fix: count { success: false } results as failures in the effect tracker (GIT-104)

[nTEG-dev]

Linear: https://linear.app/nteg-labs/issue/GIT-104

What

Makes health report failed writes as failures. Before this change, EffectTracker.track() counted any resolved promise as a success. Tracked writers that catch their own errors and resolve anyway were therefore reported as successes.

• track(): a resolved value of the form { success: false } is now recorded as a failure. The message comes from error, then message, then display. The value is still returned to the caller, and rejections are handled as before.
• Tracked writers that swallowed errors now let the tracker see them:
  • writeTriples attempts every triple, then rejects with a summary if any failed.
  • generateVariantsForScar rejects if a variant insert failed. It used Promise.allSettled and always resolved.
  • Cache writes under track() use a new rejecting writeResult(). setResult() keeps swallowing for its direct callers (check, tests).
  • The session_close transcript task now returns its save result, so { success: false } is visible.
• recordScarUsageBatch now puts the error text on its { success: false } result.

track() call-site audit (18 sites)

Call site	Path / target	Wrapped fn on error	Before	After
tools/session-close.ts:1484	scar_usage / session_close_batch	returns {success:false} (record-scar-usage-batch.ts:201-209)	counted as success	failure via {success:false}
tools/create-thread.ts:347	triple_write / thread_creation	swallows (per-triple try/catch in writeTriples)	counted as success	rejects → failure
tools/resolve-thread.ts:236	triple_write / thread_resolution	swallows (same)	counted as success	rejects → failure
tools/resolve-thread.ts:249	triple_write / thread_resolution_cascade	swallows (same)	counted as success	rejects → failure
tools/create-decision.ts:95	triple_write / decision	swallows (same)	counted as success	rejects → failure
tools/create-learning.ts:242	triple_write / learning	swallows (same)	counted as success	rejects → failure
tools/create-learning.ts:258	variant_generation / learning	swallows (Promise.allSettled, logs only)	counted as success	rejects → failure
services/cache.ts:226	cache_set / scar_search	swallows (setResult try/catch)	counted as success	rejects (writeResult) → failure
services/cache.ts:252	cache_set / decisions	swallows (same)	counted as success	rejects → failure
services/cache.ts:278	cache_set / wins	swallows (same)	counted as success	rejects → failure
services/cache.ts:316	cache_set / sessions	swallows (same)	counted as success	rejects → failure
services/cache.ts:339	cache_set / scar_usage	swallows (same)	counted as success	rejects → failure
tools/session-close.ts:1369	transcript / session_close	swallows (a {success:false} save result was dropped; the fn returned undefined)	counted as success	returns result → failure
tools/session-close.ts:1435	embedding / session_close	throws (embed throws on API errors, directPatch throws)	correct	unchanged
tools/contribute-feedback.ts:118	feedback / remote-submit	throws (submitFeedbackRemote throws on non-2xx)	correct	unchanged
services/metrics.ts:168	metrics / <tool>	throws (directUpsert)	correct	unchanged
services/metrics.ts:363	relevance_update / <session>	throws (listRecords / directUpsert)	correct	unchanged

(lib/telemetry.ts has its own unrelated track() method; it isn't the effect tracker.)

Venue check: blank Supabase (setup.sql only, no edge functions)

Driver: tests/e2e/blank-supabase.mjs (draft PR, test/GIT-83-blank-supabase-driver), flow mode.

Before (main d47a625): the netlog shows POST /rest/v1/scar_usage → 404 and POST /rest/v1/knowledge_triples → 400, but health reports them as succeeded:

  metrics              4/5 succeeded (1 failed)
  cache_set            1/1 succeeded
  triple_write         1/1 succeeded
  embedding            1/1 succeeded
  scar_usage           1/1 succeeded
  ⚠ 1 total failures across: metrics

After (this branch): the same two requests are now reported:

  cache_set            1/1 succeeded
  metrics              6/6 succeeded
  triple_write         0/1 succeeded (1 failed)
  embedding            1/1 succeeded
  scar_usage           0/1 succeeded (1 failed)
  ⚠ 2 total failures across: triple_write, scar_usage
  Recent failures:
    [scar_usage/session_close_batch] Supabase upsert error: 404 - ... Could not find the table 'public.scar_usage' ...
    [triple_write/thread_creation] 1/1 triples failed: Supabase upsert error: 400 - ... invalid input syntax for type uuid: "t-00fb1833"

The metrics 409 is the intermittent GIT-73 race. It didn't hit this run.

Expect health to show more failures after this merges. It isn't a regression: the failures already happened and are now counted. Those failures are fixed by GIT-84 (scar_usage), GIT-105 (triples) and GIT-106 (variants).

Tests

• npx tsc --noEmit -p .: no errors
• npm run test:unit (with HOME set to a temp dir): 76 files / 1224 tests passed. main is 74 / 1213, so this adds +11 tests in 2 files:
  • 6 effect-tracker outcome accounting: {success:false} counts as a failure, {success:true} and plain values count as successes, a rejection counts as a failure, plus message preference and the summary.
  • 5 tracked writers surface failures: writeTriples and generateVariantsForScar reject on a partial failure, and the tracker counts it.

Unit-test note: beforeEach(() => mock.mockReset()) combined with this repo's clearMocks/restoreMocks config made vitest 1.6 fail a passing test with the mock's error. The config already resets mocks between tests, so the new tests don't use beforeEach.

No version bump and no publish.

🤖 Generated with Claude Code