[MaxGhenis]

Docs-only follow-up to #566. It fixes the one minor finding and four nits that delta check r10 left in docs/design/open_population_evidence/op1_report.md when it approved #566 at 88b5df28. No code, test or tier file changes.

Head 961f520a: 7640fe1c makes the fixes. 5c307826 fixes six phrases that independent review docs-r1 found wrong or loose in 7640fe1c's new text, and 961f520a takes its delta check's two optional nits (see below).

What changes in the report

• r10's minor finding. The r9 mutation table's r8-harness row gave one run's property-test gains as a fixed result, though they depend on Hypothesis's draws. The row now separates them:
  • ten mutants gain the property test in each of r10's 40 seeded runs: K1, K3 to K7, S9, S12, S13 and X9;
  • five gain it only on some draws (K2, K8, S14, S15 and S16: 29, 22, 14, 25 and 20 of 40);
  • the author's run is stated as one run.
• Nit: draw-dependent cells. The r5 and r6 tables give "7 or 8" and "54 or 55" for the length bound removed (A2) and r3's A12 (A4), where r10's run gave 8 and 55. The r9 table's r6 row says eight counts have varied, each a range in the r6 table, and names the four in the row of five.
• Nit: the git diff --quiet check. §3 now says the command prints nothing either way, so its exit status is the check. Before, the sentence could be read as exiting 0 when a file differs.
• Nit: wording. §3 says the tier-slot recount, not the merge, names d8302233, and says test modules rather than test files.
• Nit: timing. Both of §3's timing statements give 8 to 20 s over the author's and reviews r8 to r10's runs, varying with host load. r10's OP1 command took 20.06 s at a one-minute load of 98.
• New review r10 section. It gives the verdict and the findings table. It also gives the 11 new streams.py mutants r10 tried: five were killed, and the six survivors fall in r9's known classes, which the exact comparison at streams.py:108 rules out.
• §5. It states that OP1's exit and harness tests import baselines/asfr.py through the baselines package, as other test modules on master do, so runs load it without displaying it.

Checks

• git diff --check is clean, and the only file changed is op1_report.md.
• I checked each new figure against r10's review. Cited lines test_exits.py:11, test_harness.py:20–21 and streams.py:108 were checked at 444f970d.
• No test names the report. Tests that scan docs/ Markdown run in CI.

Review docs-r1 (Opus 5.5, at 7640fe1c): REQUEST_CHANGES, fixed in 5c307826

It found scope clean, all five r10 items fixed as asked, and the new figures exact against r10's seeded runs and the author's run. It asked for six phrase-level fixes:

• K30 "never" killed: it was killed in none of r10's 20 seeded runs. docs-r1 found a two-step walk inside the strategy that kills it.
• "Four counts vary": eight counts have varied, and the r6 table gives each as a range.
• K25 and S23: S23 differs only for seeds of 2^1203 or more; K25 only for two seeds a multiple of 2^1203 apart, which within the strategy's range means exactly 2^1203.
• The weakening sentence: the flips and steps past bit 64 carry the test, and the swap result comes from the exact-int test's variants.
• "Every count" means every count r10's brief asked for.
• The findings row for r10's fourth nit described the problem the wrong way round.

docs-r1's delta check (at 5c307826): APPROVE WITH NITS. All six fixes hold. Its two optional nits are taken in 961f520a: K25's general case ("a multiple of 2^1203"; this body had said "exactly"), and naming the deterministic test behind the swap result.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
social-security-model	Ready	Preview	Oct 26, 2026 12:51am UTC

[MaxGhenis]

Landing audit (Microcosm Dynamics coordinator, 2026-10-26 03:59Z).

Merged at the reviewed head 961f520a02f0d7102763e0555ae4e070bf3f9df6 with --match-head-commit, as merge commit 0f69bbb2. This is #566's docs-only follow-up. It changes only docs/design/open_population_evidence/op1_report.md, and no tier or test file, so it did not need the tier-count slot.

Gates checked at 03:57Z:

• gh pr checks 589 exited 0. Tests run 40251263402 succeeded at 961f520a in every job.
• MERGEABLE, merge state CLEAN, not a draft, no changes-requested review.
• Three commits on master 444f970d, which was still master at merge time: 34 lines added and 13 removed, in the one file.

Review: one independent Claude Opus 5.5 check (docs-r1), run in-session because no Subfleet lane was open.

• It asked for six phrase-level fixes at 7640fe1c, nothing blocking.
• It approved 5c307826 with two optional nits.
• It approved 961f520a after both were taken.

It checked each changed claim against review r10's evidence. This PR closes r10's one minor finding and four nits on #566.