Skip to content

docs: handoff for the 2026-07-21/22 session (release point + method findings)#662

Closed
ligon wants to merge 1 commit into
developmentfrom
docs/handoff-2026-07-22
Closed

docs: handoff for the 2026-07-21/22 session (release point + method findings)#662
ligon wants to merge 1 commit into
developmentfrom
docs/handoff-2026-07-22

Conversation

@ligon

@ligon ligon commented Jul 22, 2026

Copy link
Copy Markdown
Owner

Session handoff. development is at fb749f0, CI green, 135 commits / 45 merges ahead of master — a release point.

Covers:

The method section is the part worth reading before dispatching agents. Each item made a failure look like a success at least once today, and the through-line is this: of seven PRs held by adversarial review, none had incorrect code. Every one had a false claim — a measurement that did not reproduce, a test that discriminated nothing, a stale duplicate, a comment asserting cleanliness. That class is invisible to CI and outlives the PR that introduced it.

Docs only.

`development` is at fb749f0, CI green, 135 commits / 45 merges ahead of
`master` -- a release point.  Records what landed, what is waiting on a human
decision, what is still open, and the method findings.

The method section is the part worth reading before dispatching agents.  Every
item in it made a failure look like a success at least once today:

  - the cache hides the bug it caused (five times; LSMS_NO_CACHE=1 is soft for
    script-path parquets);
  - WHICH CONFIG TREE you measured matters as much as the cache --
    countries_root() follows the main checkout's branch, and two "bugs"
    reported today were master's config rather than stale data;
  - a negative control must actually REMOVE the change (git stash does nothing
    to a committed change and silently reports "no difference");
  - a test that passes before AND after is not evidence;
  - zero can mean UNMEASURED rather than clean -- and a BROKEN KEY produces
    zero duplicates;
  - never trust a PR body's status table (three agents redid merged work);
  - put CONTENTS.org in the agent's BRIEF, not just in a rule it may skim.

And the through-line: of seven PRs held by adversarial review, NONE had
incorrect code.  Every one had a false claim -- a measurement that did not
reproduce, a test that discriminated nothing, a stale duplicate, a comment
asserting cleanliness.  That class is invisible to CI and outlives the PR.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@ligon

ligon commented Jul 22, 2026

Copy link
Copy Markdown
Owner Author

Closing — a handoff is a working note, not a reviewed artifact, and it does not belong in the review queue.

The file now lives in the working tree at slurm_logs/HANDOFF_2026-07-22.org (174 lines), which is where the next session will look for it. Branch docs/handoff-2026-07-22 remains on origin if anyone wants the committed copy.

Note the TL;DR is now one step stale in a good way: it said "development is a release point"; v0.9.0 has since been cut from d0d62214, with publish.yml firing on the release event. Everything else in it stands.

@ligon ligon closed this Jul 22, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant