Skip to content

Find code by asking what it does. A CLI for coding agents that uses Jev to discover relevant files and source context.

docs: retain first corrected-package result without pooling

David Zhangf331ec24d6cc
diff --git a/specs/jevgrep/README.md b/specs/jevgrep/README.md
index 0acc36f..e0e7814 100644
--- a/specs/jevgrep/README.md
+++ b/specs/jevgrep/README.md
@@ -20,6 +20,9 @@ Full merged verification and exact archive platform checks passed, followed by
all ten frozen no-call validations. [Freshness evidence](assets/source-freshness.md)
owns the archive identity, failure reproductions and verification results. Do not
transfer the previous cohort's measured results to this corrected archive.
+[Current confirmation](assets/freshness-confirmation.md) records completed cells;
+scikit-learn solved officially but exceeded baseline cost. The choices ledger has
+been consolidated against current code; final reconciliation and archive remain.
Preserve healthy retrieval semantics and the canonical skill while fixing
filesystem freshness. Keep the measured archive immutable; it matches the
diff --git a/specs/jevgrep/assets/freshness-confirmation.md b/specs/jevgrep/assets/freshness-confirmation.md
new file mode 100644
index 0000000..4d6da7d
--- /dev/null
+++ b/specs/jevgrep/assets/freshness-confirmation.md
@@ -0,0 +1,28 @@
+# Corrected-package confirmation
+
+Status: running. This is a separate prospective ten-task cohort for the
+[source-freshness correction](source-freshness.md), not a replacement of individual
+cells from the preceding accepted study. The frozen plan, archived runner and
+attempt evidence live at `evals/runs/swebench/installed-jg-freshness-v1/`.
+All saved baselines are reused without execution. No prior treatment outcome is
+pooled into this cohort.
+
+Full Sol task cost is the scored metric. Observed Jev API costs are reported
+separately and excluded; incomplete observations are subtotals, not totals.
+
+| Task | Official result | Sol | Baseline Sol | Jev observed API cost |
+| --- | --- | ---: | ---: | --- |
+| scikit-learn__scikit-learn-13124 | solved; not a cost win | $0.3434830 | $0.2944360 | known $0.046611348; total unknown |
+
+Scikit-learn returned five relevant files, implementation reading leads and test
+source. The agent then read the splitter and tests, searched documentation,
+reproduced the paired-fold problem and changed the shared random-state handling.
+It ran broader tests as well as focused regressions. Its unconditional
+`check_random_state` call retains the known invalid-seed compatibility caveat
+when shuffling is disabled. Official success does not establish that broader
+behavior unchanged, and the graded patch is not repaired after the fact.
+
+Root execution log: `/tmp/jg-freshness-v1-run.log`. Per-task grade and accounting
+receipts, complete event streams, patches and request/response evidence remain
+with the attempts. Acceptance stays unproven until all ten outcomes are retained
+and the full aggregate is evaluated.