Eighty-eight files live here. This is what's in them.
If what you actually need to know is whether a given file is still true, that's DOC_STATUS.md — it splits everything into current, historical, generated, and standards. Start there when a document surprises you.
| USER_GUIDE.md | Using the app |
| OPERATOR_GUIDE.md | Running it on your own hardware — gates, diagnostics, tuning, debugging entry points |
| runbooks/ | 38 incident procedures, one per known failure mode. Written against aura doctor --bundle fields |
| SLO.md | What the runtime promises, measured by slo/ and gated in CI |
| PLATFORM_POSTURE.md | The five deliberate platform decisions (no RBAC, no SSO, single-tenant, manual DR, hash-allowlist plugins) and what enforces each |
| ARCHITECTURE_MAP.md | Generated dependency map — subsystems, edges, boot contract |
| ONTOLOGY.md | The formal ontology: continuants, occurrents, axioms rendered as test invariants |
| TERMINOLOGY.md | Internal poetic name ↔ sober technical label |
| RUNTIME_CONTRACT.md | Generated from health_contract.py |
| FMEA.md | Generated failure-mode registry |
| ENGINEERING_ADOPTION.md | The seven clean-room adoption waves and why each landed |
| COGNITIVE_ARCHITECTURE_ADOPTION.md | What was taken from Soar and ACT-R, the equation that fitted, and the one that didn't |
| MODEL_ROSTER.md | Every model lane — LLM, ASR, embeddings — and the measurement that put it there |
| WRITING_RULES.md | The nine patterns that read as machine-written, and the make writing gate that checks for them |
This is the part of the repo that exists to keep it honest.
| REALITY_REACH.md | Physical claims: typed contracts, reachability proof, and an open ledger. Read "Current Evidence" before crediting any physical result |
| CLAIM_BOUNDARIES.md · CLAIM_SURFACE.md | What may be claimed and where the edges are |
| ABLATION_LEGIBILITY.md | Run it with pieces switched off and see the measured delta — including the no-delta results |
| BEHAVIORAL_PROOF_STANDARD.md | The bar for autonomy and novel-output claims |
*_STANDARD.md (11 files) |
One evidence bar each. Evergreen — they change when the bar changes, not when the code does |
Long-running lines of work, each with its own ledger.
- Recursive latent cortex — the flagship programme. Start at RECURSIVE_LATENT_CORTEX.md, which is the landing page and carries the claims ladder; then INTRINSIC_RECURRENCE.md (the live training front), RLC_RECONCILIATION.md (why two negative results were void), RLC_WIRING_HANDOFF.md, RLC_SPARK_EXECUTION_LEDGER.md, RLC_COMMITMENT_SEARCH.md, RLC_SPARK_LITERATURE.md, RLC_KNOWLEDGE_SOURCE_MATRIX.md, SPARK_PRETRAINING_LEGS.md
- Consciousness and integration — WHOLE_SYSTEM_PHI.md, ORGANISMAL_WORKSPACE_THEORY.md, INNER_LIGHT_TEST.md, GHOST_SUBSTRATE.md
- Autonomy and self-modification — ULYSSES_COVENANT.md, ALLOSTASIS_ENGINE.md, AUTONOMOUS_ARCHITECTURE_GOVERNOR.md, GENERAL_ENVIRONMENT_AUTONOMY.md, RSI_VALIDATION.md
- Capability maps — OMEGA_CAPABILITY_MATRIX.md, FICTIONAL_AI_CAPABILITY_MAP.md, FLIGHT_RECORDER.md, DELIBERATE_PRACTICE.md
evidence/ and every file with a date in its name are point-in-time
records. They carry a banner saying so. They are not updated to match the
present — editing a July verdict in August is editing the record, not
correcting it.
Four files here are rendered from code. A manual edit survives until the next build and then vanishes.
make architecture-map # ARCHITECTURE_MAP.md
make fmea-doc # FMEA.md
make contract-doc # RUNTIME_CONTRACT.md
python tools/reqproof/progress.py # AURA_PROGRESS.mdCheck for drift without rewriting:
python tools/render_fmea.py --check
python tools/render_health_contract.py --check