Find code by asking what it does. A CLI for coding agents that uses Jev to discover relevant files and source context.
TypeScript ai-sdkclaude-codeclicode-searchcodexcoding-agentscontext-retrievaldeveloper-toolsjevsemantic-searchtypescriptvercel-ai-gateway
docs: record outage recovery and freeze unchanged replacement cohort
David Zhang3f960d011cae
diff --git a/specs/jevgrep/README.md b/specs/jevgrep/README.mdindex a9665d8..dfe7582 100644--- a/specs/jevgrep/README.md+++ b/specs/jevgrep/README.md@@ -7,15 +7,14 @@ Jev classification through Vercel AI Gateway. This is the build plan; production## Next Agent PromptYou are implementing this spec. Status: **implementing**, updated **2026-09-26**.-The only active confirmation is `installed-jg-cpython-work-clock-v1`.-Django officially solved below its immutable baseline; the registered rule now-permits the other nine tasks in that same frozen plan. Scikit-learn also officially solved below baseline; the sequential runner is-now on Astropy, where Docker has stopped after a disk-space error. Restore the-engine and inspect the existing attempt before any continuation; no replacement.-The runner and Docker restart are still pending. Collect every terminal receipt, official grade and complete-Sol bill, then aggregate the full cohort. Never replace an attempt or rerun a-baseline. [Timing protocol and current evidence](assets/timing-protocol.md) own-this study's procedure and results.+The work-clock v1 confirmation stopped after Docker's disk-space failure during+Astropy. That attempt is failed and recovered separately; two completed wins and+seven unstarted cells remain retained. Do not resume or replace its failed cell.+Docker is recovered and disposable build cache was cleared. The new full `installed-jg-cpython-work-clock-v2` study is prepared and all ten+no-call validations pass with identical package and protocol. Execute each cell+once and grade/account all outcomes.+Never pool studies or rerun a baseline. [Timing protocol and evidence](assets/timing-protocol.md)+own the outage, recovered bills and prospective replacement rationale.Keep production retrieval and the canonical skill unchanged. The installedpackage matches the restored bundled-CPython candidate; only the prospectivediff --git a/specs/jevgrep/assets/timing-protocol.md b/specs/jevgrep/assets/timing-protocol.mdindex ded6c72..946c0ed 100644--- a/specs/jevgrep/assets/timing-protocol.md+++ b/specs/jevgrep/assets/timing-protocol.md@@ -124,7 +124,7 @@ The production patch again calls `check_random_state` unconditionally. It repeatthe compatibility issue established in the first faithful cohort: with shufflingdisabled, invalid ignored seeds accepted before the patch now raise. The officialsolve remains recorded, alongside this broader-quality caveat. No agent or-baseline is rerun. Astropy is the current active cell in the sequential runner.+baseline is rerun. Astropy was the next active cell before the outage described below.## Infrastructure interruption under investigation@@ -132,8 +132,43 @@ baseline is rerun. Astropy is the current active cell in the sequential runner.While Astropy was active, Docker Desktop stopped serving its engine. Host uptimecontinued; the backend log at 2026-09-26T11:36:02Z reports inability to write`Data/log/vm/init.log` with “no space left on device.” A subsequent host disk check-showed about 5.8 GiB available. The original runner process remains alive, but no-new Astropy events arrive. Docker restart is pending; neither Astropy nor another-cell has been replaced. Check actual process/container state before classifying-the attempt, collecting evidence, or continuing. Do not infer model failure from-this service outage or silently exclude it from the cohort.+showed about 5.8 GiB available. At that observation the original runner remained alive without new Astropy+events, and Docker restart was pending. The recovery below supersedes that+process status. Do not infer model failure from this service outage or silently+exclude it from the cohort.+++### Recovery and prospective replacement study++The original runner terminated: Astropy has `status=failed`, exit 255 and no+completed native turn. Its remaining seven cells were never started. No patch+existed in the recovered working tree. This cohort cannot satisfy acceptance and+will not be resumed or have its interrupted cell replaced.++Docker recovered after its stuck shutdown processes were cleared. Native sessions,+597 Jev request/response pairs and broker logs were copied from stopped containers+into `installed-jg-cpython-work-clock-v1/recovery/astropy__astropy-13579`, with a+hashed recovery manifest. Original attempt receipts are unchanged. Recovery-only+accounting retains known Sol **$0.2080088** (full bill unknown) and known Jev+**$0.126530544** (also not asserted complete). The two owned containers and their+network were removed only after copying evidence and checking the clean source.+No coding agent resumed. `/tmp/jg-docker-outage/` retains recovery procedure/logs.++Unused build cache older than an hour was pruned (9.721 GB); images, volumes,+baselines and research evidence were preserved. Host free space then measured+approximately 24 GiB. The preceding older-than-24-hours prune removed zero bytes.++Prepare `installed-jg-cpython-work-clock-v2` as a new full ten-task study using+the exact same production tarball, skill, runtime and work-clock procedure. The+reason is the documented host failure, not a quality or cost selection. Freeze+and validate all cells before calls. Run the full fixed cohort once; include all+its outcomes and costs, and never pool v1's two wins into it. Retain the interrupted+v1 cohort and its costs alongside the new result. Baselines remain immutable and+are executed zero additional times. No product behavior changes are introduced.+++V2 preparation and all ten no-call validations passed. The 179 installed files,+package, canonical skill, archived runner/broker/registry, task inputs, dataset+and timing policy match v1 byte-for-byte where applicable. Proof:+`/tmp/jg-work-clock-v2-freeze-proof.json`. Independent recovery review verified+all 1,198 manifest hashes and the unchanged failed receipt before paid continuation.