Skip to content

Guide §6.0: the gate skips a row only on its own written condition; a skip resolves only that row - #59

Merged
Carr1005 merged 4 commits into
mainfrom
guide-gate-skip
Oct 19, 2026
Merged

Carr1005 merged 4 commits into
mainfrom
guide-gate-skip

Conversation

@Carr1005

Copy link
Copy Markdown
Collaborator

Part 5 (issue #48, "rows that stop mattering"): the gate skips a row only when its own written "Applies only when" condition is false.

Approach

Instead of adding guards, this changes what "resolved" means. A row is resolved when it is answered, or skipped by its own written condition. Nothing else resolves a row. The existing every-row-resolved loop (step 4), the step-5 "unresolved → go back" check, and §12.11 therefore also block skipping by judgment and stopping after a skip.

What changes

  • Gate step 3: the only skip, a one-line note ("D4 is skipped: it applies only when D2 is D2a or D2b; D2 = D2c"), "a skip resolves that row only", and re-checking conditions when an earlier answer changes.
  • Steps 4–5: "resolved"; a new checklist mark not applicable (applies only when <row> is <id>; <row> = <answer>), which is invalid if the quoted condition is not false.
  • Rewritten in place: the §6.0 "Presents every row" bullet, §5.5 conditions, row order, the §0 comment, §12.11, §14.
  • §7 "Tests by option": an option lists the retired tests of the rows it makes moot.
  • Removed: the "Rows that stop mattering" column, checker check MOOT, and their §14/audit halves.
  • Lab runtime: in "Question N of M", a skipped row keeps its number. The overlay is re-pinned (4 quoted passages unchanged).

Tested (Opus 5.5)

Each probe's expected asked and skipped rows were computed from the spec, and a script compared them with the transcript (experiments/gate-skip/check_probe.py, local).

Probe Answers Result
A, agent-memory conversation-only ✅ skipped D2, D3, D4, RR3, C1; asked RR1 and RR2 between and after the skips
B, loop-eng lab no grader, then switched back ✅ skipped D4 and asked RR1; after the change, re-asked D2, then D4
C, loop-eng lab deterministic grader ✅ nothing skipped
D, agent-memory baseline ✅ no questions, every row marked
Regeneration Result
K1, K2 (da81a7d) checker 0
K3 (da81a7d) checker 0, but its no-grader line omitted the learning-loop tests; fixed by 18c4fbd
K4 (18c4fbd) checker 0; the no-grader line retires the learning-loop rules and tests

Not in this PR

When an option removes a model role, its table line doesn't say what happens to that role's "setting required" test. Seen 3 times; logged on #48.

🤖 Generated with Claude Code

Carr1005 and others added 4 commits October 19, 2026 19:36
…y when" condition; a skip resolves only that row

Changes what "resolved" means instead of adding guards: a row is resolved
when answered or skipped by its own condition; nothing else resolves a row.
The existing every-row-resolved loop, step-5 check and §12.11 therefore also
block skipping by judgment and stopping after a skip.

- Gate step 3: the one skip, a one-line note, "a skip resolves that row
  only", re-check conditions when an earlier answer changes. Steps 4, 5:
  "resolved"; new checklist mark `not applicable (…; <row> = <answer>)`,
  invalid if the quoted condition is not false.
- §6.0 bullet, §5.5, row order, §0 comment, §12.11, §14 rewritten in place.
- Removed: §7 "Rows that stop mattering" column, its checker check (MOOT),
  §14 and audit (k) halves — the conditions carry moot rows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…eeps its number; re-pin 'Written against' to base da81a7d (quoted passages unchanged, verified with grep -F)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…teria of the rows it makes moot (a skipped row has no line of its own)

Fixes a regression from da81a7d: "Rows an option makes moot are not listed
here" was read as "do not list their tests" (K3: the no-grader line kept the
learning-loop tests in force).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…uoted passages unchanged, verified with grep -F)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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