Enable Workers Logs (incident J-002 follow-up)

[klappy]

Why

Direct response to incident J-002 (transient 503s on aquifer.klappy.dev/mcp at 2026-04-22T00:17:02Z, single trace 3daca3e0-1024-4733-93af-510d65a8b7ae). Investigation 28 hours later established that:

• The deployed Worker bundle returns status: 500 from its only two error paths and status: 503 from zero. Every 503 the orchestrator sees from this service is, by mechanism, a Cloudflare-edge-injected response (Error 1101/1102, R2 transient, isolate cold-start collision, or upstream Cloudflare service blip) — not application logic.
• wrangler.toml had no [observability] block, so Workers Logs was never ingesting per-invocation logs / exceptions / CPU and wall time / subrequest counts. The data class that would have distinguished those four hypotheses for the J-002 incident does not exist and cannot be reconstructed.

This PR closes that visibility gap so the next 503 is a one-query investigation instead of a four-hypothesis exploration.

What

wrangler.toml:

• Top-level [observability] block, enabled = true, [observability.logs] with invocation_logs = true and head_sampling_rate = 1.
• Same block mirrored under [env.staging] for parity with wrangler dev --env staging and the staging branch preview at staging-aquifer-mcp.klappy.workers.dev.

odd/ledger/journal.md:

• Appends J-002 entry covering the incident: Observation, Learning, Decision (this PR), Constraint (logs are not retroactive), and four Handoffs.
  • H7 — oddkit_encode did not honor DOLCHE splitting on the planning session that produced this entry; collapsed five-section payload into one Constraint stub. Recorded for separate investigation against the oddkit worker.
  • H8 — this PR.
  • H9 — bt-servant orchestrator constructed get arguments as {compound_key: "..."} instead of three separate fields; investigate tool-description presentation layer.
  • H10 — open question whether "always enable Workers Logs on klappy.dev MCP servers" deserves its own canon constraint.

Validation after merge

Workers Builds will redeploy on push to main. Within ~1 hour of the deploy, the Workers Logs view should populate:

• Production: https://dash.cloudflare.com/b03e6ea242724c05eb97eb732cceb21d/workers/services/view/aquifer-mcp/production/observability/investigate
• Staging: https://dash.cloudflare.com/b03e6ea242724c05eb97eb732cceb21d/workers/services/view/aquifer-mcp-staging/production/observability/investigate

Quick sanity check: wrangler tail --name aquifer-mcp --format pretty should stream live invocation events; previously this returned no events because no telemetry was being captured.

Risk

Low. The change is purely additive — no code changed, no bindings changed, no compatibility flags changed. Cloudflare Workers Builds will redeploy the Worker on merge (the deployed bundle is bytes-identical to the current 2026-04-13T21:35:14Z deploy plus the observability config flag). At head_sampling_rate = 1 the maximum possible log ingest cost is 1 row per invocation — within the Workers Paid free allotment for this service's traffic level.

Mode trail

Investigation conducted under exploration → execution per the project model operating contract. J-002 entry was hand-written rather than copied verbatim from oddkit_encode output because the encoder collapsed the OLDC+H payload to a single Constraint stub (see H7); the hand-written entry follows the J-001 narrative precedent already at the bottom of the journal.

---

Note

Low Risk
Low risk: configuration-only change enabling Workers Logs for production and staging; no runtime code paths, bindings, or caching behavior are modified.

Overview
Enables Cloudflare Workers invocation logs for aquifer-mcp by adding [observability] (and mirrored [env.staging.observability]) to wrangler.toml, so transient edge-injected 5xxs can be investigated with per-invocation telemetry.

Appends a new journal entry J-002 to odd/ledger/journal.md documenting the 2026-04-22 transient 503 incident and the decision to enable Workers Logs as the follow-up action.

^{Reviewed by Cursor Bugbot for commit 2059ac0. 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	aquifer-mcp	2059ac0	Commit Preview URL Branch Preview URL	Apr 23 2026, 04:36 AM