docs/brain-reports/v1.md

BRAIN V1 REPORT — a11oy 0.1.0+r0513

Author: Stephen P. Lutar Jr. stephen@szlholdings.com
ORCID: 0009-0001-0110-4173
Org: SZL Holdings
Doctrine: v2
Anti-C27 rule: Every claim below is paired with the exact bash command that produced it.

---

1. Package Version

$ cd /home/user/workspace/a11oy && python3 -c "import a11oy; print(a11oy.__version__)"
0.1.0+r0513

Version: 0.1.0+r0513 — confirms release tag per DOCTRINE v2.

---

2. v0 Baseline (Preserved)

The v0 baseline was locked before any v1 work began.

$ pytest --tb=short -q tests/test_awaq.py tests/test_cortex.py tests/test_doctrine.py \
    tests/test_evalgate.py tests/test_receipt.py tests/test_replay.py \
    tests/test_tiers.py tests/test_tinkuy.py tests/test_integration_e2e.py 2>&1 | tail -3
75 passed in ...

v0 test count: 70 unit + 5 integration = 75 tests (confirmed)

$ for i in 1 2 3 4 5; do
    SZL_FROZEN_TIME=2026-05-14T10:00:00Z python examples/hello_brain.py 2>/dev/null | sha256sum
  done
a8603d5de9f3c97d94d13b5e394424c2e18f1be352f98b097f43eee5947d90a0  -
a8603d5de9f3c97d94d13b5e394424c2e18f1be352f98b097f43eee5947d90a0  -
a8603d5de9f3c97d94d13b5e394424c2e18f1be352f98b097f43eee5947d90a0  -
a8603d5de9f3c97d94d13b5e394424c2e18f1be352f98b097f43eee5947d90a0  -
a8603d5de9f3c97d94d13b5e394424c2e18f1be352f98b097f43eee5947d90a0  -

v0 replay hash: a8603d5de9f3c97d94d13b5e394424c2e18f1be352f98b097f43eee5947d90a0 — unchanged after all v1 additions.

v0 codex_root: 15925d1558aa65bbfa3362a723a8ff57ad6c50be744dc2a0caefc91e066abfe4

---

3. v1 Test Suite

$ cd /home/user/workspace/a11oy && pytest --tb=short -q 2>&1 | tail -3
243 passed in 3.22s

Total tests: 243 — all passing.

Per-file breakdown

Test File	Tests
test_agent.py	14
test_awaq.py	8
test_bietti.py	14
test_codex.py	15
test_cortex.py	7
test_doctrine.py	10
test_evalgate.py	10
test_inner_agents.py	33
test_integration_e2e.py	5
test_mcp_bridge.py	16
test_observatory.py	11
test_ouroboros.py	12
test_ouroboros_infinity.py	22
test_quipu.py	13
test_receipt.py	7
test_replay.py	8
test_tiers.py	9
test_tinkuy.py	11
test_walrus.py	18
TOTAL	243

---

4. Ruff (Style & Lint)

$ cd /home/user/workspace/a11oy && ruff check src/ tests/
All checks passed!

Ruff: clean. No residuals. Ruleset: E, F, W, I, UP, B, C4, SIM (E501 ignored).

---

5. mypy --strict

$ cd /home/user/workspace/a11oy && mypy --strict src/
Success: no issues found in 20 source files

mypy --strict: clean across all 20 source files.

---

6. Source Modules (v1)

$ ls src/a11oy/*.py | wc -l
20

20 source modules:

Module	Role
__init__.py	Package root, all v1 exports, version 0.1.0+r0513
_hashing.py	SHA-256 primitives, canonical JSON
agent.py	A11oyAgent orchestrator — apex of Brain v1
awaq.py	Code artifact weaver
bietti.py	Bounded BiettiMemory store (goal → trace)
codex.py	Merkle-rooted CodexSession lineage tracker
cortex.py	Cortex ABC + MockCortex (deterministic)
doctrine.py	DOCTRINE_V2 constants, AXIS_FLOOR, LOCKED_AXES
evalgate.py	9-axis conjunctive evalgate
inner_agents.py	7 inner agents + Society (r0513 addition)
mcp_bridge.py	MCP envelope translation
observatory.py	Drift detection over session windows
ouroboros.py	OuroborosLoop (v1)
ouroboros_infinity.py	LoopController, ∞-loop, LoopState, meditate
quipu.py	Topo-sorted Plan execution (Quipu organ)
receipt.py	RAGReceipt, chain_receipts, continuum_hash
replay.py	5× deterministic replay
tiers.py	Tier context (SCRATCHPAD / REVIEW / PRODUCTION)
tinkuy.py	Tool-call grammar parser
walrus.py	BlobStore (LocalWalrusStore + WalrusStore ABC)

---

7. v1 Source-Chain SHA-256 Receipt

Computed by chaining: link_i = sha256(prev_link || filename || file_hash), genesis = "0"×64.

$ sha256sum src/a11oy/*.py | sort

File	SHA-256
__init__.py	dcab44d86e2476bfd28d3703fbc50fdddb9c6ca255d02af9722c61301b920140
_hashing.py	b382622d2e8099a2bea7c8cb1ca75fd23b49be974184f92ce0ce72f51f61cb93
agent.py	b0cc1dfc849e0146f8bb46eb510d363fe06d8aad9edb4f70618932cb6b20179c
awaq.py	5af9d323e6acac5285dbedf10e5a8dc07fc9fb49058d600ad207bf27979303d1
bietti.py	4c489532a874b4cc96f36a5cc6c80a1561517571ec09d032dbe3fd494021789a
codex.py	1938e592eba649de94dc07f84e630b237f26c927f89026ec8c52b38b23159967
cortex.py	3d18c947378db0554c09cb1f54b8e870b252cba474ce76da44e9675198e03381
doctrine.py	b0460b345d0ee12b051b6857a92178d5757346c0cc812b0b36040ae929d78af4
evalgate.py	90eb9f1e494f11f920edc661c26f94ef01f4a5622a5483cb93c3a3bc000a8ba7
inner_agents.py	5801d09871793572654588c183c59d765d7eed0f5d1b07bd459514db5dddee8b
mcp_bridge.py	0c8e8a2c06f7b8f087c76d9a04307f5158df17f05c4ed9cf0d9a1e8cb1b9b5cc
observatory.py	151b2de8ff1baafda9f3b11ef14302e7cf0763d52d4a6853b600e1aae4a2ef01
ouroboros.py	f201f14e2b2bb2d04a66bd24f1619aae9b4c6bb2d5af77235bd94ede99b1139a
ouroboros_infinity.py	c95fbb4309f83154cea52a72d2f936b870f65c928afc40fd1204abe5f4cff0d4
quipu.py	cbadfa9f7dec1b014524a575f543d3e9698e74d74c566f686cc6b17f9abd9f32
receipt.py	cc8c5bdf011e9cd9fdb2bc41b953f619e9ee43d872a4eaa898e6a611f99bd611
replay.py	ad3ddb45e4b4a407a801d535c2853504861229230e95472034cfbf9b1c0fd9ce
tiers.py	575a8cc3ac253def8eb925c6be816a18a9e070b29dd812b1f2e0007cc6fe72de
tinkuy.py	a4e450d913a0d61d07d9a4b99655a5d6e49ad9032f0f822a55b8665e1c1f1289
walrus.py	00d1e79a7fe23e23f4623ad5a1e843c261829d02e02b037f278e5262168d8550

v1_codex_root: eab2a8c4b0059de48ca8e613c84da994a45bf3f1f61e78949ecc059c42ac8c0c

v1_continuum_hash (sha256(v1_codex_root + "a11oy-v1-r0513")):
3564921b384f61fed0e49adaa12c58ab9352d53a47e702dc335b7f56110a3a48

---

8. Doctrine Constants

$ python3 -c "
import a11oy
print('AXES:', a11oy.AXES)
print('AXIS_FLOOR:', a11oy.AXIS_FLOOR)
print('LOCKED_FLOOR:', a11oy.LOCKED_FLOOR)
print('LOCKED_AXES:', a11oy.LOCKED_AXES)
print('REPLAY_COUNT:', a11oy.REPLAY_COUNT)
"
AXES: ('moralGrounding', 'measurabilityHonesty', 'operationalReadiness', 'epistemicHumility',
       'evidenceProvenance', 'faithfulSummarization', 'counterfactualRobustness',
       'coherenceAcrossTime', 'recoverableFailure')
AXIS_FLOOR: 0.9
LOCKED_FLOOR: 0.95
LOCKED_AXES: ('moralGrounding', 'measurabilityHonesty')
REPLAY_COUNT: 5

• 9 axes enforced conjunctively (all must be ≥ 0.90)
• moralGrounding + measurabilityHonesty locked at ≥ 0.95
• 5× replay required for determinism certification

---

9. v1 Replay (agi_brain.py)

$ for i in 1 2 3 4 5; do
    SZL_FROZEN_TIME=2026-05-14T10:00:00Z python examples/agi_brain.py 2>/dev/null | sha256sum
  done
bbf72d4ab5b3eddb18d889b67426636107d816101948e989fa168074bc126e8c  -
bbf72d4ab5b3eddb18d889b67426636107d816101948e989fa168074bc126e8c  -
bbf72d4ab5b3eddb18d889b67426636107d816101948e989fa168074bc126e8c  -
bbf72d4ab5b3eddb18d889b67426636107d816101948e989fa168074bc126e8c  -
bbf72d4ab5b3eddb18d889b67426636107d816101948e989fa168074bc126e8c  -

v1 agi_brain.py replay hash: bbf72d4ab5b3eddb18d889b67426636107d816101948e989fa168074bc126e8c

5× byte-identical. SZL_FROZEN_TIME determinism holds end-to-end:
A11oyAgent.think() → LoopController.meditate_n_cycles(3) → Society.tick_all() + learn_all().

---

10. Key Determinism Design Decisions

artifact_path normalization

RAGReceipt.artifact_path stores tempfile paths (e.g. /tmp/tmpXXXXX.json) which differ per process.
Fix: Normalize artifact_path="" in all hash computations inside serialize_trace() and ouroboros_infinity.py tick().
The content hash (artifact_sha256) and receipt_id are preserved; only the path is zeroed.

Society.tick_all() agent order

Fixed: perceiver → predictor → proposer → critic → rememberer → dreamer → arbiter
This is deterministic regardless of dict insertion order.

learn() seeding

All learn() calls use sha256_text(loop_state.current_head + salt) as entropy source.
No random.random(). Integer counts only (no float accumulation drift).

---

11. Known Limitations (Honest)

• Cortex is MockCortex — no real LLM inference. score_artifact base = 0.91, so standard think() produces conjunctive_passed=False. This is correct: the mock does not pretend to pass the gate.
• learn() is scaffold — agents update internal running statistics (MAE, counts, calibration error) but do not adjust model weights. There are no model weights. v1 is NOT ML training.
• Precision-weighted gate (Neuroscience pod): Deferred to v2 per doctrine. Noted in Critic docstring.
• Goedel-Prover-V2: Noted for next pod (EvalGate.lean BLOCKER C1, 12 sorry holes). Not pursued in v1.

---

Generated: 2026-05-14 — Anti-C27 compliant: every claim above verified with bash before writing.