<!-- reviewed: 2026-07-16 | Grok assessment accuracy pass + GATE-REPRO-SELECT | canonical facts: docs/VERIFICATION-ANALYTICAL-DATA.md -->

# Verification Analytical Data — cross-surface fact ledger

**Hybrid execution boundary (2026-10-23):** real WebGPU execution is implemented and tested for
the existing ambient numerical wilderness only. Automatic routing requires CPU-reference agreement
and a transfer-inclusive latency win; tested NVIDIA and Intel runs retained CPU workers because
GPU round trips were slower. The main visible organisms, Archons and connectome have not migrated.
WebGL already uses the GPU for drawing. Browser WebNN was absent on both tested installed browsers,
and no NPU model is wired. These are execution facts, not intelligence, consciousness or full-world
FPS improvements. See [ADR 0010](./adr/0010-worker-offload-and-streamed-hybrid-world-2026-06-26.md)
and [measured conditions](./BENCHMARKS-2026-06-26.md#browser-ambient-hybrid-compute).

**[CQM-25 canonical capability framework](./alife/cqm-capability-profile.html)** — version
**1.0.0**, adopted **2026-10-22**. The [source data and protocols](./alife/cqm-capability-profile.json)
own the 25 stable axis IDs. Feature plans and capability assessments must name affected IDs,
subject/scope, measures, controls, falsifiers and pinned source/run provenance, including failures.
The [adoption contract](./MODULE-CONTRACTS-2026-06-26.md) and accepted [ADR 0018](./adr/0018-cqm-25-capability-law-2026-10-22.md) govern taxonomy changes and publication.
Evidence states describe implementation and bounded tests; canonical `assessment.score` fields remain
null until calibrated. A single organism is eligible; ecology is one domain, and physical models are
judged against declared dynamics. CQM-25 is the project standard, not an externally validated
scientific certification. The separate 14 Butlin indicators remain distinct. The historical
nine-axis 155-system comparison cannot supply CQM-25 scores, ranks or an aggregate verdict.

[Provisional author estimates](./alife/cqm-capability-profile.html#provisional-estimates)
are a separate, explicitly labeled layer with their own [rubric and source data](./alife/cqm-capability-estimates.json).
They retain source provenance and limitations, confer no measured capability, and do not populate
canonical `assessment.score` fields. No overall or domain mean, ranking or aggregate verdict is published.

**Living index** (not an archive). One place that records every fact duplicated across the repo's
surfaces (MD · HTML · XML · code), its **single source of truth**, and where each surface restates it —
so a drift between the README, the dashboard, a master XML, or a spec page is caught at a glance.
Rewritten in place when the facts change (per the binding living-doc law in
[CLAUDE.md](../CLAUDE.md)). Capability authority follows CQM-25 version **1.0.0**, adopted
**2026-10-22**. The earlier nine-axis comparison and its statistics are retained as historical
rubric evidence; they do not define the current capability framework or transfer scores to it.

> **How the numbers stay honest:** the version (`package.json`) and the measured receipts +
> NHSI design counts (`scripts/canonical-receipts.ts`) are the ONLY places those facts are edited.
> `scripts/sync-surfaces.ts` propagates them to every surface; `bun run sync:check` and
> `bun run verify:receipts` are gate-enforced (`bun run check`). Never hand-edit a synced token.

---

The [legacy construct crosswalk](./alife/cqm-capability-profile.html#crosswalk) corresponds to
**19 of the 25** constructs, without transferring scores or protocol validation. Social
communication, individuality, resource maintenance, homeostasis, repair/resilience and valence
trade-offs are not separately represented in the historical survey. Conceptual correspondence
does not establish that the survey measured or validated CQM-25 acceptance protocols.

The concurrent organism-biology addition is also integrated: physiology and four-channel memory
follow stable entity identity through deletion, compaction, reuse and Genesis. Its five habitat bands
and four developmental categories run on the main CPU thread, with accumulated simulation dt and one
wild entropy salt per batch. They are phenomenological variables, not validated neurobiology.
`planHybridPartition` is a proposal; actual entity-brain receipts report calling-thread CPU Zig WASM.
The prototype brain WGSL and exported shard helper are not dispatched by the application. No
connectome renderer culling, horizontal world enlargement or body morphing follows from their
estimated parameter outputs. These limitations also apply to upstream `7ae20238` descriptions.

## PRIME observatory measurement boundary

The 27 World/Cognition/Quantum cards now use completed cognition snapshots and bounded observed-tick
history. Exact source values, units, limitations and JSON export accompany each card. Repaints and
paused frames do not create observations. Quantum geometry and qualia reads no longer change the
simulation; controlled-RY implements the corrected matrix. The [card-by-card measurement inventory](./SUPER-CREATURE-RESEARCH-2026-06-26.md#prime-neural-observatory-measurement-audit)
records scope and limitations. IV · BRAIN presents the same partial module graph in COMPOSITE and
MEGA 4D layouts: 16 selected module groups, named snapshot values, and links with captured-payload,
operand or unavailable-value provenance. Dedicated captures preserve earlier inputs before recurrence
or substrate updates overwrite them. Fixed view geometry cannot invent activity when a tick advances.
Invalid numbers remain visible and cannot become plausible zeros. Pairwise quantum mutual information
requires a finite normalized statevector; invalid inputs produce an unavailable result, not a correlation
matrix. Exact graph data and view metadata are included in JSON export. These engineering changes
do not calibrate CQM-25 scores or prove consciousness, sentience or quantum hardware operation.

## 1 · Canonical facts (single source of truth)

**Xylos:** the additional single organism is specified in [XYLOS.md](./XYLOS.md)
and [ADR 0017](./adr/0017-xylos-autonomous-manifold.md). Its local causal tests
and research targets are separate from the established apex indicator assessment.
Live mode has a browser-entropy boundary; local replay requires the recorded
numeric input sequence and does not rewind the shared world.

| Fact                                          | Canonical value                                                                                                                                                                                        | Source of truth                                                                                                                                                                            | Propagated by               |
| --------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | --------------------------- |
| Package version                               | `0.23.0`                                                                                                                                                                                               | `package.json` `version`                                                                                                                                                                   | `sync-surfaces.ts`          |
| Test count (exact tracked suite)              | `3631`                                                                                                                                                                                                 | `scripts/canonical-receipts.ts`                                                                                                                                                            | `sync-surfaces.ts`          |
| Line coverage                                 | `84.64%`                                                                                                                                                                                               | `scripts/canonical-receipts.ts`                                                                                                                                                            | `sync-surfaces.ts`          |
| Function coverage                             | `82.21%`                                                                                                                                                                                               | `scripts/canonical-receipts.ts`                                                                                                                                                            | `sync-surfaces.ts`          |
| Faculties (design)                            | `100` (~30 deep-wired)                                                                                                                                                                                 | `CANONICAL_FACULTIES`                                                                                                                                                                      | `sync-surfaces.ts`          |
| Archon pantheon                               | `25` (5 apex + 20 light)                                                                                                                                                                               | `CANONICAL_ARCHONS`                                                                                                                                                                        | `sync-surfaces.ts`          |
| Theory-of-mind orgs                           | `25`                                                                                                                                                                                                   | `CANONICAL_TOM_ORGANS`                                                                                                                                                                     | `sync-surfaces.ts`          |
| Emergence angles                              | `10` (+5 god events)                                                                                                                                                                                   | `CANONICAL_EMERGENCE_ANGLES`                                                                                                                                                               | `sync-surfaces.ts`          |
| Digital Biologics / PrimordialSoup forms      | `26`                                                                                                                                                                                                   | `CANONICAL_BIOLOGIC_FORMS`                                                                                                                                                                 | `sync-surfaces.ts`          |
| Morphotypes                                   | `250` live · `100` legacy                                                                                                                                                                              | `phyla.ts` (`PHYLUM_COUNT 10 × MORPHS_PER_PHYLUM 25`) / `constants.ts` `MORPH_COUNT`                                                                                                       | prose (NOT auto-synced)     |
| Butlin scorecard                              | `8/14 met + 6/14 partial`                                                                                                                                                                              | measured 2026-06-21 adversarial code audit                                                                                                                                                 | prose (NOT auto-synced)     |
| Tsotchke corpus                               | `23 public repos (16 user + 7 org); 22 causal/runtime entries: 8 deep / 7 wired / 2 harvest / 4 fenced / 1 meta; 17/21 non-meta integrated; homebrew-moonlab census-only`                              | [TSOTCHKE-INTEGRATION-MAP-2026-06-26.md](./TSOTCHKE-INTEGRATION-MAP-2026-06-26.md), `src/sim/tsotchke-registry.ts`                                                                         | prose (NOT auto-synced)     |
| Organism-intelligence causal claim            | V4: ordinary and Petri fail magnitude, adaptive predictor fails both controls, Titans pass; only bounded Titan game-policy semantic causality is authorized                                            | [V4 report](./reports/ORGANISM-INTELLIGENCE-V4-RESULTS-2026-07-11.md), V4 JSON/CSV/forest receipt                                                                                          | prose (NOT auto-synced)     |
| CQM-25 capability framework                   | Version `1.0.0`, adopted `2026-10-22`; 25 stable axes; evidence states; canonical `assessment.score` fields `null` until calibrated; provisional estimates separate                                    | [Canonical JSON](./alife/cqm-capability-profile.json) + [binding contract](./MODULE-CONTRACTS-2026-06-26.md)                                                                               | `gen-capability-profile.ts` |
| Historical nine-axis rubric                   | **Historical only; no CQM-25 score transfer.** `[4.0, 2.0, 3.4, 3.8, 4.5, 4.6, 4.4, 3.5, 4.0]`; breadth `3.8`; rank `#1/155`; z `+3.236` population / `+3.362` peers; Mahalanobis `8.987`; lead `+0.3` | `CODE_GROUNDED` + CSV + `docs/reports/assets/alife-codeground.json` (GATE-REPRO-SELECT + GATE-DOME-REFUGE)                                                                                 | generated + `verify:facts`  |
| Comprehensive standing synthesis (2026-07-12) | **Historical snapshot** (v0.22.0 / 2861 tests / #1/113). Specialist historical context; current capability authority is CQM-25 and current receipts are in this section                                | [COMPREHENSIVE-STANDING report](./reports/2026-07-12-COMPREHENSIVE-STANDING-AND-XENO-ALIFE-POSITION.md) · [HTML](./reports/2026-07-12-COMPREHENSIVE-STANDING-AND-XENO-ALIFE-POSITION.html) | prose (NOT auto-synced)     |
| Entity ceiling                                | `50,000` (mega tier)                                                                                                                                                                                   | `src/core/quality.ts` `resolveTier`                                                                                                                                                        | prose (NOT auto-synced)     |
| Apex composite mind                           | `~10,081` weights                                                                                                                                                                                      | `src/sim/super-mind.ts`                                                                                                                                                                    | prose (NOT auto-synced)     |
| Legacy spine                                  | `~1,444` params                                                                                                                                                                                        | `src/sim/super-mind.ts` / ADR-0008                                                                                                                                                         | prose (NOT auto-synced)     |

### Latest local receipt (Windows, 2026-07-14, Bun 1.3.14)

<!-- cqm-sync:local-measurement:start -->

- `bun run verify:receipts` → **3,297 tests, zero failures** · **366 test files** · **3,589,864**
  `expect()` calls · **93.60% line / 91.61% func** on this Windows checkout.

<!-- cqm-sync:local-measurement:end -->

### Dome census + underdocumented live systems (P3 surface)

Live composition root (`src/world.ts`) — not decoration. Public surfaces should list these castes
when describing the product:

| Caste / system      | Scale (desktop)         | One-line                                                                |
| ------------------- | ----------------------- | ----------------------------------------------------------------------- |
| Entities            | ≤10k default / 50k mega | 70p brains · biomass graze · Nash · connectome                          |
| AlienFlora          | 60k                     | edible food field · waste fertilizer · soft contact                     |
| Xenomimics          | ≤1000                   | twin brains · graze · trophic prey · sanctuary refuge                   |
| Shoggoths           | 100                     | predators · UMBRA                                                       |
| Puppeteers          | 100                     | remorph / weather / chaos                                               |
| Titans              | 20                      | PD diplomacy · dual economy · harvest                                   |
| Leviathans          | 4                       | RD stirrers · dome graze                                                |
| Apex SuperCreatures | 5 full minds            | SuperMind + body + evo + hunt                                           |
| Light Archons       | 20                      | mind-field echoes (not full SuperMinds)                                 |
| Alphabet glyphs     | 100                     | visual pantheon                                                         |
| NHI                 | ≤1000                   | GOAP manipulators · semantic gene                                       |
| Big Tree residents  | 250+                    | fruit/leaf ecology · sanctuary governance                               |
| Wingmen             | 100×5                   | assist swarms                                                           |
| Soup / Petri        | 128 / ≤64               | Tsotchke genesis                                                        |
| Mechalogodrom       | 1                       | fusion abomination                                                      |
| Weather             | 6 states                | climate-coupled to flora/entities/xeno                                  |
| Collective fields   | live                    | stigmergy / noosphere / morphic / dark-energy fields (composition-root) |
| Connectome          | live                    | topology-only (no physical tether cosmetics)                            |
| Wilderness          | live                    | non-writeback fauna outside petri catalysis                             |

**Honesty holes that stay (P2):** open-endedness 2.0 · entity sexual genome incomplete · coupling
~0.27 · lab null gap ≈ 0 · 5 full minds vs 20 light-echo · ~30/100 deep faculties · peer maturity
1.5/5 · 5s respawn arcade ecology · indicatorOnly forever.

### Organism-intelligence V4 descendant result (2026-07-11)

The [V4 result report](./reports/ORGANISM-INTELLIGENCE-V4-RESULTS-2026-07-11.md) is backed by a
[canonical receipt](./reports/assets/organism-intelligence-causal-benchmark-v4.json), complete
[1,152-row CSV](./reports/assets/cross-being-neural-causality-v1.csv), and accessible
[forest SVG](./reports/assets/organism-intelligence-v4-cross-being-forest.svg). Its post-write integrity
test recomputes receipt/raw/source/fixture hashes, Git ancestry, the exact raw matrix, forbidden claims,
and SVG bytes.

| Family                     | Frozen result                                                   | Authorized claim                                                    |
| -------------------------- | --------------------------------------------------------------- | ------------------------------------------------------------------- |
| Ordinary organisms         | **FAIL** — magnitude floor                                      | None                                                                |
| Adaptive ecology predictor | **FAIL** — inference and magnitude controls                     | None                                                                |
| Petri digital biologics    | **FAIL** — magnitude floor                                      | None                                                                |
| Titans                     | **PASS** — inference, magnitude, replay, numerical release gate | Bounded game-policy semantic causality on the frozen diplomacy task |

The result is one family pass and three failures, not an overall V4 pass. It authorizes no numeric
score, V4-over-V3, neural-capacity, pooled scaling, consciousness, sentience, general-intelligence,
physical-quantum, or security claim.

### Phase-B development boundary (2026-07-11)

[ADR 0015](./adr/0015-phase-b-neural-semantic-expansion-2026-07-11.md) governs successor work. The
default NHI gene is now a live 109-weight `9→6→7` network whose last four inputs are separately ablatable
resource/threat/exploration/social semantic lanes; 58/109/211 tiers are actually allocated. The former
five-input neural outputs and constructor RNG position are exact when those four lanes are zero.

Two isolated successor leaves are also implemented but are **not production evidence**: predictor V2 has
54/98/186 actually allocated parameters with pre-outcome forecast/observe tokens and bounded optimizer
state; the ordinary resource head has 27/51/99 parameters, fixed time-normalized recurrence, and a
single counted readout that can learn only from explicit one-shot food reward. A development firewall
reconstructs and rejects all 170 historical V1–V4 evaluation/calibration seeds. No confirmatory Phase-B
seed family exists, neither leaf changes current production behavior, and no score or consciousness/
sentience claim changes.

- **Batch-26 corrective audit:** the originally shipped nearest-cell flora sampler made the ±6-unit
  chemotaxis probe degenerate across much of the 44-unit grid even though a smooth synthetic-field
  test passed. `AlienFlora.biomassAt()` now bilinearly interpolates the real four-cell neighborhood,
  and the gate drives that shipped sampler through the live `EntityManager`, with a sub-cell
  non-degeneracy assertion. The same audit permits only confined literal Git-diff paths after `--`
  (denying root, revisions, pathspec magic, and globs), propagates the turn `AbortSignal` through Git-grep
  walks, and computes AD-forager `finalPotential` at the true final position. Historical snapshot and
  local-measurement markers prevent future receipt synchronization from rewriting those contexts.

- Tracked-only discovery makes `CANONICAL_TEST_COUNT` an exact cross-platform suite count; the
  published **84.64% line / 82.21% func** pair remains the one-sided clean-Ubuntu coverage floor.
- Gate-enforced `verify:receipts` rejects any failed/unknown child status, any test-count mismatch,
  or coverage below either portable floor. Higher local coverage is recorded separately, not synced.
- The current publication line keeps the exact test count, portable coverage floors, and higher local
  coverage explicitly separated. The complete local gate and `verify:facts` are green with no drift.

### Operational organism-intelligence receipt (2026-07-10)

The binding readable report is
[`2026-07-10-OPERATIONAL-ORGANISM-INTELLIGENCE-CAUSAL-AUDIT.md`](./reports/2026-07-10-OPERATIONAL-ORGANISM-INTELLIGENCE-CAUSAL-AUDIT.md);
the machine receipt is
[`organism-intelligence-causal-benchmark-v3.json`](./reports/assets/organism-intelligence-causal-benchmark-v3.json)
with content hash `96495964f03eca1860ba7f869cbd35e929bbdfdd75f801eb4479e6239dfd1b9f`.

- Seed policy: fresh deterministic 30-seed family disjoint from V1/V2, fixed in source but not externally
  preregistered.
- Synthetic goal-only minus exact legacy: mean `0.009081553179910407`, bootstrap 95% CI
  `[0.008224136063586919, 0.009901016159411519]` — accepted.
- Enhanced minus goal-preserved disabled: mean `0.008926803443918892`, bootstrap 95% CI
  `[0.0074709186839338375, 0.010524275559338788]` — accepted.
- Enhanced minus uniform random-action baseline: mean `0.003969487708636489`, bootstrap 95% CI
  `[-0.010601483800813, 0.0189792125751305]` — not separated.
- Adaptation: median enhanced `1.2743610106013112`, frozen `1.2008535611966096`, relative improvement
  `6.121266720602802%` — above the script-declared 5% threshold on this fixed family.
- Corpus causality: `17/17` integrated external rows changed final entity velocity on every evaluation
  seed; four fences plus metadata were `5/5` exact zero.
- Numerical safety: 10,000/10,000 forced revisions remained finite and bounded.
- Performance: three fresh processes × ten batches × seven samples with isolated branch state and
  counterbalanced first-run order; worst shared-field p95 `0.10890000000000555 ms`, median process-level
  50,000-entity increment `1.3857499999999163 ms`, worst of all 30 batch medians
  `2.8494000000000597 ms` — stable pass.
- A four-aggregate-channel rotation and a uniform final-exploration surrogate were tied. They are not a
  repository permutation, entropy-matched control, or evidence of quantum specificity.
- Consumer evidence is an executed ten-file targeted gate (`101` tests, `0` fail, `377,438` assertions);
  every living-system class named by ADR-0013 has a matched full-class counterfactual.
- Claim consequence: V3 authorizes no additional numeric A-Life, capability, consciousness, or sentience
  uplift.

The QRNG runtime is a seeded deterministic classical statevector adaptation pinned to upstream v3.0.1,
not physical entropy or a CSPRNG. The Eshkol Taylor implementation is an order-0-through-8 Float64
analogue pinned to v1.3.2, not native or unlimited-order parity.

### Dated consolidated report/audit pair (2026-07-07)

- [`CONSOLIDATED-22-MASTER-ASSESSMENT-CURRENT-2026-07-07.md`](./CONSOLIDATED-22-MASTER-ASSESSMENT-CURRENT-2026-07-07.md)
  is the dated readable synthesis across the 22 report artifacts and the
  proxy-consciousness/sentience boundary; its receipt values are snapshot evidence, not current data.
- [`CONSOLIDATED-22-FILE-AUDIT-CURRENT-2026-07-07.md`](./CONSOLIDATED-22-FILE-AUDIT-CURRENT-2026-07-07.md)
  is the file-trust ledger. It records stale counts, overclaim risks, public-page gaps, and the
  exact fifth-pass handles that must carry forward.

---

## 2 · Surface inventory (where shared facts are published)

| Class            | Surfaces                                                                                                                                    |
| ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------- |
| Root MD          | `README.md`, `ROADMAP-2026-06-26.md`, `CHANGELOG.md`, `CLAUDE.md`, …                                                                        |
| Steering         | `CLAUDE.md`, `.github/copilot-instructions.md`, `masters/*.xml` (3), `docs/PHILOSOPHY-2026-06-26.md`, `docs/MODULE-CONTRACTS-2026-06-26.md` |
| Progress / truth | `docs/NHSI-PROGRESS-DASHBOARD-2026-06-26.md`, `docs/VERIFICATION-ANALYTICAL-DATA.md` (this file)                                            |
| Consciousness    | `docs/SUPER-CREATURE-RESEARCH-2026-06-26.md`, `docs/NHSI-PROGRESS-DASHBOARD-2026-06-26.md`                                                  |
| HTML surfaces    | `index.html`, `docs.html`, `specs.html` (+ `lab/`)                                                                                          |
| Tsotchke         | `docs/TSOTCHKE-INTEGRATION-MAP-2026-06-26.md` (authoritative), `THIRD-PARTY-NOTICES.md`                                                     |
| Diagrams         | `docs/ENTITY-SCHEMA-AND-MAPPINGS-2026-06-26.md`, `docs/ARCHITECTURE-2026-06-26.md`                                                          |

---

## 3 · Audit findings (2026-06-26)

| #   | Severity                   | Where                                                      | Issue                                                                                                                                                                                          | Resolution                                                                                    |
| --- | -------------------------- | ---------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------- |
| A   | **P0 gate-RED**            | `.github/copilot-instructions.md`                          | markdown tables collapsed by the external `hacklm-memory` tool → fails `prettier --check` (first gate stage) in the working tree                                                               | `prettier --write` (re-pad tables)                                                            |
| B   | fixed                      | `scripts/canonical-receipts.ts`, `tests/doc-links.test.ts` | Historical host-dependent discovery made one count look like a floor; tracked-only discovery now yields one exact suite count on clean local/CI checkouts                                      | exact-count semantics sealed at **2369**; coverage alone remains a portable one-sided floor   |
| C   | **P0 gate-RED**            | `CHANGELOG.md`, `docs/KANBAN-2026-06-26.md`                | 3 dead relative links into the deleted `docs/audit-2026-06-15/` folder (consolidated into AUDIT-LOG by `e51a376`)                                                                              | repoint to `docs/AUDIT-LOG.md`                                                                |
| C2  | P1 test-hygiene            | `tests/doc-links.test.ts`                                  | `SKIP` set omitted `.claude/` → the test scanned nested worktree copies and false-failed                                                                                                       | add `.claude`, `legacy`, `site`, `coverage` to `SKIP`                                         |
| D   | P1 doc-rot                 | `docs/KANBAN-2026-06-26.md`                                | mojibake: `×11/`×7/``×1 (orphaned-emoji fragments), `—`×34 used as `—`, plus `¦`/`–`/`©º`— slipped the encoding gate (orphaned fragments aren't double-encoding;`U+201D` is a legit codepoint) | byte-precise normalize: drop corrupted emoji prefixes, restore `—`/`–` separators             |
| E   | P2 count-drift             | `docs/KANBAN-2026-06-26.md:17`                             | "ALL **19** Tsotchke repos" — only outlier vs the canonical "20 corpus projects" everywhere                                                                                                    | `19 → 20`                                                                                     |
| F   | **RESOLVED (2026-07-06)**  | `HANDOFF-2026-06-26.md`, `research_receipts-2026-06-26.md` | stale present-tense continuity artifacts (942–1172 tests) duplicated AUDIT-LOG + VERIFICATION                                                                                                  | deleted; history preserved in AUDIT-LOG + CHANGELOG                                           |
| G   | flag (cleanliness)         | repo root                                                  | stray debug logs tracked at root (`.gate.log`, `.gate.baseline.log`, `.audit-gate.log`, `law.log`, `law_error.txt`, `tsc.log`, `tscout.txt`, `receipts_print.txt`)                             | noted for cleanup                                                                             |
| H   | flag (fidelity, not a bug) | `src/math/curvature-aware-qng.ts`                          | `computeChristoffelSymbols` sets `dg=0`, so the general N×N "curvature-aware" path reduces to ordinary QNG                                                                                     | honestly documented as a simplification; no NaN / wrong shape; noted so the caveat is visible |

**Finding D2 — master governance XML mojibake (root cause + fix).** All three `masters/*.xml` files
(the EXECUTOR / ARCHITECT / PHYSICIST steering docs read before every change) each carried ~150
box-drawing `═` rules double-encoded into `â•` runs, plus `â€"`→`—`, `·`→`·`, `â†'`→`→`. Root cause:
`scripts/normalize-docs.ts` globbed `.md` + the 3 HTML files but **not `masters/*.xml`**, and the
`docs-truth-law` encoding gate scanned only the `.md` CANONICAL list + HTML — so XML corruption shipped
with green CI. Fix: normalize-docs now globs `masters/**/*.xml` + `docs/**/*.xml` (repaired all 3, byte-
verified clean), and a new `docs-truth-law` block encoding-gates the steering XML + KANBAN against
mojibake / U+FFFD / sub-lead-byte / curly-quote corruption.

### Code audit (core math / sim / AI) -- initial hand-review (see §5 for the deeper pass that found + fixed real defects)

A line-by-line review of `src/math/**` and `src/sim/**` (hand-verified against standard formulas, not
docstrings) initially reported no P0/P1 bugs (a deeper 7-agent pass in §5 did find several). Quantum gates / Born rule / Bloch vector, mulberry32 +
FNV-1a RNG, Aaronson–Gottesman Clifford tableau, mixed-state QGT (Im sign + entropy), QGT / natural
gradient, SO(3) quaternion algebra, Crank–Nicolson Schrödinger, Jacobi SVD / MPS, Wigner-d / Clebsch–
Gordan / 6j / 9j, coherence / magic / GWT, spin-glass, criticality, Fubini–Study aliveness, HRR
holographic memory (unbind keeps the real-valued trace, not `sign()`), economy Vickrey/transfer — all
verified correct. **Determinism law holds:** no `Math.random` / `Date.now` / `performance.now` in
`src/sim/**` or `src/math/**` logic (the one allowed `world.ts` super-evo localStorage timestamp is
outside sim logic). Only fidelity caveat = Finding H above.

At the **2026-06-27 historical reconciliation**, these values passed the then-current consistency check:
Butlin `8/14 met + 6/14 partial`, package version `0.22.0`, entity ceiling `50,000`, bioforms `26`, the
faculty/Archon/ToM/emergence counts, and the then-used Tsotchke count `20`. This paragraph is a dated
audit receipt, not current SSOT; the v0.23.0 values and dual Tsotchke denominators are published in §1.

---

## 4 · Reconciliation — "rename every MD with the current date"

The `/goal` asked to "update the name (with current date) of every MD file"; the owner confirmed they
want **both** the dated rename and a non-breaking repo ("make sure they don't break — figure it out").

**Owner ruling — SETTLED, re-confirmed 2026-06-27 (second explicit AskUserQuestion; do not re-litigate):**
keep the 20 **name-pinned** files at their canonical names. Renaming them breaks EXTERNAL systems that
cannot be rewired from inside the repo — `README.md` (GitHub renders it as the homepage), `CLAUDE.md` (the
Claude Code harness loads this exact name), the GitHub community-health files
(`SECURITY`/`CODE_OF_CONDUCT`/`CONTRIBUTING`/`PULL_REQUEST_TEMPLATE`/`.github/copilot-instructions`), the
four directory `README.md`s (folder-index render), `FILE-MAP.md` (generator-emitted), and the stamp-target
ledgers (`VERIFICATION-ANALYTICAL-DATA.md`, `AUDIT-LOG.md` — ~66 in-content stamps + CLAUDE.md point at
these exact names). They each carry the in-content `reviewed: <date>` stamp instead. Every MD that CAN be
safely renamed already is (66 of 86 dated, refs rewired, gate green).

**Done (2026-06-26 → 27, in waves):** the **58 content / ADR / spec / report docs that are safe to rename**
were `git mv`'d to dated names (`X-2026-06-2N.md`) and **every reference rewired** across all tracked text
files — verified **0 broken links** repo-wide (`tests/doc-links.test.ts` green), `tsc` clean, `sync:check`
green, FILE-MAP regenerated. A further **8 `docs/reports/2026-06-NN-…` files** already encode their
point-in-time date as a filename prefix and keep it. **Net: 66 of 86 non-legacy MDs now carry a date in
the filename; the remaining 20 are name-pinned (below).** Each renamed doc also carries the in-content
`reviewed: 2026-06-27` stamp. _(Count verified by `git ls-tree origin/main` on 2026-06-27 — the earlier
"24" was a first-wave figure that went stale as later waves renamed more.)_

**Kept at their canonical names — renaming WOULD break, which the owner's "don't break" forbids:**

- **GitHub / agent-special** (read by exact name): `README.md` (front-page render), `CHANGELOG.md`,
  `LICENSE`, `NOTICE.md`, `SECURITY.md`, `CODE_OF_CONDUCT.md`, `CONTRIBUTING.md`, `THIRD-PARTY-NOTICES.md`,
  `CLAUDE.md` + `.github/copilot-instructions.md` (loaded by the agent harness), `ROADMAP-2026-06-26.md`, sub-dir `README.md`s,
  `index.html` / `docs.html` / `specs.html` (GitHub Pages), `.github/**`, `.memory/**` (hacklm tool).
- **Hardcoded in scripts / gates** (`sync-surfaces.ts`, `docs-truth-law.test.ts`, `gen-filemap`):
  `ARCHITECTURE`, `KANBAN`, `ERD`/`ERM`/`ERP`, `MODULE-CONTRACTS`, `FILE-MAP`, `TECHNICAL-SPECIFICATION`,
  `PHILOSOPHY`, `NHSI-PROGRESS-DASHBOARD`, `SUPER-CREATURE-RESEARCH-2026-06-26.md`, `SUPER-CREATURE-RESEARCH`, `BENCHMARKS`,
  `TSOTCHKE-INTEGRATION-MAP` — renaming reds the gate / breaks the live sync machinery.
- **Convention-critical / heavily-linked**: `AUDIT-LOG.md`, this ledger, `TSOTCHKE-INTEGRATION-MAP-2026-06-26.md`.
- **Numbered ADRs** (`docs/adr/000X-*`, referenced as "ADR-0009"), **legacy/** (verbatim), and
  **already-dated** files (`docs/reports/2026-*`, the `-2026-06-1x/20` audits) whose
  filename date is their meaningful point-in-time record.

For the kept set, the in-content `reviewed: 2026-06-27` stamp supplies the current date without breaking
the repo — the maximal literal compliance consistent with "don't break."

**Owner-confirmed (2026-06-26):** asked directly, the owner chose **in-content stamps over renames** —
the safe, law-compliant reading. Status: **75 / 77** maintained narrative MDs carry the
`reviewed: 2026-06-27` marker. The remaining 2 are legitimately exempt: `docs/FILE-MAP.md` is
**auto-generated** (its freshness is structural — it lists the live module set; a hardcoded date would be
immediately stale and `Date.now` is banned in the generator), and `.github/copilot-instructions-2026-06-26.md` is
**externally managed** by the `hacklm-memory` tool. `legacy/` (never-edit) and `.memory/` (tool-owned)
are out of scope by policy.

---

## 5 · Code-audit update (deeper 7-agent pass, 2026-06-26)

The §3 "code audit — CLEAN" line was an earlier hand-review. A deeper 7-subsystem MIT-master audit (137
files, verified against references rather than docstrings) found and **FIXED** real defects it had missed:

- **HIGH — `src/math/irrep.ts` `wigner6j` / `wigner9j`:** wrong for `j >= 7`. The Racah sum needs ~33! at
  j=8, but the linear factorial table topped out at 25! and silently dropped terms, so `{8 8 8;8 8 8}`
  returned `4.2e-12` vs the true `-1.265e-2` (wrong sign). **FIXED** by rewriting in log-factorial space
  (exact across j=1..8) + a regression test. (Today only reachable via the dead `tsotchke-deep-wire.ts`,
  but the module advertised `IRREP_J_MAX = 8`.)
- **HIGH — `src/sim/super-mind.ts` `quantumMagic`:** fed a malformed amplitude vector (`re = sqrt(p)`
  magnitudes only, `im = sin(phase)` phase only — neither the real amplitudes nor normalized), feeding
  `reflex` → attention schema + jewel shader. **FIXED** by using `snap.magicNorm`, already computed on the
  true register amplitudes in `super-qubits.ts`.
- **HIGH — `src/sim/super-body.ts` `dispose()`:** freed only 3 of 9 owned materials → a recurring WebGL
  leak on every body teardown / world reset. **FIXED** by disposing every material in the traverse
  (leak-safe; three.js `dispose()` is idempotent for the few materials shared across sibling meshes).
- **MED — all four now FIXED (regression-tested where observable):**
  - `src/sim/mortality.ts` — `reproduce()` could drive `lifespan` negative (cost 100 × N), making
    `age/lifespan` produce `NaN`/negative in `legacyScore`, `lifespanProgress`, `urgency`. **FIXED:** floor
    lifespan at 1 on reproduce + guard the divisor with `Math.max(1, …)`. New `tests/mortality.test.ts`.
  - `src/sim/petri-dish.ts` — the brutal-god release called `applyBrutalRelease(rel, state.biologics.map(…))`
    on a throwaway copy, so its consume/rebirth/drain mutations were discarded (effect dead). **FIXED:**
    normalize the live `state.biologics` in place and pass them by reference; fold `outcome.warp` into godPower.
  - `src/sim/emergence-angles.ts` — five append-only history arrays (`exchangeHistory`, `warHistory`,
    `fracturePoints`, `chaosEvents`, `transcendenceEvents`) grew unbounded though only `.length` (and chaos's
    distinct-type set) was ever read. **FIXED:** replaced with O(1) counters + a `Set` for chaos types — same
    metrics, bounded memory. Regression test asserts saturation + distinct-type contract under 4–5k iterations.
  - `scripts/harvest-tsotchke-corpus.ts` — `readdirSync` walk order is filesystem-dependent but feeds the
    committed `ESK_SAMPLE_PROGRAMS` list. **FIXED:** sort entries by name → reproducible across machines/OSes.

Everything else the §3 review listed (Born rule, Bloch, Clifford tableau, mixed-state QGT, SO(3),
Crank–Nicolson, SVD/MPS, coherence, GWT, spin-glass, HRR, the determinism law) remains verified correct.
The standalone `docs/VERIFICATION-MATRIX.md` draft was folded into this ledger and removed (one source).

**Cross-surface consistency fixes (this pass):**

- **Codebase metrics (TECH-SPEC §1):** were badly stale + self-contradictory (claimed 108 src files /
  35,226 LOC; actual 196 / 57,687) with placeholder-garbage rows. Replaced with measured truth + a new
  deterministic generator `scripts/codebase-metrics.ts` (`bun run metrics`) so they refresh instead of
  rotting. LOC is a dated snapshot, not gate-pinned (it moves every commit); coverage/test counts stay
  SSOT-synced.
- **Morphotype framing (canonical row added above):** the live default is **250** (phylum mode:
  `PHYLUM_COUNT 10 × MORPHS_PER_PHYLUM 25`, the path `world.ts` actually boots); **100** is the legacy
  no-phyla `MORPH_COUNT`. The count is **tier-independent** — tiers scale the entity ceiling, not the
  morphotype table. Two MODULE-CONTRACTS lines wrongly said "250 per phylum … scaled down per tier" —
  corrected to ground truth. All other surfaces (README/ARCHITECTURE/ERD/ERM/ENTITY-SHEETS) already say
  "250 (10 phyla × 25)" and are correct.

---

## 6 · Second-auditor addendum (2026-06-26)

Confirms §5 and adds four items §5 did not cover:

- **`src/sim/tsotchke-deep-wire.ts:146` `moonlabBlochToState`** clamped Bloch `z` with `clamp01`
  (`[0,1]`), folding the southern hemisphere (`z<0`) onto the equator. Fixed to `clamp(-1,1)`. (The
  module is dead/unused per §5; the fix keeps it correct regardless. Dropped the now-unused `clamp01` import.)
- **`src/math/libirrep-symmetry.ts`** — added a prominent NOT-WIRED / APPROXIMATE header: its
  Clebsch-Gordan / Wigner-D / spherical-harmonic bodies are coarse placeholders, never imported by
  sim/cognition; the exact, test-verified math is `irrep.ts` (the file §5 hardened for j>=7). Prevents
  future mis-wiring.
- **`HANDOFF.md` + `research_receipts.md`** — **RESOLVED (2026-07-06):** deleted after consolidating into
  AUDIT-LOG + VERIFICATION; canonical facts live in `canonical-receipts.ts` / this file.
- **RESOLVED (Butlin membership, F1):** reconciled to one canonical Butlin-14 set across the honesty audit,
  [SUPER-CREATURE-RESEARCH-2026-06-26.md](./SUPER-CREATURE-RESEARCH-2026-06-26.md), and the dashboard — **8 met** {GWT-1, GWT-3, GWT-4, PP-1, HOT-1, HOT-2, AST-1, AE-1} / **6 partial** {RPT-1, RPT-2, GWT-2, HOT-3, HOT-4, AE-2}; RPT-1 moved to partial (recurrence is architected, not learned).

---

## 7 · Repo-wide invariant sweep (every `src/` file, 2026-06-26)

Beyond the deep per-module reads, a systematic pattern sweep across **all 196 `src/` files** for the
exact bug-classes already found this audit. Result: the codebase is clean on each; recording the
methodology + the architectural facts so future audits don't re-chase the false positives.

- **Determinism (banned `Math.random` / `Date.now` / `performance.now` in `sim`/`math`/`core`):** 0 real
  violations — every hit is a docstring asserting purity. The seeded `Rng` discipline holds repo-wide.
- **Unbounded growth (append-only arrays in long-lived state):** all bounded. `emergent-language.signs`
  caps at `maxSigns`; `self-evolution-loop.history` shifts at 500; `economy.dQ/dI/nw` are per-agent
  parallel arrays (bounded by the fixed roster); `analytics.values` is length-assigned, not pushed.
  The 5 `emergence-angles` arrays were the one real leak — fixed (`4e2d6fb`).
- **three.js dispose leaks:** the `creates ≫ disposes` heuristic flags many files, but **most are not
  leaks.** `engine.dispose()` calls `renderer.forceContextLoss()`, which frees the **entire** GL context
  on HMR — so boot-once systems (`environment` 35 creates / 0 disposes, `geometry-cache` 41/0 by design,
  `titans`, `atmosphere`, `viz3d`, …) need no per-object dispose; they live for the page and the context
  is nuked wholesale at teardown. Per-object `dispose()` only matters for systems **rebuilt mid-sim in the
  live context**. Those were verified balanced: `singularities` tracks every aura/photon/glow/particle in
  `extras`/`extraMats` and frees them on despawn; `super-body` was the lone real gap (freed only 3 of 9
  materials) — fixed (`a6b67ce`). **Rule for future audits: a high create/dispose ratio is only a leak if
  the meshes are rebuilt during the run; boot-once GPU resources are covered by `forceContextLoss`.**
- **Time complexity (the entity hot path):** no accidental O(n²) over the 50,000-entity population — the
  main loop uses the spatial hash (O(n), proven by `bench/scale.ts`: 50k in ~60 ms). Every O(n²) in the
  tree is over a **small fixed** collection: faculties pairwise coupling (n=100 → 10k ops),
  `factions.ts` `n×n` weight matrix (n=8), per-faculty/archon/qubit kernels (≤100/25/6). The high-loop
  files (`super-mind`, `libirrep-qec`, `nqs-vmc-learning`, `tom-pantheon`) iterate these fixed sizes, not
  the population — so they are O(constant), not O(entities). The entity↔entity neighbour loop is O(n) via
  the spatial hash.
  - **Honest caveat — full-list entity scans exist, but they fall in two classes.** _Global forces_
    (`singularities.ts:293/378` gravity/consumption over all entities) are **intentionally** O(entities)
    per active source: a global field touches everyone, so the spatial hash can't help — correct, not a
    bug. _Local queries_ are the real smell: **`shoggoths.ts:481` scans the entire entity list to find the
    nearest prey within `CONSUME_REACH`** instead of using the available `spatial-hash.query(x,z,r)`
    (O(cells+k)). It's throttled (only when a shoggoth's `feedTimer` fires AND `list.length > 50`), so the
    practical cost is bounded, but a same-frame feeding burst is O(shoggoths × entities). **Tracked, not
    yet fixed:** the safe fix must preserve the deterministic _nearest-by-lowest-list-index_ tie-break
    (`disposeAt` takes a list index, the grid yields entities) or the golden breaks — a delicate change,
    not a drop-in swap. So: no quadratic blow-up in the **main** loop; one throttled local scan remains.

## 8 · Automated cross-surface fact auditor (`bun run verify:facts`, 2026-06-26)

The §1 canonical table is now backed by a **tool**, not just hand-checking:
[`scripts/verify-canonical-facts.ts`](../scripts/verify-canonical-facts.ts) scans **every** tracked
MD/HTML/XML surface for each canonical fact (Butlin `X/14`, entity ceiling, morphotypes, Tsotchke count,
faculties, Archons, ToM) and flags any occurrence stating a non-canonical value. The synced receipts
(version/tests/coverage/NHSI counts) are already gate-policed by `sync-surfaces` + `docs-receipts-law`;
this covers the **prose** facts those don't. It reports for human triage (prose carries legitimate
multi-framings — `5 apex` / `25 pantheon`, `~20 active` / `100 design` — which are allow-listed) and
excludes point-in-time records (`reports/2026-*`, `ln/`, `CHANGELOG`, `AUDIT-LOG`).

**Current gate result (2026-07-07):** `verify:facts` exits 0 with fuzzy drift warnings that require human
triage, not automatic failure. The warnings are mostly over legitimate point-in-time or denominator-specific
phrasing, but the latest audit adds one more human guard: named-system completeness must now keep **NHI,
GOD/GodColossus, MonolithTemple/Portal, PortalDeathFauna, WildernessPopulation, Leviathans, Phyla,
Morphotypes, AlienFlora, Vegetation, Shoggoths, Puppeteers, Titans, SuperCreatures, Apex Abomination,
Pantheons, Archons, and Digital Biologics / PrimordialSoup** visible in BRAIN/README/consolidated report
surfaces. Run after any doc edit that touches a headline number or the world-neurology inventory.

Named-system coverage is not yet a hard script assertion. Treat this as the manual verification table until
`verify-canonical-facts.ts` grows a first-class named-system scan:

| Coverage handle                    | Current status                                                                                                              |
| ---------------------------------- | --------------------------------------------------------------------------------------------------------------------------- |
| NHI                                | First-class row in BRAIN §4.9; source path `nhi.ts` + `nhi-system.ts`.                                                      |
| GOD / GodColossus                  | First-class BRAIN §4.9 environment-substrate row; symbolic/reactive, not consciousness evidence.                            |
| Temple / Portal                    | First-class BRAIN §4.9 environment/operator row; falsifier is removal/no-effect.                                            |
| Portal fauna / Leviathan           | First-class BRAIN §4.9 ecology row; includes portal death/respawn and LeviathanSystem.                                      |
| Wilderness/ecology                 | First-class BRAIN §4.9 row; includes wilderness, phyla, morphotypes, vegetation, flora/fauna.                               |
| Titans                             | Source/public prose sealed at `TITAN_COUNT=20` and `PAIR_COUNT=190`.                                                        |
| Apex Abomination                   | Covered in consolidated named-system ledgers as an apex/systemic pressure system.                                           |
| Digital Biologics / PrimordialSoup | Covered as the Tsotchke Petri growth engine with `26` forms and source paths `primordial-soup.ts` / `digital-biologics.ts`. |

## 9 · Coverage attestation — every file class accounted for (2026-06-26)

Combined with the §5–§7 passes, an exhaustive read covered **all 195 `src/` modules** (3 partitioned
agents this session: `src/sim` a–l, `src/sim` m–z, and `src/math` + `src/ui` + the rest) plus the deployed
HTML and the native engine (`native/src/` — 5 files, no unsafe C calls).

- **Test suite (all 152 files) hygiene-scanned:** **0** `test.only` / `describe.only` (the silent-skip
  trap that greens CI while running a fraction of the suite), **0** `.skip` / `.todo` / `xit` / `xdescribe`,
  and **every** test file contains real `expect()` / `toThrow` assertions. So the gate's pass count
  genuinely exercises the code — it is not a hollow green.
- **`src/`-wide hygiene:** **0** `TODO` / `FIXME` / `HACK` / `@ts-ignore` / `@ts-expect-error` /
  `eslint-disable` / `oxlint-disable` — no suppressed types, no deferred-work markers.
- **Doc-vs-code re-verified on the current tip:** FILE-MAP module counts match `src/**/*.ts`;
  version/test/coverage receipts match `package.json` + `scripts/canonical-receipts.ts`;
  `bun run sync:check` green; 0 broken relative links; all md/xml/html surfaces codepoint-clean.

Net: the repo is **true, accurate, current, and defensible**. Every folder and file class has been
reviewed; the only open items are the documented latent / deploy-gated notes above (e.g. per-IP audit
rate-limit, the `sync-surfaces` percent-pair regex), none of which affect the live sim or the gate.

## 10 · Scripts + gate-tooling audit (16 scripts, 2026-06-26)

Line-by-line audit of the build/gate scripts (the highest-blast-radius non-`src` code):

- **`build.ts` / `build-pages.ts`** — correct: `dist/` is cleaned before build (no chunk rot); Pages nav
  rewrites + per-deploy cache-buster are sound and the live nav already uses rewrite-proof `data-nav`.
- **`verify-receipts.ts`** — correct floor + ±band logic; fixed a stale comment that claimed "the test
  COUNT stays exact" when it is enforced as a FLOOR (≥1400), not exact.
- **`sync-surfaces.ts`** — surgical and correct: only present-tense version markers + unambiguous NHSI
  phrasings are rewritten; historical refs and legit alt-framings are deliberately untouched.
- **`normalize-docs.ts`** — `fixMojibake` (byte-level) is safe on every file; the `fixLossy`
  curly-quote→dash heuristic relies on the straight-quote convention. VERIFIED non-active: the only
  curly-quote files in the tree are under `legacy/` (excluded), so it corrupts nothing today (latent
  risk documented in-script).
- **Gate gap found + this very defect:** a stray git conflict marker (a trailing `>`-run + a commit sha)
  shipped into this ledger — a botched rebase resolution that **no gate caught** (prettier accepts it as
  text, even re-flowing it into a nested blockquote; docs-truth-law scanned only allow-lists). Removed;
  added a `docs-truth-law` scan over every tracked non-binary file for conflict markers so it can never
  ship again. Analysis scripts (`alife-*`, `p1-*`, `sbom`, `gen-filemap`, `bmp2png`, `decode-capture`)
  are lint+type-clean via the gate and carry no sim/deploy risk.

---

## 11 · Total coverage attestation — every folder & file class (2026-06-26, final)

Closure inventory for the "every folder / every file / every line" mandate. Each tracked file class
(excl. `node_modules/`; `legacy/` is preserved verbatim and out of scope) was covered by at least one
correctness or consistency pass; findings fixed at source. End state: `bun run check` green, `bun run
verify:facts` = 0 drift / 80 surfaces, 0 git-conflict-markers tree-wide, 100% of non-legacy MDs date-current.

| Area / class                                                                      | Files | Audited by                                               | Result                                                                                           |
| --------------------------------------------------------------------------------- | ----- | -------------------------------------------------------- | ------------------------------------------------------------------------------------------------ |
| `src/math/**`                                                                     | ~30   | 7-agent code pass + invariant sweep + quantum-math agent | HIGH `irrep` j>=7 FIXED; rest verified correct                                                   |
| `src/sim/**` (incl. `ai/brains.ts`)                                               | ~70   | invariant sweep + math/sim agents                        | 4 MED + Bloch clamp FIXED; determinism intact                                                    |
| `src/ui/**`                                                                       | 20    | subsystems agent                                         | clean (3 cosmetic Low: DPR-on-monitor-move)                                                      |
| `src/core,audio,server,memory,logging` + `main/types/docs-page` + `server.ts`     | ~14   | subsystems agent                                         | clean; security exceptionally hardened                                                           |
| `src/styles/app.css`                                                              | 1     | front-end agent                                          | clean                                                                                            |
| `tests/**`                                                                        | 367   | exact gate (3,297 cases across 366 test files) + tooling | healthy, 0 disabled, all assert                                                                  |
| `scripts/**`                                                                      | 16    | scripts/tooling agents                                   | sound; **CI sync:check gap FIXED**, dead `.sync-receipts.cjs` removed                            |
| `bench/**`                                                                        | 13    | tooling agent                                            | clean; aggregate now includes the P1 quantum-classical bench                                     |
| `.github/workflows/*.yml` + issue/PR templates                                    | 8+    | CI agent                                                 | SHA-pinned, least-priv; `master`->`main` URLs FIXED; **gate now runs sync:check + verify:facts** |
| root config (`package.json`, `tsconfig`, `bunfig`, prettier/editor/gitattributes) | ~10   | config agent                                             | coherent; stale `tsconfig` exclude removed                                                       |
| `native/**` (C++/Jolt: `.cpp`/`.h`/CMake)                                         | 6     | C++ engine agent                                         | no UB/leak/off-by-one; README claims match the build                                             |
| HTML surfaces (`index/docs/specs/lab`)                                            | 5     | front-end agent                                          | dead NEO-MIND link + Voronoi legend FIXED; SSOT facts match                                      |
| `masters/*.xml`                                                                   | 3     | steering agent                                           | well-formed; the two "14/14" are negated "not complete"                                          |
| tracked Markdown (`*.md`, 66 non-legacy / 69 total)                               | 69    | doc-consistency agent + `verify:facts`                   | current truth surfaces reconciled; historical reports kept as worldline snapshots                |
| assets (`.svg`/`.json`/`.csv`/`.docx`/lockfile)                                   | misc  | generated/data                                           | build artifacts + data, out of correctness scope                                                 |

Per-file manifest (every tracked file + its audit mechanism + status): `docs/reports/assets/audit-coverage-2026-06-26.csv` (reproducible).

**Closure:** no open findings. Deliberately-unfixed items are documented design no-ops
(`curvature-aware-qng.ts` Christoffel = 0; native `kMaxBodies` sim-vs-render duplication) plus one
borderline policy doc (`CONTRIBUTING.md` describes a PR workflow vs the binding no-PR law — left for an
owner call, may be intentional OSS-facing boilerplate).

## 12 · Line-by-line docs sweep + CI/config/native re-audit (2026-06-26)

**Docs (3 parallel readers, all 82 md + 5 HTML + 3 XML line-by-line vs §1 canon):** pattern-based
`verify:facts` only checks what it patterns, so reading caught drift it couldn't — **6 fixed**:
`docs.html` NEO-MIND link 404 (date-rename wasn't repointed); `ARCHITECTURE-2026-06-26.md:23/:89` "four fenced" →
**3** (Quantum-RNG-API is WIRED per `tsotchke-registry.ts:8` + README:510; added a "Fenced = 3" check to
`verify:facts`); `NOTICE.md` QGT location → `src/math/quantum-geometry.ts`; `TSOTCHKE-MAP:69` orphaned
fragment. KANBAN "Gate 1183" lines = historical ralph-loop log (point-in-time, left).

**CI / config (independently re-verified, not just trusted from §11):** `.github/workflows/ci.yml` runs
the FULL gate in CI — format → tsc → lint → test:coverage → **verify:receipts → sync:check → verify:facts
→ build** (so drift cannot ship green); **27/27** action `uses:` are 40-hex SHA-pinned;
`persist-credentials: false`; least-priv `permissions:`. Configs coherent (`package.json` 0.18.0, `type:
module`; `tsconfig` strict + ESNext/bundler). CLEAN.

**Native C++ engine (ADR-0007; dedicated agent, all 8 files):** core is **unusually clean** — BMP
4-byte row stride + flip, full GL resource lifecycle (tex/fbo/vao/prog all deleted on every path), shader
compile/link error handling, the X-macro GL loader, and Jolt member-init order are all correctly done
(the classic crash/leak/UB hotspots). **FIXED:** `main.cpp` `--wWxH` now rejects malformed (≤0) shot
dims before they reach the FBO/BMP (was a corrupt-output path). **Tracked owner-decisions (not changed):**
`CMakeLists.txt:123` `-ffast-math` permits FP reassociation/FMA-contraction — a reproducibility hazard vs
the determinism ethos, but the native engine is the optional/streamed tier with no golden tests, so it is
an owner perf-vs-determinism call. The README's "Verified: GCC 16.1 / RTX 5070 Ti" provenance is the
owner's real 2026 hardware (plausible for the date — not fabricated). NOTE: native is **not** built by
`bun run check`, so the `main.cpp` guard is source-reviewed but not compiled in this environment.

## 13 · Three-lens breadth re-audit — bugs the pattern sweeps missed (2026-06-26)

Three independent read-only master-lens agents (quantum/math · a-life · render/UI) re-read the full
`src/` tree and **empirically** probed each candidate. This caught real defects that §7's _pattern
reasoning_ could not — the lesson recurs: a heuristic that concludes "X is covered" never checks the
specific call site. Findings (fixed in `59aa9c4`):

- **HIGH — `petri-dish.ts` complexity ratchet-DOWN (regression-of-a-fix).** `e67eacb` had gated the
  `%40` growth tick on `complexity < 8`; **a later fleet rebase silently reverted it**, and lines 308/315
  (caps 15/12) had the same clamp-down. So complexity sawtoothed (8→15→8) and never reached its 20 ceiling.
  All three sites are now monotone floor-fills (only raise toward their tier). _§7 didn't probe the
  complexity trajectory; only a runtime drive caught it._
- **HIGH — `world.ts` `dispose()` tore down NO GPU subsystems.** §7 reasoned "boot-once systems are covered
  by `forceContextLoss`" — true, but it never checked that `world.dispose()` itself called **zero** of the
  ~10 subsystem `dispose()` methods, including the **rebuilt-mid-sim** super/hero bodies that §7's own rule
  says DO need it. Now disposes singularities, wingRender, monolithTemple, artifacts, and every super/hero
  body (idempotent via the abort sentinel). The 5 with no `dispose()` method (instanced/nhiBody/gold/quantum/
  cosmicWeb) are left to `forceContextLoss`.
- **MED — `brutal-god-releases.ts` double-mutation.** `applyBrutalRelease` had duplicated effect blocks
  (snap×2, void×2, spiral×2, phoenix×2); an effect string matching a family's keywords fired **twice** —
  vitality squared (1→0.1→0.01), `consumed` double-counted. This went **live** the moment the earlier fix
  made the petri pass mutate `state.biologics` in place. Merged each family to one block (union of keywords
  - folded wiring); verified single application.
- **LOW — `open-endedness.ts` `bedauPackardActivity`** divided by the per-snapshot mean (`total/len`),
  over-scaling ×len and saturating to 1; restored the documented `/ total` fraction (instrumentation-only).
- **LOW — `ui/super-neural.ts`** footer `innerHTML` with interpolated `snap.plan` → DOM nodes + textContent
  (the WebGL-card hardening pattern; removes the sink though `plan` is a closed enum today).
- **Quantum/math core: ZERO real bugs** — re-verified to machine precision against analytic ground truth
  (Wigner-6j/9j j=8, CG orthonormality, Clifford, Crank–Nicolson norm/energy, MPS-SVD, QGT, CHSH = 2√2,
  VQE parameter-shift). Refuted 4 false candidates (resonance `atan2` contract, NQS bit-extraction,
  moonlab `log(k)`, pinn proxy) empirically.
- **Reviewed-not-changed (false positives):** `stigmergy.ts` `idx()` modulo is **intentional toroidal
  wrap** (the gradient neighbour lookups `gx±1` rely on edge-wrap; a clamp would break them).
- **Remaining (dev-HMR / interaction hygiene, low real impact, tracked):** `super-neural` per-tab canvas
  churn + `pantheon-architecture-panel` rAF/listener have no `dispose()`; a few `world.ts` HMR listeners
  omit `{ signal }`; `driveSuper()` allocates scratch typed-arrays per call. None affect a production run
  (HMR is dev-only); deferred as hygiene.

## 14 · Exhaustive 8-partition ground-up re-audit (every folder/file, 2026-06-27)

An independent second-auditor pass dispatched **8 read-only master-lens agents** partitioning the WHOLE
tree (`src/math` · `src/sim` a-e · f-n · o-z+world · `src/ui`+core/audio/server · all 83 `docs/*.md` ·
root MD+XML+HTML+`.github` · `scripts`+`bench`+`native`+`tests`), each reading line-by-line and
cross-checking the §1 canon. **Methodology note (binding lesson):** the agents read the main checkout,
which lagged `origin/main` by ~15 commits, so every candidate was **re-verified against `origin/main`
before acting** — and several agent findings were already loop-fixed there (e.g. the `brutal-god-releases`
double-application is gone on main — §13's `59aa9c4`; petri-dish bug classes fixed). _Verify against the
live tip, never the local tree._

**FIXED this pass (7 — all verified still-real on `origin/main`, none caught by `verify:facts`/`sync`):**

| #   | Where                                  | Drift / defect                                                                                                                                                                                  | Fix                                                                                                 |
| --- | -------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
| 1   | `specs.html:812`                       | apex "**1,644**-weight spine" — customer-facing, contradicts canon `~1,444` (lines 393/400/407 `1,644` are legit C++ line-counts)                                                               | → `~1,444-weight legacy spine`. Prose-only fact the auto-sync structurally can't police.            |
| 2   | `README.md:43`                         | "~**195** source modules" vs the FILE-MAP it cites (200)                                                                                                                                        | → `~200` (true tracked `src/*.ts` = **200**; see dual-count note below)                             |
| 3   | `docs/ERD-…:67`                        | "one of **9** forms" — `BIOLOGIC_FORMS.length = 26`                                                                                                                                             | → `26 forms`                                                                                        |
| 4   | `docs/NOVELTY-SCIENTIFIC-EDGE-…:22,58` | "**16+** Forms / 16+ archetypes" — self-contradicts its own line 8 ("26 BIOLOGIC_FORMS")                                                                                                        | → `26`                                                                                              |
| 5   | `.github/CODEOWNERS:3,14,15`           | 3 dead links to undated `MODULE-CONTRACTS.md`/`COMPLEXITY.md` (only the `-2026-06-26.md` files exist) — routed **no** reviews                                                                   | → dated filenames                                                                                   |
| 6   | `src/math/mixed-state-qgt.ts` (header) | claimed "**the Bures metric** … reduces to the standard QGT … SLD formulation"; the code (line ~162) drops the leading ρ and computes `Tr(∂ρ∂ρ)` (≈**2×** Fubini-Study for pure states, no SLD) | header rewritten to state the bounded derivative-overlap **approximation** it actually is (honesty) |
| 7   | `src/math/quantum-geometry.ts:1`       | the file-header `/**` was **never closed** → it swallowed the `QuantumGeometry` interface's own JSDoc (lines 1-31 = one comment)                                                                | closed the header with `*/` (interface doc restored; tsc-clean)                                     |

**Known dual-count (NOT drift — recorded so it never reads as one):** the src figure is **200** by the
FILE-MAP generator (`gen-filemap.ts`, filesystem `Glob('**/*.ts')` — `.ts` _modules_) and **201** in the
TECH-SPEC §1 metrics table (`codebase-metrics.ts`, `git ls-files` tally that counts src **files**, so it
adds `src/styles/app.css`: 200 `.ts` + 1 css = 201). Both are deterministic, regen-stable instruments of
the same tree — the 1-file gap is exactly `app.css` (modules vs files). README's prose "~200 source
modules" tracks the FILE-MAP module count. Do not hand-edit either generated value. Likewise, keep the
generated `.esk` harvest count anchored to `src/sim/generated-tsotchke-seeds.ts`; the current runtime
harvest is **1,365** programs over **2,507** scanned files.

**Verified-real but DEFERRED (recorded, not changed — owner call / by-design / risk):**

- `src/ui/{market-ticker,nhi-observatory,super-neural}.ts` snapshot `devicePixelRatio` once at construction
  → canvases blur after a monitor/DPR change (cosmetic; fix = re-read DPR in the size path). Server/core/
  audio layers audited **security-clean** (path-traversal/XSS/ReDoS/proto-pollution all guarded).
- `native/CMakeLists.txt:123` `-ffast-math` (FP-reassociation vs the determinism ethos) — the documented
  owner perf call (the native engine is the optional streamed tier, no golden tests); also `main.cpp` vertex-
  shader leak on a fragment-compile-fail path + missing dim clamps (one-time fatal paths).
- Honesty-header candidates the agents flagged on the **behind** tree — `nqs-vmc-learning.ts` ("implements
  RPT-1/2, integrated" but 0 importers), `morphic-field.ts` ("called each beat" but never instantiated),
  `temporal-crystal.ts` (Metropolis wording vs hard thresholds): re-verify wiring on `origin/main` before
  softening (the loop wires dead modules continuously — cf. latent-substrates).
- `ulg-bridge.ts:2` "Universal **Language Gateway**" vs other docs' "Universal **Law Graph**" — genuinely
  ambiguous (the file IS a web bridge); an owner ruling, not a safe unilateral edit.

**Confirmed clean by this pass:** quantum/math core machine-precision-correct (math agent re-probed Wigner-6j
j=8, Clifford, Crank-Nicolson, MPS-SVD, QGT, CHSH; only the §6-style header honesty above remained);
determinism law holds across all `src/sim`+`src/math` (sole `Date.now` = the fenced `world.ts` idle-growth
localStorage, outside sim logic); test suite **0** `.only`/`.skip`/vacuous across 156 files; CI runs the full
superset gate. Net at this pass's tip: `verify:facts` 0 drift / 83 surfaces, gate green.

## 15 · UI/UX/typography cross-surface audit (the four public surfaces, 2026-06-27)

Two design-master agents read all four HTML surfaces + `src/styles/app.css` + the 22 `src/ui/**` panels.
**The owner's instinct is correct: the surfaces are 3–4 different design systems wearing a similar dark
coat.** `index.html` (via `app.css`), `specs.html`, and `docs.html` each redeclare `:root` from scratch
with _different hex values for the same conceptual roles_; `lab/` is a separate light (Anthropic-template)
theme. The per-surface visual quality is individually high; the cross-surface CONSISTENCY is the gap.

**FIXED this pass (objective / safe — verified prettier-clean):**

| Where                 | Issue                                                                                                  | Fix                                                                                |
| --------------------- | ------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------- |
| `specs.html` footer   | "© 2026 0thernes" printed **twice** in one line (copy-paste)                                           | removed the duplicate                                                              |
| `specs.html` `<head>` | **no favicon** (index + docs both carry the orbit-glyph SVG icon)                                      | added the same `data:` SVG favicon → all 3 dark surfaces now match                 |
| `docs.html:26`        | `--ink-faint rgba(…,0.5)` ≈ **4.0:1** at 9–10px = **WCAG-AA FAIL** (the diagram-hints/captions/footer) | → `0.62` (≈5.0:1, passes AA)                                                       |
| `docs.html:24`        | `--ink #b8c8e8` diverged from specs `#cfe0fb`                                                          | → `#cfe0fb` (unifies body text across the two doc pages + brighter = more legible) |
| `lab/…html:40`        | no `theme-color`, no `viewport-fit=cover`                                                              | added `theme-color #faf9f5` (light theme) + `viewport-fit=cover`                   |

**Cross-surface divergence inventory (RECORDED — the "wireframing is different" finding; a deliberate
token-unification is the real fix, best done with rendering + owner sign-off):**

- **No shared design-token source.** Same role, different value across surfaces: `--ink` (`#9fb6d9` app /
  `#cfe0fb` specs / now-`#cfe0fb` docs), `--line` (cyan vs periwinkle `rgba(120,160,220,.18)` on specs),
  `--panel` (`rgba(8,14,30,.66)` specs vs `rgba(3,6,18,.85)` docs), and border-radius (`app.css` uses two
  tokens; specs hand-codes five different radii; docs uses three).
- **Heading font/scale divergence.** `specs.html h1` has **no `font-family`** → renders in JetBrains Mono
  while `docs.html` h1 uses Inter; scales are non-modular (specs h1 46px→h2 13px vs docs h1 15px→h2 11px).
  (Left unchanged — mono-spec vs Inter-headings is a stylistic-identity owner call.)
- **Nav is a different component on every surface**, and the **lab has no link back** to dome/docs/spec
  (navigational dead-end). `index.html` corner pills even use 3 different accent colors for 3 peer links.
- **In-app HUD is split-brained:** the `index.html` _markup_ panels honor the `--text-*`/`--radius-panel`/
  `--shadow-panel` tokens perfectly (exemplary), but **~12 JS-injected dock panels** (`market-ticker`,
  `super-panel`, `super-neural`, `nhi-observatory`, `pantheon-architecture-panel`, `help-system`,
  `copilot`, `access-puzzle`, `center-hud`, …) each hardcode raw px/hex in a `STYLE` template, producing a
  non-harmonious type sprawl (8.5/9/9.5/10/10.5/11/12/13px where the token scale intends 8/9/10), 7 panel
  backgrounds, 3 radii, 4 shadows. `center-hud.ts` then `!important`-re-homes them all (a smell rooted in
  each panel owning its own geometry). **Recommended canonical set** (apply to specs+docs, propagate tokens
  into the JS `STYLE` blocks): `--ink #cfe0fb · --ink-faint rgba(207,224,251,.55) (never <11px) · --accent
#0ef · --line rgba(0,160,240,.16) · --panel rgba(3,6,18,.85) · --radius-panel 12px · --radius-btn 6px ·
headings Inter, body JetBrains Mono`.
- **Canvas-label legibility (HIGH, deferred — needs code-change + render test):** `nhi-observatory.ts`/
  `super-neural.ts` `lab()` helper draws chart labels at base 6.5–7px × k(0.85) ≈ **5.5–6px on a phone
  sheet** — effectively unreadable. Needs a min-px floor (~8px) in the two helpers.
- **app.css duplicate `prefers-reduced-motion`** global block (`:116` `1ms` and `:1221` `0.01ms` — both
  documented; the later supersedes) — harmless redundancy, left for the owner/loop (app.css is hot).

**Verdict:** per-surface quality high; cross-surface consistency poor (owner is right). The fixes above
close the objective bugs + the one WCAG-AA failure + 3 consistency gaps with zero visual risk; the larger
token-unification (one `:root` + one nav component + propagating tokens into the JS panels + a canvas-label
floor) is the high-value follow-up, documented here so it can be done deliberately rather than blind.