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.

Record paired trace evidence for restored Requests task cost

David Zhang046929657bd6
diff --git a/specs/jevgrep/assets/parity-restoration.md b/specs/jevgrep/assets/parity-restoration.md
index bab2336..8ab355a 100644
--- a/specs/jevgrep/assets/parity-restoration.md
+++ b/specs/jevgrep/assets/parity-restoration.md
@@ -113,6 +113,16 @@ Historical live answers were not retained, and the coding agent generated its
own query, so these live differences alone do not establish a port mismatch or
explain the cost increase. Full traces and patch are retained for comparison.
+The paired trace comparison places the added cost after retrieval: the first two
+restored generations cost $0.0704952 versus $0.0713406 in the accepted spike.
+After an initial passing regression, the restored agent reread test definitions,
+changed assertion style and ran six methods; the spike ran its two added methods
+once. The restored implementation patch matches baseline, while the spike's
+branch arrangement differs. The tests cover different cases, so neither test set
+strictly contains the other. These observations identify downstream work; they
+do not establish that file-list differences caused it. The saved generation
+receipts fully account for the $0.064617 increase over baseline.
+
The quality gate remains every baseline solve preserved and at least seven
solved strict Sol-cost wins in this single frozen ten-task cohort. One restored
solve and deterministic parity do not satisfy it.