pathway_memory: audit-consensus → retire wire
Some checks failed
lakehouse/auditor 1 blocking issue: todo!() macro call in tests/real-world/scrum_master_pipeline.ts

When observer's hand-review explicitly rejects the output of a
hot-swap-recommended model, the matrix's recommendation was wrong
for this context. Auto-retire the trace so future agents don't
get the same poisoned recommendation in their preamble.

crates/vectord/src/pathway_memory.rs — add `trace_uid` to
HotSwapCandidate response and populate from the matched trace.
This gives consumers single-trace precision for /pathway/retire.

tests/real-world/scrum_master_pipeline.ts:
  - HotSwapCandidate interface gains trace_uid
  - new retirePathwayTrace() helper (fire-and-forget, fall-open)
  - in the obsVerdict reject branch: if hotSwap was active AND
    the rejected model is the hot-swap-recommended one AND
    observer confidence ≥0.7, fire retire and null hotSwap so
    post-loop replay bookkeeping doesn't double-process.
  - hotSwap declared `let` (was const) so it can be nulled

Cycle verdicts ("needs different angle") don't trigger retire —
only outright rejects do. Confidence gate avoids retiring on
heuristic-fallback verdicts that come back without a confidence
number. Closes the "audit-consensus → retire" item from
HANDOVER.md.

Live-tested: insert synthetic trace → /pathway/retire by trace_uid
→ retired counter 1 → 2.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
root 2026-04-26 00:01:20 -05:00
parent e1ef185868
commit 626f18d491
2 changed files with 51 additions and 1 deletions

View File

@ -373,6 +373,11 @@ pub struct PathwayMemory {
#[derive(Debug, Serialize)]
pub struct HotSwapCandidate {
pub pathway_id: String,
/// trace_uid of the SPECIFIC trace this hot-swap recommendation
/// came from. Lets a caller call /pathway/retire with single-trace
/// precision when observer rejects the result — the audit-consensus
/// → retire wire (HANDOVER §queued, ADR-021).
pub trace_uid: String,
pub similarity: f32,
pub replay_count: u32,
pub success_rate: f32,
@ -754,6 +759,7 @@ impl PathwayMemory {
let accepted = p.ladder_attempts.iter().find(|a| a.accepted)?;
Some(HotSwapCandidate {
pathway_id: p.pathway_id.clone(),
trace_uid: p.trace_uid.clone(),
similarity,
replay_count: p.replay_count,
success_rate: p.success_rate(),

View File

@ -224,6 +224,7 @@ function buildQueryVec(taskClass: string, filePath: string, signalClass: string
interface HotSwapCandidate {
pathway_id: string;
trace_uid: string;
similarity: number;
replay_count: number;
success_rate: number;
@ -231,6 +232,25 @@ interface HotSwapCandidate {
recommended_model: string;
}
// Audit-consensus → retire wire (2026-04-25). When observer rejects the
// output of a hot-swap-recommended model, the matrix's recommendation
// was wrong for this context — retire the trace so future agents don't
// get the same poisoned recommendation in their preamble. Server-side
// retire is idempotent so duplicate calls are safe.
async function retirePathwayTrace(traceUid: string, reason: string): Promise<void> {
if (!traceUid) return;
try {
await fetch(`${GATEWAY}/vectors/pathway/retire`, {
method: "POST",
headers: { "content-type": "application/json" },
body: JSON.stringify({ trace_uid: traceUid, reason }),
signal: AbortSignal.timeout(3000),
});
} catch {
// Fire-and-forget; pathway memory is a hint store, not a hard gate.
}
}
async function queryHotSwap(taskClass: string, filePath: string, signalClass: string | null): Promise<HotSwapCandidate | null> {
try {
const query_vec = buildQueryVec(taskClass, filePath, signalClass);
@ -1361,7 +1381,9 @@ Respond with markdown. Be specific, not generic. Cite file-region + PRD-chunk-of
// service unavailable → null candidate → business as usual.
const signalClass = await lookupSignalClass(rel);
const taskClass = "scrum_review";
const hotSwap = await queryHotSwap(taskClass, rel, signalClass);
// mutable so retire-on-reject can null it after firing — prevents
// the post-loop replay bookkeeping from re-touching a retired trace.
let hotSwap: HotSwapCandidate | null = await queryHotSwap(taskClass, rel, signalClass);
// ADR-021 Phase C: pre-review enrichment. Pull aggregated bug
// fingerprints the matrix index has learned for this narrow
@ -1515,6 +1537,28 @@ Respond with markdown. Be specific, not generic. Cite file-region + PRD-chunk-of
const reason = `observer ${obsVerdict.verdict}: ${obsVerdict.notes ?? "no notes"} (conf=${obsVerdict.confidence ?? "?"})`;
history.push({ n, model: rung.model, status: "thin", chars: r.content.length, error: reason });
pathwayAttempts.push({ rung: i + 1, model: rung.model, latency_ms: attemptMs, accepted: false, reject_reason: reason });
// Audit-consensus → retire: if a hot-swap influenced THIS
// attempt (we're trying its recommended model) and observer
// explicitly rejects, the matrix recommendation is wrong for
// this context. Retire the trace so future agents don't repeat
// it. Cycle verdicts ("needs different angle") don't trigger
// retire — only outright rejects do. Confidence ≥0.7 gate
// avoids retiring on heuristic-fallback verdicts (which return
// no confidence).
if (
hotSwap &&
obsVerdict.verdict === "reject" &&
rung.model === hotSwap.recommended_model &&
(obsVerdict.confidence ?? 0) >= 0.7
) {
const retireReason = `observer reject on hot-swap replay: ${obsVerdict.notes ?? "no notes"}`;
log(` 🗑 retiring pathway ${hotSwap.trace_uid.slice(0, 8)}… (${retireReason})`);
retirePathwayTrace(hotSwap.trace_uid, retireReason);
// Null out hotSwap so the post-loop replay bookkeeping doesn't
// also try to record success/failure against the now-retired
// trace.
hotSwap = null;
}
qualityRetriesOnCurrentModel++;
if (qualityRetriesOnCurrentModel > MAX_QUALITY_RETRIES) {
log(`${reason} — quality retries exhausted on ${rung.model}, advancing fallback`);