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 Django confirmation and current cohort handoff
David Zhang8cb0688ccf1d
diff --git a/specs/jevgrep/README.md b/specs/jevgrep/README.mdindex be8299b..2c2773d 100644--- a/specs/jevgrep/README.md+++ b/specs/jevgrep/README.md@@ -7,72 +7,44 @@ 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**.-Next: collect the running Django attempt in `installed-jg-cpython-work-clock-v1`,-then grade/account and apply the registered Django-first continuation rule.-The [agent-work timing protocol](assets/timing-protocol.md) passed review and-merged Docker checks; all ten cells are frozen and no-call validated.-The [source placement experiment](assets/presentation-study.md) is terminal and-failed; its remaining nine cells stay unrun. Keep production renderer and skill unchanged.-The no-model counterfactual preserves all evidence while moving first-ranked-source inside the observed tool prefix. Real SQLite replays match the frozen oracle, and an 81-request-three-file replay proves both pipelines exceed the follow-up threshold and skip-that pass. Exact coverage and whole-pipeline limits are in the-[mechanism-first skill experiment](assets/query-framing-study.md). That candidate-failed the same Django regression at Sol $1.2353512; its remaining nine tasks stay-unrun and the production skill is unchanged. Packet-visibility comparison is-complete: initial truncation hid selected source, but a later read and explicit-recognition preceded replacing the failing regression with a type-changing test.-Do not infer another query-breadth fix. Django under the work-time protocol is-the only active paid task; do not launch a replacement attempt.-The frozen `installed-jg-cpython-parity-v1` cohort is complete and not accepted:-7/10 solves versus 8/10 baseline, six cost wins, and 35.27% lower full Sol cost.-[Confirmation evidence](assets/cpython-confirmation.md) owns all outcomes, trace-findings and quality caveats. Never rerun a baseline, replace an attempt, or treat-the lower total as compensation for a lost solve. Keep the measured CLI unchanged-while testing any query/skill hypothesis as a separately identified candidate.--The merged `bun run verify` gate passed and whole-product review's sole finding-is fixed with a clean follow-up review. Exact archive checks pass on macOS arm64-and Linux arm64/amd64. [Runtime evidence](assets/python-runtime.md) records the-artifact and interpreter limits. The architecture port is verified for the-exercised corpus; task-quality acceptance failed and remains required.--`installed-jg-reference-parity-v1` is stopped and superseded. Requests and-scikit-learn officially solved but exceeded baseline cost; Django was interrupted-with its patch and traces retained; seven cells never started. These superseded-studies remain stopped. The bundled-CPython cohort is terminal. The completed-query diagnostics made only bounded Jev calls; the separate mechanism-first plan-owns the new skill candidate and its Django-first evaluation.--The changed-policy cohort `installed-jg-final-cohort-v1` was stopped and marked-superseded, retaining completed outcomes and interrupted traces. It is not a-release acceptance cohort. Never rerun saved baselines or pool candidates.-Report Jev observed API costs separately from scored full Sol cost.--Keep the production incumbent's restored expansion, discovery inputs/order,-excerpt presentation, scoped guidance and accepted skill (only rename executable-to `jg`) unchanged. The isolated experiment changes only the recorded skill-instruction and does not promote that change into production. Required-Node-only runtime, auth, cache, filesystem protections and truthful failure-handling remain, but must not silently change healthy retrieval decisions.-The earlier deterministic and platform gates passed for the deviating port;-they do not prove parity. See [parity restoration](assets/parity-restoration.md) and the-[source-budget investigation](assets/source-budget-decision.md).-Remaining: a candidate that passes frozen quality confirmation, final choices-consolidation, whole-spec closeout review and close-spec.-See [contracts](contracts.md) for the corrected contract.--No user decision blocks starting. External verification dependencies are a working-Docker daemon, explicit Gateway credentials for live tests, retained official task-snapshots/baselines, and macOS access for native install verification. Missing live-inputs do not block deterministic development, but do block the corresponding-acceptance claim. Do not repair credentials or regenerate an existing baseline.--Implement one focused slice at a time, running its narrow tests and recording its-artifact. Use write-tests for behavioral tests and review before completing an-implementation pass. Update this section, slice status, decisions and unresolved-verification before ending each pass. The named commands are implemented; their-recorded evidence does not substitute for the open quality gate.+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. The sequential runner has+started scikit-learn. 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.++Keep production retrieval and the canonical skill unchanged. The installed+package matches the restored bundled-CPython candidate; only the prospective+benchmark timing protocol differs. The failed query and presentation experiments+have not earned promotion. Their remaining cells stay unrun. Prior outcomes+cannot be pooled into the active cohort.++Evidence and boundaries:++- [Runtime verification](assets/python-runtime.md): merged default tests,+ whole-product review, and exact archive checks on macOS arm64 and Linux+ arm64/amd64 passed. Interpreter parity is established for the exercised corpus,+ not all possible inputs.+- [Restoration record](assets/parity-restoration.md): frozen reference, known+ deviations and controlled request/output/recovery comparisons.+- [Completed first confirmation](assets/cpython-confirmation.md): rejected for+ losing a baseline solve despite lower total cost. Its traces remain evidence.+- [Query trial](assets/query-framing-study.md) and+ [presentation trial](assets/presentation-study.md): rejected experiments,+ source-parity probes and limits on causal explanations.++Priority: finish this frozen quality confirmation; address any failing contract+without weakening acceptance; run whole-spec closeout review; consolidate choices+against the final code; then close-spec. Slice 08 remains open. No publication or+release tag is authorized for the intentional `0.0.0` development checkpoint.+Report observed Jev costs separately; they never enter scored Sol task cost.++No user decision blocks continuing. Use the authorized credentials and retained+Docker snapshots; do not repair credentials or regenerate baselines. Preserve+user-edited skill files and local historical evidence. Stage only explicit files,+keeping deprecated personal-repository evals out of Git.- [x] [01 — Reference and real HTTP fixture](slices/01-reference.md)- [x] [02 — Bundled parser parity](slices/02-parser.md)diff --git a/specs/jevgrep/assets/parity-restoration.md b/specs/jevgrep/assets/parity-restoration.mdindex 703ebea..8d42db8 100644--- a/specs/jevgrep/assets/parity-restoration.md+++ b/specs/jevgrep/assets/parity-restoration.md@@ -4,9 +4,9 @@ This record preserves the pre-CPython restoration work and its supersededcohorts. The bundled-CPython port subsequently passed integration, platform andwhole-product checks; [runtime evidence](python-runtime.md) owns that coverage.Its completed quality cohort failed acceptance, as recorded in-[confirmation evidence](cpython-confirmation.md). The separate-[mechanism-first skill experiment](query-framing-study.md) owns the active trial;-none of the stopped studies below is reactivated.+[confirmation evidence](cpython-confirmation.md). The separate query and presentation experiments also failed.+[The prospective timing protocol](timing-protocol.md) owns the active faithful+confirmation; none of the stopped studies below is reactivated.The user rejected unvalidated architecture changes during production porting.The accepted reference is the source closure and skill bound bydiff --git a/specs/jevgrep/assets/timing-protocol.md b/specs/jevgrep/assets/timing-protocol.mdindex 7bc4ae2..f519694 100644--- a/specs/jevgrep/assets/timing-protocol.md+++ b/specs/jevgrep/assets/timing-protocol.md@@ -71,7 +71,18 @@ Study `installed-jg-cpython-work-clock-v1` is frozen and all ten no-call checkspass. Its 179 installed files match the original bundled-CPython cohort exactly;the skill hash is`1a6b29ca8855d3538883c6dada289db68dd2dfca90f466e39407902dc000dd50`.-Proof: `/tmp/jg-work-clock-freeze-proof.json`. Django is running first under the-registered rule. Collect its terminal receipt, official grade and complete bill-before any continuation. No other cell has started. Production CLI and canonical-skill remain unchanged.+Proof: `/tmp/jg-work-clock-freeze-proof.json`.++Django completed with valid timing evidence and officially resolved (one solve,+zero infrastructure/errors). All 37 Sol requests are accounted for: **$1.1409928**+versus the saved **$1.5055472** baseline. Jev's known observed subtotal is+**$0.148011822**; its complete total remains unknown (776 cost responses for 777+client calls) and is excluded from scoring.++The receipt records 359.286 seconds wall time, 222.354 charged work and 136.933+credited retrieval. This run fits the old 900-second wall limit too: its success+must not be attributed to gaining more time. Query and agent trajectory vary.+The official pass and lower complete Sol bill satisfy the registered continuation+rule, so the sequential runner has started the remaining nine tasks in this same+plan. No Django replacement or baseline execution occurs. Production CLI and+canonical skill remain unchanged; full-cohort acceptance is still unproven.diff --git a/specs/jevgrep/slices/08-release.md b/specs/jevgrep/slices/08-release.mdindex 77d1cc0..8d835ec 100644--- a/specs/jevgrep/slices/08-release.md+++ b/specs/jevgrep/slices/08-release.md@@ -1,6 +1,6 @@# Confirm quality and the release artifact-Status: frozen confirmation complete and rejected; solve preservation fails on Django and only six cost wins are verified. Depends on: 07. Read [contracts](../contracts.md) first.+Status: open. The first faithful confirmation was rejected; the new frozen work-clock cohort is running after its Django-first continuation gate passed. Depends on: 07. Read [contracts](../contracts.md) first.## Contract and owner@@ -28,8 +28,9 @@ One frozen release-candidate dossier.[the runner workflow](../../../evals/implementation/swebench/installed.md); nativemacOS installed smoke command documented by the harness.-Commands are implementation targets. Add them in this slice; do not imply they-already exist. CLI transcripts replace visual/screenshot gates for this product.+These commands are implemented. Use the frozen installed runner to finish the+open quality gate and retain its evidence. CLI transcripts replace visual/screenshot+gates for this product.## Verification and verdict