matrix: judge-gated Shape B inject — closes lift-suite tail issues

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>
This commit is contained in:
root
2026-04-30 19:38:12 -05:00
co-authored by Claude Opus 4.7
parent 247e36e687
commit 5a3364f539
5 changed files with 313 additions and 15 deletions
+88 -5
View File
@@ -187,7 +187,7 @@ func TestInjectPlaybookMisses_AddsMissingAnswers(t *testing.T) {
},
},
}
out, injected := InjectPlaybookMisses(results, hits, 0)
out, injected := InjectPlaybookMisses("test query", results, hits, 0, nil)
if injected != 1 {
t.Fatalf("expected 1 injected, got %d", injected)
}
@@ -240,7 +240,7 @@ func TestInjectPlaybookMisses_SkipsAnswersAlreadyPresent(t *testing.T) {
},
},
}
out, injected := InjectPlaybookMisses(results, hits, 0)
out, injected := InjectPlaybookMisses("test query", results, hits, 0, nil)
if injected != 0 {
t.Errorf("expected 0 injected (answer already present), got %d", injected)
}
@@ -266,7 +266,7 @@ func TestInjectPlaybookMisses_DedupesPerAnswer(t *testing.T) {
Entry: PlaybookEntry{QueryText: "q2", AnswerID: "w-99", AnswerCorpus: "workers", Score: 1.0},
},
}
out, injected := InjectPlaybookMisses(results, hits, 0.5) // explicit loose threshold so 0.30 hits qualify
out, injected := InjectPlaybookMisses("test query", results, hits, 0.5, nil) // explicit loose threshold so 0.30 hits qualify
if injected != 1 {
t.Errorf("expected 1 injection (deduped), got %d", injected)
}
@@ -280,6 +280,89 @@ func TestInjectPlaybookMisses_DedupesPerAnswer(t *testing.T) {
}
}
// TestInjectPlaybookMisses_GateRejectsCandidate locks the judge-gate
// path (OPEN item #1, closed by this commit). When the InjectGate
// returns false on a candidate, the candidate is skipped — even if
// distance would otherwise allow it. Closes the lift-suite tail
// issues where embedding said "tight" but a judge said "wrong domain."
func TestInjectPlaybookMisses_GateRejectsCandidate(t *testing.T) {
results := []Result{{ID: "w-1", Corpus: "workers", Distance: 0.30}}
hits := []PlaybookHit{
{
PlaybookID: "pb-x",
Distance: 0.10, // tight in cosine — would inject without gate
Entry: PlaybookEntry{
QueryText: "recorded crane operator query",
AnswerID: "w-99", AnswerCorpus: "workers", Score: 1.0,
},
},
}
rejectAll := InjectGateFunc(func(string, PlaybookHit) bool { return false })
out, injected := InjectPlaybookMisses("forklift loader query", results, hits, 0, rejectAll)
if injected != 0 {
t.Errorf("rejectAll gate should skip injection, got %d injected", injected)
}
if len(out) != 1 {
t.Errorf("results should be unchanged at len=1, got %d", len(out))
}
}
// TestInjectPlaybookMisses_GateApprovesCandidate locks the
// always-approve gate path: behavior matches nil-gate (current
// distance-only filter). Useful for tests that want to assert
// "judge-gate API is wired" without an actual decision.
func TestInjectPlaybookMisses_GateApprovesCandidate(t *testing.T) {
results := []Result{{ID: "w-1", Corpus: "workers", Distance: 0.30}}
hits := []PlaybookHit{
{
PlaybookID: "pb-x",
Distance: 0.10,
Entry: PlaybookEntry{
QueryText: "x", AnswerID: "w-99", AnswerCorpus: "workers", Score: 1.0,
},
},
}
approveAll := InjectGateFunc(func(string, PlaybookHit) bool { return true })
out, injected := InjectPlaybookMisses("test query", results, hits, 0, approveAll)
if injected != 1 {
t.Errorf("approveAll gate should inject, got %d", injected)
}
if len(out) != 2 {
t.Errorf("results should grow to 2, got %d", len(out))
}
}
// TestInjectPlaybookMisses_GateSeesCorrectQuery locks the gate's
// query+hit visibility — the gate must receive the CURRENT search's
// query (not the recorded one) so it can judge the (current_query,
// candidate) pair. The recorded query lives on hit.Entry.QueryText.
func TestInjectPlaybookMisses_GateSeesCorrectQuery(t *testing.T) {
results := []Result{{ID: "w-1", Corpus: "workers", Distance: 0.30}}
hits := []PlaybookHit{
{
PlaybookID: "pb-x",
Distance: 0.10,
Entry: PlaybookEntry{
QueryText: "RECORDED",
AnswerID: "w-99", AnswerCorpus: "workers", Score: 1.0,
},
},
}
var seenQuery, seenRecordedQuery string
gate := InjectGateFunc(func(q string, h PlaybookHit) bool {
seenQuery = q
seenRecordedQuery = h.Entry.QueryText
return true
})
_, _ = InjectPlaybookMisses("CURRENT", results, hits, 0, gate)
if seenQuery != "CURRENT" {
t.Errorf("gate received query=%q, want CURRENT", seenQuery)
}
if seenRecordedQuery != "RECORDED" {
t.Errorf("gate received recorded=%q, want RECORDED", seenRecordedQuery)
}
}
// TestInjectPlaybookMisses_RespectsInjectThreshold locks the
// cross-pollination defense added after run #003: hits whose playbook
// distance exceeds the inject threshold are skipped, preventing the
@@ -303,7 +386,7 @@ func TestInjectPlaybookMisses_RespectsInjectThreshold(t *testing.T) {
},
}
// Default threshold (0 → DefaultPlaybookMaxInjectDistance = 0.20)
out, injected := InjectPlaybookMisses(results, hits, 0)
out, injected := InjectPlaybookMisses("test query", results, hits, 0, nil)
if injected != 1 {
t.Errorf("expected 1 injection (only the tight hit qualifies), got %d", injected)
}
@@ -324,7 +407,7 @@ func TestInjectPlaybookMisses_RespectsInjectThreshold(t *testing.T) {
// TestInjectPlaybookMisses_EmptyHits is a fast-path no-op check.
func TestInjectPlaybookMisses_EmptyHits(t *testing.T) {
results := []Result{{ID: "w-1", Corpus: "workers", Distance: 0.30}}
out, injected := InjectPlaybookMisses(results, nil, 0)
out, injected := InjectPlaybookMisses("test query", results, nil, 0, nil)
if injected != 0 {
t.Errorf("expected 0 injection, got %d", injected)
}