Same shape of proof as embed_parity.sh for the embed endpoint:
take the just-shipped Go port (ca142b9) and validate it against
the actual production data the Rust legacy emits, not just unit-
test fixtures. Locks the cross-runtime parity that operators
running mixed pipelines depend on.
scripts/cutover/audit_baselines_validate.go:
- Reads /home/profit/lakehouse/data/_kb/audit_baselines.jsonl
- Parses every entry via the Go AuditBaseline struct
- Round-trips the last entry: encode → decode → field-by-field
equality check (catches any silently-dropped JSON keys)
- Calls LoadLastBaseline against the live file (proves the public
API works on real shapes, not just inline parsing)
- Computes BuildAuditDriftTable(first → last) — full-window
lineage drift over the captured baselines
Live-data probe results (reports/cutover/audit_baselines_roundtrip.md):
- 7 entries parse without error
- Round-trip is byte-equal on every metric + every header field
- Drift table fires the expected verdicts:
- p2_evidence_rows 12→82 (+583%) → warn (above 20% threshold)
- p3_accepted/partial/rejected/human 0→non-zero → warn (the
zero-baseline edge case TestBuildAuditDriftTable_ZeroBaseline
was designed to lock — verified now firing on real history)
- p4_* metrics +0% → ok (stable across the window)
What this does NOT prove (documented in the report): the Go-side
audit-FULL pipeline that PRODUCES baselines doesn't exist yet.
Only the load/append/drift substrate is ported. Operators running
audit-full from Go would still need a metric-collection pass —
that's a separate port deliberately not in this wave.
reports/cutover/SUMMARY.md gains a new row alongside the embed
parity entries; cutover-prep verification log keeps the
discipline of "verified against real data, not just fixtures."
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The original OPEN #2 line called for "SFT export pipeline +
audit_baselines lineage." Commit 7bb432f shipped the SFT export.
This commit ports the audit_baselines half — the longitudinal
drift signal that distinguishes "metrics shifted because the world
changed" from "metrics shifted because we broke something."
Mirrors Rust scripts/distillation/audit_full.ts's substrate:
- LoadLastBaseline(path) reads the most recent entry from
data/_kb/audit_baselines.jsonl. Returns (nil, nil) on missing
file (first run), errors on truncated last line (partial-write
detection — operators don't lose drift signal silently).
- AppendBaseline(path, baseline) appends one entry as a JSON line.
Atomic at the line level via bufio + O_APPEND. Creates the
parent directory if missing.
- BuildAuditDriftTable(prior, current, threshold) computes
per-metric drift. flag values mirror Rust exactly: first_run,
ok, warn. DefaultDriftWarnThreshold = 0.20 = Rust's 20%.
- FormatAuditDriftTable renders a fixed-width text grid for
stdout dumps in audit-full runs.
Edge cases handled:
- Zero-baseline: prior=0 means no division — PctChange stays nil.
current=0 → ok (no change). current>0 → warn (zero→nonzero is
always notable, never silently fine).
- New metric in current: flagged first_run, not "0%-change".
Operators see "this is a new signal we haven't tracked before."
- Sort: stable by metric name for deterministic JSON output and
clean CI diffs.
Generic on metric name (vs Rust's pinned p2_evidence_rows etc.):
the Rust phase numbering doesn't translate to Go directly. The
AuditBaselineRustCompat constant pins the Rust names so operators
running both runtimes use the same labels, which makes drift
comparison meaningful across the two pipelines.
13 new tests covering: missing file, last-line-wins, blank-line
tolerance, malformed-line errors, append round-trip, append-to-
existing, schema validation, first-run, threshold boundary,
zero-baseline, new-metric-in-current, sort-by-metric stability,
formatter output rendering.
OPEN #2's "audit_baselines lineage" half now closed. The
distillation package surface is at parity with the Rust pipeline:
scorer, scored runs, SFT export, audit baselines all available
on the Go side.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Follow-up to b216b7e (which shipped the SFT export substrate). This
commit ports the synthesis logic, completing the migration:
- SynthesizeSft(scored, ev, recordedAt, sftID) → *SftSample
Mirrors the Rust synthesizeSft byte-for-byte. Returns nil for
extraction-class records + empty-text records (same skip
semantics as Rust).
- LoadEvidenceByRunID(scoredPath, cache) reads the paired evidence
JSONL (path derived by /scored-runs/ → /evidence/ replacement).
Per-call cache so multiple scored-runs files in the same dir
don't reload the same evidence.
- buildInstruction maps source_file stem → per-class instruction
template. All 8 templates (scrum_reviews, mode_experiments,
auto_apply, audits, observer_reviews, contract_analyses,
outcomes, default) match Rust output exactly so a/b validation
between runtimes can diff JSONL byte-for-byte.
- stemFromSourceFile strips data/_kb/ prefix + .jsonl suffix.
- ExportSft now writes data/distilled/sft/sft_export.jsonl with
the synthesized samples (DryRun=true skips file write).
Per-class templates verified by 8-case sub-test:
- scrum_reviews → "Review the file '...' against the PRD..."
- mode_experiments → "Run task_class='...' for file..."
- auto_apply → "Auto-apply: emit a 6-line surgical patch..."
- audits with phase: prefix → strips to bare phase name
- observer_reviews → "Observer-review the latest attempt..."
- contract_analyses with permit: prefix → strips to permit ID
- outcomes → "Run scenario; report per-event outcome..."
- unknown source → "Source 'X' run; produce the appropriate output"
Caveat documented inline: contract_analyses uses ev.metadata.contractor
in Rust to produce "Analyze contractor 'X' for permit 'Y'" when
present. Go's EvidenceRecord doesn't carry a free-form metadata bag
yet, so we always emit the no-contractor form. Operators needing
contractor-aware instructions can extend EvidenceRecord with an
explicit Metadata field (separate ADR).
Test additions (5 new):
- TestSynthesizeSft_PerSourceClass: 8 sub-cases, one per template
- TestSynthesizeSft_RejectsExtraction: extraction-role records skipped
- TestSynthesizeSft_RejectsEmptyText: empty/whitespace text skipped
- TestSynthesizeSft_ContextAssembly: matrix + pathway + model
context string formatting matches Rust " · " join
- TestExportSft_FullPort_WritesJSONL: end-to-end fixture, asserts
output contains expected instruction + omits firewalled records
Pre-existing TestExportSft_PartialPort_FirewallFires renamed +
updated to TestExportSft_FirewallFiresBeforeEvidenceLoad — reflects
the new contract that records passing the firewall but lacking
evidence land in "not-instructable" rather than being silently
exported. Honest semantics shift documented in the test.
OPEN #2 now fully closed (was: substrate-only). The synthesis path
no longer requires the Rust pipeline to be invoked — Go-side
operators can run the full distillation export end-to-end.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Substantial wave addressing all 4 prior OPEN items. Three closed in
full, one partially (the speculative half deliberately deferred).
OPEN #1 — Periodic fresh→main index merge (FULL):
- POST /v1/vectors/index/{src}/merge with {dest, clear_source}
- Idempotent on re-runs (existing-in-dest items skipped)
- internal/vectord/index.go: new Index.IDs() snapshot method +
i.ids tracker field as canonical ID set, independent of meta
map's nil-vs-{} sparseness (was a real bug — IDs() backed by meta
alone missed items added with nil metadata)
- 4 cmd-level integration tests (happy path drain+clear, dim
mismatch, dest not found, self-merge rejection) + 1 unit test
- DecodeIndex backward-compat: old envelopes restore i.ids from
meta keys (best effort; new items going forward use the tracker)
OPEN #2 — Distillation SFT export (SUBSTRATE):
- internal/distillation/sft_export.go ports the load-bearing half:
IsSftNever predicate + ListScoredRunFiles (data/scored-runs/YYYY/
MM/DD walk) + LoadScoredRunsFromFile + partial ExportSft.
- Synthesis (instruction/input/response generation) deferred to a
separate wave — too big for this session, but the substrate
makes the next wave a port-not-design exercise.
- TestSftNever_PinsExpectedSet locks the contamination firewall
set: if a future commit adds/removes from SftNever, this test
fails — forcing the change through review.
- 5 new tests; firewall fires end-to-end through the partial port.
OPEN #3 — Distribution drift via PSI (FULL):
- internal/drift/drift.go: ComputeDistributionDrift via Population
Stability Index. Standard finance/risk metric, well-defined
verdict tiers (stable < 0.10, minor 0.10–0.25, major ≥ 0.25).
- Equal-width bucketing over combined min/max so neither dist
falls outside; epsilon-clamping for empty buckets so log doesn't
blow up. Per-bucket breakdown for drilldown.
- Pairs with the existing ComputeScorerDrift: scorer drift is
categorical, distribution drift is continuous. Different shapes,
same package.
- 7 new tests covering identical-is-stable, hard-shift-is-major,
moderate-detected-not-stable, empty-inputs-safe, all-identical-
safe, bucket-counts-conserved, num-buckets-clamping.
OPEN #4 — Ops nice-to-haves (PARTIAL — wall-clock done, others
deferred):
- (a) Real-time wall-clock for stress harness: per-phase elapsed
time logged to stdout as it runs (`[stress] phase NAME starting
(T+12.3s)` + `[stress] phase NAME done — 8.5s (T+20.8s)`).
Output.PhaseTimings + Output.TotalElapsedMs in JSON.
- (b) chatd fixture-mode S3 mock + (c) liberal-paraphrase
calibration: not actioned — no fired trigger, would be
speculative. Documented as deferred-until-need rather than
ignored. Per the project's discipline ("don't add features
beyond what the task requires").
OPEN list now empty / steady-state. Future items will land as
production triggers fire.
Build + vet + tests green; 18 new tests across the 4 closures.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Real wire-up gap discovered post-scrum: Demand.Role was already
extracted at every call site in multi_coord_stress (44 occurrences,
both contract-driven and LLM-parsed inbox-triggered paths), but
neither matrixSearch nor playbookRecord accepted role in their
signatures. Cross-role gate (real_001..real_004 work) was bypassed
for the entire multi-coord harness — recordings and queries went
through with empty role, gate fell back to lenient behavior.
Fix:
- matrixSearchReq gains query_role field
- matrixSearch signature: (..., query, role string, ...)
- tracedSearch wrapper gains role param + emits it in span input
metadata for Langfuse visibility
- playbookRecord signature: (..., query, role, ...) — body emits
role only when non-empty (preserves backward compat at API)
- 14 call sites updated:
contract-driven Demand loops → d.Role
LLM-parsed inbox path → parsed.Role (qwen2.5 already extracts it)
swap path (warehouseDemand) → warehouseDemand.Role
reissue path → ev.Role (captured at original event time)
fresh-verify (resume snippet, no role concept) → ""
Build clean, vet clean, all tests pass. Cross-role gate now fires
end-to-end across the multi-coord harness — matches the playbook_lift
harness's coverage from the original real_001 fix.
This closes the symmetric gap to scripts/playbook_lift's existing
wire-through. Both production-shape harnesses now exercise the role
gate; future reality tests automatically inherit the protection.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
5 explicit-negation queries ("Need Forklift Operators in Aurora IL,
NOT in Detroit", "excluding Cornerstone Fabrication roster", etc.)
through the standard playbook_lift harness. Goal: characterize
whether the substrate has negation handling or silently treats
"NOT X" as "X".
Headline: substrate has zero negation handling. Cosine on dense
embeddings tokenizes "NOT in Detroit" identical to "in Detroit"
plus noise — there is no logical-quantifier representation in the
embedding space. This is a structural property of dense embeddings,
not a substrate bug.
Per-query observations:
- Q1 (Aurora IL, NOT Detroit): all top-10 rated 1-2/5 by judge
- Q2 (NOT Beacon Freight): top-1 rated 4/5 — accidentally OK
because role+city signal pulled non-Beacon worker naturally
- Q3 (excluding Cornerstone): unanimous 1/5 across top-10
- Q4 (NOT Detroit-area): all top-10 rated 1-2/5
- Q5 (exclude Heritage Foods): top-1 rated 4/5 — accidentally OK
The judge IS the safety net: when retrieval can't honor the
constraint, the judge refuses to approve any result. That's the
honesty signal — `discovery=0` for the run aggregates it.
No code change. The architectural answer for production is:
- UI surfaces an "exclude" affordance that populates ExcludeIDs
(already supported, added in multi-coord stress 200-worker swap)
- Coordinators don't type natural-language negation — they click
- Substrate's role: surface honesty signal (judge ratings) + don't
pretend to honor unparseable constraints
Adding NL-negation handling at the substrate level would be product
debt — it would let coordinators type sloppier queries that
silently fail when the LLM extractor misses a phrasing. Don't ship
until production traffic demonstrates demand for it.
Findings: reports/reality-tests/real_005_findings.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3-lineage scrum review of the role-gate work (commits 7f2f112..0331288)
ran Opus 4.7 / Kimi K2.6 / Qwen3-coder via scripts/scrum_review.sh.
All three flagged the same edge case: the homegrown plural-stripper
in roleNormalize would collapse non-plural-s tokens like "Sales" →
"Sale", "Logistics" → "Logistic", "Operations" → "Operation". In a
staffing domain those are real role names; the silent normalization
would have caused false role-equality matches and re-opened the
cross-role bleed for those clusters.
Fix:
- nonPluralSWords allowlist for known staffing-domain non-plural-s
tokens (sales, logistics, operations, facilities, premises, news,
physics, economics, mathematics, analytics).
- Last-word-only stripping ("Sales Associate" stays whole; only
"Associates" head noun is plural-checked).
- -ss ending check so "Press Operator", "Boss" don't lose their s.
- strings.ToLower + strings.TrimSpace replace the homegrown rune-
loop ASCII normalizer (Opus INFO — minor cleanup, folded in).
Tests:
- TestRoleNormalize_NonPluralS: 18 cases covering the allowlist,
-ss ending, real plurals (Operators → Operator, Boxes → Box),
multi-word real plurals (Forklift Operators → forklift operator),
whitespace/case tolerance.
- TestRoleEqual_NonPluralS: gate-level pairing — proves equal-
shape allowlisted tokens compare equal AND that "Sales" ≠ "Sale"
(the original bug shape).
- Existing TestRoleEqual_PluralAndCase still green (refactor
preserved behavior).
Other scrum findings dispositioned (not actioned):
- Opus WARN on empty-role fail-open semantics: documented
backward-compat behavior; production path closes via opt-in LLM
extractor (real_004).
- Opus INFO on unsynchronized package-global cache map: harness is
single-goroutine; add sync.Mutex when/if it parallelizes.
- Opus INFO on parallel constructor (NewPlaybookEntryWithRole vs
optional arg): API smell only, both forms preserved.
- Kimi 2 BLOCKs (NewPlaybookEntryWithRole missing, ApplyPlaybookBoost
signature breakage): FALSE positives. Pre-push smoke chain green
on 0331288, both symbols + all call sites compile clean. Matches
feedback_cross_lineage_review.md's documented Kimi truncation
behavior — Kimi BLOCKs warrant trace verification before action.
Disposition (local): reports/scrum/_evidence/2026-04-30/verdicts/role_gate_v1_disposition.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
real_003 left a known-weak hole: shorthand-style queries
("{count} {role} {city} {state} ...") have no separator between
role and city, so a regex can't reliably extract — leaving the
cross-role gate disabled when both record AND query are shorthand.
This commit adds a roleExtractor with regex-first + LLM fallback:
- Regex first (fast, deterministic) — handles need + client_first +
looking from real_003b. ~75% of styles, no LLM cost paid.
- LLM fallback when regex returns empty AND model is configured —
Ollama-shape /api/chat with format=json, schema-tight prompt,
temperature 0. ~1-3s on local qwen2.5.
- Per-process cache — paraphrase + rejudge passes reuse the same
query 4× per run; cache prevents 4× LLM cost.
- Off-by-default — opt-in via -llm-role-extract flag (CLI) and
LLM_ROLE_EXTRACT=1 env var (harness wrapper). real_003b shipping
config unchanged unless explicitly enabled.
8 new tests in scripts/playbook_lift/main_test.go:
- TestRoleExtractor_RegexFirst: LLM not called when regex matches
- TestRoleExtractor_LLMFallback: shorthand goes to LLM
- TestRoleExtractor_LLMOffLeavesEmpty: opt-in default preserved
- TestRoleExtractor_Cache: 3 calls = 1 LLM hit
- TestRoleExtractor_NilSafe: nil receiver runs regex only
- TestExtractRoleViaLLM_HTTPError + _BadJSON: failure paths
- TestRoleExtractor_ClosesCrossRoleShorthandBleed: synthetic
witness for the real_003 scenario — both record + query are
shorthand, regex returns "" for both, LLM produces DIFFERENT
role tokens for CNC vs Forklift, so matrix gate's cross-role
rejection (locked separately in
TestInjectPlaybookMisses_RoleGateRejectsCrossRole) fires
correctly. This is the load-bearing verification.
Reality test real_004 ran the same 40-query stress as real_003 with
LLM extraction on. Cross-style same-role boosts fired correctly
across all 4 styles for Loaders + Packers + Shipping Clerk clusters
(including shorthand → other-style transfer). No cross-role bleed
observed. The reality test alone can't be a clean "with vs without"
comparison (HNSW build is non-deterministic across runs, and
real_004 stochastics didn't trigger a shorthand recording at all),
which is why the unit-test witness exists.
Production note (in real_004_findings.md): LLM extraction is for
reality-test coverage of arbitrary query shapes. Production should
extract role at INGEST time (when the inbox parser already runs an
LLM) and pass already-resolved role through requests — same shape
as multi_coord_stress's existing Demand{Role: ...} model. The hot
path should never need the harness extractor's per-query LLM cost.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Stress-tests the role gate with 40 queries (10 fill_events rows × 4
styles): need, client_first, looking, shorthand. Each row's role +
client + city stays the same; only the surface phrasing changes.
real_003 (original extractor) confirmed the shorthand-vs-shorthand
failure mode: CNC Operator shorthand recording leaked w-2404 onto
Forklift Operator shorthand query within the same Beacon Freight
Detroit cluster. Both record + query had empty role (extractor
returns "" for shorthand because there's no separator between role
and city), gate disabled, distance check passed, bleed fired.
Fix: extended extractRoleFromNeed to handle client_first
("{client} needs N {role} in...") and looking ("Looking for N
{role} at...") patterns. Shorthand left intentionally unmatched —
"Forklift Operator Detroit" is shape-indistinguishable from
"Forklift" + "Operator Detroit" without an LLM extractor or known-
cities lookup.
real_003b (extended extractor) verifies bleed closed across all 4
styles for this dataset. Forklift Operator queries keep w-2136 (the
cold-pass-correct match) regardless of which style the query came
in. Same-role boosts now fire correctly across styles — a CNC
Operator recording made in `looking` style boosts the CNC need-form
query.
scripts/cutover/gen_real_queries.go: added -styles flag with values
need|client_first|looking|shorthand|all (default need preserves
real_001/002 behavior). Tests/reality/real_coord_queries_v2.txt is
the 40-query stress file.
scripts/playbook_lift/main_test.go: 10 sub-tests lock the four
documented patterns + shorthand limitation + lift-suite-style
queries (no clean role, returns empty as expected).
Aggregate metrics:
- real_003 (original): disc=7, lift=7, boost=14, meanΔ=-0.108
- real_003b (extended): disc=11, lift=10, boost=31, meanΔ=-0.202
The growth reflects more LEGITIMATE same-role same-cluster transfer
firing across styles, not bleed (verified by per-cluster bleed
table — Forklift Operator queries unchanged across all 4 styles).
Known limitation documented in real_003_findings.md: same-cluster,
same-role queries in shorthand still embed close enough that a
shorthand recording could bleed onto a different-role shorthand
query if both record + query strip role. Closing this requires
LLM extraction or known-cities lookup at record + query time.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
real_001 surfaced same-client+city queries bleeding across roles:
Q#2 (Forklift Operator @ Beacon Freight Detroit) recorded e-6193
in the playbook corpus. Q#5 (Pickers same client+city) and Q#10
(CNC Operator same client+city) embedded within 0.13-0.18 cosine of
Q#2's query — well inside the 0.20 inject threshold — so e-6193
injected on both, demoting the cold-pass-correct workers.
Root cause: the inject distance threshold isn't tight enough on
the same-client+city cluster. Cosine collapses queries that share
city + client + count-token + time-token regardless of role. The
existing judge gate is per-injection at record time and doesn't
fire at retrieve time.
Fix: structural role gate in front of both Shape A boost and
Shape B inject. PlaybookEntry gains Role; SearchRequest gains
QueryRole. When both are non-empty and differ under roleEqual's
case+plural normalization, the entry is rejected before BoostFactor
or judge-gate logic runs.
Backward-compat: empty role on either side disables the gate —
preserves behavior for the lift suite's free-form multi-constraint
queries that have no clean single role. Caller-supplied (not
inferred), so existing recordings unaffected.
Wire-through:
- internal/matrix/playbook.go: Role field, NewPlaybookEntryWithRole,
roleEqual helper with plural+case normalization
- internal/matrix/retrieve.go: QueryRole on SearchRequest, threaded
to both ApplyPlaybookBoost + InjectPlaybookMisses
- cmd/matrixd/main.go: role on POST /matrix/playbooks/record + bulk
- scripts/playbook_lift/main.go: extractRoleFromNeed regex pulls
role from "Need N {role}{s} in" queries (the fill_events shape);
free-form queries fall back to empty (gate disabled)
Tests (5 new):
- TestInjectPlaybookMisses_RoleGateRejectsCrossRole: exact Q#10
scenario (distance 0.135, recorded "Forklift Operator", query
"CNC Operator") — locks the bleed at unit level
- TestInjectPlaybookMisses_RoleGateAllowsSameRole: Forklift Operator
recording fires on Forklift Operators query (plural normalization)
- TestInjectPlaybookMisses_RoleGateBackwardCompat: empty Role on
either side = gate disabled, preserves current behavior
- TestApplyPlaybookBoost_RoleGateRejectsCrossRole: Shape A defense
in depth — boost doesn't fire on cross-role even when answer is
in cold top-K
- TestRoleEqual_PluralAndCase: case + -s + -es plural normalization
Verification (real_002, same query set as real_001):
- Q#5 Pickers @ Beacon Freight: e-6193 → e-8499 (no bleed)
- Q#10 CNC Operator @ Beacon Freight: e-6193 → w-2404 (no bleed)
- Discoveries + lifts unchanged at 2 each (same-role lift still fires)
- Mean Δdist tightens from -0.127 to -0.040 (boosts no longer
pulling distances through the floor on cross-role mismatches)
Findings: reports/reality-tests/real_002_findings.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First retrieval probe with non-synthetic query distribution. Pulls
N rows from /home/profit/lakehouse/data/datasets/fill_events.parquet
(real-shape demand data) and translates each to the natural language
a coordinator would type: "Need {count} {role}s in {city} {state}
starting at {at} for {client}".
Headline: 8/10 cold-pass top-1 = judge-best on real distribution.
Substrate works on queries it was never trained for. v2-moe + workers
corpus carry the load.
Surfaced finding (the real value of running this): same-client+city
queries cluster, and Shape A's distance boost bleeds across roles
within the cluster. Q#2 (Forklift @ Beacon Freight Detroit) records
e-6193 in the playbook corpus. Q#5 (Pickers same client+city) and
Q#10 (CNC Operator same client+city) inherit e-6193 at warm top-1
even though:
- Neither query has its own recorded playbook.
- Neither warm pass triggers a Shape B inject (boosted=0).
- The roles are different staffing categories.
Q#10 specifically demoted the cold-pass-correct w-3759 (judge rating
4 at rank 0) for a worker who was approved by the judge for a
different role on a different query.
Why the lift suite missed it: synthetic queries use 7 disjoint
scenario buckets (forklift+OSHA+WI / CDL+IL / etc.). Real demand
clusters on (client, city). The cluster doesn't exist in the
synthetic distribution.
Why the judge gate doesn't catch it: the gate (5a3364f) is
per-injection at record time. After approval the worker rides Shape A
distance boosts on all later same-cluster queries with no second
gate call.
Becomes new OPEN #1. Fix candidate: role-scoped playbook corpus
metadata + Shape A boost gate on role match. Cheap; doesn't need
new judge calls.
Files:
- scripts/cutover/gen_real_queries.go: parquet → coordinator NL
- tests/reality/real_coord_queries.txt: 10 generated queries
- reports/reality-tests/playbook_lift_real_001.md: harness output
- reports/reality-tests/real_001_findings.md: the reading
Repro:
go run scripts/cutover/gen_real_queries.go -limit 10 > tests/reality/real_coord_queries.txt
QUERIES_FILE=tests/reality/real_coord_queries.txt RUN_ID=real_001 \
WITH_PARAPHRASE=0 WITH_REJUDGE=0 ./scripts/playbook_lift.sh
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First concrete cutover artifact: scripts/cutover/embed_parity.sh
brings up Go embedd + gateway alongside the live Rust gateway,
hits both /ai/embed and /v1/embed with the same forced model, and
emits a per-date verdict report under reports/cutover/.
Why embed first: the parity invariant is one math identity (cosine
sim of vectors against same input). Retrieve has thousands of edge
cases. If embed parity holds, all downstream vector consumers
inherit confidence; if it doesn't, we catch it in 30s instead of
after a flip.
Verdict 2026-04-30: 5/5 samples cosine=1.000000 with model forced
to nomic-embed-text (v1). Same with nomic-embed-text-v2-moe (both
Ollamas have it loaded). Math is provably equivalent across the
gateway plumbing.
Drift catalog (reports/cutover/SUMMARY.md):
- URL: Rust /ai/embed vs Go /v1/embed
- Wire: Rust {embeddings, dimensions} (plural) vs Go {vectors,
dimension} (singular). Wire-format adapter is the only real
cutover work for this endpoint.
- L2 norm: Rust unit vectors (~1.0); Go raw Ollama (~20-23). Same
direction (cos=1.0); harmless under cosine-distance HNSW (which
is Go vectord's default), but worth fixing in internal/embed/
before extending to euclidean indexes.
reports/cutover/ now tracked (joined the scrum/ + reality-tests/
exemptions in .gitignore).
Next probe: /v1/matrix/retrieve ↔ Rust /vectors/hybrid for the
real user-facing retrieve path. Embed parity gives that probe a
clean foundation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`jq --arg` and `curl --data-binary @-` both read stdin/argv-bound
buffers. Diffs >~128KB blow past the kernel's argv limit even when
piped via stdin (because we still build `body` as a shell variable
first, then feed it to curl). Voice-ai full bundle was 156K and
hit it.
Switch to writing user/system/body to mktemp files, jq reads via
--rawfile, curl reads via @file. Same on-the-wire shape, no argv
involvement. Cleanup with rm at the end.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds langfuseMiddleware in internal/shared so every daemon's
shared.Run gets free production-traffic trace visibility when
LANGFUSE_URL + LANGFUSE_PUBLIC_KEY + LANGFUSE_SECRET_KEY are set.
Same env names + file shape as the multi_coord_stress driver, so
operators ship one /etc/lakehouse/langfuse.env across the deploy.
Wiring is auth-gated: middleware runs INSIDE the RequireAuth group,
so 401s from credential-stuffing don't pollute traces. /health is
exempt so LB probes don't either. Missing env vars → nil client →
middleware is a passthrough no-op (fail-open per ADR-005 5.1).
Bundled deploy:
- langfuse.env.example template (mode 0640, root:lakehouse)
- 11 systemd units gain `EnvironmentFile=-/etc/lakehouse/langfuse.env`
(leading - so missing file = OK)
- REPLICATION.md bootstrap section documents setup
Tests (4): nil passthrough, /health bypass, real-request emission,
status-writer wrapping. All green.
STATE_OF_PLAY OPEN list: 5 rows → 4 rows.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Lift suite run #004 left two unresolved tail issues:
- Q6 ("Forklift loader") ↔ Q7 ("Hazmat warehouse, cold storage")
swap recordings as warm top-1 because their embeddings are within
0.20 cosine of each other. Distance gate can't tell them apart.
- Q9 + Q15 lose paraphrase recovery when qwen2.5 rephrases past the
0.20 threshold. Distance says "drift too far"; sometimes the drift
is real (skip), sometimes the paraphrase is still on-domain (don't
want to skip).
Multi-coord run #008's judge re-rating proved the LLM can
distinguish: Q3 crane case landed at distance 0.23 (looks tight)
but rating 1 (irrelevant). The judge sees domain mismatch the
embedder doesn't.
This commit lifts that pattern into the matrix substrate. Shape B
inject now optionally routes every candidate through a judge gate
before the rank insert lands. Distance + judge BOTH have to approve.
internal/matrix/playbook.go:
- InjectPlaybookMisses signature gains a query string + an
optional InjectGate. nil gate preserves pre-judge-gating
behavior (current tests already pass with nil).
- New InjectGate interface + InjectGateFunc adapter for tests
and non-LLM callers.
- Per-candidate gate.Approve(query, hit) call inserted between
the dedup and the inject. Rejected candidates skip silently;
injected count reflects post-gate decision.
internal/matrix/judge.go (new, ~140 lines):
- LLMJudgeGate calls an Ollama-shape /api/chat endpoint with the
same 1-5 staffing-rubric prompt that worked in multi_coord
run #008. fail-closed on HTTP/JSON errors (don't inject if
judge can't speak — better miss than wrong-domain).
- NewLLMJudgeGate returns nil when URL or Model is empty,
matching InjectGate's nil-means-no-judge semantics.
internal/matrix/retrieve.go:
- SearchRequest gains JudgeURL, JudgeModel, JudgeMinRating
fields. Run() builds an LLMJudgeGate when set; passes nil
otherwise. Backward compatible — existing callers see no
behavior change.
Tests:
- TestInjectPlaybookMisses_GateRejectsCandidate (rejectAll → 0
injected, even with tight distance)
- TestInjectPlaybookMisses_GateApprovesCandidate (approveAll →
same as nil-gate behavior)
- TestInjectPlaybookMisses_GateSeesCorrectQuery (gate receives
CURRENT query + RECORDED query separately so it can score
the (current, candidate) pair)
- All 5 existing inject tests updated to new signature
go test ./internal/matrix → all 8 inject tests pass.
go test ./internal/matrix ./internal/shared ./cmd/{matrixd,
queryd,pathwayd,observerd} → all green.
STATE_OF_PLAY:
- OPEN item #1 (judge-gated injection) closed.
- DO NOT RELITIGATE adds the substrate-level judge-gate lock.
- OPEN list now 5 rows (was 6).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sprint 4 row removed (shipped: a59ef5b systemd + 54a05d9 docker).
ADR-006 row already dropped on the previous STATE update.
Two lift-suite tail items (Q6↔Q7 adjacent-query, Q9/Q15 liberal-
paraphrase) consolidated into one "judge-gated playbook injection"
row — both are downstream of the same fix (let the judge approve
before Shape B inserts). Captures the design lineage from
multi-coord run #008's judge-rating pattern.
Three items folded into a single "operational nice-to-haves" row:
real-time clock, chatd fixture storage half, liberal-paraphrase
calibration. None are product-blocking; each lights up when
someone hits its specific trigger.
Reorder reflects leverage on the active product theory (multi-
coord staffing co-pilot via the 5-loop substrate), not effort:
1. Judge-gated injection (lift quality + lift-tail closure)
2. Wider Langfuse instrumentation (production observability)
3. Fresh→main merge (operational hygiene as the corpus grows)
4. Distillation full port (production dependency, not yet)
5. Drift quantification (research)
6. Operational nice-to-haves
Lead-in note added: "Items move to closed when the work demands
them, not on a calendar." Locks intent against future drift toward
a sprawling todo list.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Parallel deploy target to the systemd units that landed in a59ef5b.
Single image carries all 11 daemons; docker-compose runs one
container per daemon with the same dependency graph as the systemd
units. Useful when systemd isn't available (Mac dev, remote VMs
without root) or when isolation to a private docker network is
preferred.
Dockerfile (multi-stage):
- Builder: golang:1.25-bookworm. DuckDB cgo needs gcc + glibc;
alpine's musl doesn't link the official duckdb-go bindings cleanly.
- Runtime: debian:bookworm-slim — same libc, much smaller surface.
Adds ca-certificates (outbound HTTPS to OpenRouter/OpenCode/Kimi),
curl + jq (in-container healthchecks + smoke probes), tini (PID 1
signal forwarding so docker stop sends SIGTERM to the daemon, not
to a wrapper).
- Single image, multiple binaries. Ships all 11 cmd/* + 3 scripts/
(staffing_workers, playbook_lift, multi_coord_stress) so deployed
stacks can run reality tests against themselves.
- Non-root runtime user (uid 999 lakehouse). Layout matches
/usr/local/bin/lakehouse/<daemon> from REPLICATION.md.
- ENTRYPOINT=tini; no default CMD — operators / compose pick
which daemon explicitly.
docker-compose.yml (11 services):
- Same dependency graph as deploy/systemd/. depends_on with
service_healthy condition matches Requires= equivalents:
catalogd → storaged
ingestd → storaged + catalogd
queryd → catalogd
matrixd → embedd + vectord
- Gateway uses bare depends_on (no health condition) — Wants=
equivalent so single-upstream restart doesn't cascade.
- chatd has per-provider env_file entries (one each for
ollama_cloud, openrouter, opencode, kimi) — missing files are
silently OK, matching the systemd unit's EnvironmentFile=- list.
- Persistent state on the lakehouse-state named volume; commented
driver_opts shows how to bind to a host path for off-volume
backups.
.dockerignore:
- Excludes bin/ + reports/ + data/ + git metadata + .env files.
- Especially excludes lakehouse.toml/secrets-go.toml/auth.env so
local dev configs don't accidentally bake into a published image.
REPLICATION.md gains a Docker section between systemd setup and
the logs section. Ten-line copy-paste from "git clone" to
"docker compose up -d", plus a docker-vs-systemd differences
table covering process supervision, logs, restart policy, file
ownership, host networking quirks, and backup targets.
Validation: docker compose config --quiet → exit 0 (with
placeholder env files in place).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ADR-003 locked the auth substrate; ADR-006 ratifies the operator
playbook + adds two implementation pieces needed for Sprint 4
deployment: env-resolved tokens and dual-token rotation.
Six decisions locked in docs/DECISIONS.md:
- 6.1: Non-loopback bind requires auth.token (mechanical gate at
shared.Run, already implemented; this ratifies it).
- 6.2: Token from env, not TOML. /etc/lakehouse/auth.env (mode 0600)
loaded by systemd EnvironmentFile=. New TokenEnv field on
AuthConfig defaults to "AUTH_TOKEN".
- 6.3: AllowedIPs for inter-service same-trust-domain; Token for
cross-trust-boundary (gateway ↔ external).
- 6.4: /health stays unauthenticated; everything else under
shared.Run is gated. Already implemented; ratified here.
- 6.5: Token rotation is dual-token. New SecondaryTokens []string
on AuthConfig — both primary and any secondary pass auth
during the rotation window. Implemented in this commit.
- 6.6: TLS terminates at the network edge (nginx/Caddy), not
in-process. Daemons stay HTTP-only; internal traffic stays
on private subnets per Decision 6.3.
Implementation:
- internal/shared/config.go: AuthConfig gains TokenEnv +
SecondaryTokens fields. New resolveAuthFromEnv() called by
LoadConfig fills Token from os.Getenv(TokenEnv) when Token is
empty. TokenEnv defaults to "AUTH_TOKEN" so the happy path needs
no TOML config.
- internal/shared/auth.go: RequireAuth pre-encodes Bearer headers
for primary + every secondary token; per-request constant-time
compare walks the slice. Fast path is 1 compare (primary).
Tests:
- TestLoadConfig_AuthTokenFromEnv (3 sub-tests): default env name,
custom token_env, explicit Token wins over env.
- TestRequireAuth_SecondaryTokenAccepted: both primary + secondary
tokens pass during rotation window.
- TestRequireAuth_SecondaryTokensOnly: only-secondary path works
for the case where primary was just promoted-to-empty mid-rotation.
go test ./internal/shared all green; existing auth_test.go
unchanged (constant-time compare path preserved).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Cross-lineage scrum on bundle 87cbd10..f971e64 (3,652 lines)
produced 4 actionable findings, all defensive hardening.
1. (Opus WARN) internal/langfuse/client.go:queue
Synchronous Flush at maxBatch threshold blocked the calling
goroutine for the full 5s HTTP timeout when Langfuse hiccupped,
defeating the "best-effort, never blocks calling path" contract
in the package doc. Now fire-and-forget via goroutine.
2. (Opus + Kimi convergent) cmd/observerd/main.go:handleInbox
- Free-form priority string was accepted; "nonsense" passed
through unchecked. Now closed enum: urgent|high|medium|low (+
empty defaults to medium). Tested: TestInbox_RejectsBadPriority.
- No size cap on body, only emptiness check; multi-MB payloads
would bloat observer's ring + JSONL. Now 8 KiB cap returns 413.
Tested: TestInbox_RejectsOversizedBody.
- Subject/sender/tag concatenated into InputSummary without
newline stripping; embedded \n could corrupt JSONL line-based
parsers. New sanitizeInboxField strips \r\n + caps at 256 chars
before interpolation.
3. (Opus INFO) scripts/multi_coord_stress/main.go
Removed dead `must[T]` generic — tracedSearch took over the
fail-fast role for matrix searches, so the helper became unused.
4. (Opus INFO) scripts/multi_coord_stress/main.go:Event
`JudgeRating int` collapsed "judge errored" and "judge said
unrated" both to 0. Changed to *int — nil = errored, 1-5 =
verdict. judgeInboxResult still returns 0 on error; caller
gates on > 0 before assigning.
Dismissed (with rationale):
- Opus WARN ExcludeIDs ordering: verified by code read — filter
applies after sort + before top-K truncation as documented;
no slot waste possible.
- Opus INFO 10 prior-run reports contradict #011: those are
point-in-time snapshots; intentional history.
- Kimi INFO Langfuse error suppression: design intent (best-effort
per package doc).
- Kimi INFO contract schema validation: defer until contract count
grows enough to make hand-edit drift a real risk.
- Kimi INFO paraphrase prompt duplicated across lift + multi_coord:
defer (lift to internal/paraphrase/ when a third consumer appears).
- Qwen HOLD: single-line, no actionable finding.
go test ./cmd/observerd ./internal/langfuse all green; multi_coord
driver builds clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pre-push hook caught the regression — the smoke hardcoded
MODEL = "nomic-embed-text" and the bump to nomic-embed-text-v2-moe
in 4da32ad failed the gate.
Fix: glob-match the family prefix (nomic-embed-text*). Both v1 and
v2-moe are 768d drop-ins; the property the smoke is locking is
dim + distinct-vectors, not the exact model variant. Operators
swap the variant in lakehouse.toml without needing to touch the
smoke.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Anchor was last touched at v4 split-threshold; since then the
multi-coord stress harness landed end-to-end across 11 commits.
Future sessions reading this file need to see the verified state,
not derive from git log.
Major additions:
- New "Multi-coordinator stress test (Phase 1 → 3)" section in
VERIFIED WORKING. 11-row capability table covering per-coord
playbook isolation, diversity metrics, paraphrase handover,
ExcludeIDs swap, fresh-resume two-tier, inbox endpoints, LLM
demand parsing, judge re-rating, Langfuse tracing.
- Substrate-gains list under that section: ExcludeIDs on
SearchRequest, observer.SourceInbox + /observer/inbox,
internal/langfuse client, embedd default bumped to v2-moe,
two-tier fresh_workers index pattern.
- Last-verified bumped to 16:42 CDT on the run #011 anchor.
DO NOT RELITIGATE expanded with five new locks:
1. Boost / inject use SEPARATE thresholds (0.5 / 0.20)
2. Multi-coord product theory is empirically VALIDATED
3. Fresh content uses two-tier indexing (fresh_workers)
4. embedd.default_model = nomic-embed-text-v2-moe (don't downgrade)
5. Inbox flow: parse + search + judge + trace
6. Langfuse Go-side client lives at internal/langfuse/
OPEN list refresh:
- Removed: re-judge metric (shipped as b13b5cd), adjacent-query as
separate item (folded into a single "judge-approves-before-inject"
follow-up), liberal-paraphrase (kept).
- Added: real-time 48-hour clock, wider Langfuse instrumentation,
periodic fresh→main merge job.
RECENT VERIFIED WAVE table extended with 11 commits (b13b5cd..5d49967).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 1c-only tracing (commit 7e6431e) was the proof-of-concept.
This commit threads tracing through every phase: baseline / fresh-
resume / inbox burst / surge / swap / merge / handover (verbatim +
paraphrase) / split / reissue. Each phase is a parent span; each
matrix.search / LLM call inside is a child span.
Refactor:
- One run-level trace is created at driver startup.
- New startPhase(name, hour, meta) helper emits a phase span as a
child of the run trace; subsequent emitSpan calls nest under it.
- New tracedSearch(spanName, query, corpora, ...) wraps matrixSearch
with span emission. Every search call site replaced with this so
the input/output JSON (query, corpora, k, playbook, exclude_n →
top-K ids, top1 distance, boost/inject counts) lands in Langfuse.
- Phase 4b's paraphrase generation also emits llm.paraphrase spans.
- Phase 1c's existing inline span emission converted to use the new
helpers (no more inboxTraceID variable).
Run #011 result: trace landed at http://localhost:3001 with 111
observations attached. Span breakdown:
phase.* parents: 9 (one per phase that ran)
matrix.search.baseline: 10
matrix.search.fresh_verify: 3 (top-1 confirmed for all 3 fresh)
observerd.inbox.record: 6
llm.parse_demand: 6
matrix.search.inbox: 6
llm.judge_top1: 6
matrix.search.surge: 12
matrix.search.swap_orig: 1
matrix.search.swap_replace: 1
matrix.search.merge: 6
matrix.search.handover_verbatim: 4
llm.paraphrase: 4
matrix.search.handover_paraphrase: 4
matrix.search.split: 4
matrix.search.reissue: 12
matrix.search.reissue_retrieval_only: 12
─────────────
Total: 111
Browse: http://localhost:3001 → Traces → "multi_coord_stress run"
Each phase is a collapsible section showing per-call timing and
input/output JSON. Operators can drill into any single retrieval
to see exactly what query was issued and what came back.
All other metrics held: diversity 0.026, determinism 1.000,
verbatim handover 4/4, paraphrase handover 4/4, fresh-resume 3/3
at top-1 (two-tier index), 200-worker swap Jaccard 0.000.
This is the FULL TEST J asked for — every action in the run
visible in Langfuse, full input/output drilldown.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Runs #003-#009 surfaced the same finding: fresh workers added
mid-run to the main 'workers' vectord index (5K items) reliably
*absorbed* (HTTP 200) but failed to *surface* in semantic queries
even with content-matching prompts. Distances on the verify queries
sat at 0.25-0.65 against existing workers; fresh items were beyond
top-K. Better embedder (v2-moe) didn't help — distances got TIGHTER
on existing items, pushing fresh items further out of reach.
Root cause: coder/hnsw incremental adds to a populated graph land
in poorly-connected regions and disappear from search traversal.
Known property of HNSW post-build adds; not a bug.
Fix: two-tier index pattern (canonical NRT search architecture).
Fresh content goes to a small "hot" corpus (fresh_workers); main
queries include it in the corpora list and merge results. Hot corpus
has no recall crowding because it's tiny; periodic batch job (post-
G3) merges it into the main index.
Implementation:
- ensureFreshIndex(hc, gw, name, dim) — idempotent POST
/v1/vectors/index. 409 from re-create treated as "already there."
- ingestFreshWorker now takes idx parameter so callers can target
fresh_workers instead of workers.
- multi_coord_stress phase 1b creates fresh_workers index + ingests
3 fresh workers there + searches verifyCorpora=[workers,
ethereal_workers, fresh_workers].
Run #010 result:
fresh-001 (Senior tower crane rigger NCCCO Chicago)
top-1: fresh-001 from fresh_workers, distance 0.143
fresh-002 (Bilingual Spanish/English OSHA trainer Indianapolis)
top-1: fresh-002 from fresh_workers, distance 0.146
fresh-003 (FAA Part 107 drone surveyor Chicago)
top-1: fresh-003 from fresh_workers, distance 0.129
3/3 fresh workers surface at top-1 — the absorption-but-not-
findable issue from runs #003-#009 is closed.
All other metrics held: diversity 0.007, determinism 1.000,
verbatim handover 4/4, paraphrase handover 4/4, swap Jaccard 0.000,
inbox burst all 6 events accepted + traced to Langfuse.
This is the final structural fix for the multi-coord stress
suite. Phase 3 is feature-complete.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Rust side has Langfuse tracing already (gateway/v1/langfuse_trace.rs);
this commit lands Go-side parity so the multi-coord stress harness can
emit traces visible at http://localhost:3001.
internal/langfuse/client.go:
- Minimal Trace + Span + Flush API mirroring what the Rust emitter
uses. Auth: Basic over public_key:secret_key.
- Best-effort posture: errors are slog.Warn'd, never block calling
paths. Same fail-open as observerd's persistor (ADR-005 Decision
5.1) — observability is a witness, not a gate.
- Events buffered until 50, then auto-flushed; explicit Flush() at
process exit.
- Each Trace/Span returns its id so callers can build hierarchies.
multi_coord_stress driver wiring:
- New --langfuse-env flag (default /etc/lakehouse/langfuse.env).
Empty / missing / unparseable file → skip tracing with a logged
warning; run still proceeds.
- Phase 1c (inbox burst) now emits one parent trace + 4 spans per
inbox event:
1. observerd.inbox.record (post to /v1/observer/inbox)
2. llm.parse_demand (qwen2.5 → structured fields)
3. matrix.search (parsed query → top-K)
4. llm.judge_top1 (rate top-1 vs original body)
Each span carries input/output JSON + start/end times so the
Langfuse UI shows a full waterfall per event.
Run #009 result:
Trace landed: "multi_coord_stress phase 1c inbox burst"
Observations attached: 24 (= 6 events × 4 spans)
Tags: stress, phase-1c, inbox
Browseable at http://localhost:3001 by tag query.
Other harness metrics: diversity 0.016, determinism 1.000,
verbatim handover 4/4, paraphrase handover 4/4 — all unchanged
by the tracing addition (best-effort post in parallel).
Phase 1c is the proof-of-concept; future commits can wrap other
phases (baseline / merge / handover / split) in traces too. Once
that's done, the entire stress run becomes scrubbable in Langfuse
without grepping the events JSON.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Run #007 surfaced a tradeoff: LLM-parsed inbox queries produce much
tighter cosine distances (0.05-0.10 in three cases) but lose the
"system has no good match" signal that high-distance results give.
A coordinator UI showing only distance can't tell wrong-domain
matches apart from real ones.
Fix: judge re-rates top-1 against the ORIGINAL inbox body (not the
LLM-parsed query). Coordinators see both:
- distance: how close was retrieval in vector space
- rating: does this person actually fit the original ask
The pair tells the honest story.
Run #008 result on the 6 inbox events:
Demand Top-1 Distance Rating Reading
─────────────────────────────────────────────────────────────
Forklift Cleveland w-3573 0.29 4 Strong
Production Indy e-1764 0.41 3 Adjacent
Crane Chicago e-7798 0.23 1 TIGHT BUT WRONG
Bilingual safety Indy w-3918 0.05 5 Perfect
Drone Chicago e-1058 0.06 5 Perfect (verify e-1058)
Warehouse Milwaukee w-460 0.32 4 Strong
The crane-Chicago case is the architectural-honesty signal at work:
distance 0.23 says "tight match" but the judge says rating 1 reading
the original body. A coordinator seeing only distance would ship the
wrong worker; coordinator seeing distance+rating sees the disagreement
and escalates.
Net distribution: 5/6 rated 3+ (acceptable→perfect), 1/6 rated 1
(irrelevant despite tight cosine). The substrate-honesty signal is
recovered without losing the LLM-parse quality wins.
Cost: 6 extra judge calls (~9s on qwen2.5). Production amortizes
when judge runs only on top-1 of high-priority inbox events; the
search-cost-vs-quality tradeoff lives in the priority gate.
Implementation:
- New JudgeRating int field on Event (omitempty so non-judged
events stay clean in JSON)
- New judgeInboxResult helper, reusing the same prompt structure as
playbook_lift's judgeRate. The two could share an internal package
if a third judge consumer appears.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaced the hard-coded DemandQuery on inbox events with an actual
LLM call: each email/SMS body is parsed by qwen2.5 (format=json,
schema-anchored) into structured {role, count, location, certs,
skills, shift}. The driver then composes a query string from those
fields and runs matrix.search.
This is the real-product flow that the Phase 3 stress test was
asking for: real bodies → real LLM parsing → real search. Before
this commit, the DemandQuery was my hand-crafted string, which
made the inbox phase trivial.
Run #007 result vs #006 (same bodies, parser swapped):
All 6 inbox events parsed cleanly — qwen2.5 nailed:
"Need 50 forklift operators in Cleveland OH for Monday day
shift. OSHA-30 + active forklift cert required."
→ {role:"forklift operator", count:50, location:"Cleveland, OH",
certs:["OSHA-30","active forklift cert"], skills:[], shift:"day"}
Other 5 similarly faithful (indy stayed as "indy", count
defaulted to 1 when unspecified, no hallucinated fields).
LLM-parsed queries produced TIGHTER matches than hard-coded:
Demand #006 dist #007 dist Δ
Crane Chicago 0.499 0.093 -82%
Drone Chicago 0.707 0.073 -90%
Bilingual safety 0.240 0.048 -80%
Forklift Cleveland 0.330 0.273 -17%
Production Indy 0.260 0.399 +53%
Warehouse Milwaukee 0.458 0.420 -8%
Three matches landed at distance < 0.10 — verbatim-replay-tight
territory. Structured queries embed sharper than conversational
hand-crafted strings.
Other metrics unchanged: diversity 0.000, determinism 1.000,
verbatim handover 4/4, paraphrase handover 4/4.
Tradeoff worth flagging: the drone-Chicago case dropped from
distance 0.71 (clear "we don't have one") to 0.07 (confident match
returned). The OOD honesty signal weakens when LLM-parsed structure
makes any closest-neighbor look tight. Future Phase 4 work: judge
re-rates the top match before surfacing, so coordinators see "your
demand was for X but the closest match scored 2/5" rather than just
the worker ID + distance.
Substrate cost: +6 LLM calls per inbox burst (~9s on qwen2.5).
Production would amortize via a small dedicated parser model.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 3 ask: real-world inbox-style event injection during the stress
test. Coordinators in production receive emails + SMS that trigger
contract responses; the substrate has to RECORD these signals AND
react with a search using the embedded demand. This commit lands the
endpoint and exercises it end-to-end in the stress harness.
observerd surface:
- New POST /observer/inbox route — accepts {type, sender, subject,
body, priority, tag} and records as ObservedOp with
Source=SourceInbox. Type must be email|sms; body required;
priority defaults to medium. The handler ONLY records — downstream
triggers (search, ingest, etc.) are the caller's concern, recorded
separately. Keeps the witness role pure.
- New observer.SourceInbox = "inbox" alongside SourceMCP /
SourceScenario / SourceWorkflow.
- Three contract tests on the new route (happy path / bad type / empty
body), router-mount test extended, all green.
Stress harness phase 1c (Hour 9):
- 6 inbox events fire in priority order (urgent → high → medium):
2 urgent emails (forklift Cleveland, production Indianapolis)
1 high email (crane Chicago)
1 high sms (bilingual safety Indianapolis)
1 medium sms (drone Chicago)
1 medium email (warehouse Milwaukee FYI)
- Each event:
1. POSTs to /v1/observer/inbox (recorded by observerd)
2. Triggers matrix.search using a parsed demand (the demand
extraction is hard-coded for now; production needs a small
LLM to parse from body)
3. Captures both as events in the run JSON
Run #006 result (with v2-moe embedder + all phases including inbox):
Diversity:
Same-role-across-contracts Jaccard = 0.000 (n=9)
Different-roles-same-contract Jaccard = 0.046 (n=18)
Determinism: 1.000
Verbatim handover: 4/4 (100%)
Paraphrase handover: 4/4 (100%)
Inbox burst:
6/6 events accepted by observerd (200 status, all recorded)
6/6 triggered searches produced distinct top-1 worker IDs
distance distribution: 0.24 (Indy production) → 0.71 (Chicago
drone surveyor — honest stretch since drones aren't in the
5K-worker corpus, system surfaces closest neighbor at high
distance rather than fabricating)
The drone-Chicago case is the architectural-honesty signal: when
the demand asks for a specialist NOT in the roster, the system
returns the closest semantic neighbor with a distance that flags
"this is a stretch." Coordinators reading distances see "we don't
have a great match here" rather than a confident wrong answer.
Total events captured: 67 (was 61 pre-inbox).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Local Ollama has three embedding models loaded:
nomic-embed-text:latest 137M 768d (previous default)
nomic-embed-text-v2-moe:latest 475M 768d (this commit's default)
qwen3-embedding:latest 7.6B 4096d (would require dim change)
v2-moe is a drop-in upgrade — same 768 dim, 3.5× more params, MoE
architecture. Workers index doesn't need rebuilding, just future ingests
embed with the stronger model.
Run #005 result on the multi-coord stress suite:
Diversity (same-role-across-contracts): 0.080 → 0.000 (n=9)
→ MoE is more discriminating: zero worker overlap across
Milwaukee / Indianapolis / Chicago for shared role names.
The geo + cert + skill context fully separates worker pools.
Different-roles-same-contract: 0.013 → 0.036 (still ~96% diff)
Determinism: 1.000 (unchanged)
Verbatim handover: 4/4 (100%)
Paraphrase handover: 4/4 (100%)
200-worker swap: Jaccard 0.000 (unchanged — still perfect)
Fresh-resume verify: STILL doesn't surface fresh workers in top-8.
With v2-moe, distances increased (top-1 = 0.43–0.65 vs v1's 0.25–0.39)
— the embedder is MORE discriminating, but the fresh worker's
vector still doesn't outrank the 8th-best existing worker. Now
suspect of being an HNSW post-build add issue (coder/hnsw
incremental adds can land in hard-to-reach graph regions, not an
embedder problem). Better embedder didn't fix it; needs a
different strategy: full index rebuild after fresh adds, or
explicit playbook-layer score boost for fresh workers, or
hybrid (keyword + semantic) retrieval. Phase 3 investigation.
Cost: ingest is ~5× slower (workers 20s→100s; ethereal 35s→112s).
Acceptable for the quality jump on diversity. Real production with
incremental ingest won't pay this once-per-deploy.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three Phase 2 additions land in this commit:
1. matrix.SearchRequest gains ExcludeIDs ([]string) — filters specific
worker IDs out of results post-retrieval, AND skips them at the
playbook boost+inject step (so excluded answers can't sneak back
via Shape B). Real-world driver: coordinator placed N workers,
client asks for replacements, system needs alternatives, not the
same N. Threaded through retrieve.go after merge but before
metadata filter so excluded IDs don't waste post-filter top-K slots.
2. New harness phase 2b: 200-worker swap simulation. Captures the
top-K from alpha's warehouse query, then re-issues with
exclude_ids=<placed>. Result Jaccard(orig, swap) measures whether
the substrate finds genuine alternatives.
3. New harness phase 1b: fresh-resume mid-run injection. Three new
workers ingested via /v1/embed + /v1/vectors/index/workers/add,
then verified findable via semantic queries matching resume content.
Plus Hour labels on every event (operational narrative: 0/6/12/18/
24/30/36/42/48) and a refactor of captureEvent to take hour as a
param.
Run #003 + #004 results (5K workers + 10K ethereal):
Diversity (#004):
Same-role-across-contracts Jaccard = 0.080 (n=9)
Different-roles-same-contract Jaccard = 0.013 (n=18)
Determinism: 1.000 (#004 unchanged)
Verbatim handover: 4/4 = 100%
Paraphrase handover: 4/4 = 100%
Phase 2b — 200-worker swap (Jaccard 0.000):
8 originally-placed workers fully replaced by 8 alternatives.
ExcludeIDs substrate change works end-to-end — boost AND inject
both honor the exclusion, so excluded workers don't return via
the playbook either.
Phase 1b — fresh-resume injection: REAL PRODUCT FINDING.
Substrate ABSORPTION is fine — 3 /v1/vectors/index/workers/add
calls at 200 status, 3 vectors persisted. But none of the 3
fresh workers surfaced in top-8 even with semantic queries
matching their resume content (e.g. "Senior tower crane rigger
NCCCO Chicago" vs fresh-001's resume "Senior rigger with 12
years tower-crane signaling..." NCCCO + Chicago).
Top-1 came from existing workers at distance ~0.25; fresh
workers' distances must be > 0.25, pushing them past rank 8.
Cause: dense retrieval at 5000+ workers means many existing
profiles cluster near any specific query in cosine space;
nomic-embed-text-v2 (137M) introduces enough noise that a
fresh worker doesn't reliably outrank them just because the
text content overlaps.
Workarounds (Phase 3 work): (a) hybrid retrieval (keyword +
semantic), (b) playbook-layer score boost for fresh adds,
(c) larger embedder. Documented in run #004 report.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 1 had two known gaps: (1) the 3 contracts had zero shared role
names, so same-role-across-contracts Jaccard was vacuous (n=0); (2)
the verbatim handover at 100% was the trivial case, not the hard
learning test (paraphrased queries against another coord's playbook).
Both fixed in this commit.
Contract redesign — all 3 contracts now share warehouse worker /
admin assistant / heavy equipment operator roles, plus a unique
specialist per contract (industrial electrician / bilingual safety
coord / drone surveyor — the "specialist not on the standard roster"
case from J's spec). Counts and skill mixes vary per region.
New driver phase 4b — paraphrase handover. Bob runs qwen2.5-paraphrased
versions of Alice's contract queries against Alice's playbook
namespace. Tests whether institutional memory propagates across
coordinators AND across natural wording variation that Bob would
introduce when running Alice's contract.
Run #002 result (5K workers + 10K ethereal_workers, 4 demand × 3
coords + paraphrase handover):
Diversity (the question J asked: locking or cycling?):
Same-role-across-contracts Jaccard = 0.119 (n=9)
→ 88% of workers DIFFER across regions for the same role name.
Milwaukee warehouse vs Indianapolis warehouse vs Chicago
warehouse pull mostly distinct top-K from the same population.
The system locks into geo+cert+skill context, not cycling.
Different-roles-same-contract Jaccard = 0.004 (n=18)
→ role-specific retrieval works (unchanged from Phase 1).
Determinism: Jaccard = 1.000 (n=12) — unchanged.
Learning:
Verbatim handover 4/4 = 100% (trivial case, expected)
Paraphrase handover 4/4 = 100% (HARD case — passes!)
Of those 4 paraphrase recoveries:
- 2 used boost (Alice's recording was already in Bob's
paraphrase top-K; ApplyPlaybookBoost re-ranked to top-1)
- 2 used Shape B inject (recording wasn't in Bob's
paraphrase top-K; InjectPlaybookMisses brought it in)
The boost/inject mix is healthy — both paths are used and both
produce correct top-1s. Multi-coord institutional memory propagation
is empirically working under wording variation.
Sample warehouse worker top-1s across contracts (proves diversity):
alice / Milwaukee → w-713
bob / Indianapolis → e-8447
carol / Chicago → e-7145
Three different workers from the same 15K-person population,
selected on geo+cert+skill context.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three coordinators (alice / bob / carol) with three contracts
(Milwaukee distribution / Indianapolis manufacturing / Chicago
construction). 7-phase scenario runner: baseline → surge → merge →
handover → split → reissue → analysis. Each coord has a separate
playbook namespace (playbook_{name}) so institutional memory stays
isolated by default but transferable on demand.
Phase 1 deliberately skips the 48-hour clock, email/SMS endpoints,
and Langfuse tracing — those are Phase 2/3.
Run #001 (52 events, 4 queries × 3 coords × 2 demand flavors):
Diversity:
Different-roles-same-contract Jaccard = 0.004 (n=18)
→ role-specific retrieval is working perfectly. Different
roles within one contract pull totally different worker
pools. System is NOT cycling; locks into per-role retrieval.
Same-role-across-contracts Jaccard = N/A (n=0)
→ TEST-DESIGN ISSUE: the 3 contracts use distinct role names
per industry (warehouse worker / production worker / general
laborer), so no exact-name overlaps exist. Phase 2 should
either share at least one role across contracts OR add a
skill-based diversity metric.
Determinism: Jaccard = 1.000 (n=12)
→ HNSW + Ollama retrieval is fully deterministic on identical
query text. coder/hnsw + nomic-embed-text are stable.
Learning: handover hit rate = 4/4 = 100%
→ Bob inherits Alice's recordings perfectly when bob runs
identical queries with alice's playbook namespace. CAVEAT:
this tests the trivial verbatim case, not paraphrase handover.
The harder test (bob runs paraphrased queries with alice's
playbook) is Phase 2 work.
Per-event capture in JSON: every matrix.search response is logged
with phase / coordinator / contract / role / query / top-K IDs +
distances + per-corpus counts + boosted/injected counts. Reviewable
via:
jq '.events[] | select(.phase == "merge")'
jq '.events[] | select(.coordinator == "alice")'
jq '.events[] | select(.role == "warehouse worker")'
Notable finding from per-event: carol's "general laborer" and "crane
operator" queries both surface w-1009 as top-1, with crane operator
at distance 0.098 (very tight) and general laborer at 0.297. The
system found a worker who legitimately covers both roles — realistic
for small construction crews.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The rank-based "lift" metric (warm-top-1 == cold-judge-best) doesn't
distinguish "Shape B surfaced a strictly-better answer" from "Shape B
shuffled ranks but quality is unchanged" from "Shape B replaced a good
answer with a wrong one." This commit adds Pass 4: judge warm top-1
with the same prompt as cold ratings, then bucket the comparison.
Implementation:
- New --with-rejudge driver flag (default off).
- New WITH_REJUDGE harness env (default 1, on for prod runs).
- queryRun gains WarmTop1Metadata (cached during Pass 2 for the
rejudge call) + WarmTop1Rating *int (nil-distinguishable; nil = no
rejudge, 0..5 = rating).
- summary gains RejudgeAttempted, QualityLifted, QualityNeutral,
QualityRegressed (counts of warm-rating > / == / < cold-rating).
- Markdown headline gains a Quality block when rejudge ran.
- ~21 extra judge calls (~30s on qwen2.5).
Run #005 result (split inject threshold 0.20 + paraphrase + rejudge):
Quality lifted 5 / 21 (24%) — 3× +2 rating, 2× +1 rating
Quality neutral 13 / 21 (62%) — includes OOD queries holding 1
Quality regressed 3 / 21 (14%)
Net rating delta +3 across 21 queries (+0.14 average)
The 5 lifts were all rating-2 cold replaced with rating-3 or rating-4
warm — Shape B took mediocre matches and substituted substantively
better ones. The 3 regressions were small (-1, -1, -3).
Q11 is the cautionary tale: cold top-1 "production line worker"
(rating 4) got replaced by Q1's recorded "forklift OSHA-30 operator"
e-5729 (rating 1). Adjacent-domain cross-pollination — production
worker and forklift operator embed within 0.20 cosine because both
are warehouse-adjacent staffing queries, even though the judge
correctly distinguishes them. The split-threshold defense (0.5 boost
/ 0.20 inject) catches OOD cross-pollination (Q19/Q20/Q21 all stayed
neutral at rating 1) but not adjacent-domain cross-pollination.
Net product verdict: working, net-positive on quality, but the worst
case (Q11 4→1) is customer-visible and warrants a tighter inject
threshold OR an additional gate beyond cosine distance. Filed in
STATE_OF_PLAY OPEN as a follow-up.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Reality test table extends from #001-#003 to #001-#004; v4 row marked
as "the honest configuration" because OOD cross-pollination is gone.
- Shape B section gains the split-threshold rationale (boost safe at
loose, inject structurally riskier so tighter).
- Verbatim drop framing rewritten — v3→v4 is configuration evolution,
not regression.
- OPEN: closed "Shape B cap/decay" + the conditional Q15 boost-math
item (Shape B + split threshold addressed both). Replaced with two
finer-grained follow-ups: adjacent-query Q6↔Q7 swap (might be
correct, verify with v4 re-judge metric) and liberal-paraphrase
recovery loss (Q9/Q15 missed because qwen2.5 drifted >0.20).
- RECENT VERIFIED WAVE adds 94fc3b6 + 67d1957.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Run #003 surfaced rampant cross-pollination: Q2's "OSHA-30 forklift
Wisconsin" recording (w-4435) became warm top-1 for Q19 (dental
hygienist), Q20 (RN), Q21 (software engineer), and 6 other unrelated
staffing queries. Cause: InjectPlaybookMisses inherited the same
DefaultPlaybookMaxDistance (0.5) as the boost path, but inject is
structurally riskier than boost — boost only re-ranks results that
already retrieved on their own merits, while inject FORCES a result
into top-K, so a loose match cross-pollinates wrong-domain answers.
Empirical motivation from v3:
Implied playbook hit distances for cross-pollinated cases: 0.20-0.46
Implied distances for the 6/6 paraphrase recoveries: 0.23-0.30
Threshold of 0.20 should keep most paraphrases, kill the OOD bleed.
Implementation:
- New DefaultPlaybookMaxInjectDistance = 0.20 in playbook.go.
- New PlaybookMaxInjectDistance field on SearchRequest (override).
- InjectPlaybookMisses signature gains maxInjectDist param; hits whose
Distance exceeds it are skipped (boost path may still re-rank them).
- TestInjectPlaybookMisses_RespectsInjectThreshold locks the contract
with one tight + one loose hit, asserting only the tight one injects.
- Existing tests pass explicit threshold (0 = default for tight tests,
0.5 for the dedupe test which uses 0.30 hits).
Run #004 result on identical queries with the split threshold:
Verbatim discovery 8 (vs v3's 6 — judge variance, separate)
Verbatim lift 6 / 8 (75%)
Paraphrase top-1 6 / 8 (75%)
Paraphrase any-rank in K 6 / 8
OOD queries Q19/Q20/Q21 ALL show warm top-1 = cold top-1 (no
injection) — cross-pollination eliminated where it was wrong-direction.
Mean Δ top-1 distance dropped from -0.164 (v3, distorted) to -0.071
(v4, comparable to v1's -0.053).
Two paraphrases missed in v4 (Q9, Q15) were ones where qwen2.5
rephrased liberally enough to drift past 0.20 — Q9: "Inventory
specialist..." → "Individual needed for inventory management..." and
Q15: "Engaged warehouse associate..." → "Warehouse associate currently
engaged with a robust history...". The system correctly refusing to
inject when it's not confident is the right product behavior; the
boost path still re-ranks recorded answers when they appear in regular
retrieval.
The Q6 ↔ Q7 cross-pollination ("Forklift-certified loader" ↔
"Hazmat warehouse worker") is legitimate — these are genuinely similar
staffing queries and the judge ranks both directions as plausible.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Reality test section now spans v1/v2/v3 across one table — the
product story (boost-only verbatim → paraphrase gap → Shape B
closes the gap) is legible without reading the reports.
- Verbatim-lift drop v1→v3 (7→2) explicitly framed as
cross-pollination, NOT regression — and filed as v4 re-judge metric
in OPEN.
- "DO NOT RELITIGATE" gains: Shape B is the stance now (don't revert
to boost-only); local_judge stays on qwen2.5 (don't bump to qwen3.5
for cleanliness — vision-SSM cost geometry).
- OPEN list: removed the now-closed paraphrase v2 row + the boost-math
Q15 row (Shape B may have addressed it; flagged for verify after v4).
Added v4 re-judge metric and Shape B injection cap/decay design call.
- RECENT VERIFIED WAVE adds the four new commits past 6c02c90
(2c71d1c, 9ce067b, e9822f0, 154a72e).
- Matrix indexer §5/5 component description now references
InjectPlaybookMisses + the run #002→#003 evidence chain.
- [models] tier registry comment locks the local_judge=qwen2.5 choice
with the rationale inline.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The v0 boost-only stance documented in internal/matrix/playbook.go:22-27
("the boost only re-ranks results that ALREADY surfaced from the regular
retrieval") couldn't promote recorded answers that dropped out of a
paraphrase's top-K. playbook_lift_002 surfaced exactly that gap: 0/2
paraphrase recoveries because the recorded answers weren't in regular
retrieval at all (rank=-1).
Shape B: when warm-pass retrieval doesn't surface a playbook hit's
answer, inject a synthetic Result for it directly. Distance =
playbook_hit_distance × BoostFactor — same formula as the boost path so
injections land in comparable distance space. Caller re-sorts +
truncates after both boost and inject have run.
Result on playbook_lift_003 (Shape B + paraphrase pass):
Verbatim discovery 6
Verbatim lift 2 / 6
**Paraphrase top-1** **6 / 6**
Paraphrase any-rank in K 6 / 6
Mean Δ top-1 distance -0.1637 (warm closer than cold)
Every paraphrase the judge generated landed the v1-recorded answer at
top-1 of the new query's results. The learning property holds — cosine
on embed(paraphrase) finds the recorded query's vector within
DefaultPlaybookMaxDistance (0.5), and Shape B injects the answer.
Verbatim lift dropped from v1's 7/8 because Shape B cross-pollinates
recorded answers across queries. w-4435 (Q2's recording) appears as
warm top-1 for several other queries because their embeddings are
within the playbook hit threshold of "OSHA-30 forklift Wisconsin." This
is a feature, not a bug — the matrix layer's purpose is to share
knowledge across queries — but the lift metric only counts "warm top-1
== cold judge best," so cross-pollinated lifts don't register. A v3
metric would re-judge warm pass to measure true judge improvement.
Tests:
- TestInjectPlaybookMisses_AddsMissingAnswers — primary claim
- TestInjectPlaybookMisses_SkipsAnswersAlreadyPresent — no double-inject
- TestInjectPlaybookMisses_DedupesPerAnswer — multi-hit same answer
- TestInjectPlaybookMisses_EmptyHits — fast-path no-op
Driver fix: ParaphraseRecordedRank int → *int. The `omitempty` int
silently dropped rank=0 (top-1, the WANTED value) from JSON, making the
v003 report show "null" instead of "0" for every successful recovery.
Pointer keeps nil/rank-0 distinguishable.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds an opt-in Pass 3 to the lift driver: for each query whose Pass 1
recorded a playbook, ask the judge to rephrase the query, then re-query
with playbook=true and check whether the recorded answer surfaces in
top-K. This is the test the v1 report's caveat #3 explicitly flagged
as the actual learning-property gate (not the cheap verbatim case).
Implementation:
- New flag --with-paraphrase on the driver (default off).
- New WITH_PARAPHRASE env in the harness (default 1, on for prod runs).
- New paraphrase_* fields on queryRun + summary, // 0 fallback in jq so
re-rendering verbatim-only evidence stays clean.
- generateParaphrase() calls the same judge model with format=json and
a tight schema; temperature=0.5 for variance without domain drift.
- Markdown report adds a paraphrase per-query table (only when the
pass ran) and an honesty caveat about judge-also-rephrases coupling.
Run #002 result (reports/reality-tests/playbook_lift_002.{json,md}):
Verbatim lift 2/2 (100% — Q7 + Q13, both stable from v1)
Paraphrase top-1 0/2
Paraphrase any-rank in K 0/2
Both paraphrases dropped the recorded answer OUT of top-K entirely
(rank=-1). This isn't a paraphrase-quality problem — qwen2.5's outputs
preserved intent ("Hazmat-certified warehouse worker comfortable with
cold storage" → "Warehouse worker with Hazmat certification and
experience in cold storage"). It's the v0 boost-only stance documented
in internal/matrix/playbook.go:22-27: the boost only re-ranks results
that ALREADY surfaced from regular retrieval. If paraphrase's cosine
retrieval doesn't include the recorded answer in top-K, no boost can
promote it.
The "Shape B" upgrade mentioned in the playbook.go comment — inject
playbook hits directly even when they weren't in the top-K — is what
would close this gap. The reality test surfaced exactly the gap the
docs warned about. Worth filing as the next product gate.
Run-to-run variance also visible: v1 had 8 discoveries, v2 had 2.
HNSW insertion order + judge variance both contribute. Stability of
Q7 and Q13 across both runs (lifted in v1 AND v2) is the most reliable
signal in the dataset.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
TestWorkflowRun_AllProvenanceRecordedPostRun proves that
handleWorkflowRun records ObservedOps only AFTER runner.Run returns,
not interleaved with node execution.
The test pauses inside a node via a controlled channel, samples
observer.Store mid-run (must be 0), unblocks, then samples again
(must be N). If a future commit adds per-node streaming (e.g.
runner.NodeHook firing as each node finishes), n1's record would
appear before the unblock and the first assertion fires.
This is intentional test-as-spec lock. Closing the streaming gap is
deferred per the ADR ("acceptable for short workflows; streaming
callback is the right shape when workflows get longer") — but if
someone later adds the streaming callback without updating the ADR,
this test catches it in `go test` instead of leaving the doc and
code drifted.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes the OPEN item from STATE_OF_PLAY. Required because observerd is
now on the prod-realistic data path via the lift harness boot (b2e45f7),
so the next consumer (scrum runner / distillation rebuild / production
workflow) needs the fail-safe rationale locked, not implicit.
The Rust "verdict:accept on crash" anti-pattern doesn't translate
one-to-one to the Go observer (witness, not gate). But four adjacent
fail-safe decisions are real and live:
5.1 Persist failure is logged-not-fatal; ring is in-flight source of
truth. Persist-required mode deferred to a future opt-in ADR.
5.2 Mode failure → Success=false, no panic-swallow path. The runner
catches mode errors and surfaces them via node.Error; downstream
consumers see failures explicitly rather than as fake successes
(the Rust anti-pattern surface).
5.3 One row per node, recorded post-run. A workflow with N nodes
produces N audit rows, never a per-workflow catch-all that
survives partial crashes. Known gap: recording happens after
runner.Run returns (acceptable for short workflows; streaming
callback is the right shape when workflows get longer).
5.4 /observer/event accepts on full ring (oldest evicted). Refusing
to write would translate every burst into client errors — wrong
direction for an audit witness.
Mostly ratifies existing behavior; cross-checked claims against
actual code (caught one error in Decision 5.3 draft — recording is
post-run-batched, not per-node-as-it-completes — and the ADR now
states reality).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Cross-lineage scrum on b2e45f7 produced 1 convergent + 3 single-reviewer
findings worth fixing. All apply.
1. (Opus WARN + Qwen INFO convergent) scripts/playbook_lift.sh: replace
sleep 2.5 in SQL probe with active polling up to 5s. refresh_every=1s
is a lower bound; under load the manifest may not be visible in a
fixed sleep, which would 4xx the probe and abort the reality run.
2. (Opus INFO) scripts/playbook_lift.sh: report template glued
"env JUDGE_MODEL" + value as "env JUDGE_MODELqwen2.5:latest" with no
separator. Replaced two :+/:- substitution chains with a single
JUDGE_SOURCE variable computed once at the top of the harness.
3. (Opus INFO) scripts/staffing_workers/main.go: -id-prefix "" silently
allowed, defeating the flag's purpose (cross-corpus collision prevent).
Now log.Fatal at startup with explicit hint.
4. (Opus WARN) cmd/{pathwayd,observerd}/main_test.go: newTestRouter
returned http.Handler then re-cast to chi.Router for chi.Walk.
Returning chi.Router directly satisfies http.Handler AND avoids an
assertion that would panic if future middleware wraps the router.
Dismissed (with rationale):
- Kimi INFO hardcoded MinIO endpoint: harness is local-by-design.
- Kimi WARN matrixd accepts 502/500: documented; real retriever needs
real upstreams the test doesn't spin up.
- Qwen INFO queryd string.Contains: brittle but very low risk; restating
through typed-error path would couple without adding signal.
go test ./cmd/{matrixd,queryd,pathwayd,observerd} all green.
Verdicts at reports/scrum/_evidence/2026-04-30/verdicts/lift_001_*.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The 5-loop substrate's load-bearing gate is verified — playbook +
matrix indexer give the results we're looking for. Per the report's
rubric, lift ≥ 50% of discoveries means matrix is doing real work;
7/8 = 87.5% blew through that.
Harness was structurally hiding bugs behind a 5-daemon stripped boot.
Expanding to the full 10-daemon prod stack surfaced 7 fixes in cascade:
1. driver→matrixd: {"query": ...} → {"query_text": ...} field name
2. harness temp toml missing [s3] → wrong default bucket → catalogd
rehydrate 500 on first call
3. harness→queryd SQL probe: {"q": ...} → {"sql": ...} field name
4. expand boot from 5 → 10 daemons in dep-ordered launch
5. add SQL surface probe (3-row CSV ingest → COUNT(*)=3 assertion)
6. candidates corpus was synthetic SWE-tech (Swift/iOS, Scala/Spark) —
wrong domain for staffing queries; replaced with ethereal_workers
(10K rows, real staffing schema, "e-" id prefix to avoid collision
with workers' "w-"). staffing_workers driver gains -index-name +
-id-prefix flags so the same binary serves both corpora
7. local_judge qwen3.5:latest is a vision-SSM 256K-ctx build running
~30s per judge call against the lift loop; reverted to
qwen2.5:latest (~1s/call, 30× faster, held lift theory)
Each contract drift (1, 3) is now locked into a cmd/<bin>/main_test.go
so future drift fires in `go test`, not in a reality run. R-005 closed:
- cmd/matrixd/main_test.go (new) — playbook record drift detector +
score bounds + 6 routes mounted
- cmd/queryd/main_test.go — wrong-field-name drift detector
- cmd/pathwayd/main_test.go (new) — 9 routes + add round-trip + retire
- cmd/observerd/main_test.go (new) — 4 routes + invalid-op + unknown-mode
`go test ./cmd/{matrixd,queryd,pathwayd,observerd}` all green.
Reality test results (reports/reality-tests/playbook_lift_001.{json,md}):
Queries 21 (staffing-domain, 7 categories)
Discoveries 8 (judge ≠ cosine top-1)
Lifts 7/8 (87.5%)
Boosts triggered 9
Mean Δ distance -0.053 (warm closer than cold)
OOD honesty dental/RN/SWE rated 1, no fake matches
Cross-corpus boosts confirmed (e- ↔ w- swaps in lifts)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 4-bundle review (128KB diff) hit "Argument list too long" when
curl --data was passed the body as a literal arg. Pipe via stdin
with --data-binary @- instead. Lifts the practical bundle size from
~30KB to whatever fits in process memory.
Caught while running the harness scrum on golangLAKEHOUSE today —
the bigger Phase A+B harness diff (4566 lines) tripped it.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- SPEC §1 component table: add chatd row marked DONE; replaces
Rust gateway's v1::ollama_cloud / openrouter / opencode adapters
+ the aibridge crate.
- SPEC §3.9 — chatd shipped: 5-provider routing (ollama, ollama_cloud,
openrouter, opencode, kimi) by model-name prefix or :cloud suffix.
Captures the Anthropic 4.7 temperature-deprecation quirk + the
local-Ollama think=false default that the playbook_lift judge
needed. Mentions scrum_review.sh as the reusable cross-lineage
vehicle eating chatd's own /v1/chat.
- SPEC §3.10 — local-review-harness sibling tool: separate repo at
git.agentview.dev/profit/local-review-harness, MVP shipped today.
Documents the cross-pollination plan for when both substrates
stabilize (chatd as the harness's LLM backend; harness findings
as Lakehouse pathway-memory drift signal; .memory/known-risks
as a matrix corpus). Explicit "don't re-port" so future Claudes
don't try to absorb the harness into Lakehouse.
- STATE_OF_PLAY.md: SIBLING TOOLS section with 1-line summary
+ pointer to SPEC §3.10.
No code changes. just verify still PASS — touched only docs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bash driver wrapping /v1/chat for Opus + Kimi + Qwen3-coder review
runs. Used today to scrum the 4-phase wave (1,624 LoC of chatd +
config-refactor + Rust cleanup) and caught 2 BLOCKs + 2 WARNs.
Usage:
./scripts/scrum_review.sh <bundle.diff> <bundle_label>
Output: reports/scrum/_evidence/<DATE>/verdicts/<bundle>_<reviewer>.md
verbatim, per the evidence-only convention. Per-reviewer latency +
token counts captured in the report header.
System prompt enforces the BLOCK/WARN/INFO + WHERE/WHAT/WHY shape
per feedback_cross_lineage_review.md — leads with verdict, no
preamble (Kimi tends to spend tokens thinking otherwise).
Reviewer fleet matches project_golang_lakehouse.md "Scrum routing":
- opencode/claude-opus-4-7
- openrouter/moonshotai/kimi-k2-0905
- openrouter/qwen/qwen3-coder
This is the first dogfood of chatd as the scrum vehicle — eats its
own /v1/chat dispatcher.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3-lineage scrum (Opus 4.7 / Kimi K2.6 / Qwen3-coder) on today's wave
landed 4 real findings (2 BLOCK + 2 WARN) and 2 INFO touch-ups.
Verbatim verdicts + disposition table at:
reports/scrum/_evidence/2026-04-30/
B-1 (BLOCK Opus + INFO Kimi convergent) — ResolveKey API:
collapse from 3-arg (envVar, envFileName, envFilePath) to 2-arg
(envVar, envFilePath). Pre-fix every chatd caller passed the env
var name twice; if operator renamed *_key_env in lakehouse.toml
while keeping the canonical KEY= line in the .env file, fallback
silently missed.
B-2 (WARN Opus + WARN Kimi convergent) — handleProviders probe:
drop the synthesize-then-Resolve probe; look up by name directly
via Registry.Available(name). Prior probe synthesized "<name>/probe"
model strings and routed through Resolve, fragile to any future
routing rule (e.g. cloud-suffix special case).
B-3 (BLOCK Opus single — verified by trace + end-to-end probe) —
OllamaCloud.Chat StripPrefix used "cloud" but registry routes
"ollama_cloud/<m>". Result: upstream got the prefixed model name
and 400'd. Smoke missed it because chatd_smoke runs without
ollama_cloud registered. Now strips the right prefix; new
TestOllamaCloud_StripsCorrectPrefix locks both prefix + suffix
cases. Verified live: ollama_cloud/deepseek-v3.2 round-trips
cleanly through the real ollama.com endpoint.
B-4 (WARN Opus single) — Ollama finishReason: read done_reason
field instead of inferring from done bool alone. Newer Ollama
reports done=true with done_reason="length" on truncation; the
prior code mapped that to "stop" and lost the truncation signal
the playbook_lift judge needs to retry. New
TestFinishReasonFromOllama_PrefersDoneReason covers the fallback
ladder.
INFOs:
- B-5: replace hand-rolled insertion sort in Registry.Names with
sort.Strings (Opus called the "avoid sort import" comment a
false economy — correct).
- A-1: clarify the playbook_lift.sh comment around -judge "" arg
passing (Opus noted the comment said "env priority" but didn't
reflect that the empty arg also passes through the Go driver's
resolution chain).
False positives dismissed (3, documented in disposition.md):
- Kimi: TestMaybeDowngrade_WithConfigList wrong assertion (test IS
correct per design — model excluded from weak list = strong = downgrade)
- Qwen: nil-deref claim (defensive code already handles nil)
- Opus: qwen3.5:latest doesn't exist on Ollama hub (true on the
public hub but local install has it)
just verify: PASS. chatd_smoke 6/6 PASS. New regression tests:
3 (B-2, B-3, B-4 each get a focused test).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
new cmd/chatd on :3220 routes /v1/chat to the right provider based
on model-name prefix or :cloud suffix. closes the architectural gap
named in lakehouse.toml [models]: tiers map to model IDs, but until
phase 4 there was no service that could actually CALL those models
from go.
routing rules (registry.Resolve):
ollama/<m> → local Ollama (prefix stripped)
ollama_cloud/<m> → Ollama Cloud
<m>:cloud → Ollama Cloud (suffix variant — kimi-k2.6:cloud)
openrouter/<v>/<m> → OpenRouter (prefix stripped, OpenAI-compat)
opencode/<m> → OpenCode unified Zen+Go
kimi/<m> → Kimi For Coding (api.kimi.com/coding/v1)
bare names → local Ollama (default)
provider implementations:
- internal/chat/types.go Provider interface, Request/Response, errors
- internal/chat/registry.go prefix + :cloud suffix dispatch
- internal/chat/ollama.go local Ollama via /api/chat (think=false default)
- internal/chat/ollama_cloud.go Ollama Cloud via /api/generate (Bearer auth)
- internal/chat/openai_compat.go shared OpenAI Chat Completions for the
OpenRouter/OpenCode/Kimi family
- internal/chat/builder.go BuildRegistry from BuilderInput;
ResolveKey reads env then .env file fallback
config:
- ChatdConfig in internal/shared/config.go with bind, ollama_url,
per-provider key env names + .env fallback paths, timeout
- Gateway gains chatd_url + /v1/chat + /v1/chat/* routes
- lakehouse.toml [chatd] block with /etc/lakehouse/<provider>.env defaults
tests (19 in internal/chat):
- registry: prefix + :cloud + errors + telemetry + provider listing
- ollama: happy path + prefix strip + format=json + 500 mapping +
flatten_messages
- openai_compat: happy path + format=json + 429 mapping + zero-choices
think=false default in ollama + ollama_cloud — local hot path skips
reasoning, low-budget callers (the playbook_lift judge at max_tokens=10)
get direct answers instead of empty content + done_reason=length.
proven via chatd_smoke acceptance.
acceptance gate: scripts/chatd_smoke.sh — 6/6 PASS:
1. /v1/chat/providers lists exactly registered providers (1 in dev mode)
2. bare model → ollama default with content + token counts + latency
3. explicit ollama/<m> → prefix stripped at upstream
4. <m>:cloud without ollama_cloud registered → 404 (no silent fall-through)
5. unknown/<m> → falls through to default → upstream 502 (no prefix rewrite)
6. missing model field → 400
just verify: PASS (vet + 30 packages × short tests + 9 smokes).
chatd_smoke is a domain smoke (not in just verify, mirrors matrix /
observer / pathway pattern).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
migrate the reality-test harness's judge-model default from a
hardcoded "qwen3.5:latest" string to cfg.Models.LocalJudge.
resolution priority: explicit -judge flag > $JUDGE_MODEL env >
cfg.Models.LocalJudge from lakehouse.toml > hardcoded fallback.
bumping the judge for run #N+1 now means editing one line in
lakehouse.toml [models].local_judge — no Go file or shell script
edits required.
changes:
- scripts/playbook_lift/main.go: -config flag added, judge default
flips to "" so resolution chain runs. Imports internal/shared for
config loader.
- scripts/playbook_lift.sh: JUDGE_MODEL no longer defaulted in bash;
EFFECTIVE_JUDGE resolved by mirror-of-the-Go-chain (env > config
grep > qwen3.5:latest fallback). Used for the Ollama presence
check + report header. Pre-flight grep avoids requiring jq just
to read the toml.
- reports/reality-tests/README.md: documents the 4-step priority
chain.
verified all 4 paths produce the expected judge:
- config (no env): qwen3.5:latest (from lakehouse.toml)
- env override: env wins
- flag override: flag wins over env
- missing config: DefaultConfig fallback still gives qwen3.5:latest
just verify PASS.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
migrate the strong-model auto-downgrade gate from a hardcoded weak
list to cfg.Models.WeakModels. backward compatible: existing API
preserved, callers that don't migrate keep using DefaultWeakModels.
changes:
- internal/matrix/downgrade.go: split IsWeakModel into rule-based
base (`:free` suffix/infix) + literal-list lookup. New
IsWeakModelInList(model, list) takes the config-supplied list.
DowngradeInput grows a WeakModels field; nil falls back to
DefaultWeakModels (preserves pre-phase-2 behavior).
- internal/workflow/modes.go: add MatrixDowngradeWithWeakList(list)
factory mirroring MatrixSearch's pattern. Plain MatrixDowngrade
kept for backward compat.
- cmd/matrixd/main.go: handlers struct holds weakModels populated
from cfg.Models.WeakModels at startup; handleDowngrade threads it
into every DowngradeInput.
- cmd/observerd/main.go: registerBuiltinModes accepts weakModels
and uses the factory variant. observerd reads cfg.Models.WeakModels
in main().
end-to-end verified: downgrade + matrix + observer + workflow smokes
all pass. Existing TestMaybeDowngrade_TruthTable + TestIsWeakModel
unchanged (backward compat). Two new tests cover the config path:
- TestIsWeakModelInList — covers rule + literal + empty + nil
- TestMaybeDowngrade_WithConfigList — verifies cfg list overrides
default
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
root