_u64() as u8,
+                imm: rng.range_i64(-100000, 100000),
+            })
+            .collect();
+        ResolutionInput {
+            world: w,
+            program: RuneProgram { id: world_model::ProgramId(seed), tokens, seed },
+            contexts: standard_executors(seed, 3),
+            contract_seed: seed,
+            perturbation_seed: seed,
+        }
+    }
+
+    #[test]
+    fn execution_is_deterministic() {
+        let cfg = EngineConfig::reference();
+        for s in 0..200 {
+            let input = random_input(s);
+            let a = execute(&cfg, &input);
+            let b = execute(&cfg, &input);
+            assert_eq!(canonical(&a), canonical(&b), "nondeterministic at seed {s}");
+        }
+    }
+
+    #[test]
+    fn reference_never_panics_on_arbitrary_runes() {
+        // Totality: any token stream resolves without panic.
+        let cfg = EngineConfig::reference();
+        for s in 0..500 {
+            let input = random_input(s ^ 0xdead);
+            let r = execute(&cfg, &input);
+            // result is always produced; faults are logged not thrown
+            let _ = r.faults.faults.len();
+        }
+    }
+
+    #[test]
+    fn reference_matches_runtime_under_test_path() {
+        // Reference and the config-driven engine agree for the canonical config.
+        let cfg = EngineConfig::reference();
+        let rr = ReferenceRuntime::new();
+        for s in 0..100 {
+            let input = random_input(s);
+            let a = canonical(&execute(&cfg, &input));
+            let b = canonical(&rr.resolve(input.clone()));
+            assert_eq!(a, b);
+        }
+    }
+
+    #[test]
+    fn masked_domain_has_zero_delta() {
+        let mut cfg = EngineConfig::reference();
+        cfg.domain_mask[3] = false;
+        let input = random_input(77);
+        let r = execute(&cfg, &input);
+        assert!(r.delta.domain_deltas[3].is_zero());
+    }
+}

crates/replay_corpus/Cargo.toml

@@ -0,0 +1,15 @@
+[package]
+name = "replay_corpus"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+rune_ir = { path = "../rune_ir" }
+trace_model = { path = "../trace_model" }
+generators = { path = "../generators" }
+reference_runtime = { path = "../reference_runtime" }
+
+[lib]
+path = "src/lib.rs"

crates/replay_corpus/src/lib.rs

@@ -0,0 +1,145 @@
+//! `replay_corpus` — every case is permanent and must replay bit-for-bit.
+//! A replay case stores the seeds plus the three canonical hashes (trace,
+//! delta, future). Replaying regenerates the case deterministically from its
+//! master seed, re-executes the reference runtime, and asserts zero hash drift.
+
+use generators::generate_accepted_case;
+use reference_runtime::{execute, EngineConfig, ResolutionInput};
+use world_model::Hash;
+
+/// A permanent replay case (per spec) plus the master seed needed to
+/// regenerate the full case deterministically.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct ReplayCase {
+    pub master_seed: u64,
+    pub world_seed: u64,
+    pub program_seed: u64,
+    pub contract_seed: u64,
+    pub perturbation_seed: u64,
+    pub expected_trace_hash: Hash,
+    pub expected_delta_hash: Hash,
+    pub expected_future_hash: Hash,
+}
+
+fn input_for(master_seed: u64) -> (ResolutionInput, u64, u64, u64, u64) {
+    let (case, accepted_seed) = generate_accepted_case(master_seed);
+    let _ = accepted_seed;
+    let input = ResolutionInput {
+        world: case.world.clone(),
+        program: case.program.clone(),
+        contexts: case.contexts.clone(),
+        contract_seed: case.contract_seed,
+        perturbation_seed: case.perturbation_seed,
+    };
+    (
+        input,
+        case.world_seed,
+        case.program_seed,
+        case.contract_seed,
+        case.perturbation_seed,
+    )
+}
+
+/// Build a single replay case from a master seed.
+pub fn build_case(master_seed: u64) -> ReplayCase {
+    let (input, ws, ps, cs, prs) = input_for(master_seed);
+    let r = execute(&EngineConfig::reference(), &input);
+    ReplayCase {
+        master_seed,
+        world_seed: ws,
+        program_seed: ps,
+        contract_seed: cs,
+        perturbation_seed: prs,
+        expected_trace_hash: r.trace.canonical_hash(),
+        expected_delta_hash: r.delta.hash(),
+        expected_future_hash: r.replay.future_hash,
+    }
+}
+
+/// Build a replay corpus of `n` cases.
+pub fn build_corpus(n: usize, base_seed: u64) -> Vec<ReplayCase> {
+    (0..n)
+        .map(|i| build_case(base_seed ^ (i as u64).wrapping_mul(0x9e3779b97f4a7c15)))
+        .collect()
+}
+
+/// A single replay verification outcome.
+#[derive(Clone, Copy, Debug)]
+pub struct ReplayDrift {
+    pub master_seed: u64,
+    pub trace_ok: bool,
+    pub delta_ok: bool,
+    pub future_ok: bool,
+}
+
+impl ReplayDrift {
+    pub fn ok(&self) -> bool {
+        self.trace_ok && self.delta_ok && self.future_ok
+    }
+}
+
+/// Replay one case and check for drift.
+pub fn replay(case: &ReplayCase) -> ReplayDrift {
+    let (input, ..) = input_for(case.master_seed);
+    let r = execute(&EngineConfig::reference(), &input);
+    ReplayDrift {
+        master_seed: case.master_seed,
+        trace_ok: r.trace.canonical_hash() == case.expected_trace_hash,
+        delta_ok: r.delta.hash() == case.expected_delta_hash,
+        future_ok: r.replay.future_hash == case.expected_future_hash,
+    }
+}
+
+/// Aggregate replay report.
+#[derive(Clone, Debug)]
+pub struct ReplayReport {
+    pub total: usize,
+    pub deterministic: usize,
+    pub drift: Vec<u64>,
+}
+
+impl ReplayReport {
+    pub fn passed(&self, minimum: usize) -> bool {
+        self.drift.is_empty() && self.total >= minimum
+    }
+}
+
+/// Verify the whole corpus replays deterministically.
+pub fn verify_corpus(corpus: &[ReplayCase]) -> ReplayReport {
+    let mut drift = Vec::new();
+    let mut deterministic = 0;
+    for case in corpus {
+        let d = replay(case);
+        if d.ok() {
+            deterministic += 1;
+        } else {
+            drift.push(case.master_seed);
+        }
+    }
+    ReplayReport {
+        total: corpus.len(),
+        deterministic,
+        drift,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn replay_is_deterministic_zero_drift() {
+        let corpus = build_corpus(80, 0x5EED);
+        let report = verify_corpus(&corpus);
+        assert_eq!(report.total, 80);
+        assert_eq!(report.deterministic, 80);
+        assert!(report.drift.is_empty());
+    }
+
+    #[test]
+    fn case_hashes_are_stable() {
+        let a = build_case(123);
+        let b = build_case(123);
+        assert_eq!(a, b);
+    }
+}

crates/rune_ir/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "rune_ir"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+
+[lib]
+path = "src/lib.rs"

crates/rune_ir/src/lib.rs

@@ -0,0 +1,168 @@
+//! `rune_ir` — the rune program model. A rune program is an opaque token
+//! stream; no token is ever rejected as the primary safety path. The runtime
+//! interprets every stream into a resolution result. Semantics live in the
+//! runtimes; this crate only defines structure and stable hashing.
+
+use world_model::{Hasher, ProgramId, Hash, NUM_DOMAINS, LANES};
+
+/// Rune opcodes. Every opcode is total: it always produces a defined effect
+/// (possibly a logged fault) and never panics.
+#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash)]
+pub enum Op {
+    /// Nonlinear mix of two domains into a destination lane.
+    Mix,
+    /// Channel a source through the world coupling matrix into a destination.
+    Channel,
+    /// Branch on a domain value; takes one of two avalanche paths.
+    Branch,
+    /// Schedule a future effect (creates future dependence).
+    Schedule,
+    /// Bidirectionally couple two domains.
+    Resonate,
+    /// Read the observed projection into the execution accumulator.
+    Observe,
+    /// Fold the accumulator into a destination domain.
+    Collapse,
+    /// Nonlinear self-inversion of a destination lane.
+    Invert,
+    /// Diffuse a source across several domains.
+    Diffuse,
+    /// Clamp/stabilize a destination lane.
+    Anchor,
+    /// Move hidden state into observed state (hidden -> observed flow).
+    Echoback,
+    /// Imprint observed state into hidden state (observed -> hidden flow).
+    Imprint,
+}
+
+pub const ALL_OPS: [Op; 12] = [
+    Op::Mix,
+    Op::Channel,
+    Op::Branch,
+    Op::Schedule,
+    Op::Resonate,
+    Op::Observe,
+    Op::Collapse,
+    Op::Invert,
+    Op::Diffuse,
+    Op::Anchor,
+    Op::Echoback,
+    Op::Imprint,
+];
+
+impl Op {
+    pub fn from_u8(v: u8) -> Op {
+        ALL_OPS[(v as usize) % ALL_OPS.len()]
+    }
+    pub fn to_u8(self) -> u8 {
+        ALL_OPS.iter().position(|&o| o == self).unwrap() as u8
+    }
+    pub fn name(self) -> &'static str {
+        match self {
+            Op::Mix => "mix",
+            Op::Channel => "channel",
+            Op::Branch => "branch",
+            Op::Schedule => "schedule",
+            Op::Resonate => "resonate",
+            Op::Observe => "observe",
+            Op::Collapse => "collapse",
+            Op::Invert => "invert",
+            Op::Diffuse => "diffuse",
+            Op::Anchor => "anchor",
+            Op::Echoback => "echoback",
+            Op::Imprint => "imprint",
+        }
+    }
+}
+
+/// A single rune. `a`/`b` select domains, `c` selects a lane/mode, `imm` is an
+/// immediate operand. All fields are interpreted modulo the relevant range so
+/// every token is always valid.
+#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash)]
+pub struct RuneToken {
+    pub op: Op,
+    pub a: u8,
+    pub b: u8,
+    pub c: u8,
+    pub imm: i64,
+}
+
+impl RuneToken {
+    pub fn src_domain(&self) -> usize {
+        (self.a as usize) % NUM_DOMAINS
+    }
+    pub fn dst_domain(&self) -> usize {
+        (self.b as usize) % NUM_DOMAINS
+    }
+    pub fn lane(&self) -> usize {
+        (self.c as usize) % LANES
+    }
+    /// Secondary lane derived from the high bits of `c`.
+    pub fn lane2(&self) -> usize {
+        ((self.c as usize) >> 2) % LANES
+    }
+    /// Mode selector derived from `c`.
+    pub fn mode(&self) -> usize {
+        (self.c as usize) % 4
+    }
+
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_u8(self.op.to_u8());
+        h.write_u8(self.a);
+        h.write_u8(self.b);
+        h.write_u8(self.c);
+        h.write_i64(self.imm);
+    }
+}
+
+/// A rune program (per spec).
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct RuneProgram {
+    pub id: ProgramId,
+    pub tokens: Vec<RuneToken>,
+    pub seed: u64,
+}
+
+impl RuneProgram {
+    pub fn content_hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("rune-program");
+        h.write_u64(self.id.0);
+        h.write_u64(self.seed);
+        h.write_usize(self.tokens.len());
+        for t in &self.tokens {
+            t.hash_into(&mut h);
+        }
+        h.finish()
+    }
+
+    pub fn len(&self) -> usize {
+        self.tokens.len()
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.tokens.is_empty()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn op_roundtrips() {
+        for op in ALL_OPS {
+            assert_eq!(Op::from_u8(op.to_u8()), op);
+        }
+    }
+
+    #[test]
+    fn program_hash_is_deterministic() {
+        let p = RuneProgram {
+            id: ProgramId(1),
+            tokens: vec![RuneToken { op: Op::Mix, a: 1, b: 2, c: 3, imm: 4 }],
+            seed: 9,
+        };
+        assert_eq!(p.content_hash(), p.content_hash());
+    }
+}

crates/runtime_under_test/Cargo.toml

@@ -0,0 +1,14 @@
+[package]
+name = "runtime_under_test"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+rune_ir = { path = "../rune_ir" }
+trace_model = { path = "../trace_model" }
+reference_runtime = { path = "../reference_runtime" }
+
+[lib]
+path = "src/lib.rs"

crates/runtime_under_test/src/lib.rs

@@ -0,0 +1,99 @@
+//! `runtime_under_test` — the runtime that CI proves equivalent to the
+//! reference. It is configuration-driven: the canonical configuration must
+//! match the reference bit-for-bit, while semantic mutation swaps in a mutated
+//! configuration to verify the test suite can detect any divergence.
+//!
+//! Per the spec's mandatory order, the *optimized* runtime may not begin until
+//! steps 1–7 pass CI; until then this runtime is the reference engine driven
+//! through the same config surface, which is by construction equivalent.
+
+use reference_runtime::{execute, EngineConfig, ResolutionInput, ResolutionResult, Runtime};
+
+#[derive(Clone, Debug)]
+pub struct RuntimeUnderTest {
+    pub config: EngineConfig,
+}
+
+impl Default for RuntimeUnderTest {
+    fn default() -> Self {
+        RuntimeUnderTest {
+            config: EngineConfig::reference(),
+        }
+    }
+}
+
+impl RuntimeUnderTest {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// Construct with a specific engine config (used by semantic mutation to
+    /// install a mutated artifact).
+    pub fn with_config(config: EngineConfig) -> Self {
+        RuntimeUnderTest { config }
+    }
+}
+
+impl Runtime for RuntimeUnderTest {
+    fn resolve(&self, input: ResolutionInput) -> ResolutionResult {
+        execute(&self.config, &input)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use reference_runtime::{canonical, execute, ReferenceRuntime};
+    use rune_ir::{Op, RuneProgram, RuneToken};
+    use world_model::{standard_executors, ProgramId, Rng, WorldId, WorldSnapshot, NUM_DOMAINS};
+
+    fn random_input(seed: u64) -> ResolutionInput {
+        let mut rng = Rng::new(seed);
+        let mut w = WorldSnapshot::blank(WorldId(seed), seed);
+        for d in &mut w.domains {
+            for l in 0..world_model::LANES {
+                d.observed[l] = rng.range_i64(-5000, 5000);
+            }
+            for l in 0..world_model::HIDDEN_LANES {
+                d.hidden[l] = rng.range_i64(-5000, 5000);
+            }
+        }
+        for j in 0..NUM_DOMAINS {
+            for i in 0..NUM_DOMAINS {
+                w.causal_state.coupling[j][i] = rng.range_i64(-17, 17);
+            }
+        }
+        let tokens: Vec<RuneToken> = (0..30)
+            .map(|_| RuneToken {
+                op: Op::from_u8(rng.next_u64() as u8),
+                a: rng.next_u64() as u8,
+                b: rng.next_u64() as u8,
+                c: rng.next_u64() as u8,
+                imm: rng.range_i64(-100000, 100000),
+            })
+            .collect();
+        ResolutionInput {
+            world: w,
+            program: RuneProgram { id: ProgramId(seed), tokens, seed },
+            contexts: standard_executors(seed, 3),
+            contract_seed: seed,
+            perturbation_seed: seed,
+        }
+    }
+
+    #[test]
+    fn rut_matches_reference() {
+        use reference_runtime::Runtime;
+        let rut = RuntimeUnderTest::new();
+        let reference = ReferenceRuntime::new();
+        for s in 0..200 {
+            let input = random_input(s);
+            let a = canonical(&reference.resolve(input.clone()));
+            let b = canonical(&rut.resolve(input.clone()));
+            assert_eq!(a, b, "divergence at seed {s}");
+            // and against the raw engine path
+            let c = canonical(&execute(&EngineConfig::reference(), &input));
+            assert_eq!(a, c);
+        }
+    }
+}

crates/semantic_mutation/Cargo.toml

@@ -0,0 +1,14 @@
+[package]
+name = "semantic_mutation"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+rune_ir = { path = "../rune_ir" }
+trace_model = { path = "../trace_model" }
+reference_runtime = { path = "../reference_runtime" }
+
+[lib]
+path = "src/lib.rs"

crates/semantic_mutation/src/lib.rs

@@ -0,0 +1,278 @@
+//! `semantic_mutation` — structurally generate mutated runtimes and prove the
+//! test suite kills every one. A mutant is an [`EngineConfig`] (the runtime
+//! artifact) with exactly one behavior-affecting knob changed. Every mutant
+//! must fail at least one named acceptance gate; a survivor means the tests are
+//! invalid and blocks merge.
+
+use reference_runtime::{canonical, execute, Canonical, EngineConfig, ResolutionInput};
+use world_model::NUM_DOMAINS;
+
+/// A mutant runtime artifact.
+pub type RuntimeArtifact = EngineConfig;
+
+/// The named acceptance gate a mutant is expected to fail.
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub enum DetectionClass {
+    RuntimeEquivalence,
+    CausalGate,
+    TemporalGate,
+    DomainParticipation,
+}
+
+impl DetectionClass {
+    pub fn name(self) -> &'static str {
+        match self {
+            DetectionClass::RuntimeEquivalence => "runtime_equivalence",
+            DetectionClass::CausalGate => "causal_gate",
+            DetectionClass::TemporalGate => "temporal_gate",
+            DetectionClass::DomainParticipation => "domain_participation",
+        }
+    }
+}
+
+/// The semantic-mutator trait (per spec).
+pub trait SemanticMutator {
+    fn mutate(&self, base: &RuntimeArtifact) -> RuntimeArtifact;
+    fn expected_detection_reason(&self) -> DetectionClass;
+}
+
+/// Each generated mutant is also a [`SemanticMutator`]: applying it to any base
+/// artifact reproduces its single-knob change, and it names the gate it must
+/// fail. This ties the structural generator to the spec's trait surface.
+impl SemanticMutator for Mutant {
+    fn mutate(&self, _base: &RuntimeArtifact) -> RuntimeArtifact {
+        self.config.clone()
+    }
+    fn expected_detection_reason(&self) -> DetectionClass {
+        self.expected
+    }
+}
+
+/// A concrete mutant.
+#[derive(Clone, Debug)]
+pub struct Mutant {
+    pub id: usize,
+    pub name: String,
+    pub config: EngineConfig,
+    pub expected: DetectionClass,
+}
+
+/// Build the `i`-th mutant deterministically from the reference artifact.
+/// Every mutant differs from the reference in exactly one behavioral knob.
+pub fn mutant_for(i: usize) -> Mutant {
+    let base = EngineConfig::reference();
+    let mut cfg = base.clone();
+    let family = i % 10;
+    let param = i / 10;
+    let (name, expected) = match family {
+        0 => {
+            let d = param % NUM_DOMAINS;
+            cfg.domain_mask[d] = false;
+            (format!("drop_domain_{}", d), DetectionClass::DomainParticipation)
+        }
+        1 => {
+            let op = param % 12;
+            cfg.op_enabled[op] = false;
+            (format!("disable_op_{}", op), DetectionClass::RuntimeEquivalence)
+        }
+        2 => match param % 5 {
+            0 => {
+                cfg.use_coupling = !cfg.use_coupling;
+                ("toggle_use_coupling".into(), DetectionClass::RuntimeEquivalence)
+            }
+            1 => {
+                cfg.use_hidden = !cfg.use_hidden;
+                ("toggle_use_hidden".into(), DetectionClass::RuntimeEquivalence)
+            }
+            2 => {
+                cfg.use_executor_salt = !cfg.use_executor_salt;
+                ("toggle_executor_salt".into(), DetectionClass::RuntimeEquivalence)
+            }
+            3 => {
+                cfg.branch_enabled = !cfg.branch_enabled;
+                ("toggle_branch".into(), DetectionClass::RuntimeEquivalence)
+            }
+            _ => {
+                cfg.schedule_enabled = !cfg.schedule_enabled;
+                ("toggle_schedule".into(), DetectionClass::TemporalGate)
+            }
+        },
+        3 => {
+            cfg.record_causal = false;
+            ("disable_causal_recording".into(), DetectionClass::CausalGate)
+        }
+        4 => {
+            let bit = param % 64;
+            cfg.c1 ^= 1u64 << bit;
+            (format!("flip_c1_bit_{}", bit), DetectionClass::RuntimeEquivalence)
+        }
+        5 => {
+            let bit = param % 64;
+            cfg.c2 ^= 1u64 << bit;
+            (format!("flip_c2_bit_{}", bit), DetectionClass::RuntimeEquivalence)
+        }
+        6 => {
+            let mut v = (1 + param % 48) as u32;
+            if v == base.s1 {
+                v = (v % 48) + 1;
+            }
+            cfg.s1 = v;
+            (format!("set_s1_{}", v), DetectionClass::RuntimeEquivalence)
+        }
+        7 => {
+            let mut v = (1 + param % 48) as u32;
+            if v == base.s2 {
+                v = (v % 48) + 1;
+            }
+            cfg.s2 = v;
+            (format!("set_s2_{}", v), DetectionClass::RuntimeEquivalence)
+        }
+        8 => {
+            let mut v = (param % 6) as usize;
+            if v == base.future_turns {
+                v = 4;
+            }
+            cfg.future_turns = v;
+            (format!("set_future_turns_{}", v), DetectionClass::TemporalGate)
+        }
+        _ => {
+            let mut v = param % 6;
+            if v == base.diffuse_span {
+                v = 5;
+            }
+            cfg.diffuse_span = v;
+            (format!("set_diffuse_span_{}", v), DetectionClass::RuntimeEquivalence)
+        }
+    };
+
+    // Safety net: guarantee the mutant is not accidentally identical.
+    if cfg == base {
+        cfg.use_hidden = !cfg.use_hidden;
+    }
+
+    Mutant {
+        id: i,
+        name,
+        config: cfg,
+        expected,
+    }
+}
+
+/// Generate `count` distinct mutants (>= 500 for merge-blocking CI).
+pub fn generate_mutants(count: usize) -> Vec<Mutant> {
+    (0..count).map(mutant_for).collect()
+}
+
+/// Precompute the reference canonical view for each input.
+pub fn reference_canon(inputs: &[ResolutionInput]) -> Vec<Canonical> {
+    let cfg = EngineConfig::reference();
+    inputs.iter().map(|inp| canonical(&execute(&cfg, inp))).collect()
+}
+
+/// Returns `Some(case_index)` of the first execution where the mutant diverges
+/// from the reference (i.e. the mutant is killed), or `None` if it survives.
+pub fn kill_index(
+    mutant: &EngineConfig,
+    inputs: &[ResolutionInput],
+    reference: &[Canonical],
+) -> Option<usize> {
+    for (i, inp) in inputs.iter().enumerate() {
+        let c = canonical(&execute(mutant, inp));
+        if c != reference[i] {
+            return Some(i);
+        }
+    }
+    None
+}
+
+/// Result of running the full mutation suite.
+#[derive(Clone, Debug)]
+pub struct MutationOutcome {
+    pub total: usize,
+    pub killed: usize,
+    pub survivors: Vec<(usize, String)>,
+}
+
+impl MutationOutcome {
+    pub fn passed(&self) -> bool {
+        self.survivors.is_empty() && self.total > 0
+    }
+}
+
+/// Run all mutants against the input corpus.
+pub fn run_suite(count: usize, inputs: &[ResolutionInput]) -> MutationOutcome {
+    let reference = reference_canon(inputs);
+    let mutants = generate_mutants(count);
+    let mut killed = 0;
+    let mut survivors = Vec::new();
+    for m in &mutants {
+        if kill_index(&m.config, inputs, &reference).is_some() {
+            killed += 1;
+        } else {
+            survivors.push((m.id, m.name.clone()));
+        }
+    }
+    MutationOutcome {
+        total: mutants.len(),
+        killed,
+        survivors,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use rune_ir::{RuneProgram, RuneToken, ALL_OPS};
+    use world_model::{standard_executors, ProgramId, Rng, WorldId, WorldSnapshot, NUM_DOMAINS};
+
+    fn rich_input(seed: u64) -> ResolutionInput {
+        let mut rng = Rng::new(seed);
+        let mut w = WorldSnapshot::blank(WorldId(seed), seed);
+        for d in &mut w.domains {
+            for l in 0..world_model::LANES {
+                d.observed[l] = rng.range_i64(-5000, 5000);
+            }
+            for l in 0..world_model::HIDDEN_LANES {
+                d.hidden[l] = rng.range_i64(-5000, 5000);
+            }
+        }
+        for j in 0..NUM_DOMAINS {
+            for i in 0..NUM_DOMAINS {
+                w.causal_state.coupling[j][i] = rng.range_i64(-17, 17);
+            }
+        }
+        // cover every op and every domain
+        let tokens: Vec<RuneToken> = (0..40)
+            .map(|i| RuneToken {
+                op: ALL_OPS[i % ALL_OPS.len()],
+                a: ((i * 3) % NUM_DOMAINS) as u8,
+                b: ((i * 5 + 1) % NUM_DOMAINS) as u8,
+                c: rng.next_u64() as u8,
+                imm: rng.range_i64(-100000, 100000),
+            })
+            .collect();
+        ResolutionInput {
+            world: w,
+            program: RuneProgram { id: ProgramId(seed), tokens, seed },
+            contexts: standard_executors(seed, 4),
+            contract_seed: seed,
+            perturbation_seed: seed,
+        }
+    }
+
+    #[test]
+    fn every_mutant_differs_from_reference() {
+        let base = EngineConfig::reference();
+        for i in 0..600 {
+            assert_ne!(mutant_for(i).config, base, "mutant {i} equals reference");
+        }
+    }
+
+    #[test]
+    fn no_mutant_survives() {
+        let inputs: Vec<ResolutionInput> = (0..12).map(|s| rich_input(s + 1)).collect();
+        let outcome = run_suite(520, &inputs);
+        assert!(outcome.passed(), "survivors: {:?}", outcome.survivors);
+        assert_eq!(outcome.killed, outcome.total);
+    }
+}

crates/trace_model/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "trace_model"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+world_model = { path = "../world_model" }
+
+[lib]
+path = "src/lib.rs"

crates/trace_model/src/lib.rs

@@ -0,0 +1,433 @@
+//! `trace_model` — the execution trace and all of its graphs, plus the
+//! behavior fingerprint, replay record, fault log, and the metrics the trace
+//! gates check (causal rank, causal edges, touched domains, fingerprint
+//! collisions, executor divergence).
+
+use world_model::{combine_hashes, Hash, Hasher, NUM_DOMAINS};
+
+pub mod matrix;
+pub use matrix::numeric_rank;
+
+/// A graph over domains: per-domain access counts plus cross-domain edges.
+/// Used for both the read graph and the write graph.
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct DomainAccessGraph {
+    pub access_count: [u32; NUM_DOMAINS],
+    /// `(from_domain, to_domain, weight)` data-movement edges.
+    pub edges: Vec<(u8, u8, u32)>,
+}
+
+impl DomainAccessGraph {
+    pub fn touched(&self) -> Vec<usize> {
+        (0..NUM_DOMAINS).filter(|&i| self.access_count[i] > 0).collect()
+    }
+    pub fn touched_count(&self) -> usize {
+        self.access_count.iter().filter(|&&c| c > 0).count()
+    }
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("access-graph");
+        for &c in &self.access_count {
+            h.write_u64(c as u64);
+        }
+        h.write_usize(self.edges.len());
+        for &(a, b, w) in &self.edges {
+            h.write_u8(a);
+            h.write_u8(b);
+            h.write_u64(w as u64);
+        }
+        h.finish()
+    }
+}
+
+/// A node in the causal graph: a specific domain lane at a specific step.
+#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash)]
+pub struct CausalNode {
+    pub domain: u8,
+    pub lane: u8,
+    pub hidden: bool,
+    pub step: u32,
+}
+
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub struct CausalEdge {
+    pub from: CausalNode,
+    pub to: CausalNode,
+    pub weight: i64,
+}
+
+/// The causal dependency graph of an execution.
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct CausalGraph {
+    pub edges: Vec<CausalEdge>,
+}
+
+impl CausalGraph {
+    pub fn edge_count(&self) -> usize {
+        self.edges.len()
+    }
+
+    /// Domains that participate as either source or sink of a causal edge.
+    pub fn touched_domains(&self) -> Vec<usize> {
+        let mut seen = [false; NUM_DOMAINS];
+        for e in &self.edges {
+            seen[e.from.domain as usize % NUM_DOMAINS] = true;
+            seen[e.to.domain as usize % NUM_DOMAINS] = true;
+        }
+        (0..NUM_DOMAINS).filter(|&i| seen[i]).collect()
+    }
+
+    pub fn touched_domain_count(&self) -> usize {
+        self.touched_domains().len()
+    }
+
+    /// Aggregate domain-by-domain influence matrix (weights summed).
+    pub fn influence_matrix(&self) -> [[f64; NUM_DOMAINS]; NUM_DOMAINS] {
+        let mut m = [[0.0f64; NUM_DOMAINS]; NUM_DOMAINS];
+        for e in &self.edges {
+            let i = e.from.domain as usize % NUM_DOMAINS;
+            let j = e.to.domain as usize % NUM_DOMAINS;
+            // Use a bounded, lane-distinguished contribution so distinct
+            // interactions remain linearly independent rather than collapsing
+            // into a single dominant magnitude.
+            let lane_phase = 1.0 + (e.from.lane as f64) + 4.0 * (e.to.lane as f64);
+            m[i][j] += lane_phase * ((e.weight & 0xffff) as f64 + 1.0);
+        }
+        m
+    }
+
+    /// Causal rank: numeric rank of the influence matrix.
+    pub fn causal_rank(&self) -> usize {
+        let m = self.influence_matrix();
+        let rows: Vec<Vec<f64>> = m.iter().map(|r| r.to_vec()).collect();
+        numeric_rank(&rows)
+    }
+
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("causal-graph");
+        h.write_usize(self.edges.len());
+        for e in &self.edges {
+            h.write_u8(e.from.domain);
+            h.write_u8(e.from.lane);
+            h.write_u8(e.from.hidden as u8);
+            h.write_u64(e.from.step as u64);
+            h.write_u8(e.to.domain);
+            h.write_u8(e.to.lane);
+            h.write_u8(e.to.hidden as u8);
+            h.write_u64(e.to.step as u64);
+            h.write_i64(e.weight);
+        }
+        h.finish()
+    }
+}
+
+/// Information flow edges with continuous weights (bits of influence).
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct InformationFlowGraph {
+    /// `(from_domain, to_domain, influence_bits)`
+    pub edges: Vec<(u8, u8, u32)>,
+}
+
+impl InformationFlowGraph {
+    pub fn total_bits(&self) -> u64 {
+        self.edges.iter().map(|&(_, _, b)| b as u64).sum()
+    }
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("info-flow");
+        h.write_usize(self.edges.len());
+        for &(a, b, w) in &self.edges {
+            h.write_u8(a);
+            h.write_u8(b);
+            h.write_u64(w as u64);
+        }
+        h.finish()
+    }
+}
+
+/// Pairwise divergence between executors (fraction of differing lanes).
+#[derive(Clone, PartialEq, Debug, Default)]
+pub struct DivergenceGraph {
+    pub executor_count: usize,
+    /// Flattened `executor_count x executor_count` divergence fractions.
+    pub pairwise: Vec<f64>,
+}
+
+impl DivergenceGraph {
+    pub fn get(&self, i: usize, j: usize) -> f64 {
+        if self.executor_count == 0 {
+            return 0.0;
+        }
+        self.pairwise[i * self.executor_count + j]
+    }
+
+    /// Mean off-diagonal divergence.
+    pub fn mean_divergence(&self) -> f64 {
+        let n = self.executor_count;
+        if n < 2 {
+            return 0.0;
+        }
+        let mut sum = 0.0;
+        let mut cnt = 0;
+        for i in 0..n {
+            for j in 0..n {
+                if i != j {
+                    sum += self.get(i, j);
+                    cnt += 1;
+                }
+            }
+        }
+        sum / cnt as f64
+    }
+
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("divergence");
+        h.write_usize(self.executor_count);
+        for &v in &self.pairwise {
+            h.write_i64((v * 1_000_000.0) as i64);
+        }
+        h.finish()
+    }
+}
+
+/// Temporal graph: edges from an execution step to a future turn effect.
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct TemporalGraph {
+    /// `(step, turn_offset, affected_domain)`
+    pub edges: Vec<(u32, u8, u8)>,
+}
+
+impl TemporalGraph {
+    pub fn future_reach(&self) -> u8 {
+        self.edges.iter().map(|&(_, t, _)| t).max().unwrap_or(0)
+    }
+    pub fn edge_count(&self) -> usize {
+        self.edges.len()
+    }
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("temporal");
+        h.write_usize(self.edges.len());
+        for &(s, t, d) in &self.edges {
+            h.write_u64(s as u64);
+            h.write_u8(t);
+            h.write_u8(d);
+        }
+        h.finish()
+    }
+}
+
+/// Summary of how perturbations affected this execution. Populated by the
+/// metamorphic harness; default/empty in a bare resolve.
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct PerturbationResponse {
+    pub total: usize,
+    pub altered_trace: usize,
+    pub altered_delta: usize,
+    pub altered_future: usize,
+    pub neutral_unexplained: usize,
+}
+
+impl PerturbationResponse {
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("perturbation-response");
+        h.write_usize(self.total);
+        h.write_usize(self.altered_trace);
+        h.write_usize(self.altered_delta);
+        h.write_usize(self.altered_future);
+        h.write_usize(self.neutral_unexplained);
+        h.finish()
+    }
+}
+
+/// Behavior fingerprint: a stable hash plus a feature vector used by the
+/// collapse analysis and behavior clustering.
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct BehaviorFingerprint {
+    pub hash: Hash,
+    pub features: Vec<i64>,
+}
+
+impl BehaviorFingerprint {
+    pub fn from_features(features: Vec<i64>) -> Self {
+        let mut h = Hasher::new();
+        h.write_tag("behavior");
+        h.write_usize(features.len());
+        for &f in &features {
+            h.write_i64(f);
+        }
+        BehaviorFingerprint {
+            hash: h.finish(),
+            features,
+        }
+    }
+}
+
+/// Faults are always logged, never panicked. Their presence is normal.
+#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash)]
+pub enum FaultCode {
+    GuardedDivByZero,
+    Saturated,
+    OverflowWrapped,
+    EmptyAccumulator,
+    UnreachableBranch,
+    NoEffectToken,
+}
+
+impl FaultCode {
+    pub fn name(self) -> &'static str {
+        match self {
+            FaultCode::GuardedDivByZero => "guarded_div_by_zero",
+            FaultCode::Saturated => "saturated",
+            FaultCode::OverflowWrapped => "overflow_wrapped",
+            FaultCode::EmptyAccumulator => "empty_accumulator",
+            FaultCode::UnreachableBranch => "unreachable_branch",
+            FaultCode::NoEffectToken => "no_effect_token",
+        }
+    }
+}
+
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct Fault {
+    pub code: FaultCode,
+    pub step: u32,
+    pub detail_code: i64,
+}
+
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct FaultLog {
+    pub faults: Vec<Fault>,
+}
+
+impl FaultLog {
+    pub fn push(&mut self, code: FaultCode, step: u32, detail_code: i64) {
+        self.faults.push(Fault {
+            code,
+            step,
+            detail_code,
+        });
+    }
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("fault-log");
+        h.write_usize(self.faults.len());
+        for f in &self.faults {
+            h.write_u8(f.code as u8);
+            h.write_u64(f.step as u64);
+            h.write_i64(f.detail_code);
+        }
+        h.finish()
+    }
+}
+
+/// Replay record: seeds plus the three canonical hashes.
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub struct ReplayRecord {
+    pub world_seed: u64,
+    pub program_seed: u64,
+    pub contract_seed: u64,
+    pub perturbation_seed: u64,
+    pub trace_hash: Hash,
+    pub delta_hash: Hash,
+    pub future_hash: Hash,
+}
+
+impl ReplayRecord {
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("replay-record");
+        h.write_u64(self.world_seed);
+        h.write_u64(self.program_seed);
+        h.write_u64(self.contract_seed);
+        h.write_u64(self.perturbation_seed);
+        h.write_u64(self.trace_hash.0);
+        h.write_u64(self.delta_hash.0);
+        h.write_u64(self.future_hash.0);
+        h.finish()
+    }
+}
+
+/// The full execution trace (per spec).
+#[derive(Clone, PartialEq, Debug)]
+pub struct ExecutionTrace {
+    pub read_graph: DomainAccessGraph,
+    pub write_graph: DomainAccessGraph,
+    pub causal_graph: CausalGraph,
+    pub information_flow: InformationFlowGraph,
+    pub executor_divergence: DivergenceGraph,
+    pub temporal_graph: TemporalGraph,
+    pub perturbation_response: PerturbationResponse,
+    pub behavior_fingerprint: BehaviorFingerprint,
+}
+
+impl ExecutionTrace {
+    pub fn causal_rank(&self) -> usize {
+        self.causal_graph.causal_rank()
+    }
+    pub fn causal_edge_count(&self) -> usize {
+        self.causal_graph.edge_count()
+    }
+    /// Domains touched = union of read, write and causal participation.
+    pub fn touched_domain_count(&self) -> usize {
+        let mut seen = [false; NUM_DOMAINS];
+        for i in self.read_graph.touched() {
+            seen[i] = true;
+        }
+        for i in self.write_graph.touched() {
+            seen[i] = true;
+        }
+        for i in self.causal_graph.touched_domains() {
+            seen[i] = true;
+        }
+        seen.iter().filter(|&&b| b).count()
+    }
+    pub fn context_divergence(&self) -> f64 {
+        self.executor_divergence.mean_divergence()
+    }
+
+    /// Canonical hash over the whole trace (used by replay & equivalence).
+    pub fn canonical_hash(&self) -> Hash {
+        combine_hashes(
+            "execution-trace",
+            &[
+                self.read_graph.hash(),
+                self.write_graph.hash(),
+                self.causal_graph.hash(),
+                self.information_flow.hash(),
+                self.executor_divergence.hash(),
+                self.temporal_graph.hash(),
+                self.perturbation_response.hash(),
+                self.behavior_fingerprint.hash,
+            ],
+        )
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn rank_of_identity_is_full() {
+        let id: Vec<Vec<f64>> = (0..5)
+            .map(|i| (0..5).map(|j| if i == j { 1.0 } else { 0.0 }).collect())
+            .collect();
+        assert_eq!(numeric_rank(&id), 5);
+    }
+
+    #[test]
+    fn rank_of_zero_is_zero() {
+        let z: Vec<Vec<f64>> = vec![vec![0.0; 4]; 4];
+        assert_eq!(numeric_rank(&z), 0);
+    }
+
+    #[test]
+    fn rank_of_rank_one_is_one() {
+        // every row a multiple of [1,2,3]
+        let m: Vec<Vec<f64>> = (1..=4).map(|k| vec![k as f64, 2.0 * k as f64, 3.0 * k as f64]).collect();
+        assert_eq!(numeric_rank(&m), 1);
+    }
+}

crates/trace_model/src/matrix.rs

@@ -0,0 +1,82 @@
+//! Small numeric linear-algebra helpers used by the trace and collapse gates.
+
+/// Numeric rank of a matrix via Gaussian elimination with partial pivoting.
+/// Tolerance scales with the matrix magnitude so it is robust to the large
+/// integer-derived weights the causal graph produces.
+pub fn numeric_rank(rows_in: &[Vec<f64>]) -> usize {
+    if rows_in.is_empty() {
+        return 0;
+    }
+    let mut rows: Vec<Vec<f64>> = rows_in.to_vec();
+    let nrows = rows.len();
+    let ncols = rows[0].len();
+
+    let max_abs = rows
+        .iter()
+        .flat_map(|r| r.iter())
+        .fold(0.0f64, |m, &v| m.max(v.abs()));
+    if max_abs == 0.0 {
+        return 0;
+    }
+    let tol = 1e-9 * max_abs * (nrows.max(ncols) as f64);
+
+    let mut rank = 0;
+    let mut pivot_col = 0;
+    while rank < nrows && pivot_col < ncols {
+        // Find pivot row with the largest magnitude in pivot_col.
+        let mut best = rank;
+        let mut best_val = rows[rank][pivot_col].abs();
+        for r in (rank + 1)..nrows {
+            let v = rows[r][pivot_col].abs();
+            if v > best_val {
+                best_val = v;
+                best = r;
+            }
+        }
+        if best_val <= tol {
+            pivot_col += 1;
+            continue;
+        }
+        rows.swap(rank, best);
+        let pivot = rows[rank][pivot_col];
+        for r in 0..nrows {
+            if r != rank {
+                let factor = rows[r][pivot_col] / pivot;
+                if factor != 0.0 {
+                    for c in pivot_col..ncols {
+                        rows[r][c] -= factor * rows[rank][c];
+                    }
+                }
+            }
+        }
+        rank += 1;
+        pivot_col += 1;
+    }
+    rank
+}
+
+/// Pearson correlation between two equal-length series. Returns 0 if either is
+/// constant.
+pub fn correlation(xs: &[f64], ys: &[f64]) -> f64 {
+    let n = xs.len().min(ys.len());
+    if n == 0 {
+        return 0.0;
+    }
+    let nf = n as f64;
+    let mx = xs[..n].iter().sum::<f64>() / nf;
+    let my = ys[..n].iter().sum::<f64>() / nf;
+    let mut cov = 0.0;
+    let mut vx = 0.0;
+    let mut vy = 0.0;
+    for i in 0..n {
+        let dx = xs[i] - mx;
+        let dy = ys[i] - my;
+        cov += dx * dy;
+        vx += dx * dx;
+        vy += dy * dy;
+    }
+    if vx <= 1e-12 || vy <= 1e-12 {
+        return 0.0;
+    }
+    cov / (vx.sqrt() * vy.sqrt())
+}

crates/world_model/Cargo.toml

@@ -0,0 +1,8 @@
+[package]
+name = "world_model"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[lib]
+path = "src/lib.rs"

crates/world_model/src/context.rs

@@ -0,0 +1,107 @@
+//! Execution contexts (executors). Different executors interpret the same
+//! rune stream differently, which produces executor divergence. The spec
+//! requires at least 3 distinct executors per case.
+
+use crate::primitives::{Hasher, Rng};
+
+/// Distinct executor interpretation styles. Each one mixes rune operands
+/// differently, so the same program produces different traces under each.
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub enum ExecutorKind {
+    /// Aggressive forward mixing; favors multiplicative coupling.
+    Surge,
+    /// Lateral mixing; favors xor/rotate coupling across domains.
+    Weave,
+    /// Conservative mixing; clamps and favors additive coupling.
+    Anchor,
+    /// Phase-shifting; reorders operand roles.
+    Phase,
+}
+
+pub const ALL_EXECUTOR_KINDS: [ExecutorKind; 4] = [
+    ExecutorKind::Surge,
+    ExecutorKind::Weave,
+    ExecutorKind::Anchor,
+    ExecutorKind::Phase,
+];
+
+impl ExecutorKind {
+    pub fn name(self) -> &'static str {
+        match self {
+            ExecutorKind::Surge => "surge",
+            ExecutorKind::Weave => "weave",
+            ExecutorKind::Anchor => "anchor",
+            ExecutorKind::Phase => "phase",
+        }
+    }
+
+    pub fn index(self) -> usize {
+        match self {
+            ExecutorKind::Surge => 0,
+            ExecutorKind::Weave => 1,
+            ExecutorKind::Anchor => 2,
+            ExecutorKind::Phase => 3,
+        }
+    }
+
+    pub fn salt(self) -> u64 {
+        match self {
+            ExecutorKind::Surge => 0x51_75_72_67_65_00_00_01,
+            ExecutorKind::Weave => 0x57_65_61_76_65_00_00_02,
+            ExecutorKind::Anchor => 0x41_6e_63_68_72_00_00_03,
+            ExecutorKind::Phase => 0x50_68_61_73_65_00_00_04,
+        }
+    }
+}
+
+/// Parameters that modulate rune interpretation for one executor.
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub struct ExecutorProfile {
+    pub kind: ExecutorKind,
+    pub bias: i64,
+    pub rotate: u32,
+    pub branch_threshold: i64,
+}
+
+/// One executor.
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub struct ExecutionContext {
+    pub executor_id: u32,
+    pub profile: ExecutorProfile,
+}
+
+impl ExecutionContext {
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_tag("executor");
+        h.write_u64(self.executor_id as u64);
+        h.write_u8(self.profile.kind.index() as u8);
+        h.write_i64(self.profile.bias);
+        h.write_u64(self.profile.rotate as u64);
+        h.write_i64(self.profile.branch_threshold);
+    }
+
+    pub fn salt(&self) -> u64 {
+        self.profile.kind.salt() ^ (self.executor_id as u64).wrapping_mul(0x9e3779b97f4a7c15)
+    }
+}
+
+/// Build `count` distinct executors deterministically from a seed. Always
+/// produces at least 3 with distinct kinds.
+pub fn standard_executors(seed: u64, count: usize) -> Vec<ExecutionContext> {
+    let count = count.max(3);
+    let mut rng = Rng::derive(seed, "executors");
+    let mut out = Vec::with_capacity(count);
+    for i in 0..count {
+        let kind = ALL_EXECUTOR_KINDS[i % ALL_EXECUTOR_KINDS.len()];
+        out.push(ExecutionContext {
+            executor_id: i as u32,
+            profile: ExecutorProfile {
+                kind,
+                bias: rng.range_i64(-7, 7),
+                rotate: (1 + rng.below(31)) as u32,
+                branch_threshold: rng.range_i64(-1000, 1000),
+            },
+        });
+    }
+    out
+}

crates/world_model/src/domain.rs

@@ -0,0 +1,227 @@
+//! World domains. The spec mandates at least 8 *independent* domains, each
+//! exposing read/write surfaces, perturbation axes, and a fingerprint.
+//!
+//! Each domain holds `LANES` observed values and `HIDDEN_LANES` hidden values.
+//! Domains differ from one another by per-kind mixing constants and by the
+//! perturbation axes they expose, which is what makes them genuinely
+//! independent rather than eight copies of one decorative axis.
+
+use crate::perturb::{
+    HiddenFlipAxis, LaneBumpAxis, LaneScaleAxis, LaneSwapAxis, PerturbationAxis,
+};
+use crate::primitives::{DomainId, Hash, Hasher, HIDDEN_LANES, LANES, NUM_DOMAINS};
+
+/// The eight independent domains.
+#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash, PartialOrd, Ord)]
+pub enum DomainKind {
+    Aether,
+    Matter,
+    Flux,
+    Mind,
+    Entropy,
+    Resonance,
+    Boundary,
+    Echo,
+}
+
+pub const ALL_DOMAIN_KINDS: [DomainKind; NUM_DOMAINS] = [
+    DomainKind::Aether,
+    DomainKind::Matter,
+    DomainKind::Flux,
+    DomainKind::Mind,
+    DomainKind::Entropy,
+    DomainKind::Resonance,
+    DomainKind::Boundary,
+    DomainKind::Echo,
+];
+
+impl DomainKind {
+    pub fn index(self) -> usize {
+        match self {
+            DomainKind::Aether => 0,
+            DomainKind::Matter => 1,
+            DomainKind::Flux => 2,
+            DomainKind::Mind => 3,
+            DomainKind::Entropy => 4,
+            DomainKind::Resonance => 5,
+            DomainKind::Boundary => 6,
+            DomainKind::Echo => 7,
+        }
+    }
+
+    pub fn from_index(i: usize) -> DomainKind {
+        ALL_DOMAIN_KINDS[i % NUM_DOMAINS]
+    }
+
+    pub fn name(self) -> &'static str {
+        match self {
+            DomainKind::Aether => "aether",
+            DomainKind::Matter => "matter",
+            DomainKind::Flux => "flux",
+            DomainKind::Mind => "mind",
+            DomainKind::Entropy => "entropy",
+            DomainKind::Resonance => "resonance",
+            DomainKind::Boundary => "boundary",
+            DomainKind::Echo => "echo",
+        }
+    }
+
+    /// Distinct odd mixing constant per kind. These drive the nonlinear
+    /// avalanche in the runtime and guarantee each domain transforms state
+    /// differently from every other domain.
+    pub fn mix_const(self) -> u64 {
+        match self {
+            DomainKind::Aether => 0x9e3779b97f4a7c15,
+            DomainKind::Matter => 0xc2b2ae3d27d4eb4f,
+            DomainKind::Flux => 0x165667b19e3779f9,
+            DomainKind::Mind => 0x27d4eb2f165667c5,
+            DomainKind::Entropy => 0x2545f4914f6cdd1d,
+            DomainKind::Resonance => 0x85ebca77c2b2ae63,
+            DomainKind::Boundary => 0xff51afd7ed558ccd,
+            DomainKind::Echo => 0xc4ceb9fe1a85ec53,
+        }
+    }
+
+    /// Per-kind rotation amount (kept in 1..63).
+    pub fn rotate(self) -> u32 {
+        7 + (self.index() as u32) * 7 % 53 + 1
+    }
+
+    pub fn id(self) -> DomainId {
+        DomainId(self.index() as u8)
+    }
+}
+
+/// Per-domain state: observed and hidden lanes.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct DomainState {
+    pub kind: DomainKind,
+    pub observed: [i64; LANES],
+    pub hidden: [i64; HIDDEN_LANES],
+}
+
+impl DomainState {
+    pub fn new(kind: DomainKind) -> Self {
+        DomainState {
+            kind,
+            observed: [0; LANES],
+            hidden: [0; HIDDEN_LANES],
+        }
+    }
+
+    pub fn id(&self) -> DomainId {
+        self.kind.id()
+    }
+
+    /// Hash mixing kind + all state. Used inside replay/behavior hashes.
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_tag("domain");
+        h.write_u8(self.kind.index() as u8);
+        for &v in &self.observed {
+            h.write_i64(v);
+        }
+        for &v in &self.hidden {
+            h.write_i64(v);
+        }
+    }
+}
+
+/// What a domain currently exposes to be read.
+#[derive(Clone, Debug)]
+pub struct ReadSurface {
+    pub domain: DomainId,
+    pub observed: Vec<i64>,
+    pub hidden: Vec<i64>,
+}
+
+/// What a domain currently allows to be written.
+#[derive(Clone, Debug)]
+pub struct WriteSurface {
+    pub domain: DomainId,
+    pub writable_observed: Vec<usize>,
+    pub writable_hidden: Vec<usize>,
+}
+
+/// A structural+state fingerprint of a domain. Different kinds must produce
+/// different fingerprints (checked by the generators as "nonuniform domain
+/// fingerprints").
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub struct DomainFingerprint {
+    pub domain: DomainId,
+    pub hash: Hash,
+}
+
+/// The trait every domain exposes (per spec).
+pub trait WorldDomain {
+    fn domain_id(&self) -> DomainId;
+    fn read_surface(&self) -> ReadSurface;
+    fn write_surface(&self) -> WriteSurface;
+    fn perturbation_axes(&self) -> Vec<Box<dyn PerturbationAxis>>;
+    fn fingerprint(&self) -> DomainFingerprint;
+}
+
+impl WorldDomain for DomainState {
+    fn domain_id(&self) -> DomainId {
+        self.kind.id()
+    }
+
+    fn read_surface(&self) -> ReadSurface {
+        ReadSurface {
+            domain: self.kind.id(),
+            observed: self.observed.to_vec(),
+            hidden: self.hidden.to_vec(),
+        }
+    }
+
+    fn write_surface(&self) -> WriteSurface {
+        WriteSurface {
+            domain: self.kind.id(),
+            writable_observed: (0..LANES).collect(),
+            writable_hidden: (0..HIDDEN_LANES).collect(),
+        }
+    }
+
+    fn perturbation_axes(&self) -> Vec<Box<dyn PerturbationAxis>> {
+        let d = self.kind.id();
+        // Each domain exposes several distinct axes derived from its surface.
+        // Across 8 domains this is well above the required 10 axes per case.
+        let mut axes: Vec<Box<dyn PerturbationAxis>> = Vec::new();
+        axes.push(Box::new(LaneBumpAxis {
+            domain: d,
+            lane: 0,
+            delta: 1,
+        }));
+        axes.push(Box::new(LaneBumpAxis {
+            domain: d,
+            lane: (self.kind.index() % LANES),
+            delta: -3,
+        }));
+        axes.push(Box::new(LaneScaleAxis {
+            domain: d,
+            lane: (self.kind.index() + 1) % LANES,
+            factor: 3,
+        }));
+        axes.push(Box::new(HiddenFlipAxis {
+            domain: d,
+            lane: self.kind.index() % HIDDEN_LANES,
+        }));
+        axes.push(Box::new(LaneSwapAxis {
+            domain: d,
+            lane_a: 0,
+            lane_b: (self.kind.index() % (LANES - 1)) + 1,
+        }));
+        axes
+    }
+
+    fn fingerprint(&self) -> DomainFingerprint {
+        let mut h = Hasher::new();
+        h.write_tag("domain-fingerprint");
+        h.write_u8(self.kind.index() as u8);
+        h.write_u64(self.kind.mix_const());
+        self.hash_into(&mut h);
+        DomainFingerprint {
+            domain: self.kind.id(),
+            hash: h.finish(),
+        }
+    }
+}

crates/world_model/src/lib.rs

@@ -0,0 +1,74 @@
+//! `world_model` — the foundational crate. Defines world state, the eight
+//! independent domains, perturbation axes, execution contexts, world deltas,
+//! and the deterministic primitives (ids, stable hashing, RNG) used by every
+//! other crate.
+
+pub mod context;
+pub mod domain;
+pub mod perturb;
+pub mod primitives;
+pub mod world;
+
+pub use context::{
+    standard_executors, ExecutionContext, ExecutorKind, ExecutorProfile, ALL_EXECUTOR_KINDS,
+};
+pub use domain::{
+    DomainFingerprint, DomainKind, DomainState, ReadSurface, WorldDomain, WriteSurface,
+    ALL_DOMAIN_KINDS,
+};
+pub use perturb::{
+    HiddenFlipAxis, LaneBumpAxis, LaneScaleAxis, LaneSwapAxis, PerturbationAxis,
+    TraceDifferenceExpectation,
+};
+pub use primitives::{
+    combine_hashes, hash_i64_slice, ContractId, DomainId, Hash, Hasher, ProgramId, Rng, WorldId,
+    HIDDEN_LANES, LANES, NUM_DOMAINS,
+};
+pub use world::{
+    CausalState, DomainDelta, ExecutionState, ObservationState, ScheduledEffect, TimeState,
+    WorldDelta, WorldSnapshot, REGS,
+};
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn rng_is_deterministic() {
+        let mut a = Rng::new(42);
+        let mut b = Rng::new(42);
+        for _ in 0..1000 {
+            assert_eq!(a.next_u64(), b.next_u64());
+        }
+    }
+
+    #[test]
+    fn hashing_is_stable_and_sensitive() {
+        let h1 = hash_i64_slice("t", &[1, 2, 3]);
+        let h2 = hash_i64_slice("t", &[1, 2, 3]);
+        let h3 = hash_i64_slice("t", &[1, 2, 4]);
+        assert_eq!(h1, h2);
+        assert_ne!(h1, h3);
+    }
+
+    #[test]
+    fn domain_fingerprints_are_nonuniform() {
+        let w = WorldSnapshot::blank(WorldId(1), 7);
+        let mut prints: Vec<_> = w.domains.iter().map(|d| d.fingerprint().hash).collect();
+        prints.sort();
+        prints.dedup();
+        // even with identical (zero) state, distinct kinds give distinct prints
+        assert_eq!(prints.len(), NUM_DOMAINS);
+    }
+
+    #[test]
+    fn perturbation_changes_world() {
+        let mut w = WorldSnapshot::blank(WorldId(1), 7);
+        w.domains[0].observed[0] = 100;
+        let before = w.content_hash();
+        let axis = LaneBumpAxis { domain: DomainId(0), lane: 0, delta: 5 };
+        let w2 = axis.apply(&w);
+        assert_ne!(before, w2.content_hash());
+        assert_eq!(w2.domains[0].observed[0], 105);
+    }
+}

crates/world_model/src/perturb.rs

@@ -0,0 +1,152 @@
+//! Perturbation axes. Perturbations are derived from domain surfaces (not a
+//! fixed global list) and each declares what trace/world difference it is
+//! expected to cause. The metamorphic gates check that these expectations
+//! actually hold across the corpus.
+
+use crate::primitives::DomainId;
+use crate::world::WorldSnapshot;
+
+/// What difference a perturbation is expected to produce.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct TraceDifferenceExpectation {
+    pub expect_trace_change: bool,
+    pub expect_delta_change: bool,
+    pub expect_future_change: bool,
+    /// If the perturbation is allowed to be observationally neutral, this
+    /// explains why (spec allows <=5% neutral *with explanation*).
+    pub neutral_explanation: Option<&'static str>,
+}
+
+impl TraceDifferenceExpectation {
+    pub fn active() -> Self {
+        TraceDifferenceExpectation {
+            expect_trace_change: true,
+            expect_delta_change: true,
+            expect_future_change: true,
+            neutral_explanation: None,
+        }
+    }
+}
+
+/// A perturbation axis derived from a domain surface.
+pub trait PerturbationAxis {
+    fn name(&self) -> String;
+    fn target(&self) -> DomainId;
+    fn apply(&self, world: &WorldSnapshot) -> WorldSnapshot;
+    fn expected_trace_difference(&self) -> TraceDifferenceExpectation;
+}
+
+fn domain_mut<'a>(world: &'a mut WorldSnapshot, d: DomainId) -> &'a mut crate::domain::DomainState {
+    &mut world.domains[d.0 as usize]
+}
+
+/// Add a delta to an observed lane.
+pub struct LaneBumpAxis {
+    pub domain: DomainId,
+    pub lane: usize,
+    pub delta: i64,
+}
+
+impl PerturbationAxis for LaneBumpAxis {
+    fn name(&self) -> String {
+        format!("bump(d{},l{},{:+})", self.domain.0, self.lane, self.delta)
+    }
+    fn target(&self) -> DomainId {
+        self.domain
+    }
+    fn apply(&self, world: &WorldSnapshot) -> WorldSnapshot {
+        let mut w = world.clone();
+        let ds = domain_mut(&mut w, self.domain);
+        ds.observed[self.lane] = ds.observed[self.lane].wrapping_add(self.delta);
+        w.mark_perturbed();
+        w
+    }
+    fn expected_trace_difference(&self) -> TraceDifferenceExpectation {
+        TraceDifferenceExpectation::active()
+    }
+}
+
+/// Multiply an observed lane by a factor.
+pub struct LaneScaleAxis {
+    pub domain: DomainId,
+    pub lane: usize,
+    pub factor: i64,
+}
+
+impl PerturbationAxis for LaneScaleAxis {
+    fn name(&self) -> String {
+        format!("scale(d{},l{},x{})", self.domain.0, self.lane, self.factor)
+    }
+    fn target(&self) -> DomainId {
+        self.domain
+    }
+    fn apply(&self, world: &WorldSnapshot) -> WorldSnapshot {
+        let mut w = world.clone();
+        let ds = domain_mut(&mut w, self.domain);
+        ds.observed[self.lane] = ds.observed[self.lane].wrapping_mul(self.factor).wrapping_add(1);
+        w.mark_perturbed();
+        w
+    }
+    fn expected_trace_difference(&self) -> TraceDifferenceExpectation {
+        TraceDifferenceExpectation::active()
+    }
+}
+
+/// Flip the sign / toggle a hidden lane. Hidden changes may be observationally
+/// neutral on the immediate delta but should still influence future turns.
+pub struct HiddenFlipAxis {
+    pub domain: DomainId,
+    pub lane: usize,
+}
+
+impl PerturbationAxis for HiddenFlipAxis {
+    fn name(&self) -> String {
+        format!("hidden(d{},l{})", self.domain.0, self.lane)
+    }
+    fn target(&self) -> DomainId {
+        self.domain
+    }
+    fn apply(&self, world: &WorldSnapshot) -> WorldSnapshot {
+        let mut w = world.clone();
+        let ds = domain_mut(&mut w, self.domain);
+        ds.hidden[self.lane] = !ds.hidden[self.lane].wrapping_add(0x5bd1e995);
+        w.mark_perturbed();
+        w
+    }
+    fn expected_trace_difference(&self) -> TraceDifferenceExpectation {
+        // Hidden state feeds the runtime, so we still expect change; but it is
+        // permitted to be observationally neutral on the immediate delta.
+        TraceDifferenceExpectation {
+            expect_trace_change: true,
+            expect_delta_change: false,
+            expect_future_change: true,
+            neutral_explanation: Some("hidden lane influences future, not immediate observed delta"),
+        }
+    }
+}
+
+/// Swap two observed lanes.
+pub struct LaneSwapAxis {
+    pub domain: DomainId,
+    pub lane_a: usize,
+    pub lane_b: usize,
+}
+
+impl PerturbationAxis for LaneSwapAxis {
+    fn name(&self) -> String {
+        format!("swap(d{},l{}<->l{})", self.domain.0, self.lane_a, self.lane_b)
+    }
+    fn target(&self) -> DomainId {
+        self.domain
+    }
+    fn apply(&self, world: &WorldSnapshot) -> WorldSnapshot {
+        let mut w = world.clone();
+        let ds = domain_mut(&mut w, self.domain);
+        ds.observed.swap(self.lane_a, self.lane_b);
+        w.mark_perturbed();
+        w
+    }
+    fn expected_trace_difference(&self) -> TraceDifferenceExpectation {
+        TraceDifferenceExpectation::active()
+    }
+}

crates/world_model/src/primitives.rs

@@ -0,0 +1,214 @@
+//! Deterministic primitives shared across the whole framework: stable ids,
+//! a stable content hash, and a deterministic RNG. Everything here is
+//! reproducible from a seed so that replay is bit-exact.
+
+use std::fmt;
+
+/// Number of independent world domains. The spec mandates >= 8.
+pub const NUM_DOMAINS: usize = 8;
+/// Observed value lanes per domain.
+pub const LANES: usize = 4;
+/// Hidden (unobserved) value lanes per domain. These create the
+/// hidden/observed state divergence the spec requires.
+pub const HIDDEN_LANES: usize = 2;
+
+macro_rules! id_type {
+    ($name:ident, $inner:ty) => {
+        #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+        pub struct $name(pub $inner);
+        impl fmt::Debug for $name {
+            fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+                write!(f, "{}({})", stringify!($name), self.0)
+            }
+        }
+        impl fmt::Display for $name {
+            fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+                write!(f, "{}", self.0)
+            }
+        }
+    };
+}
+
+id_type!(WorldId, u64);
+id_type!(ProgramId, u64);
+id_type!(ContractId, u64);
+
+/// Identifies one of the [`NUM_DOMAINS`] domains.
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct DomainId(pub u8);
+
+impl fmt::Debug for DomainId {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "DomainId({})", self.0)
+    }
+}
+
+/// A stable 64-bit content hash. Used for canonical comparison, replay
+/// hashes, and behavior fingerprints. Implemented with FNV-1a so the value
+/// is identical across machines and runs.
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Default)]
+pub struct Hash(pub u64);
+
+impl fmt::Debug for Hash {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "Hash({:016x})", self.0)
+    }
+}
+
+impl fmt::Display for Hash {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{:016x}", self.0)
+    }
+}
+
+const FNV_OFFSET: u64 = 0xcbf29ce484222325;
+const FNV_PRIME: u64 = 0x100000001b3;
+
+/// Streaming stable hasher (FNV-1a, 64 bit).
+#[derive(Clone, Copy)]
+pub struct Hasher {
+    state: u64,
+}
+
+impl Default for Hasher {
+    fn default() -> Self {
+        Hasher { state: FNV_OFFSET }
+    }
+}
+
+impl Hasher {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    #[inline]
+    pub fn write_u8(&mut self, b: u8) {
+        self.state ^= b as u64;
+        self.state = self.state.wrapping_mul(FNV_PRIME);
+    }
+
+    #[inline]
+    pub fn write_u64(&mut self, v: u64) {
+        for i in 0..8 {
+            self.write_u8(((v >> (i * 8)) & 0xff) as u8);
+        }
+    }
+
+    #[inline]
+    pub fn write_i64(&mut self, v: i64) {
+        self.write_u64(v as u64);
+    }
+
+    #[inline]
+    pub fn write_usize(&mut self, v: usize) {
+        self.write_u64(v as u64);
+    }
+
+    #[inline]
+    pub fn write_bytes(&mut self, bytes: &[u8]) {
+        for &b in bytes {
+            self.write_u8(b);
+        }
+    }
+
+    /// Mix in a label so structurally different streams that happen to share
+    /// numbers do not collide.
+    #[inline]
+    pub fn write_tag(&mut self, tag: &str) {
+        self.write_bytes(tag.as_bytes());
+        self.write_u8(0xff);
+    }
+
+    #[inline]
+    pub fn finish(&self) -> Hash {
+        Hash(self.state)
+    }
+}
+
+/// Hash a slice of i64 with a tag.
+pub fn hash_i64_slice(tag: &str, vals: &[i64]) -> Hash {
+    let mut h = Hasher::new();
+    h.write_tag(tag);
+    h.write_usize(vals.len());
+    for &v in vals {
+        h.write_i64(v);
+    }
+    h.finish()
+}
+
+/// Combine several hashes into one (order sensitive).
+pub fn combine_hashes(tag: &str, hashes: &[Hash]) -> Hash {
+    let mut h = Hasher::new();
+    h.write_tag(tag);
+    for hh in hashes {
+        h.write_u64(hh.0);
+    }
+    h.finish()
+}
+
+/// Deterministic SplitMix64 RNG. Fully reproducible from a seed.
+#[derive(Clone, Copy, Debug)]
+pub struct Rng {
+    state: u64,
+}
+
+impl Rng {
+    pub fn new(seed: u64) -> Self {
+        // Avoid the trivial all-zero state.
+        Rng {
+            state: seed ^ 0x9e3779b97f4a7c15,
+        }
+    }
+
+    /// Derive a sub-stream from a seed and a label, so independent concerns
+    /// never accidentally share a stream.
+    pub fn derive(seed: u64, tag: &str) -> Self {
+        let mut h = Hasher::new();
+        h.write_tag(tag);
+        h.write_u64(seed);
+        Rng::new(h.finish().0)
+    }
+
+    #[inline]
+    pub fn next_u64(&mut self) -> u64 {
+        self.state = self.state.wrapping_add(0x9e3779b97f4a7c15);
+        let mut z = self.state;
+        z = (z ^ (z >> 30)).wrapping_mul(0xbf58476d1ce4e5b9);
+        z = (z ^ (z >> 27)).wrapping_mul(0x94d049bb133111eb);
+        z ^ (z >> 31)
+    }
+
+    #[inline]
+    pub fn next_i64(&mut self) -> i64 {
+        self.next_u64() as i64
+    }
+
+    /// Uniform-ish integer in `[0, n)`.
+    #[inline]
+    pub fn below(&mut self, n: usize) -> usize {
+        if n == 0 {
+            return 0;
+        }
+        (self.next_u64() % (n as u64)) as usize
+    }
+
+    /// Integer in `[lo, hi]` inclusive.
+    #[inline]
+    pub fn range_i64(&mut self, lo: i64, hi: i64) -> i64 {
+        if hi <= lo {
+            return lo;
+        }
+        let span = (hi - lo) as u64 + 1;
+        lo + (self.next_u64() % span) as i64
+    }
+
+    #[inline]
+    pub fn chance(&mut self, p: f64) -> bool {
+        (self.next_u64() as f64 / u64::MAX as f64) < p
+    }
+
+    #[inline]
+    pub fn next_bool(&mut self) -> bool {
+        self.next_u64() & 1 == 1
+    }
+}

crates/world_model/src/world.rs

@@ -0,0 +1,265 @@
+//! The world snapshot and its constituent state machines.
+
+use crate::domain::{DomainState, ALL_DOMAIN_KINDS};
+use crate::primitives::{
+    DomainId, Hash, Hasher, Rng, HIDDEN_LANES, LANES, NUM_DOMAINS,
+};
+
+/// Number of world-level execution accumulator registers.
+pub const REGS: usize = 4;
+
+/// Cross-domain coupling. A dense `NUM_DOMAINS x NUM_DOMAINS` weight matrix
+/// that governs how a change in one domain propagates into others when turns
+/// advance. Generated per world; high rank by construction.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct CausalState {
+    pub coupling: [[i64; NUM_DOMAINS]; NUM_DOMAINS],
+}
+
+impl CausalState {
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_tag("causal");
+        for row in &self.coupling {
+            for &v in row {
+                h.write_i64(v);
+            }
+        }
+    }
+}
+
+/// Which lanes are observable, plus a deterministic observation-noise seed.
+/// Drives the divergence between hidden ground truth and observed projection.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct ObservationState {
+    pub visible: [[bool; LANES]; NUM_DOMAINS],
+    pub noise_seed: u64,
+}
+
+impl ObservationState {
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_tag("observation");
+        for row in &self.visible {
+            for &b in row {
+                h.write_u8(b as u8);
+            }
+        }
+        h.write_u64(self.noise_seed);
+    }
+}
+
+/// World-level execution accumulators carried across runes.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct ExecutionState {
+    pub accumulator: [i64; REGS],
+}
+
+impl ExecutionState {
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_tag("execstate");
+        for &v in &self.accumulator {
+            h.write_i64(v);
+        }
+    }
+}
+
+/// A future effect scheduled by execution; resolved when turns advance.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct ScheduledEffect {
+    pub turn_offset: u8,
+    pub domain: DomainId,
+    pub lane: usize,
+    pub hidden: bool,
+    pub value: i64,
+}
+
+/// Temporal state: pending scheduled effects create genuine future dependence.
+#[derive(Clone, PartialEq, Eq, Debug, Default)]
+pub struct TimeState {
+    pub pending: Vec<ScheduledEffect>,
+}
+
+impl TimeState {
+    pub fn hash_into(&self, h: &mut Hasher) {
+        h.write_tag("time");
+        h.write_usize(self.pending.len());
+        for e in &self.pending {
+            h.write_u8(e.turn_offset);
+            h.write_u8(e.domain.0);
+            h.write_usize(e.lane);
+            h.write_u8(e.hidden as u8);
+            h.write_i64(e.value);
+        }
+    }
+}
+
+/// The full world snapshot (per spec).
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct WorldSnapshot {
+    pub id: crate::primitives::WorldId,
+    pub turn: u64,
+    pub domains: Vec<DomainState>,
+    pub causal_state: CausalState,
+    pub observation_state: ObservationState,
+    pub execution_state: ExecutionState,
+    pub time_state: TimeState,
+    pub seed: u64,
+    /// Provenance counter for perturbations; never consumed by the runtime.
+    pub perturb_nonce: u64,
+}
+
+impl WorldSnapshot {
+    /// An all-zero baseline world (the generators fill it with real state).
+    pub fn blank(id: crate::primitives::WorldId, seed: u64) -> Self {
+        let domains = ALL_DOMAIN_KINDS.iter().map(|&k| DomainState::new(k)).collect();
+        WorldSnapshot {
+            id,
+            turn: 0,
+            domains,
+            causal_state: CausalState {
+                coupling: [[0; NUM_DOMAINS]; NUM_DOMAINS],
+            },
+            observation_state: ObservationState {
+                visible: [[true; LANES]; NUM_DOMAINS],
+                noise_seed: seed ^ 0xa5a5a5a5,
+            },
+            execution_state: ExecutionState {
+                accumulator: [0; REGS],
+            },
+            time_state: TimeState::default(),
+            seed,
+            perturb_nonce: 0,
+        }
+    }
+
+    pub fn mark_perturbed(&mut self) {
+        self.perturb_nonce = self.perturb_nonce.wrapping_add(1);
+    }
+
+    pub fn domain(&self, d: DomainId) -> &DomainState {
+        &self.domains[d.0 as usize]
+    }
+
+    pub fn domain_mut(&mut self, d: DomainId) -> &mut DomainState {
+        &mut self.domains[d.0 as usize]
+    }
+
+    /// Observed projection: only visible observed lanes, with deterministic
+    /// observation noise. Hidden lanes are excluded entirely. This is what an
+    /// outside observer can measure, and differs from ground truth.
+    pub fn observed_projection(&self) -> Vec<i64> {
+        let mut rng = Rng::new(self.observation_state.noise_seed ^ self.turn);
+        let mut out = Vec::with_capacity(NUM_DOMAINS * LANES);
+        for (di, d) in self.domains.iter().enumerate() {
+            for lane in 0..LANES {
+                if self.observation_state.visible[di][lane] {
+                    let noise = rng.range_i64(-1, 1);
+                    out.push(d.observed[lane].wrapping_add(noise));
+                } else {
+                    out.push(0);
+                }
+            }
+        }
+        out
+    }
+
+    /// Ground-truth state vector (observed + hidden), no noise. Used by the
+    /// runtime and by canonical hashing.
+    pub fn ground_truth(&self) -> Vec<i64> {
+        let mut out = Vec::with_capacity(NUM_DOMAINS * (LANES + HIDDEN_LANES));
+        for d in &self.domains {
+            out.extend_from_slice(&d.observed);
+            out.extend_from_slice(&d.hidden);
+        }
+        out
+    }
+
+    pub fn content_hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("world");
+        h.write_u64(self.id.0);
+        h.write_u64(self.turn);
+        h.write_u64(self.seed);
+        for d in &self.domains {
+            d.hash_into(&mut h);
+        }
+        self.causal_state.hash_into(&mut h);
+        self.observation_state.hash_into(&mut h);
+        self.execution_state.hash_into(&mut h);
+        self.time_state.hash_into(&mut h);
+        h.finish()
+    }
+}
+
+/// Difference of a single domain (after - before, wrapping).
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct DomainDelta {
+    pub domain: DomainId,
+    pub observed: [i64; LANES],
+    pub hidden: [i64; HIDDEN_LANES],
+}
+
+impl DomainDelta {
+    pub fn is_zero(&self) -> bool {
+        self.observed.iter().all(|&v| v == 0) && self.hidden.iter().all(|&v| v == 0)
+    }
+}
+
+/// The change produced by an execution.
+#[derive(Clone, PartialEq, Eq, Debug)]
+pub struct WorldDelta {
+    pub domain_deltas: Vec<DomainDelta>,
+    pub turn_advance: u64,
+}
+
+impl WorldDelta {
+    /// Compute `after - before`.
+    pub fn between(before: &WorldSnapshot, after: &WorldSnapshot) -> WorldDelta {
+        let mut deltas = Vec::with_capacity(NUM_DOMAINS);
+        for i in 0..NUM_DOMAINS {
+            let b = &before.domains[i];
+            let a = &after.domains[i];
+            let mut observed = [0i64; LANES];
+            let mut hidden = [0i64; HIDDEN_LANES];
+            for l in 0..LANES {
+                observed[l] = a.observed[l].wrapping_sub(b.observed[l]);
+            }
+            for l in 0..HIDDEN_LANES {
+                hidden[l] = a.hidden[l].wrapping_sub(b.hidden[l]);
+            }
+            deltas.push(DomainDelta {
+                domain: b.id(),
+                observed,
+                hidden,
+            });
+        }
+        WorldDelta {
+            domain_deltas: deltas,
+            turn_advance: after.turn.wrapping_sub(before.turn),
+        }
+    }
+
+    /// Domains that actually changed.
+    pub fn touched_domains(&self) -> Vec<DomainId> {
+        self.domain_deltas
+            .iter()
+            .filter(|d| !d.is_zero())
+            .map(|d| d.domain)
+            .collect()
+    }
+
+    pub fn hash(&self) -> Hash {
+        let mut h = Hasher::new();
+        h.write_tag("world-delta");
+        h.write_u64(self.turn_advance);
+        for d in &self.domain_deltas {
+            h.write_u8(d.domain.0);
+            for &v in &d.observed {
+                h.write_i64(v);
+            }
+            for &v in &d.hidden {
+                h.write_i64(v);
+            }
+        }
+        h.finish()
+    }
+}

plan.md

@@ -0,0 +1,284 @@
+ChatGPT3:26 PM
+Rust Phase 0/1 Implementor Specification
+Objective
+Build the testing framework first, then the runtime.
+
+No spell content is accepted until CI proves the system resists collapse into:
+
+single score
+single resource
+single effect axis
+single executor behavior
+single rune behavior
+single hidden damage formula
+decorative world state
+Hard CI Gates
+Minimum per full CI run:
+
+Generated worlds:              50,000
+Generated rune programs:       250,000
+Executions:                    1,000,000
+Perturbations per execution:   10
+Semantic mutants per run:      500
+Replay corpus cases:           10,000 minimum
+Reference/runtime comparison:  100% of executions
+Fast CI may run 10%, but merge-blocking CI must run full gates.
+
+Repository Layout
+crates/
+  world_model/
+  rune_ir/
+  trace_model/
+  generators/
+  reference_runtime/
+  runtime_under_test/
+  collapse_analysis/
+  semantic_mutation/
+  replay_corpus/
+  ci_reports/
+Implementation order is mandatory:
+
+1. world_model
+2. trace_model
+3. generators
+4. collapse_analysis
+5. semantic_mutation
+6. replay_corpus
+7. reference_runtime
+8. runtime_under_test
+The optimized runtime may not begin before steps 1–7 pass CI.
+
+World Model
+pub struct WorldSnapshot {
+    pub id: WorldId,
+    pub turn: u64,
+    pub domains: Vec<DomainState>,
+    pub causal_state: CausalState,
+    pub observation_state: ObservationState,
+    pub execution_state: ExecutionState,
+    pub time_state: TimeState,
+    pub seed: u64,
+}
+Minimum domain count:
+
+8 independent world domains
+Each domain must expose:
+
+pub trait WorldDomain {
+    fn domain_id(&self) -> DomainId;
+    fn read_surface(&self) -> ReadSurface;
+    fn write_surface(&self) -> WriteSurface;
+    fn perturbation_axes(&self) -> Vec<PerturbationAxis>;
+    fn fingerprint(&self) -> DomainFingerprint;
+}
+Domain acceptance:
+
+Each domain appears in ≥ 35% of traces.
+Each domain influences execution in ≥ 20% of corpus.
+Each domain is mutated by execution in ≥ 20% of corpus.
+Removing any domain reduces corpus behavioral diversity by ≥ 10%.
+Merging any two domains loses ≥ 8% predictive accuracy.
+Rune Program Model
+pub struct RuneProgram {
+    pub id: ProgramId,
+    pub tokens: Vec<RuneToken>,
+    pub seed: u64,
+}
+No rune stream may be rejected as the primary safety path.
+
+Execution always returns:
+
+pub struct ResolutionResult {
+    pub delta: WorldDelta,
+    pub trace: ExecutionTrace,
+    pub faults: FaultLog,
+    pub replay: ReplayRecord,
+}
+Engine crashes, panics, undefined Rust behavior, or unlogged failures fail CI.
+
+Trace Model
+pub struct ExecutionTrace {
+    pub read_graph: DomainAccessGraph,
+    pub write_graph: DomainAccessGraph,
+    pub causal_graph: CausalGraph,
+    pub information_flow: InformationFlowGraph,
+    pub executor_divergence: DivergenceGraph,
+    pub temporal_graph: TemporalGraph,
+    pub perturbation_response: PerturbationResponse,
+    pub behavior_fingerprint: BehaviorFingerprint,
+}
+Trace gates:
+
+Median causal edges per execution: ≥ 24
+95% of executions causal rank: ≥ 6
+Median touched domains per execution: ≥ 4
+95% of executions touched domains: ≥ 3
+Behavior fingerprint collision rate: < 5%
+Largest behavior cluster: < 2% of corpus
+Generator Requirements
+Generators must reject flat cases.
+
+pub struct GeneratedCase {
+    pub world: WorldSnapshot,
+    pub program: RuneProgram,
+    pub contexts: Vec<ExecutionContext>,
+    pub contract: SemanticContract,
+    pub perturbations: Vec<PerturbedCase>,
+}
+Generated case gates:
+
+estimated_causal_rank ≥ 6
+domain_entropy ≥ configured minimum
+perturbation_axes ≥ 10
+executor_count ≥ 3
+future_dependence present within 3 turns
+nonzero hidden/observed state divergence
+nonuniform domain fingerprints
+Semantic Contract
+pub struct SemanticContract {
+    pub min_causal_rank: usize,
+    pub min_domain_participation: usize,
+    pub min_future_sensitivity: f64,
+    pub min_context_divergence: f64,
+    pub max_compressibility: f64,
+}
+Default thresholds:
+
+min_causal_rank:              6
+min_domain_participation:     4
+min_future_sensitivity:       0.50
+min_context_divergence:       0.40
+max_compressibility:          0.70
+A case passes only if measured trace behavior satisfies its contract.
+
+Metamorphic Testing
+For every base execution, produce 10 perturbations.
+
+Perturbations are generated from domain surfaces, not a fixed list.
+
+pub trait PerturbationAxis {
+    fn apply(&self, world: &WorldSnapshot) -> WorldSnapshot;
+    fn expected_trace_difference(&self) -> TraceDifferenceExpectation;
+}
+Metamorphic gates:
+
+≥ 90% perturbations alter trace
+≥ 75% perturbations alter world delta
+≥ 50% perturbations alter state within 3 future turns
+≤ 5% perturbations may be observationally neutral without explanation
+Collapse Analysis
+The framework must attempt compression attacks.
+
+pub trait CollapseAttack {
+    fn compress(&self, corpus: &BehaviorCorpus) -> CompressedModel;
+    fn report(&self) -> CollapseReport;
+}
+Required attack families:
+
+domain removal
+domain merging
+constant folding
+causal edge deletion
+state aliasing
+latent factor modeling
+behavior clustering
+surrogate prediction
+temporal flattening
+observation flattening
+executor identity erasure
+Collapse gates:
+
+Best 1-factor model predicts < 40%
+Best 2-factor model predicts < 55%
+Best 4-factor model predicts < 70%
+No single domain explains > 30% outcome variance
+No pair of domains explains > 55%
+Compressed model loses ≥ 35% trace information
+If a smaller model predicts above these thresholds, fail CI.
+
+Semantic Mutation
+Mutants are generated structurally from the model.
+
+pub trait SemanticMutator {
+    fn mutate(&self, runtime: RuntimeArtifact) -> RuntimeArtifact;
+    fn expected_detection_reason(&self) -> DetectionClass;
+}
+Mutation gates:
+
+500 semantic mutants minimum
+0 surviving mutants
+Every mutant must fail at least one named acceptance gate
+Survivor report blocks merge
+A mutant surviving means the tests are invalid, not that the mutant is acceptable.
+
+Reference Runtime
+The reference runtime is the executable spec.
+
+pub trait Runtime {
+    fn resolve(&self, input: ResolutionInput) -> ResolutionResult;
+}
+Every execution runs:
+
+let expected = reference.resolve(input.clone());
+let actual = runtime_under_test.resolve(input);
+
+assert_eq!(canonical(expected), canonical(actual));
+Canonical comparison includes:
+
+world delta
+trace
+faults
+replay hash
+future-state hash over 3 turns
+Replay Corpus
+Every failure becomes permanent.
+
+pub struct ReplayCase {
+    pub world_seed: u64,
+    pub program_seed: u64,
+    pub contract_seed: u64,
+    pub perturbation_seed: u64,
+    pub expected_trace_hash: Hash,
+    pub expected_delta_hash: Hash,
+    pub expected_future_hash: Hash,
+}
+Replay gates:
+
+10,000 cases minimum
+100% deterministic replay
+0 hash drift unless migration explicitly updates corpus
+Reports Required Per CI Run
+Generate machine-readable JSON and human-readable markdown:
+
+domain_participation_report
+causal_rank_report
+compression_resistance_report
+metamorphic_response_report
+mutation_survivor_report
+runtime_equivalence_report
+replay_report
+coverage_report
+Merge blocked unless all reports pass.
+
+Absolute Rejection Conditions
+Reject if:
+
+A smaller model predicts behavior above thresholds.
+Any domain is decorative.
+Any domain is read-only or write-only across corpus.
+Most programs share the same behavior fingerprint.
+Most outcomes reduce to one numeric axis.
+Reference/runtime differ.
+Replay is nondeterministic.
+Any semantic mutant survives.
+Trace evidence cannot explain causality.
+Phase 0/1 Completion
+Phase 0/1 is complete only when:
+
+the adversarial framework exists first
+the reference runtime passes it
+the optimized runtime matches the reference runtime
+collapse attacks fail to simplify the universe
+mutation tests kill every generated simplification
+generated programs produce diverse, causal, replayable behavior
+No spell list. No templates. No cosmetic runes. The deliverable is a Rust engine whose tests make a fake universe fail.