Gap ledger: observation-inference (entrypoint label, plan limits) and scheduling-supervision (restored cron tasks) - #1158

Scope

• What this PR changes, in one or two sentences: This PR closes the three ADOPT rows of the Claude Code practice gap ledger on two small layer pages: observation-inference (slot usage-cost-monitoring: A3-0090 and B2-0005) and scheduling-supervision (slot scheduled-tasks: A5-0024). The command center dispatched the ledger on 2026-10-15, and each row was confirmed by a second-family gpt-6-astra re-check. The plan asks for one PR per layer page; these two are small, so they share one commit.
• Base commit: 420452059
• Lane: lane:foundation
• Owned paths touched:
  • the practice skill's slots.json, observation-inference.md and scheduling-supervision.md;
  • the slot manifest (only sha256_when_labeled changes);
  • adoption/templates/claude.settings.template.json and observability/collector/claude-settings.json.example;
  • catalogs/foundation/upstream-surface-dispositions.json;
  • docs/decisions/2026-10-02-new-wsl-client-configuration.md (a dated addendum: one more wired piece, 403 to 404);
  • tests/test_native_practice_ledger_obs_sched.py (new);
  • manifests/evidence.json.

A3-0090: label every Claude Code metric with the session entrypoint.

• Change: OTEL_METRICS_INCLUDE_ENTRYPOINT: "true" goes into the template and the collector example, which the tests treat as a pair. The new disposition row is adopt-pending, with the template as carrier.
• Inert on 2604 today: the collector's privacy filter keeps a fixed list of metric labels without app.entrypoint. That holds in the repository profile (observability/collector/collector.yaml:412,425) and in the live profile, which is byte-identical to the install plan's copy (:415,428).
• Host diff (the command center's act): proposed to the command center, with the collector keep-list step first, then the settings key, a Prometheus check and the inverse. Both the patched and the unpatched profile pass otelcol-contrib 0.162.0 validate.
• Follow-up: the repository-side collector.yaml change belongs to the observability lane, because that label list is a privacy control.

B2-0005: a record only. claude -p, Agent SDK and third-party harness usage under a plan still draws on the plan's limits; the programmatic credit announced for 2026-06-15 is paused.

• The same page now also carries a 2026-10-07 update: Max and Team plans get monthly API credits. They cover a linked Console organization's API-key usage only and leave plan limits unchanged (support article 17154008).

A5-0024 (a correction): CronCreate tasks come back on --resume or --continue and carry into a background session, so the resume and background paths now run CronList first and cancel stale tasks with CronDelete.

• Measured on this host, 2026-10-15: tab session 3d36653d created task 60ad686c at 01:35:01Z. Its background copy 33888fe1 (--resume … --bg) logged scheduled_task_fire for that task at 07:36:00Z.
• Untested: backgrounding from inside a session is documented for /loop tasks only.

SOTA sources

• https://code.claude.com/docs/en/env-vars.md:510, fetched 2026-10-15T10:33:04Z, sha256 54351937…
• https://code.claude.com/docs/en/monitoring-usage.md:156, :577-584, fetched 10:33:04Z, sha256 72374067…
• https://code.claude.com/docs/en/scheduled-tasks.md:11, :159-160, :228, :230, fetched 10:33:04Z, sha256 8cb643f4…
• https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan (updates of 2026-06-15 and 2026-10-07), fetched 10:33:04Z. Each response carries its own nonce, so its hash differs on every fetch.
• https://support.claude.com/en/articles/17154008-monthly-api-credits-for-max-and-team-plans, fetched 10:37:03Z, sha256 17185d57…
• opentelemetry-collector-contrib 0.162.0 (otelcol-contrib validate) and Claude Code 2.1.300.
• The measurement: evidence/artifacts/d4-bg-trial-20261015/scheduled-task-carry.json, read from the two sessions' own transcripts.

Evidence-class table

Local commands run

$ python3 -m unittest tests.test_native_practice_ledger_obs_sched   # base 420452059
FAILED (6 of 6), e.g. AssertionError: None != 'true'
$ python3 -m unittest tests.test_native_practice_ledger_obs_sched tests.test_new_wsl_client_config tests.test_native_practice_*
OK
$ python3 scripts/validate.py
rc 0; {"components": 71, "hashed_files": 12056, "profiles": 4, "receipts": 240, "status": "passed"}

tests.test_upstream_surface_watch has four CachedArtifactTests failures, identical on base. #1155 fixes them.

Digest custody fold, now squashed into 719f9586

Digest custody, before any read: the scheduled-tasks notes cited a /var/tmp findings file and a digest prefix of it, and the usage notes and tests cited a /var/tmp host-diff path. The measurement is now kept in the repository at evidence/artifacts/d4-bg-trial-20261015/scheduled-task-carry.json, and the host steps are named as proposed to the command center. test_the_rows_cite_no_scratch_path_or_unpublished_digest guards both rows.

Squash (719f9586), from the GPT read at 8c03cbe

• P1, digest custody: the custody regression in tests/test_native_practice_ledger_obs_sched.py held a real prefix of an unpublished digest as a literal. It now checks both rows for the scratch roots and the private findings files' name, and holds no private value. A synthetic negative control plants a scratch reference with a synthetic digest, and the check catches it.
• One commit on the same base: the branch is squashed onto 420452059 as 719f9586, following the custody precedent of Split adaptive paper sells within the durable order envelope #1091, ci: rebalance native module shards from measured Actions timings #1148 and Build frozen Jev trading case sets and paired scorer #1150. No commit in the branch keeps the prefix. The 7- and 8-character prefixes occur 0 times in git log -p 420452059..719f9586, 0 times in the head tree, and 0 times in this description (counts only).
• Replaced commits: ed5297ed8 and 8c03cbe5f may stay reachable by SHA until GitHub's garbage collection. A purge is a support request that only the owner can make.
• Results: the module plus tests.test_native_practice_snapshot_provenance and tests.test_published_digest_privacy ran 32 tests, OK (1 skipped); validate rc 0.

Rebase (a0318bd8) for landing

• On the command center's landing cue, the commit was rebased once from 719f9586 (base 420452059) onto main 8053f14e7, as a0318bd8. The commit message is byte-identical, and the delta is resolution-only:
  • claude-native-slot-manifest.json: main's copy, with sha256_when_labeled set to the merged slots.json.
  • manifests/evidence.json: main's copy, with the branch's 10 changed files re-registered (both new rows kept).
  • component_matrix.py, new_host_grand_list.py and final_stack.py --check all exit 0, so no generated report changed.
• Results at a0318bd8: the commands above (the obs-sched module, tests.test_new_wsl_client_config and tests.test_native_practice_*) ran 350 tests, OK (3 skipped); validate rc 0.
• The 7- and 8-character digest prefixes occur 0 times in git log -p 8053f14e7..a0318bd8 and 0 times in the head tree (counts only).

Follow-ups

• From the GPT read at 8c03cbe (P3s, left to the command center's ruling):
  • app.entrypoint is still dropped at the collector boundary (observability/collector/collector.yaml:412, :425; new-wsl-install-plan-20261002/config/otel.yaml:415, :428). The pinned collector filters attributes before grouping (open-telemetry/opentelemetry-collector-contrib@ae8c5075 processor/transformprocessor/internal/metrics/func_aggregate_on_attributes_metrics.go:51-56). The observability change must update both lists in both profiles and verify the exported series.
  • The resume route lists and cancels stale tasks, but the ordering is unmeasured: restore, stale-task cancellation, expired one-shots, recurring expiry and background recovery are not exercised.
  • docs/decisions/2026-10-09-claude-code-gap-resolution.md:62 cites template line 27 for CLAUDE_CODE_SUBAGENT_MODEL, which is line 28 at this head.
• From the GPT fix1 read at 719f958 (ACK, 0 blockers; the three P3s above repeat, and one is new):
  • The synthetic negative control in tests/test_native_practice_ledger_obs_sched.py:106-122 checks its own marker expression and never calls the row assertion, so it would still pass if that check were a no-op. The guard itself rejects the planted reference, so this is test hardening. Fix: run the same guard callable in the positive and negative cases, and cite the supported unittest API in the correction record (python/cpython@cbc944f4 Lib/unittest/case.py:740, :1176; Doc/library/unittest.rst:995).
• The repository collector.yaml keep-list for app.entrypoint is the observability lane's change; this PR's test fails as soon as that list changes, so the row is updated in the same commit.
• Transcripts record entrypoint as cli for both tab and --bg sessions. If the metrics carry the same value, the label will not separate background sessions.
• The open practice-row PRs all rewrite slots.json and sha256_when_labeled, so a manifest rebase follows whichever lands first.

Checklist

• No GitHub Actions changed.
• No workflows changed.
• No secrets are printed, logged or committed.
• No new paid hosting, subscription or billing surface.
• Peer-owned untracked files and worktrees were preserved.

🤖 Generated with Claude Code