68 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
root
|
8ec43e0721 |
phase 1.6 Gate 3b: deepface integration design doc (3 options + recommendation)
Per docs/PHASE_1_6_BIPA_GATES.md Gate 3b. Three viable paths for
populating BiometricCollection.classifications, sized + tradeoff'd:
Option A — Python subprocess per upload (no daemon)
~80 LOC, 0.5-1 day. Smallest integration. Reintroduces a Python
dependency the 2026-05-02 sidecar drop deliberately removed.
Option B — ONNX models in Rust (no Python at all)
~200-400 LOC + model-build pipeline, 5-7 days. Fully consistent
with sidecar drop. Need pre-trained models with appropriate
licenses (or train ourselves, multi-week). Adds face detection
preprocessing in Rust.
Option C — Defer; classifications field stays None
0.25 day. BIPA-safest position; substrate is forward-compatible.
Forces the question "do we actually need classifications?" to be
answered by a real product requirement, not by spec inertia.
Recommendation: **Option C (defer)**, conditional on confirming the
product requirement. Reasoning:
- All BIPA-load-bearing surfaces (consent + audit + retention +
erasure) ship without classifications
- Riskiest BIPA position is collecting demographic-derived data
without a documented business purpose
- Substrate accommodates A or B later in 1-3 days if real demand
surfaces
Open questions for J at the bottom of the doc — pick A/B/C is the
gating decision before any engineering happens.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
f1fa6e4e61 |
phase 1.6 Gate 3a: photo upload endpoint with consent gate
Per docs/PHASE_1_6_BIPA_GATES.md §1 Gate 3 (consent-gate substrate).
Deepface classification (Gate 3b) deferred to its own session — needs
Python subprocess design conversation after the 2026-05-02 sidecar drop.
What ships:
shared/types.rs:
- new BiometricCollection sub-struct: data_path, template_hash,
collected_at, consent_version_hash, classifications (Option<JSON>)
- SubjectManifest gains biometric_collection: Option<BiometricCollection>
with #[serde(default)] so existing on-disk manifests parse and
re-emit without drift
catalogd/biometric_endpoint.rs (NEW, ~600 LOC):
POST /subject/{candidate_id}/photo
- Auth: X-Lakehouse-Legal-Token, constant-time-eq compared against
same legal token file as /audit. Same 32-byte minimum.
- Content-Type: must be image/jpeg or image/png (415 otherwise)
- Body: raw image bytes, max 10MB
- 401: missing or wrong token
- 404: subject not registered
- 403: consent.biometric.status != "given" (returns current status)
- 403: subject status in {Withdrawn, Erased, RetentionExpired}
- 200: writes photo to data/biometric/uploads/<sanitized_id>/<ts>.<ext>
with mode 0700 dir + 0600 file, updates SubjectManifest with
BiometricCollection record, appends audit row
(kind="biometric_collection", purpose="photo_upload"), returns
UploadResponse with template_hash + audit_row_hmac.
Logic split: pure async fn process_upload() takes the headers-as-args
so unit tests exercise every branch without HTTP machinery; the
axum handler is just glue. 10 tests covering all 4 reject paths +
happy path + repeated uploads chaining + structural assertion that
the quarantine path is NOT under data/headshots/ (synthetic faces).
gateway/main.rs:
Mounts /biometric on the same condition as /audit — only when the
SubjectAuditWriter is present AND the legal token loads. Storage
root configurable via LH_BIOMETRIC_STORAGE_ROOT (default
./data/biometric/uploads).
Live verification on the running gateway (post-restart):
- GET /biometric/health → "biometric endpoint ready"
- POST without token → 401 auth_failed
- POST with token, no consent → 403 consent_required (status=NeverCollected)
- Flipped WORKER-2 to consent=given, POST → 200 with hash + path
- File at data/biometric/uploads/WORKER-2/<ts>.jpg, mode 0600
- Manifest biometric_collection field reflects the upload
- Audit row chain links cleanly off the prior validator_lookup row
- GET /audit/subject/WORKER-2 returns chain_verified=true, 2 rows
- Cross-runtime parity probe still 6/6 byte-identical post-change
Phase 1.6 status table updated: Gate 3a DONE, Gate 3b (deepface)
deferred. Calendar bottleneck remains counsel review of items 1/2/5/6.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
c7aa607ae4 |
phase 1.6 BIPA: scrum-driven fixes
Per 2026-05-03 phase_1_6_bipa_gates scrum (13 findings, 0 convergent).
1 BLOCK verified false positive, 4 real fixes shipped:
False positive (verified):
- opus BLOCK on attest:55 — claimed `set -uo pipefail` without `-e`
makes the post-python3 `if [ $? -ne 0 ]` check unreachable. Verified
WRONG: `X=$(false); echo $?` prints 1. Bash propagates command-
substitution exit through $? on the assignment line. The check IS
the python3 exit gate. Inline comment added to the script noting
the false positive so future scrums don't re-flag.
Real fixes:
1. opus WARN attestation:18 — schema fingerprint hashed names ONLY,
missing column-type changes. A column repurposed to hold base64
photo bytes under its existing name would pass undetected. Now
hashes "name<TAB>type<TAB>nullable=bool" per row. Re-run produced
evidence SHA-256 1fdcc9f1... (vs old 230fffeb..., reflecting the
broader fingerprint scope).
2. opus WARN gate_4_test:60 — definition regex didn't catch
object-literal property forms (`const t = { FEMALE_NAMES: [...] }`)
or TypeScript class fields (`class L { public NAMES_X: string[] = [] }`).
Added two new patterns + a regression test
(Gate 4: object-literal and class-field bypasses are caught) that
exercises 5 bypass forms. 4/4 tests green; 1 minor regex tweak
needed mid-fix to handle single-line class bodies.
3. kimi WARN python3-reliance — script assumed pyarrow installed and
would emit a stack trace into the attestation if not. Added
`python3 -c "import pyarrow"` gate at top with clean install
instructions on failure.
4. opus INFO PHASE_1_6:200 — item 7 (training) silently dropped from
blocking set with bare "deferred" rationale. Now explicitly states
the deferral is conditional on small operator population (J + 1-2
named ops); item 7 re-promotes to blocking if population grows.
⚖ COUNSEL marker added.
Skipped (acceptable as ⚖ COUNSEL placeholders by design):
- kimi WARN consent template:30-day-SLA (counsel decides number)
- kimi WARN consent template:email-placeholder (counsel supplies)
- kimi WARN parquet absence (env override exists; redeployment-aware)
- kimi INFO runbook manual-erasure (marked TODO when /erase ships)
- qwen INFO doc path/status nits (already addressed by file moves)
Tests: 4/4 Gate 4 absence test (incl. new bypass-coverage), 3/3
attestation evidence checks pass on live data.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
4708717f6b |
phase 1.6 BIPA gates — engineering wave (4 of 7 staged)
Per docs/PHASE_1_6_BIPA_GATES.md. Status table now reflects:
DONE (engineering-only, no counsel dependency):
- Gate 4: name→ethnicity inference removed from mcp-server.
Removal note in search.html:3372 + new Bun absence test
(mcp-server/phase_1_6_gate_4.test.ts) with 3 assertions:
walker actually scans files, regex catches synthetic positives,
no offending DEFINITION patterns in any .html/.ts/.js source.
3/3 pass.
ENG-DONE, signature pending:
- §2 attestation: scripts/staffing/attest_pre_identityd_biometric_state.sh
runs three checks against the live state:
1. workers_500k.parquet schema has no biometric/photo/face/image col
2. data/_kb/*.jsonl + pathway state contain no base64 image magic
bytes (JPEG /9j/, PNG iVBOR), no data:image/* MIME prefixes,
no field-name patterns ("photo", "biometric", "deepface_*")
3. data/headshots/manifest.jsonl is entirely synthetic-tagged
3/3 evidence checks pass on the live data dir. Generates a
signed-by-operator+counsel attestation document committed at
docs/attestations/BIPA_PRE_IDENTITYD_ATTESTATION_2026-05-03.md
with SHA-256 of the evidence summary so post-signature tampering
is detectable.
ENG-STAGED, awaiting counsel review:
- Gate 1 retention schedule scaffold at
docs/policies/consent/biometric_retention_schedule_v1.md (BIPA
§15(a)). Engineering facts (categories, 18-month operational
ceiling vs 3-year statutory cap, destruction procedure pointer
to Gate 5 runbook) plus ⚖ COUNSEL markers for the binding text.
- Gate 2 consent template scaffold at
docs/policies/consent/biometric_consent_template_v1.md (BIPA
§15(b)(1)-(3)). Required disclosures + plain-language summary +
withdrawal procedure + the structured fields the consent UI must
post to identityd.
- Gate 5 destruction runbook at docs/runbooks/BIPA_DESTRUCTION_RUNBOOK.md.
Triggers, pre-destruction checks (incl. chain-verified gate via
/audit/subject/{id}), procedure (legal-tier endpoint), automatic
audit row append (subject_audit.v1 with kind=biometric_erasure),
backup-window disclosure, monthly reporting cadence, audit-trail
attestation procedure cross-referencing the cross-runtime parity
probe.
BLOCKED on engineering design:
- Gate 3 photo-upload endpoint. Requires identityd photo intake
design + deepface integration scope. Deferred to its own session.
DEFERRED:
- §3 employee training material. Gate 5 runbook §7 may serve as
substrate; counsel decides whether a separate program is needed.
Calendar bottleneck is now counsel review. Engineering can stage no
further deliverables until either (a) Gate 3's design conversation
happens or (b) counsel completes review of items 1/2/5/6.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
ed1fcd3c26 |
specs: pathway_memory v1 + subject_manifests_on_catalogd v1
Two specifications addressing the framing J asked for after reading
the llms3.com blog: standardize what we have so future work doesn't
drift, and apply the local-first thesis to the audit problem instead
of the over-scoped SaaS-tier identity service.
PATHWAY_MEMORY_SPEC.md (~400 lines):
Documents the existing crates/vectord/src/pathway_memory.rs as a
spec — the third metadata layer alongside catalogd's data metadata
and playbook_memory's operational memory. Defines:
- PathwayTrace wire format
- pathway_id = SHA256(task_class | file_prefix | signal_class)
- file_prefix algorithm (first 2 path segments)
- pathway_vec: 32-bucket bag-of-tokens hash, fixed dim per spec
- Lifecycle: insert → revise → replay → probation gate retire
- Mem0 versioning (trace_uid + parent_trace_uid + version chain)
- Access patterns: query_for_hotswap / query_by_vec / list_versions
- PII risk surface (reducer_summary + final_verdict)
- Spec boundary: stable in v1 vs implementation-specific
No new architecture. Descriptive, not prescriptive.
SUBJECT_MANIFESTS_ON_CATALOGD.md (~400 lines):
The local-first audit-trail spec. Adds a fourth manifest type to
catalogd alongside dataset/view/tombstone/profile. NOT a separate
identity daemon. NOT Vault/KMS/dual-control JWT. Builds on
primitives catalogd already ships:
- SubjectManifest at data/_catalog/subjects/<id>.json
- Per-subject HMAC-chained audit JSONL
- Daily retention sweep using existing tombstone primitives
- Vertical-aware routing (healthcare → local-only)
- Legal-tier credential separate from gateway internal auth
~4 days estimated implementation effort vs 17-20 days for the
IDENTITY_SERVICE_DESIGN approach. Same defensibility for the
staffing-client launch window. Strictly additive to compatibility
with the v3 design if SOC2 Type II becomes a contract requirement.
These are SPECS — what the system already does (pathway) and what's
the smallest local-first thing that addresses the audit need
(subject manifests). Not 9-phase plans. Not new daemons.
The pathway spec is descriptive: writing down what exists so the
next person doesn't reinvent it. The subject-manifests spec is
prescriptive: J greenlights, implementation is days not weeks.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
dbcd05c5c5 |
audit docs: deprecation headers — over-scoped for local-only deployment
Today's PRD-line-70 reframe (everything runs locally) means the audit-trail docs I drafted earlier this session are over-engineered for J's actual deployment model. They were sized for SaaS-tier infra (Vault/KMS/S3 Object Lock/dual-control JWT/separate Postgres) — appropriate for a multi-tenant cloud service, wrong for a single-box local install. Adding clear deprecation headers so future sessions don't read these as authoritative and propose another 17-20 day plan involving cloud infrastructure that would re-violate PRD line 70. What STAYS valid (preserved in headers): - The legal use case (John Martinez worked example) - The IL/IN jurisdictional surface (counsel checklist) - The Phase 1 + 1.5 discovery findings (PII flow paths file:line) - Phase 1.6 BIPA gates (when real photos arrive) What's OVER-SCOPED (flagged in headers): - The 9-phase implementation plan - The identity service design (Vault/KMS/dual-control) Future v2 of these docs needs to be sized for local single-box: a few hundred LOC of local writers + signed local audit file, not 17-20 days of distributed-systems design. No code changes. Just doc-level guardrails for future scope drift. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
cd440d4cee |
audit phase 1.6: BIPA pre-launch gates — block identity-service backfill
Per IDENTITY_SERVICE_DESIGN v3 §5 Step 0, Phase 1.6 is hard
prerequisite to identityd backfill. This doc specifies the 5 gates +
2 supporting deliverables that must ship before real-photo intake.
Five gates (BIPA §15 compliance):
1. Public retention schedule — counsel writes; engineering files+hash
2. Informed written consent — counsel writes template; engineering
wires identityd consent-status enforcement
3. Photo-upload endpoint with consent enforcement — POST /v1/identity/
subjects/{id}/photo with hard 403 when biometric_consent_status
!= 'given'; quarantined storage path; deepface output isolated
to identityd subjects table (not synthetic-face manifest)
4. Deprecate name → ethnicity inference (mcp-server/search.html
lookup tables removed; Phase 1.5 §1B finding closed)
5. Destruction runbook — operator-facing; ties to identityd
/erase endpoint with biometric-specific erasure path; daily
sweep job for biometric_retention_until expiry
Plus:
- Cryptographic attestation that no biometric data exists
pre-identityd (per v3-B11) — defends against
infrastructure-as-notice plaintiff argument
- Employee BIPA-handling training acknowledgment
Engineering effort: ~4-5 days (one week to stage everything ready).
Counsel effort: ~3-6 weeks calendar (review cycles dominate).
Calendar bottleneck is counsel, not engineering.
Phase 1.6 exit = 7 checked gates + signoffs. Until done, identityd
backfill cannot proceed (per identity service design v3 §5 Step 0).
5 open questions for J + counsel: photo-upload UX, consent
mechanism (DocuSign/click/paper), named operator list, named
counsel for sign-off, public privacy policy URL.
No code changes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
8129ddd883 |
identity service: v3 amendments — second-pass scrum BUILD-WITH-CHANGES
Re-scrummed v2 across opus + kimi + gemini. All 3 verdict:
BUILD-WITH-CHANGES. v1 blockers verified RESOLVED. 12 new v2
findings folded as v3 amendments in §12.
Convergent v2 findings (≥2 reviewers):
v3-A1: mTLS CA root must NOT live in identityd (opus + gemini).
v3 fix: Vault PKI for CA, identityd as intermediate.
v3-A2: Dual-control public key registry must be tamper-evident
(opus + gemini). v3 fix: Vault KV with separate access
policies + server-issued nonces for replay protection.
Single-reviewer v3 amendments (10 more):
- B1: Step 8 fallback-to-SQL needs explicit 14-day time bound
- B2: NER drop-on-detect needs Prometheus alerting
- B3: legal-tier notification transport spec'd (signed Slack/email,
no PII in body, failure non-blocking)
- B4: Step 6 human review SLA flagged — ~7 months at 500/day for
~100k unknown rows; operational decision needed
- B5: Memory zeroing in Go is best-effort (Rust uses zeroize crate);
documented as not cryptographic-grade
- B6: purpose_definitions needs versioning + emergency revocation
(purpose_versions + purpose_revocations tables)
- B7: Cache invalidation needs erasure_generation atomicity
(subjects.erasure_generation int; gateway rejects stale-gen
cache hits) — replaces best-effort pub/sub
- B8: 15-min cooling-off period for dual-control issuance to
prevent emergency-bypass culture
- B9: NER calibrated test set with target recall ≥99.5% on
synthetic adversarial PII
- B10: S3 Object Lock in separate AWS account with write-only IAM;
root credentials held by external party
- B11: BIPA infrastructure-as-notice attestation in Phase 1.6 doc
- B12: Backup retention vs ciphertext-deletion erasure window
documented in RTBF runbook
Estimate revised v2 12-15d → v3 17-20d. Worth it — the cost is what
buys "I would build this" from 3 independent senior security
architects across 3 model lineages.
Must-have v3 items (block implementation): A1, A2, B1, B6, B7, B11.
Should-have (ship in Phase 5 if calendar tight): B2-B5, B8-B10, B12.
Re-scrum NOT recommended for v3 — diminishing returns; must-have
items are concrete fixes with clear acceptance criteria.
No code changes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
298fadce41 |
identity service: v2 — fold cross-lineage scrum findings + 4 'would not build' blocker fixes
Scrummed v1 across opus + kimi + gemini lineages via the new model fleet. 3/3 reviewers said 'I would NOT build v1 as written.' 4 convergent blockers, all resolved in v2: 1. Migration order wrong — backfill before validation creates dark database; if backfill bug, no production traffic catches it. v2 inserts BIPA-prereq Step 0 + shadow-write before backfill + shadow-read before cutover. 9-step migration with cryptographic attestation of completeness at quarantine. 2. Master key on disk + legal token static file = 'security theater' per all 3. v2: HashiCorp Vault Transit / AWS KMS for KEK (not sealed file). Legal token: split-secret short-lived JWT (max 24h), dual-control issuance (J + counsel both sign), revocable in <60s. 3. consent_status='inferred_existing' is BIPA prima facie violation (kimi+gemini explicit). v2 backfill uses 'pending_backfill_review'; biometric data NEVER backfilled — separate consent stream. 4. Healthcare default 'general' = HIPAA exposure window for every misclassified subject. v2 default 'unknown' with fail-closed routing (treat unknown as healthcare-equivalent until classified by manual review). Auto-escalation to healthcare on resume_text pattern match. Plus 12 single-reviewer additions: - mTLS mandatory between gateway↔identityd (kimi) - External anchor for audit chain: S3 Object Lock 7-year compliance mode, hourly + on-event commits (all 3) - Audit-log signing key separate from encryption KEK (opus) - Field-level authorization via purpose_definitions table (kimi) - Per-row encryption keys deferred to Phase 7 (kimi simplification) - pii_access_log itself needs legal-tier read auth (opus) - Synchronous cache invalidation pub/sub on RTBF (opus) - Outbound NER pass for Langfuse defense-in-depth (opus TOCTOU) - model_version_hash per decision row (gemini) - /vertical minimal-disclosure endpoint (kimi HIPAA min-necessary) - Auto-escalation healthcare on resume_text pattern (kimi) - Rate limiting + token revocation list (opus) - Oracle tests in audit_parity.sh (kimi SOC2 CC4.1) Architecturally simplified per scrum: - Per-row encryption keys deferred to Phase 7 (single DEK + HSM- wrapped KEK + ciphertext deletion is equivalent practical erasure with less complexity) - PDF render deferred (JSON ships first) - Training-safe export deferred (not critical path) Estimated effort revised 8-10 → 12-15 days. Worth it — every addition was a 3/3-reviewer convergent finding. Re-scrum recommended before implementation starts to verify v2 addresses the v1 blockers. No code changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
565ea4b32a |
audit phase 2: IDENTITY_SERVICE_DESIGN.md — full design doc
Incorporates J's confirmed answers (2026-05-03): - separate daemon (identityd) on :3225 / :4225 - signed JSON with PDF render for legal export - legal-only credential separate from admin token - Langfuse self-hosted (drops cross-border concern) - EU placeholder fields, not enforced - healthcare vertical routing — local-only models for healthcare PHI - training-safe export with hashed pseudonyms Plus Phase 1 + 1.5 findings + scrum-driven priorities: - UUID v7 candidate_id (drops kimi enumeration risk) - per-row encryption with per-subject keys (crypto-erasure target) - pii_access_log with Merkle-style integrity hash chain (FRE 901) - subject_id top-level promotion in all JSONL sinks - Langfuse boundary redaction layer (scrum C2 priority) - adverse-impact comparator pool in audit response (scrum C3) - BIPA-specific consent + retention metadata (scrum C4) - vertical detection at gateway boundary (J answer 10) Implementation single-language: Go (one identityd, both runtimes call it via HTTP). Postgres backing store, isolated schema. Master key in sealed file v1, vault migration path documented. 8-step migration path: stand up empty → backfill from parquet → behind feature flag → cut over reads incrementally → quarantine PII columns in workers_500k. Each step its own commit + gate + rollback. 6 open questions for J before implementation: master key location, Postgres shared vs isolated, vertical backfill default, legal token issuance procedure, crypto-erasure sweep cadence, EU enforcement timeline. Estimated 8-10 working days total. Largest single phase in the audit program. No code changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
fd429f4185 |
audit phase 1.5: BIPA schema audit + outcomes.jsonl content sample
Two follow-up walks per AUDIT_PHASE_1_DISCOVERY §10/C4 + gemini scrum flag. Read-only. No code changes. BIPA findings: - scripts/staffing/tag_face_pool.py uses deepface to extract gender + race + age from face images. Output persists to data/headshots/ manifest.jsonl. For synthetic faces this is fine; for real candidate photos this becomes a regulated biometric database (740 ILCS 14/10). - mcp-server/index.ts:1408 ComfyUI prompt EXPLICITLY embeds protected attributes (age + race + gender) into model prompt — system-level encoding of protected-attribute features into AI workflow. - mcp-server/search.html:3375-3432 has hard-coded FEMALE_NAMES / MALE_NAMES / NAMES_HISPANIC / SURNAMES_* lookup tables — name-based ethnicity inference. Title VII / disparate-impact risk separate from BIPA. - data/headshots/manifest.jsonl is TRACKED IN GIT today (synthetic classifications). For real photos, this would be biometric data in version control — serious failure. - No consent flow, no public retention schedule, no deletion procedure, no employee training documented. All required by BIPA §15(a)/(b) before real-photo intake. outcomes.jsonl sample: - 39/101 rows persist candidate names in fills[*].name field today - Sample names: "Carmen I. Garcia", "Jamal Z. Jones", "Jacob N. Patel" (synthetic but real shape) - 0 hits for "culture fit" / "communication" / etc proxy phrases — synthetic data doesn't generate them. When real models reason about real candidates, they will. Append-only persistence makes RTBF cryptographic-erasure-only. Recommends Phase 1.6 (NEW) — BIPA pre-launch gates between Phase 1.5 and Phase 2: BIPA_COMPLIANCE_POLICY.md, consent gate at upload endpoint, quarantine real-photo classifications to data/biometric/, deprecate name->ethnicity lookup tables, unit test that synthetic manifest stays synthetic. 4-8 hours of design + one code commit. 5 open questions for J: where do real photos enter, will deepface tagging path stay for real photos, consent UX, retention duration floor, designated privacy officer. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
64bda21614 |
audit PRD: J answered 5 open questions — fold into §10, revise phase plan
Conversation 2026-05-03 — J confirmed: - Photos/video YES → BIPA in full force ($1k-$5k per violation) - Langfuse self-hosted → drops GDPR Art. 44 cross-border concern - EU not in scope now but placeholder needed → design EU-compatible - Healthcare vertical YES → HIPAA BAA needed with model providers, PHI redaction at gateway boundary OR local-only routing for those requests, vertical-detection at boundary is Phase 2 requirement - Training/RAG MAY re-run on outcomes → design as if it will, training- safe export interface needed, crypto-erasure becomes load-bearing evidence chain §10 updated with answered/pending status per question. New §10.5 "Effect on phase plan" introduces: - Phase 1.5 (NEW) — BIPA photo/video schema audit + Langfuse boundary scoping + outcomes.jsonl content sample, BEFORE Phase 2 design - Phase 2 design must now include: EU-placeholder fields, vertical detection, training-safe export, BIPA consent metadata - Phase 9 rehearsal must cover discrimination + BIPA + healthcare PHI 3 questions still pending J's call before Phase 2 design ships: identity service daemon vs in-process, JSON vs signed PDF for legal export, audit endpoint auth model. No code changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
627a5f0c3d |
audit phase 1: §10 scrum-review findings + walk back §1F over-claim
Ran cross-lineage scrum on the discovery doc with the new model fleet
(opus + kimi-k2.6 + gemini-3-flash via Go gateway :4110, custom
"senior security architect" prompt). 3/3 reviewers responded with
substantive 800-1200 word reviews. Saved at /tmp/audit_scrum/.
5 convergent findings (≥2 reviewers) added as §10/C1-C5:
C1. §1F matrix-indexer "good for audit defensibility" claim is over-
claimed — walked back in TL;DR. Trace bodies unverified; treat as
SUSPECTED PII sink until §8.1 sampling completes.
C2. §1E (Langfuse) is the most dangerous leak — fix FIRST, ahead of
view-routing. Boundary-crossing leak (GDPR Art. 44 / CPRA sale /
SOC2 disposal). All 3 reviewers converge on this priority.
C3. Discrimination defense requires the FULL CANDIDATE POOL, not just
fills. EEOC UGESP (1978): need adverse-impact stats on everyone
who could have been picked. Phase 1 worked example missed this.
C4. BIPA / biometric exposure understated in findings (in PRD §10.5
but not translated to actionables). $1k-$5k per-violation regime.
C5. candidate_id must be promoted to top-level field in all JSONL
sinks. Grepping natural-language strings is not defensible audit
strategy. 3/3 reviewers converge.
11 single-reviewer high-value catches added as §10 single-reviewer
section: opus on LLM provider egress (8th PII path), Art. 22 right-
to-explanation, special-category data, DPIA/ROPA/DPA inventory; kimi
on sequential ID enumeration risk, Langfuse retention config, CCPA
de-identified-in-place vs crypto-shred, Bun common-mode failure,
cryptographic audit-trail integrity (Merkle/FRE 901), HIPAA BAA,
revised SELECT * effort estimate; gemini on data residency, "culture
fit" reasoning proxies, comparator-pool snapshot.
§9 reordered: sample first → defense-layer second → Langfuse
boundary third (was view-routing first per original draft;
boundary-crossing leak is higher priority per scrum).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
505ea93726 |
audit phase 1: discovery walk complete — subject + PII surface map
Read-only walk of both runtimes per AUDIT_TRAIL_PRD.md §8 phase 1.
Fills "UNKNOWN" cells in PRD §3 + §7 with file:line evidence.
Headline findings:
- candidates_safe + workers_safe views EXIST as a defense layer but
are BYPASSED — tool registry SQL templates query raw tables
- PII traverses 7+ persistence/transmission paths per fill scenario:
SQL → tool_result → LogEntry → /v1/respond → Langfuse → outcomes.jsonl
→ overseer_corrections.jsonl
- candidate_id is stable but co-located with PII in workers_500k.parquet
(no separate identity service)
- /audit/subject/{id} endpoint does not exist
- Append-only persistence is universal — RTBF requires crypto-erasure
- Pathway memory is structurally subject-agnostic in fingerprints
(defensive); trace bodies may leak PII (needs sampling)
- Go side mirrors Rust PII shape — parity in the leak too
- Worked example (John Martinez audit today): NOT POSSIBLE to produce
complete-and-defensible response
Recommends 4 cheap high-value moves before Phase 2 design starts:
defense-layer enforcement (rewrite 3 SQL templates to _safe views),
sample state.json/Langfuse to confirm pathway memory is clean, walk
Bun mcp-server tool surface, schema-audit for protected-attribute
proxies. None are commitments — J's call.
No code changes in this commit. Companion to AUDIT_TRAIL_PRD.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
b2d717ae44 |
audit PRD: add §10.5 jurisdictional surface (IL + IN, federal, SOC2)
J flagged that the staffing system targets Chicago + Indiana — added a jurisdictional checklist section to the audit-trail PRD so counsel has a working starting point. Covered: - Federal: Title VII, ADEA, ADA, EEOC, OFCCP, FCRA, Section 1981 - Illinois: BIPA (high risk if any candidate photos), AI Video Interview Act (820 ILCS 42), Illinois Human Rights Act (broader than Title VII), PIPA breach notification, Day and Temporary Labor Services Act (directly applies — staffing industry-specific recordkeeping), Cook County + City of Chicago Human Rights Ordinances (additional protected classes including source of income, parental status, credit history) - Indiana: Data Breach Disclosure, Civil Rights Law (lighter than IL), Genetic Information Privacy Act - SOC 2 Type II as the typical SaaS sale gate (Privacy + Security TSCs most relevant; 6-9 month effort to first report) - HIPAA / PCI / ISO 27001 noted as out of current scope but flagged Phase reordering implications captured: - BIPA risk on real candidate photos may need to be resolved BEFORE audit-trail work (class-action exposure) - SOC 2 Type II prep runs in parallel, not after - IL Day and Temporary Labor Services recordkeeping may override our proposed 4-year retention SLA 7 open questions added that counsel must answer before the §8 phases can be locked in. Document is explicit (multiple times) that this is NOT legal advice — a research-grade checklist for J's counsel conversation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
c170ebc86e |
docs: AUDIT_TRAIL_PRD — production-readiness gate for staffing client
J flagged that smoke + parity tests prove the surface compiles, NOT that an audit response can be produced for a specific person — and the staffing client won't sign without defensible discrimination-claim response capability. New docs/AUDIT_TRAIL_PRD.md captures: - worked example: John Martinez at Warehouse B requests audit - subject audit response output format (per-decision row schema) - surface map: where decisions happen today, where the gaps are - PII handling rules (tokenization, protected-attribute exclusion, inferred-attribute risk) - identity service design intent (separate daemon, audited reads) - retention + right-to-be-forgotten policy intent - 9-phase implementation sequence with explicit per-phase exit criteria - cross-runtime requirement (both Rust + Go must satisfy) - 7 open questions blocking phase 2+ that need J's call STATE_OF_PLAY + PRD updated with explicit "production-ready blocker" section pointing at the new doc. The "substrate is shipped" framing gets a caveat: substrate ≠ production-ready until audit phase 9 exits. No code changes. This is the planning artifact J asked for before we start building. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
5368aca4d4 |
docs: sync ADR-019 + PRD + DECISIONS with 2026-05-02 substrate changes
ADR-019: closed the "re-bench when 10M corpus exists" follow-up. Added "Follow-up: 10M re-bench (2026-05-02)" section with the post-fix numbers (search ~20ms warm / ~46ms cold, doc-fetch ~5ms post-btree). Documented the lance-bench-bypassing-IndexMeta bug + 2-layer fix + gauntlet (7 unit + 12 sanitize + 10 smoke probes). Reframes the strategic question as "Lance vs Parquet+HNSW-with-spilling" since HNSW doesn't fit RAM at 10M. DECISIONS: added ADR-022 — drop Python sidecar from Rust hot path. Captures the rationale (236× embed perf gap was pure overhead), co-shipped LRU cache, dev-only Python that survives, cross-runtime parity verification, and the operator runbook signal (ps -ef ABSENT post-deploy). PRD: updated AI Boundary table line + aibridge crate description to reflect direct Ollama path (was: Python FastAPI sidecar → Ollama). Both lines reference ADR-022 for the full rationale. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
41b0a99ed2 |
chore: add real content that was sitting untracked
Surfaced by today's untracked-files audit. None of these are accidents —
multiple are referenced by name in CLAUDE.md and memory files but were
never added.
Categories:
- docs/PHASE_AUDIT_GUIDE.md (106 LOC) — Claude Code phase audit guidance
- ops/systemd/lakehouse-langfuse-bridge.service — Langfuse bridge unit
- package.json — top-level npm manifest
- scripts/e2e_pipeline_check.sh + production_smoke.sh — real test scripts
- reports/kimi/audit-last-week*.md — the "Two reports live" CLAUDE.md cites
- tests/multi-agent/scenarios/ — 44 staffing scenarios (cutover decision A)
- tests/multi-agent/playbooks/ — 102 playbook records
- tests/battery/, tests/agent_test/PRD.md, tests/real-world/* — real tests
- sidecar/sidecar/{lab_ui,pipeline_lab}.py — 888 LOC dev-only UIs that
remain in service post-sidecar-drop (commit ba928b1 explicitly kept them)
Sensitivity check: scenarios use synthetic company names ("Heritage Foods",
"Cornerstone Fabrication"); audit reports describe code findings only;
no PII or secrets surfaced.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
c5654d417c |
docs: pointer to ARCHITECTURE_COMPARISON.md source in golangLAKEHOUSE
Some checks failed
lakehouse/auditor 18 blocking issues: cloud: claim not backed — "Verified end-to-end against persistent Go stack on :4110:"
Per J's request: the parallel-runtime comparison is a living source
file maintained at /home/profit/golangLAKEHOUSE/docs/ARCHITECTURE_COMPARISON.md.
This file is a pointer reachable from the Rust repo's docs/ so the
comparison is discoverable from either side.
Doesn't contain authoritative content — just the link + a quick
status summary + update guidance ('source lives in golangLAKEHOUSE,
don't drift two copies').
|
||
|
root
|
2cac64636c |
docs: PHASES tracker — mark Phases 42/43/44/45 complete
Today's work shipped four Phase closures (Truth Layer, Validation Pipeline, Caller Migration, Doc-Drift Detection); the canonical tracker now reflects them. Foundation for production switchover (real Chicago data replaces synthetic test data soon). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
d11632a6fa |
staffing: recon + synthetic-data gap report (Phase 0, no implementation)
Some checks failed
lakehouse/auditor 13 blocking issues: cloud: claim not backed — "Phase 8 done-criteria (per spec):"
Spec mandates these two docs before any staffing audit runner ships:
docs/recon/staffing-lakehouse-distillation-recon.md
reports/staffing/synthetic-data-gap-report.md
NO distillation core touched. Distillation v1.0.0 (commit e7636f2,
tag distillation-v1.0.0) remains the stable substrate. Staffing
work is consumer-only.
Recon findings (12 sections, ~5KB):
- Existing staffing schemas in crates/validator/staffing/* are scaffolds
(FillValidator schema-shape only; worker-existence/status/geo TODOs)
- Synthetic data spans 6+ shapes across 9 parquet files
(~625k worker-shape rows + 1k candidate-shape rows)
- PII detection lives in shared/pii.rs but enforcement at query
time is unverified — the LLM may have been seeing raw PII via
workers_500k_v8 vector corpus
- 44 scenarios + 64 playbook_lessons = ~108 RAG candidates
- No structured fill-event log exists; scenarios+lessons are
retrospective, not queryable per-event records
- workers_500k.phone is int (should be string — leading-zero loss)
- client_workerskjkk.parquet is a typo file (160 rows, sibling of
client_workersi.parquet)
- PRD §158 claims Phase 19 closed playbook write-only gap — unverified
Gap report findings (9 sections, ~6KB):
- 4 BLOCKING gaps requiring J decisions before audit ships:
A. Generate fill_events.parquet from scenarios + lessons?
B. Build views/{candidates,workers,jobs}_safe with PII masking?
C. Delete client_workerskjkk.parquet typo file?
D. Fix workers_500k.phone type (int → string)?
- 5 SOFT gaps the audit can run with (will be reported as findings)
- 3 NON-gaps (data sufficient as-is)
- Recommendation: NO new synthetic data needed; only normalization
of what already exists, contingent on J approval of A-D
Up-front commitments:
- Distillation v1.0.0 substrate untouched (verified by audit-full
running clean before+after each staffing change)
- All synthetic-data modifications via deterministic scripts under
scripts/staffing/, never hand-edit
- Every staffing artifact carries canonical sha256 provenance back
to source parquet/scenario/lesson
- _safe views are the source of truth for LLM-facing text; raw
parquets never directly fed into corpus builds
Phase 1 unblocks AFTER J reviews both docs and approves audit scope
+ the 4 gap-fix decisions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
73f242e3e4 |
distillation: Phase 9 — release freeze and operator handoff
Final phase. Adds:
scripts/distillation/release_freeze.ts ~330 lines, 6 release gates
docs/distillation/operator-handoff.md durable cold-start operator doc
docs/distillation/recovery-runbook.md failure-mode runbook by symptom
scripts/distillation/distill.ts +release-freeze subcommand
The release_freeze orchestrator runs every gate the system has:
1. Clean git state (tolerates auto-regenerated reports)
2. Full test suite (bun test tests/distillation auditor/schemas/distillation)
3. Phase commit verification (every Phase 0-8 commit resolves)
4. Acceptance gate (22-invariant fixture E2E)
5. audit-full (Phases 0-7 verified + drift detection)
6. Tag availability check (distillation-v1.0.0 not yet existing)
Outputs:
reports/distillation/release-freeze.md human-readable manifest
reports/distillation/release-manifest.json machine-readable manifest
Manifest captures:
- git_head + git_branch + released_at
- phase→commit map for all 9 commits (Phase 0+1+2 scaffold through Phase 8 audit)
- dataset counts at freeze (RAG/SFT/Preference/evidence/scored/quarantined)
- latest audit baseline row
- per-gate pass/fail with detail
Operator handoff doc covers:
- phase map with commits + report locations
- known-good commands
- how to rerun audit-full + inspect drift
- how to restore from last-good (git checkout distillation-v1.0.0)
- how to add future phases without contaminating corpus
- what NOT to modify casually (with file:reason mapping)
- cumulative commits at v1.0.0
Recovery runbook covers, by symptom:
- audit-full exit non-zero (per-phase diagnostics)
- drift table flags warn (intentional vs regression)
- acceptance fail vs audit-full pass divergence
- run-all empty exports (counter-bisection order)
- hash mismatch on identical input (determinism violation; CRITICAL)
- replay logs growing unbounded (rotation guidance)
- nuclear restore via git checkout distillation-v1.0.0
Spec constraints (per now.md Phase 9):
- DO NOT add new intelligence features ✓ (zero new logic)
- DO NOT change scoring/export logic ✓ (zero touches)
- DO NOT weaken gates ✓ (gates only added, never relaxed beyond the
auto-regen tolerance documented in checkCleanGit)
- DO NOT retrain anything ✓ (no model touches)
CLI:
./scripts/distill release-freeze # exit 0 = release-ready
Tag creation deferred to operator confirmation (the release-freeze
report prints the exact `git tag` command). Per CLAUDE.md guidance,
destructive/visible operations like tags require explicit user
authorization.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
27b1d27605 |
distillation: Phase 0 recon + Phase 1 schemas + Phase 2 transforms scaffold
Some checks failed
lakehouse/auditor 9 blocking issues: todo!() macro call in tests/real-world/scrum_master_pipeline.ts
Phase 0 — docs/recon/local-distillation-recon.md
Inventories the 23 KB JSONL streams + 20 vector corpora + auditor's
kb_index.ts as substrate for the now.md distillation pipeline. Maps
spec modules to existing producers, identifies real gaps, lists 9
schemas to formalize. ZERO implementation in recon — gating doc only.
Phase 1 — auditor/schemas/distillation/
9 schemas + foundation types + 48 tests passing in 502ms:
types.ts shared validators + canonicalSha256
evidence_record.ts EVIDENCE_SCHEMA_VERSION=1, ModelRole enum
scored_run.ts 4 categories pinned, anchor_grounding ∈ [0,1]
receipt.ts git_sha 40-char, sha256 file refs, validation_pass:bool
playbook.ts non-empty source_run_ids + acceptance_criteria
scratchpad_summary.ts validation_status enum, hash sha256
model_ledger.ts success_rate ∈ [0,1], sample_count ≥ 1
rag_sample.ts success_score ∈ {accepted, partially_accepted}
sft_sample.ts quality_score MUST be 'accepted' (no leak)
preference_sample.ts chosen != rejected, source_run_ids must differ
evidence_record.test.ts 10 tests, JSON-fixture round-trip
schemas.test.ts 30 tests, inline fixtures
realdata.test.ts 8 tests, real-JSONL probe
Real-data validation probe (one of the 3 notables from recon):
46 rows across 7 sources, 100% pass. distilled_facts/procedures alive.
Report at data/_kb/realdata_validation_report.md (also written by the
test). Confirms schema fits existing producers without migration.
Phase 2 scaffold — scripts/distillation/transforms.ts
Promoted PROBES from realdata.test.ts into a real TRANSFORMS array
covering 12 source streams (8 Tier 1 validated + 4 Tier 2 from
recon's untested-streams list). Pure functions: no I/O, no model
calls, no clock reads. Caller supplies recorded_at + sig_hash so
materializer is deterministic by construction.
Spec non-negotiables enforced at schema layer (defense in depth):
- provenance{source_file, sig_hash, recorded_at} required everywhere
- schema_version mismatch hard-rejects (forward-compat gate)
- SFT no-leak: validateSftSample REJECTS partially_accepted, rejected,
needs_human_review — three explicit tests
- Every score has WHY (reasons non-empty)
- Every playbook traces to source (source_run_ids non-empty)
- Every preference has WHY (reason non-empty)
- Receipts substantive (git_sha 40-char, sha256 64-char, validation_pass:bool)
Branch carries uncommitted auditor rebuild work (mode.rs + modes.toml
+ inference.ts + static.ts) blocked on upstream Ollama Cloud kimi-k2
500 ISE; held pending recon-driven design decisions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
f753e11157 |
docs: SCRUM_MASTER_SPEC timeline — productization wave + verified live state
Some checks failed
lakehouse/auditor 9 blocking issues: todo!() macro call in tests/real-world/scrum_master_pipeline.ts
Splits the existing 04-25/26 section into two waves: - experiment wave (mode-runner build-out, pre-productization) - productization wave (OpenAI-compat, Archon, answers corpus, staffing native runner, multi-corpus + downgrade gate, observer paid escalation, /v1/chat → observer event wiring) Adds verified-live block at the end with the numbers a fresh session needs to anchor on: pathway memory 88 traces / 11 successful replays at 100% (probation gate crossed), strong-model auto-downgrade firing on grok-4.1-fast, and the auditor blind spot at static.ts:117 (now fixed in 107a682). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
2dbc8dbc83 |
v1/mode: model-aware enrichment downgrade + 3 corpora + variance harness
Some checks failed
lakehouse/auditor 1 blocking issue: todo!() macro call in tests/real-world/scrum_master_pipeline.ts
Pass 5 (5 reps × 4 conditions × 1 file on grok-4.1-fast) showed composing matrix corpora is anti-additive on strong models — composed lakehouse_arch + symbols LOST 5/5 head-to-head vs codereview_isolation (Δ −1.8 grounded findings, p=0.031). Default flips to isolation; matrix path now auto- downgrades when the resolved model is strong. Mode runner: - matrix_corpus is Vec<String> (string OR array via deserialize_string_or_vec) - top_k=6 from each corpus, merge by score, take top 8 globally - chunk tag prefers doc_id over source so reviewer sees [adr:009] vs [lakehouse_arch] - is_weak_model() gate auto-downgrades codereview_lakehouse → codereview_isolation for strong models (default-strong; weak = :free suffix or local last-resort) - LH_FORCE_FULL_ENRICHMENT=1 bypasses for diagnostic runs - EnrichmentSources.downgraded_from records when the gate fires Three corpora indexed via /vectors/index (5849 chunks total): - lakehouse_arch_v1 — ADRs + phases + PRD + scrum spec (93 docs, 2119 chunks) - scrum_findings_v1 — past scrum_reviews.jsonl (168 docs, 1260 chunks; EXCLUDED from defaults — 24% out-of-bounds line citations from cross-file drift) - lakehouse_symbols_v1 — regex-extracted pub items + /// docs (656 docs, 2470 chunks) Experiment infra: - scripts/build_*_corpus.ts — re-runnable when source content changes - scripts/mode_pass5_variance_paid.ts — N reps × M conditions on one file - scripts/mode_pass5_summarize.ts — mean ± σ + head-to-head, parser handles numbered + path-with-line + path-with-symbol finding tables - scripts/mode_compare.ts — groups by mode|corpus when sweeps span corpora - scripts/mode_experiment.ts — default model bumped to x-ai/grok-4.1-fast, --corpus flag for per-call override Decisions + open follow-ups: docs/MODE_RUNNER_TUNING_PLAN.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
52bb216c2d |
mode_compare: grounding check + control flag + emoji-tolerant section detection
Some checks failed
lakehouse/auditor 1 blocking issue: todo!() macro call in tests/real-world/scrum_master_pipeline.ts
Three fixes after the playbook_only confabulation surfaced in 2026-04-26 experiment (8 'findings' on a 333-line file all citing lines 378-945 — fully fabricated from pathway-memory pattern names). (1) Aggregator regex bug — section detection failed on emoji-prefixed markdown headers like `## 🔎 Ranked Findings`. The original regex required word chars right after #{1,3}\s+, so the patches table header `## 🛠️ Concrete Patch Suggestions` was never recognized as a stop boundary, double-counting every finding. Fix: allow non-letter chars (emoji/space) between # and the keyword. (2) Grounding check — for each finding row in the response, extract backtick-quoted symbols + cited line numbers; verify symbols exist in the actual focus file and lines fall within EOF. Computes grounded/total ratio per mode. Surfaces 'OOB' (out-of-bounds) count explicitly so confabulation is visible at a glance. Confirms what hand-grading found: codereview_playbook_only's 8 findings on service.rs were 1/8 grounded with 7 OOB. (3) Control mode tagging — codereview_null and codereview_playbook_only are designed as falsifiers (baseline / lossy ceiling) and their numerical wins should never be read as recommendations. Output marks them with ⚗ glyph + warning footer. Per-mode aggregate is now sorted by groundedness, not raw count. Per-mode-vs-lakehouse comparison uses grounded findings, not raw — so confabulation can no longer score a "win". Updated SCRUM_MASTER_SPEC.md with refactor timeline pointing at the 2026-04-25/26 commits (observer fix, relevance filter, retire wire, mode router, enrichment runner, parameterized experiment). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
e1ef185868 |
docs: add MATRIX_AGENT_HANDOVER notes + cross-link from SCRUM_MASTER_SPEC
Some checks failed
lakehouse/auditor 1 blocking issue: todo!() macro call in tests/real-world/scrum_master_pipeline.ts
The matrix-agent-validated repo split + Ansible deploy pipeline are otherwise undocumented in this repo. This new doc explains: - the relationship between lakehouse and matrix-agent-validated - where the playbook lives and what it provisions - the critical distinction: matrix-test (10.111.129.50 Incus container) is the REAL destination, while 192.168.1.145 is a smoke-test VPS only (partial deploy, no services, do not treat as production) - what landed today (observer fix, HANDOVER.md render, relevance filter) Added a cross-link block at the top of SCRUM_MASTER_SPEC.md so the canonical scrum handoff doc points at the new MATRIX_AGENT_HANDOVER doc. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
fd92a9a0d0 |
docs: SCRUM_MASTER_SPEC.md — single handoff artifact for the scrum loop
Some checks failed
lakehouse/auditor 1 blocking issue: todo!() macro call in tests/real-world/scrum_master_pipeline.ts
Fresh-session artifact so work is recoverable if the branch is reopened
in a new Claude Code session without context. Covers:
- 9-rung ladder (kimi-k2:1t through local qwen3.5:latest)
- tree-split reducer (files >6KB sharded + map→reduce)
- schema_v4 KB rows in data/_kb/scrum_reviews.jsonl
- auto-applier 5 hardened gates (confidence, size, cargo-green,
warning-count, rationale-diff)
- pathway_memory (ADR-021) — narrow fingerprint + hot-swap gate +
semantic-correctness layer (SemanticFlag, BugFingerprint)
- HTTP surface on gateway (/vectors/pathway/*)
- current state (12 traces, 11 fingerprints, 0 hot-swaps — probation)
- commit history on scrum/auto-apply-19814 since iter-5 baseline
- how-to-run (env vars, service restarts)
- where things live (code pointers table)
- known gotchas (LLM Team mode registry, restart requirements)
Paired updates (not in this commit, live outside the repo):
- /home/profit/CLAUDE.md — active workstream pointer + notes
- /root/.claude/skills/read-mem/SKILL.md — SCRUM_MASTER_SPEC.md added
to the loading list + ADR-021 glossary
- memory/project_scrum_pipeline.md — updated with iter-9 state
- memory/feedback_semantic_correctness_via_matrix.md — updated with
end-to-end proof evidence
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
92df0e930a |
ADR-021: semantic-correctness layer on pathway_memory
Spec for the compounding-bug-grammar insight from J's feedback on the queryd/delta.rs unit-mismatch fix (86901f8). Adds three proposed fields to PathwayTrace (semantic_flags, type_hints_used, bug_fingerprints), 9 initial SemanticFlag variants, and the truth::evaluate review-time task_class pattern that reuses existing primitives instead of building a type-inference engine. Implementation pending approval on the flag set and fingerprint shape. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
39a2856851 |
docs: rewrite PR #10 description to drop unfalsifiable metric claims
Some checks failed
lakehouse/auditor 1 blocking issue: cloud: claim not backed — "journal event verified live (total_events_created 0→1 after probe)."
Auditor correctly flagged the '3 → 6' score claim as unbacked by diff (consensus: 3/3 not-backed). The claim referenced scrum_reviews.jsonl — an external metric file — which the auditor cannot verify against source changes alone. Rewrote the PR body to only claim what's directly verifiable from the diff (committed tests, committed code paths, committed startup logging). Trajectory data remains in docs/SCRUM_LOOP_NOTES.md for historical reference but is no longer asserted as fact in the PR body. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
root
|
21fd3b9c61 |
Scrum-driven fixes: P5-001 auth wired, P42-001 truth evaluator, P9-001 journal on ingest
Some checks failed
lakehouse/auditor 2 blocking issues: cloud: claim not backed — "| **P9-001** (partial) | `crates/ingestd/src/service.rs` | **3 → 6** ↑↑↑ | `journal.record_ing
Apply the highest-confidence findings from the Phase 0→42 forensic sweep
after four scrum-master iterations under the adversarial prompt. Each fix
is independently validated by a later scrum iteration scoring the same
file higher under the same bar.
Code changes
────────────
P5-001 — crates/gateway/src/auth.rs + main.rs
api_key_auth was marked #[allow(dead_code)] and never wrapped around
the router, so `[auth] enabled=true` logged a green message and
enforced nothing. Now wired via from_fn_with_state, with constant-time
header compare and /health exempted for LB probes.
P42-001 — crates/truth/src/lib.rs
TruthStore::check() ignored RuleCondition entirely — signature looked
like enforcement, body returned every action unconditionally. Added
evaluate(task_class, ctx) that actually walks FieldEquals / FieldEmpty /
FieldGreater / Always against a serde_json::Value via dot-path lookup.
check() kept for back-compat. Tests 14 → 24 (10 new exercising real
pass/fail semantics). serde_json moved to [dependencies].
P9-001 (partial) — crates/ingestd/src/service.rs
Added Optional<Journal> to IngestState + a journal.record_ingest() call
on /ingest/file success. Gateway wires it with `journal.clone()` before
the /journal nest consumes the original. First-ever internal mutation
journal event verified live (total_events_created 0→1 after probe).
Iter-4 scrum scored these files higher under same prompt:
ingestd/src/service.rs 3 → 6 (P9-001 visible)
truth/src/lib.rs 3 → 4 (P42-001 visible)
gateway/src/auth.rs 3 → 4 (P5-001 visible)
gateway/src/execution_loop 4 → 6 (indirect)
storaged/src/federation 3 → 4 (indirect)
Infrastructure additions
────────────────────────
* tests/real-world/scrum_master_pipeline.ts
- cloud-first ladder: kimi-k2:1t → deepseek-v3.1:671b → mistral-large-3:675b
→ gpt-oss:120b → devstral-2:123b → qwen3.5:397b (deep final thinker)
- LH_SCRUM_FORENSIC env: injects SCRUM_FORENSIC_PROMPT.md as adversarial preamble
- LH_SCRUM_PROPOSAL env: per-iter fix-wave doc override
- Confidence extraction (markdown + JSON), schema v4 KB rows with:
verdict, critical_failures_count, verified_components_count,
missing_components_count, output_format, gradient_tier
- Model trust profile written per file-accept to data/_kb/model_trust.jsonl
- Fire-and-forget POST to observer /event so by_source.scrum appears in /stats
* mcp-server/observer.ts — unchanged in shape, confirmed receiving scrum events
* ui/ — new Visual Control Plane on :3950
- Bun.serve with /data/{services,reviews,metrics,trust,overrides,findings,file,refactor_signals,search,logs/:svc,scrum_log}
- Views: MAP (D3 graph, 5 overlays) / TRACE (per-file iter timeline) /
TRAJECTORY (refactor signals + reverse index search) / METRICS (explainers
with SOURCE + GOOD lines) / KB (card grid with tooltips) / CONSOLE (per-service
journalctl tail, tabs for gateway/sidecar/observer/mcp/ctx7/auditor/langfuse)
- tryFetch always attempts JSON.parse (fix for observer returning JSON without content-type)
- renderNodeContext primitive-vs-object guard (fix for gateway /health string)
* docs/SCRUM_FIX_WAVE.md — iter-specific scope directing the scrum
* docs/SCRUM_FORENSIC_PROMPT.md — adversarial audit prompt (verdict/critical/verified schema)
* docs/SCRUM_LOOP_NOTES.md — iteration observations + fix-next-loop queue
* docs/SYSTEM_EVOLUTION_LAYERS.md — Layers 1-10 roadmap (trust profiling, execution DNA, drift sentinel, etc)
Measurements across iterations
──────────────────────────────
iter 1 (soft prompt, gpt-oss:120b): mean score 5.00/10
iter 3 (forensic, kimi-k2:1t): mean score 3.56/10 (−1.44 — bar raised)
iter 4 (same bar, post fixes): mean score 4.00/10 (+0.44 — fixes landed)
Score movement iter3→iter4: ↑5 ↓1 =12
21/21 first-attempt accept by kimi-k2:1t in iter 4
20/21 emitted forensic JSON (richer signal than markdown)
16 verified_components captured (proof-of-life, new metric)
Permission Gradient distribution: 0 auto · 16 dry_run · 4 sim · 1 block
Observer loop: by_source {scrum: 21, langfuse: 1985, phase24_audit: 1}
v1/usage: 224 requests, 477K tokens, all tracked
Signal classes per file (iter 3 → iter 4):
CONVERGING: 1 (ingestd/service.rs — fix clearly landed)
LOOPING: 4 (catalogd/registry, main, queryd/service, vectord/index_registry)
ORBITING: 1 (truth — novel findings surfacing as surface ones fix)
PLATEAU: 9 (scores flat with high confidence — diminishing returns)
MIXED: 6
Loop thesis status
──────────────────
A file's score rises only when the scrum confirms a real fix landed.
No false positives yet across 3 iterations. Fixes applied to 3 files all
raised their independent scores under the same adversarial prompt. Loop
is measurable, not hand-wavy.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
4251e94531 |
Update PHASES.md: Phase 41 + Guard fixes
- Phase 41: ProfileType enum, per-type endpoints - Guard: scrumaudit.py, fixed watcher.sh + pr-reviewer.md |
||
|
root
|
55f8e0fe6e |
Phase 40: Routing Engine + Policy
- RoutingEngine with RouteDecision (model_pattern → provider) - config/routing.toml: rules, fallback chain, cost gating - Per-provider Usage tracking in /v1/usage response - 12 gateway tests green |
||
|
root
|
e27a17e950 |
Phase 39: Provider Adapter Refactor
- ProviderAdapter trait with chat(), embed(), unload(), health() - OllamaAdapter wrapping existing AiClient - OpenRouterAdapter for openrouter.ai API integration - provider_key() routing by model prefix (openrouter/*, etc) |
||
|
root
|
21e8015b60 |
Phase 37: Hot-swap async + Phase 38: Universal API skeleton
- JobTracker extended with JobType::ProfileActivation + Embed - activate_profile returns job_id immediately, work spawns in background - /v1/chat, /v1/usage, /v1/sessions endpoints (OpenAI-compatible) - Langfuse trace integration (Phase 40 early deliverable) - 12 gateway unit tests green, curl gates pass |
||
| 7c1745611a |
Audit pipeline PR #9: determinism + fact extraction + verifier gate + KB stats + context injection (PR #9)
Bundles PR #9's work for the audit pipeline: - N=3 consensus on cloud inference (gpt-oss:120b parallel) with qwen3-coder:480b tie-breaker - audit_discrepancies.jsonl logs N-run disagreements - scrum_master reviews route through llm_team fact extraction; source="scrum_review" - Verifier-gated persistence: drops INCORRECT, keeps UNVERIFIABLE/UNCHECKED; schema_version:2 - scrum_master_reviewed flag on accepted reviews - auditor/kb_stats.ts: on-demand observability script - claim_parser history/proof pattern class (verified-on-PR, was-flipping, the-proven-X) - claim_parser quoted-string guard (mirrors static.ts fix) - fact_extractor project context injection via docs/AUDITOR_CONTEXT.md - Fixed verifier-verdict parser to handle multiple gemma2 output formats Empirical: 3-run determinism test on unchanged PR #9 SHA showed 7/7 warn findings stable; block count oscillation eliminated; llm_team quality scores 8-9 on context-injected extract runs. See PR #9 for full run-by-run commit history. |
|||
|
profit
|
2a4b81bf48 |
Phase 45 (first slice): DocRef + doc_refs field on PlaybookEntry
Phase J keeps asking for: playbooks know which external docs they
used, get flagged when those docs drift. This commit ships the data
model; context7 bridge + drift check endpoints land in follow-ups.
Added to crates/vectord/src/playbook_memory.rs:
- pub struct DocRef { tool, version_seen, snippet_hash, source_url,
seen_at } — one external doc reference
- PlaybookEntry.doc_refs: Vec<DocRef> — empty on legacy entries,
serde default ensures pre-Phase-45 persisted state loads cleanly
- PlaybookEntry.doc_drift_flagged_at: Option<String> — set by the
(future) drift-check code when context7 reports newer version
- PlaybookEntry.doc_drift_reviewed_at: Option<String> — set by
human via /resolve endpoint after reviewing the diagnosis
- impl Default for PlaybookEntry — collapses most test-helper
constructors from 17 explicit fields to 6-9 fields +
..Default::default()
Updated SeedPlaybookRequest + RevisePlaybookRequest (service.rs) to
accept optional doc_refs: the seed/revise endpoints already take the
field, downstream drift detection (Phase 45.2) consumes it.
Docs: docs/CONTROL_PLANE_PRD.md gains full Phase 45 spec with gate
criteria, non-goals, and risk notes.
Tests: 51/51 vectord lib tests green (same count as before, field
additions are backward-compat).
Memory: project_doc_drift_vision.md written so this keeps coming
back to the front of mind.
Next slices (same phase): context7 HTTP bridge in mcp-server,
/vectors/playbook_memory/doc_drift/check/{id} endpoint, overview-
model drift synthesis writing to data/_kb/doc_drift_corrections.jsonl,
boost exclusion for flagged+unreviewed entries.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
profit
|
6316433062 |
Phase 40 scope: Langfuse + Gitea MCP recovery as named deliverables
J flagged that a prior version of this stack had Langfuse traces piping into the observer + Gitea MCP for repo ops — lost. Adding these as explicit Phase 40 deliverables alongside routing engine + Gemini/Claude adapters. Findings during scope-check: - Langfuse container is already running (Up 2 days, langfuse:2, localhost:3001 healthcheck passes) - mcp-server/tracing.ts + package.json already have SDK wired - Credentials pk-lf-staffing / sk-lf-staffing-secret (from env) - Gitea MCP binary still installed at gitea-mcp@0.0.10 So recovery here is mostly re-connecting existing infra: 1. Add Rust-side Langfuse client for /v1/chat tracing (gateway currently bypasses tracing, mcp-server already has it) 2. Wire Langfuse → observer :3800 pipe 3. Register Gitea MCP in mcp-server/index.ts tool list Each landing as part of Phase 40 when the routing engine ships. |
||
|
profit
|
f44b6b3e6b |
Control-plane pivot: Phase 38-44 plan + bot scaffold
Direction shift 2026-04-22: docs/CONTROL_PLANE_PRD.md becomes the long-horizon architecture target. Existing Lakehouse (docs/PRD.md, Phases 0-37) is preserved as the reference implementation and first consumer. New 6-layer architecture: L1 Universal API /v1/chat /v1/usage /v1/sessions /v1/tools /v1/context L2 Routing & Policy Engine (rules, fallback chains, cost gating) L3 Provider Adapter Layer (Ollama + OpenRouter + Gemini + Claude) L4 Knowledge + Memory + Playbooks (already built) L5 Execution Loop (scenarios + bot/cycle.ts instances) L6 Observability + token accounting Phases 38-44 sequenced with detailed per-phase specs in the PRD. Current scope: staffing domain (synthetic workers_500k, contracts, emails, SMS, playbooks). DevOps (Terraform/Ansible) is long-horizon target — architecture-compatible but not current. Files added: - docs/CONTROL_PLANE_PRD.md — 6-layer architecture, Phase 38-44 sequencing with staffing-first Truth Layer + Validation pipeline - bot/ — manual-only PR bot scaffold. First consumer test-bed for /v1/chat (Phase 38). Mem0-aligned ADD/UPDATE/NOOP apply semantics; KB feedback loop reads prior cycles on same gap and injects into cloud prompt so bot cycles compound like scenario.ts runs do. - tests/multi-agent/run_stress.ts — the 6-task diverse stress test referenced in the previous commit but missing from its staging Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
profit
|
5b1fcf6d27 |
Phase 28-36 body of work
Accumulated since a6f12e2 (Phase 21 Rust port + Phase 27 versioning): - Phase 36: embed_semaphore on VectorState (permits=1) serializes seed embed calls — prevents sidecar socket collisions under concurrent /seed stress load - Phase 31+: run_stress.ts 6-task diverse stress scaffolding; run_e2e_rated.ts + orchestrator.ts tightening - Catalog dedupe cleanup: 16 duplicate manifests removed; canonical candidates.parquet (10.5MB -> 76KB) + placements.parquet (1.2MB -> 11KB) regenerated post-dedupe; fresh manifests for active datasets - vectord: harness EvalSet refinements (+181), agent portfolio rotation + ingest triggers (+158), autotune + rag adjustments - catalogd/storaged/ingestd/mcp-server: misc tightening - docs: Phase 28-36 PRD entries + DECISIONS ADR additions; control-plane pivot banner added to top of docs/PRD.md (pointing at docs/CONTROL_PLANE_PRD.md which lands in next commit) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
profit
|
a6f12e2609 |
Phase 21 Rust port + Phase 27 playbook versioning + doc-sync
Phase 21 — Rust port of scratchpad + tree-split primitives (companion to
the 2026-04-21 TS shipment). New crates/aibridge modules:
context.rs — estimate_tokens (chars/4 ceil), context_window_for,
assert_context_budget returning a BudgetCheck with
numeric diagnostics on both success and overflow.
Windows table mirrors config/models.json.
continuation.rs — generate_continuable<G: TextGenerator>. Handles the
two failure modes: empty-response from thinking
models (geometric 2x budget backoff up to budget_cap)
and truncated-non-empty (continuation with partial
as scratchpad). is_structurally_complete balances
braces then JSON.parse-checks. Guards the degen case
"all retries empty, don't loop on empty partial".
tree_split.rs — generate_tree_split map->reduce with running
scratchpad. Per-shard + reduce-prompt go through
assert_context_budget first; loud-fails rather than
silently truncating. Oldest-digest-first scratchpad
truncation at scratchpad_budget (default 6000 t).
TextGenerator trait (native async-fn-in-trait, edition 2024). AiClient
implements it; ScriptedGenerator test double lets tests inject canned
sequences without a live Ollama.
GenerateRequest gained think: Option<bool> — forwards to sidecar for
per-call hidden-reasoning opt-out on hot-path JSON emitters. Three
existing callsites updated (rag.rs x2, service.rs hybrid answer).
Phase 27 — Playbook versioning. PlaybookEntry gained four optional
fields (all #[serde(default)] so pre-Phase-27 state loads as roots):
version u32, default 1
parent_id Option<String>, previous version's playbook_id
superseded_at Option<String>, set when newer version replaces
superseded_by Option<String>, the playbook_id that replaced
New methods:
revise_entry(parent_id, new_entry) — appends new version, stamps
superseded_at+superseded_by on parent, inherits parent_id and sets
version = parent + 1 on the new entry. Rejects revising a retired
or already-superseded parent (tip-of-chain is the only valid
revise target).
history(playbook_id) — returns full chain root->tip from any node.
Walks parent_id back to root, then superseded_by forward to tip.
Cycle-safe.
Superseded entries excluded from boost (same rule as retired): filter
in compute_boost_for_filtered_with_role (both active-entries prefilter
and geo-filtered path), rebuild_geo_index, and upsert_entry's existing-
idx search. status_counts returns (total, retired, superseded, failures);
/status JSON reports active = total - retired - superseded.
Endpoints:
POST /vectors/playbook_memory/revise
GET /vectors/playbook_memory/history/{id}
Doc-sync — PHASES.md + PRD.md drifted from git after Phases 24-26
shipped. Fixes applied:
- Phase 24 marked shipped (commit b95dd86) with detail of observer
HTTP ingest + scenario outcome streaming. PRD "NOT YET WIRED"
rewritten to reflect shipped state.
- Phase 25 (validity windows, commit e0a843d) added to PHASES +
PRD.
- Phase 26 (Mem0 upsert + Letta hot cache, commit 640db8c) added.
- Phase 27 entry added to both docs.
- Phase 19.6 time decay corrected: was documented as "deferred",
actually wired via BOOST_HALF_LIFE_DAYS = 30.0 in playbook_memory.rs.
- Phase E/Phase 8 tombstone-at-compaction limit note updated —
Phase E.2 closed it.
Tests: 8 new version_tests in vectord (chain-metadata stamping,
retired/superseded parent rejection, boost exclusion, history from
root/tip/middle, legacy default round-trip, status counts). 25 new
aibridge tests (context/continuation/tree_split). Workspace total
145 green (was 120).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
root
|
137aed64fb |
Coherence pass — PRD/PHASES updates, config snapshot wired, unit tests
J flagged the audit: "make sure everything flows coherently, no pseudocode or unnecessary patches or ignoring any particular part of what we built." This is that pass. PRD.md updates: - Phase 19 refinement block — geo-filter + role-prefilter WIRED with citation density numbers (0.32 → 1.38, and 2 → 28 on same scenario). - Phase 20 rewrite — mistral dropped, qwen3.5 + qwen3 local hot path, think:false as the key mechanical finding, kimi-k2.6 upgrade path. - Phase 21 status block — think plumbing + cloud executor routing added after original commit. - Phase 22 item B (cloud rescue) — pivot sanitizer, rescue verified 1/3 on stress_01. - Phase 23 NEW — staffer identity + tool_level + competence-weighted retrieval + kb_staffer_report. Auto-discovered worker labels called out with real numbers (Rachel Lewis 12× across 4 staffers). - Phase 24 NEW — Observer/Autotune integration gap DOCUMENTED, not fixed. Observer has been idle at 0 ops for 3600+ cycles because scenarios hit gateway:3100 directly, bypassing MCP:3700 which the observer wraps. This is the honest "we're not using it in these tests" signal J surfaced. Fix deferred; gap visible now. PHASES.md: - Appended Phases 20-23 as checked, Phase 24 as unchecked gap. - Updated footer count: 102 unit tests across all layers. - Latest line updated with 14× citation lift + 46.4pt tool-asymmetry finding. scenario.ts: - snapshotConfig() was defined but never called. Now fires at every scenario start with a stable sha256 hash over the active model set + tool_level + cloud flags. config_snapshots.jsonl finally populates, which the error_corrections diff path needs to work correctly. kb.test.ts (new): 4 signature invariant tests — stability across unrelated fields (date, contract, staffer), sensitivity to role/city/ count changes, digest shape. All pass under `bun test`. service.rs: 6 Rust extractor tests for extract_target_geo + extract_target_role — basic, missing-state-returns-none, word boundary (civilian != city), multi-word role, absent role, quoted value parse. All pass under `cargo test -p vectord --lib extractor_tests`. Dangling items now honestly documented rather than silently pending: - Chunking cache (config/models.json SPEC, not wired) — flagged - Playbook versioning (SPEC, not wired) — flagged - Observer integration (WIRED but disconnected) — new Phase 24 |
||
|
root
|
9c1400d738 |
Phase 22 — Internal Knowledge Library (KB)
Meta-layer over Phase 19 playbook_memory. Phase 19 answers "which WORKERS worked for this event"; KB answers "which CONFIG worked for this playbook signature" — model choice, budget hints, pathway notes, error corrections. tests/multi-agent/kb.ts: - computeSignature(): stable sha256 hash of the (kind, role, count, city, state) tuple sequence. Same scenario shape → same sig. - indexRun(): extracts sig, embeds spec digest via sidecar, appends outcome record, upserts signature to data/_kb/signatures.jsonl. - findNeighbors(): cosine-ranks the k most-similar signatures from prior runs for a target spec. - detectErrorCorrections(): scans outcomes for same-sig fail→succeed pairs, diffs the model set, logs to error_corrections.jsonl. - recommendFor(): feeds target digest + k-NN neighbors + recent corrections to the overview model, gets back a structured JSON recommendation (top_models, budget_hints, pathway_notes), appends to pathway_recommendations.jsonl. JSON-shape constrained so the executor can inherit it mechanically. - loadRecommendation(): at scenario start, pulls newest rec matching this sig (or nearest). scenario.ts: - Reads KB recommendation at startup (alongside prior lessons). - Injects pathway_notes into guidanceFor() executor context. - After retrospective, indexes the run + synthesizes next rec. Cold-start behavior: first run with no history writes a low-confidence "no prior data" rec so the signal that something was attempted is captured. Second run gets "low confidence, 0 neighbors" until a third distinct sig gives the embedder something to compare against — hence the upcoming scenario generator. VERIFIED: - data/_kb/ populated after one scenario run: 1 outcome (sig=4674…, 4/5 ok, 16 turns total), 1 signature, 2 recs (cold + post-run). - Recommendation JSON-parsed cleanly from gpt-oss:20b overview model. PRD Phase 22 added with file layout, cycle description, and the rationale for file-based MVP → Rust port progression that matches how Phase 21 primitives shipped. What's NOT here yet (batched follow-ups per J's request, tested between each): - Lift the k=10 hybrid_search cap to adaptive k=max(count*5, 20) - Scenario generator to bulk-populate KB with varied signatures - Rust re-weighting: push playbook_memory success signal INTO hybrid_search scoring, not just post-hoc boost |
||
|
root
|
0c4868c191 |
qwen3.5 executor + continuation primitive + think:false
Three coupled fixes that together turned the Riverfront Steel scenario
from 0/5 (mistral) to 4/5 (qwen3.5) with T3 flagging real staffing
concerns rather than linter advice.
MODEL SWAP
- Executor: mistral → qwen3.5:latest (9.7B, 262K ctx, thinking).
mistral's decoder emitted malformed JSON on complex SQL filters
regardless of prompt; J called it — stop using mistral.
- Reviewer: qwen2.5 → qwen3:latest (40K ctx)
- Applied to scenario.ts, orchestrator.ts, network_proving.ts,
run_e2e_rated.ts
CONTINUATION PRIMITIVE (agent.ts)
- generateContinuable(): empty-response → geometric backoff retry;
truncated-JSON → continue from partial as scratchpad; bounded by
budget cap + max_continuations. No more "bump max_tokens until it
stops truncating" tourniquet.
- generateTreeSplit(): map-reduce for oversized input corpora with
running scratchpad digest, reduce pass for final synthesis.
- Empty text no longer throws — it's a signal to continuable that
thinking ate the budget.
think:false FOR HOT PATH
- qwen3.5 burned ~650 tokens of hidden thinking for trivial JSON
emission. For executor/reviewer/draft: think:false. For T3/T4/T5
overseers: thinking stays on (that's the point).
- Sidecar generate endpoint accepts `think` bool, passes through to
Ollama's /api/generate.
VERIFIED OUTCOMES
Riverfront Steel 2026-04-21, qwen3.5+continuable+think:false:
08:00 baseline_fill 3/3 4 turns
10:30 recurring 2/2 3 turns (1 playbook citation)
12:15 expansion 0/5 drift-aborted (5-fill orchestration
problem, separate work)
14:00 emergency 4/4 3 turns (1 citation)
15:45 misplacement 1/1 3 turns
→ T3 caught Patrick Ross double-booking across events
→ T3 flagged forklift cert drift on the event that failed
→ Cross-day lesson proposed "maintain buffer of ≥3 emergency
candidates, pre-fetch certs for expansion, booking system
cross-check" — real staffing advice, not generic linter output
PRD PHASE 21 rewritten to reflect the actual primitive shape (two-
call map-reduce with scratchpad glue) instead of the tourniquet
approach originally documented. Rust port queued for next sprint.
scripts/ab_t3_test.sh: A/B harness that chains B→C→D runs and emits
tests/multi-agent/playbooks/ab_scorecard.json.
|
||
|
root
|
6e7ca1830e |
Phase 21 foundation — context stability + chunking pipeline
PRD: add Phase 20 (model matrix, wired) and Phase 21 (context stability, partial). Phase 21 exists because LLM Team hit this exact wall — running multi-model ranking on large context silently truncated, rankings degraded, no pipeline caught it. The stable answer: every agent call goes through a budget check against the model's declared context_window minus safety_margin, with a declared overflow_policy when the check fails. config/models.json: - context_window + context_budget per tier - overflow_policies block: summarize_oldest_tool_results_via_t3, chunk_lessons_via_cosine_topk, two_pass_map_reduce, escalate_to_kimi_k2_1t_or_split_decision - chunking_cache spec (data/_chunk_cache/, corpus-hash keyed) agent.ts: - estimateTokens() chars/4 biased safe ~15% - CONTEXT_WINDOWS table (fallback; prod reads models.json) - assertContextBudget() — throws on overflow with exact numbers, can bypass with bypass_budget:true for callers with their own policy - Wired into generate() and generateCloud() so EVERY call is checked scenario.ts: - T3 lesson archive to data/_playbook_lessons/*.json (the old /vectors/playbook_memory/seed path was silently failing with HTTP 400 because it requires 'fill: Role xN in City, ST' operation shape) - loadPriorLessons() at scenario start — filters by city/state match, date-sorted, takes top-3 - prior_lessons.json archived per-run (honest signal for A/B) - guidanceFor() injects up to 2 prior lessons (≤500 chars each) into the executor's per-event context - Retrospective shows explicit "Prior lessons loaded: N" line Verified: mistral correctly rejects a 150K-char prompt (7532 tokens over), gpt-oss:120b accepts it with 90K headroom. The enforcement is in-band on every call now, not an afterthought. Full chunking service (Rust) remains deferred to the sprint this feeds: crates/aibridge/src/budget.rs + chunk.rs + storaged/chunk_cache.rs |
||
|
root
|
13b01fee9f |
ADR-021: Sparse data trust path — start with nothing, earn everything
The staffing company said: 'we don't have any of that data.' They're right. We showed a demo with 18-field profiles and they have a name and a phone number. This ADR documents the trust path: Phase 1 (Day 1): Work with name + phone + role. That's enough. Phase 2 (Week 1-4): Timesheets → reliability. Calls → history. Phase 3 (Month 2+): AI starts helping with real earned data. Key principles: - Never show empty fields or 0% bars - Show what's THERE, not what's missing - Trust indicators: 'based on 3 placements' not just 'Reliability: 87%' - The system earns data by being useful, not by demanding it upfront Also created sparse_workers dataset (200 workers, 74% have role, 34% have notes, 5 have ONLY name+phone) for realistic testing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
root
|
937569d188 |
ADR-020: Universal ID mapping — fix the flat embedding identity problem
THE REAL PROBLEM: Every new data source produces different doc_id prefixes in vector indexes (W-, W500K-, W5K-, CAND-). Hybrid search had to hardcode strip_prefix for each one. New datasets broke hybrid until someone added another prefix. This violates "any data source without pre-defined schemas." THE FIX: IndexMeta.id_prefix — the catalog records what prefix each index uses. Hybrid search reads it and strips automatically. Legacy indexes fall back to heuristic stripping. New indexes can set id_prefix=None to use raw IDs (no prefix, no stripping needed). This means: ingest a new dataset, embed it, hybrid search works immediately without code changes. The system is truly source-agnostic. Also: full ADR document at docs/ADR-020-universal-id-mapping.md with the three options considered and rationale for the chosen approach. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
root
|
296bdaa746 |
PRD: hybrid search is operational, Ethereal data integrated
Status updated to reflect hybrid SQL+vector search, IVF_PQ 0.97 recall, 10K Ethereal worker profiles, autonomous agent validation. Query Paths section updated with the shipped hybrid endpoint and its verified zero-hallucination results from the staffing simulation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
root
|
3bc82833ac |
Update PRD + PHASES.md — reflect 8-commit 2026-04-17 push
PRD status line: "Phases 0-18 shipped; hybrid operational; scheduled ingest live; PDF OCR live; entering horizon items." PHASES.md: federation L2 items marked complete, Phase 16.2 (autotune agent), Phase 17 VRAM gate, MySQL connector, Phase 18 (hybrid Lance), scheduled ingest, PDF OCR all documented with dates and measurements. Stats updated: 52+ unit tests, 13 crates, 19 ADRs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
root
|
4e1c400f5d |
Phase E.2: Compaction integrates tombstones — physical deletion closes GDPR loop
Phase E gave us soft-delete at query time (tombstones hide rows via a
DataFusion filter view). This completes the invariant: after compact,
tombstoned rows are PHYSICALLY absent from the parquet on disk.
delta::compact changes:
- Signature adds tombstones: &[Tombstone]
- After merging base + deltas, apply_tombstone_filter builds a
BooleanArray keep-mask per batch (True where row_key_value is NOT
in the tombstone set) and applies arrow::compute::filter_record_batch
- Supports Utf8, Int32, Int64 key columns (matches refresh.rs coverage
for pg- and csv-derived schemas)
- CompactResult gains tombstones_applied + rows_dropped_by_tombstones
- Caller clears tombstone store on success
Critical correctness fix surfaced during E2E testing:
The original Phase 8 compact concatenated N independent Parquet byte
streams from record_batch_to_parquet() — each with its own footer.
Parquet readers only see the FIRST footer's data; the rest is invisible.
Latent since Phase 8 shipped; triggered by tombstone-filtering produc-
ing multiple batches. Corrupted candidates.parquet on first test run
(restored from UI fixture copy — good argument for test data in repo).
Fix:
- Single ArrowWriter per compaction, writes every batch into one
properly-footered Parquet
- Snappy compression to match ingest defaults (otherwise rewrite
inflated file 3× — 10.5MB → 34MB — because no compression was set)
- Verify-before-swap: parse written buf back to confirm row count
matches expected; refuses to overwrite base_key if verification fails
- Write to {base_key}.compact-{ts}.tmp first, then to base_key; delete
temp; only then delete delta files. Any error along the way leaves
the original base intact.
TombstoneStore::clear(dataset) drops all tombstone batch files and
evicts the per-dataset AppendLog from cache. Called after successful
compact.
QueryEngine::catalog() accessor exposes the Registry so queryd
handlers can reach the tombstone store without routing through gateway
state.
E2E on candidates (100K rows, 15 cols):
- Baseline: 10.59 MB, 100000 rows
- Tombstone CAND-000001/2/3 (soft-delete): 99997 visible, 100000 raw
- Compact: tombstones_applied=3, rows_dropped=3, final_rows=99997
- Post: 10.72 MB (Snappy), valid parquet (1 row_group), 99997 rows
- Restart: persists, tombstones list empty, __raw__candidates also
99997 (the 3 IDs are physically gone from disk)
PRD invariant close: deletion is now actually deletion, not just
masking. GDPR erasure request → tombstone + schedule compact → data
gone.
Deferred:
- Compact-all-datasets cron (currently manual per-dataset via
POST /query/compact)
- Compaction of tombstone batch files themselves (they grow at
flush_threshold=1 per tombstone; TombstoneStore::compact exists
but not auto-called)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
