promote: main → prod (retire indexSource interpreter + cleanup)

[klappy]

Promotion: main → prod

Promotes the following merged-to-main work to production deployment:

• refactor(telemetry): retire indexSource interpreter — fetchers report, telemetry tallies #141 — refactor(telemetry): retire indexSource interpreter — fetchers report, telemetry tallies (the main refactor — recordFetch + cacheStats + double7/double8, deletes _indexSource interpreter)
• feat(audit): add oddkit_audit MCP action — Phase 2 PR-2.3a #143 — Bugbot follow-up: removes unused shortKey export from tracing.ts
• chore(deploy): empty commit to force CF Workers Builds rebuild #144 — empty commit (was an attempt to force a CF rebuild before realizing prod is the source of production deploys, not main)

Why a separate promotion

Per docs/decisions/D0001-prod-branch-is-production.md, the prod branch is the sole source of production deployments. Merging to main lands code; promoting to prod ships it. I missed that distinction in the prior steps and was watching the live worker waiting for a deploy that wasn't going to happen until this promotion.

Companion canon already merged

klappy/klappy.dev PR #144 (table updates for cache_hits/cache_lookups doubles, retirement of blob9 cache_tier row) merged earlier. Both code and canon land in production together with this promotion.

Post-merge verification

After CF Workers Builds completes:

-- Hit rate over last 30 min — will return real numbers once the new bundle is live
SELECT SUM(cache_hits) AS hits, SUM(cache_lookups) AS lookups,
       SUM(cache_hits) * 1.0 / NULLIF(SUM(cache_lookups), 0) AS hit_rate
FROM oddkit_telemetry
WHERE timestamp > NOW() - INTERVAL '30' MINUTE;

And the trace envelope on any real action will switch from index_source to fetches[] + cacheStats.

---

Note

Medium Risk
Medium risk because it changes the telemetry/tracing schema (retiring blob9 and adding new doubles) and refactors how storage/cache access is recorded, which can affect observability dashboards and downstream queries.

Overview
Retires the single-value telemetry field cache_tier (blob9) and replaces it with cache effectiveness metrics: cache_hits and cache_lookups (double7/double8), sourced from per-request fetch recording.

Refactors RequestTracer to stop interpreting a “winning” cache tier (indexSource) and instead record per-fetch facts via recordFetch, exposing fetches[] and cacheStats in trace output; index.ts now passes tracer.cacheStats into recordTelemetry after streaming responses complete.

Updates the baseline fetcher instrumentation and integration tests to emit/validate the new fetch-based tracing and updated Analytics Engine schema.

^{Reviewed by Cursor Bugbot for commit 1635142. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	oddkit	1635142	Commit Preview URL Branch Preview URL	Apr 26 2026, 09:16 PM

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: New fetchCount getter is never used anywhere
  • Removed the unused fetchCount getter from RequestTracer in workers/src/tracing.ts since it had zero callers in production or tests.

Preview (1635142936)

diff --git a/workers/src/tracing.ts b/workers/src/tracing.ts
--- a/workers/src/tracing.ts
+++ b/workers/src/tracing.ts
@@ -156,9 +156,4 @@
   get spanCount(): number {
     return this.spans.length;
   }
-
-  /** Number of recorded fetches. */
-  get fetchCount(): number {
-    return this.fetches.length;
-  }
 }

_{You can send follow-ups to the cloud agent here.}

^{Reviewed by Cursor Bugbot for commit 1635142. Configure here.}

[cursor]

New fetchCount getter is never used anywhere

Low Severity

The newly added fetchCount getter on RequestTracer has zero callers — not in production code, not in tests. Its analog spanCount is exercised in the test file (line 671), but fetchCount is never referenced. The test at line 663 uses json.fetches.length instead, bypassing the getter entirely.

 

^{Reviewed by Cursor Bugbot for commit 1635142. Configure here.}