mirror of
https://github.com/PerpetualSoftware/pad.git
synced 2026-09-11 13:28:57 +00:00
xarmian
1e40c972ae
Codex round 14 was a prune pass rather than a defect hunt: 695 comment lines to 243 code lines in internal/events and internal/idspace is a ratio worth looking at, and my day-51 close records what happens when review prose grows under a diff. MOSTLY DECLINED, and the reason matters more than the decision. The list would have stripped this diff to terse definitions -- deleting bug archaeology, rollout tradeoffs and codex-round attributions -- in files whose EXISTING comments are dense narrative and carry exactly those things, "(codex round 6)" attributions included. It also named several blocks on code this diff never touched. Applying it would have made the new code inconsistent with the file it lives in, which is a style change to someone else's codebase wearing a review's clothes. APPLIED, because both match lessons this team already recorded rather than a reviewer's preference: - The numeric-base-vs-travelling-epoch argument existed in FOUR copies, in both packages. The lead's requirement was that the asymmetry declare itself in both buses' comments -- not that the argument be written four times. It now lives once, in internal/idspace's package comment (the shared code both halves depend on), and the four sites state the asymmetry in a sentence and point there. Four copies is four things to update when the deferred follow-on lands. - "Costs at most ONE drop per instance per roll" was a countable claim in prose with nothing asserting it. A comment asserting countable behaviour belongs in the suite: later messages in the same generation must leave the buffers alone, or "bounded" in that sentence is doing no work. Honest measurement, since the point of the round was volume: the net is +47 -41. The prune did not shrink the diff; it removed three copies of one argument and turned one claim into an assertion. Claude-Session: https://claude.ai/code/session_01JVDBKbgn3Xt7ndW1YoYd8X