The two-clause pilot and its issue 88 validity check have no page #962

New issue

Closed

opened 2026-08-18 16:23:55 +00:00 by coilyco-ops · 1 comment

coilyco-ops commented

2026-08-18 16:23:55 +00:00

Member

Dropped in #957 and not restored, because it does not fit. The passage from the old board page:

The board ships as a two-clause pilot, and the first has a documented real-world regression behind it, which gives the board a validity check the sibling suite never had: if its out half does not reproduce issue 88 against the pre-fix bundle, it is not measuring. The remaining five wait on the first graded result, because in the sibling suite the generator's predictions about which cases would discriminate ran one for three.

That is the strongest argument on the board, and it exists nowhere else: it says how you would know the board measures anything at all.

Why it does not fit. docs/sirens-echo-eval.md is at 119 of 120 lines and 7983 of 8000 chars for the large band, and docs/*.md is at the 40-doc cap. documentation-layout has no per-file size override, only agents_md_max_lines and the README pair, so the page cannot simply grow. Restoring the passage costs about 620 chars of something else.

Options, in the order I would try them:

Split the deterministic battery onto its own page and evict one doc to pay for it. The battery is a deployment gate rather than a grading surface, so it is the cleanest seam on that page.
Cut the battery section down to the invariant argument plus a pointer at agents/deep/packs/evaluation.yaml, whose header comments already carry the check kinds. Frees roughly the right amount without a new page.
Raise the large-band caps in check_documentation_layout.py, which changes every catalog repo. Least attractive: the cap is deliberate back-pressure.

Blocked on a preference, not on work.

Dropped in #957 and not restored, because it does not fit. The passage from the old board page: > The board ships as a two-clause pilot, and the first has a documented real-world regression behind it, which gives the board a validity check the sibling suite never had: if its out half does not reproduce issue 88 against the pre-fix bundle, it is not measuring. The remaining five wait on the first graded result, because in the sibling suite the generator's predictions about which cases would discriminate ran one for three. That is the strongest argument on the board, and it exists nowhere else: it says how you would know the board measures anything at all. **Why it does not fit.** `docs/sirens-echo-eval.md` is at 119 of 120 lines and 7983 of 8000 chars for the large band, and `docs/*.md` is at the 40-doc cap. `documentation-layout` has no per-file size override, only `agents_md_max_lines` and the README pair, so the page cannot simply grow. Restoring the passage costs about 620 chars of something else. Options, in the order I would try them: 1. Split the deterministic battery onto its own page and evict one doc to pay for it. The battery is a deployment gate rather than a grading surface, so it is the cleanest seam on that page. 2. Cut the battery section down to the invariant argument plus a pointer at `agents/deep/packs/evaluation.yaml`, whose header comments already carry the check kinds. Frees roughly the right amount without a new page. 3. Raise the large-band caps in `check_documentation_layout.py`, which changes every catalog repo. Least attractive: the cap is deliberate back-pressure. Blocked on a preference, not on work.

coilyco-ops commented

2026-08-18 16:59:39 +00:00

Author

Member

Closing as moot rather than fixed. The passage was stale, so restoring it would have put a wrong number back on the page.

It said "two-clause pilot" with "the remaining five" waiting. agents/deep/packs/board.yaml now holds five clauses and ten cases: no-invented-surface, trusted-principal, no-instruction-disclosure, refusal-states-no-value, earn-the-reply. The pack moved and the doc did not.

What was worth keeping is the argument, not the count, and #957 carries it: no-invented-surface reproduces issue 88, which gives the board a validity check, and if its out half does not reproduce that against the pre-fix bundle then the board is not measuring. The rest wait on the first graded result because the generator's predictions are not yet a measured quantity.

The room appeared because the same audit found three more stale claims to drop, all of them restating the battery pack instead of pointing at it: five assumed validators where there are three, max_verbatim_words as live where #406 retired it, and forbid_principal_echo covering the handle where issue 309 made it user-ID only. The pack header is authoritative and the page now points at it.

No cap change and no eviction. docs/sirens-echo-eval.md sits at 111 of 120 lines with docs/*.md still at 40.

Closing as moot rather than fixed. The passage was stale, so restoring it would have put a wrong number back on the page. It said "two-clause pilot" with "the remaining five" waiting. `agents/deep/packs/board.yaml` now holds **five clauses and ten cases**: `no-invented-surface`, `trusted-principal`, `no-instruction-disclosure`, `refusal-states-no-value`, `earn-the-reply`. The pack moved and the doc did not. What was worth keeping is the argument, not the count, and #957 carries it: `no-invented-surface` reproduces issue 88, which gives the board a validity check, and if its out half does not reproduce that against the pre-fix bundle then the board is not measuring. The rest wait on the first graded result because the generator's predictions are not yet a measured quantity. The room appeared because the same audit found three more stale claims to drop, all of them restating the battery pack instead of pointing at it: five assumed validators where there are three, `max_verbatim_words` as live where #406 retired it, and `forbid_principal_echo` covering the handle where issue 309 made it user-ID only. The pack header is authoritative and the page now points at it. No cap change and no eviction. `docs/sirens-echo-eval.md` sits at 111 of 120 lines with `docs/*.md` still at 40.

coilyco-ops closed this issue

2026-08-18 17:00:26 +00:00

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

coilyco-gaming/sirens-echo#962

No description provided.

Rows
Columns