[realmarcin]

ProSMART was recommended for a Ramachandran-distribution Z-score that its documented comparative scores do not establish. This PR adds dated recommendation and assumption successors, corrects its score definitions and comparison modes, and preserves predecessor records and frozen EVAL/QDS evidence.

The new X-ray before/after workflow and 1SAR case study separate implementation agreement, final-model evidence, and change from the actual input. Fixed source bundles are admitted by verified bytes before conclusions are drawn. Coordinate, inventory and reflection-retention arithmetic is recomputed; historical oracle observations retain their provenance limits. The T01–T17 tools guide distinguishes documented capabilities, local implementations, retained evidence and applicable interpretation, including the limits of free superposition for placement validation.

Fresh review and fixes:

• An independent fresh-context review covered the full diff from 6bd461b5e9f005646431cd373e97ffad3aa5d80d to e8769e6366736196a42909445f0beeffae6ef598. It produced new issues P2: distinguish Gemmi superposition from experimental-frame placement validation #950/P2: bind fixed control conclusions to their admitted source bytes #951 and a duplicate P2: 1SAR case study invents Servalcat free-flag and R-factor mechanisms #934 denominator finding; all were triaged in scope and fixed.
• Commit 2f52b15 corrects placement guidance, binds the three retained control sources, adds regressions, and qualifies the historical reflection count. Fresh fix-diff review and PRIMARY verification caught two prose residuals, recorded under P2: distinguish Gemmi superposition from experimental-frame placement validation #950/P2: correct unsupported task 001 review evidence before PR acceptance #926 and corrected in d039ebd.
• A separate fresh read-only review of that final correction found no actionable findings. No in-scope finding remains unresolved. Across the complete PR review record, 32 finding issues are accepted and fixed; Review disposition: alleged ProSMART historical T05 dependency omission #923–Review disposition: dated backlog stamp is not a false current claim #925 were rejected and closed as not planned.

Validation at final head d039ebd8786e7b90d1e76197a418ce352fec6548:

• PRIMARY ran uv run --locked --extra benchmark -- bash scripts/validate.sh on Python 3.12.11: exit 0 in 124.87 seconds, with a clean, unchanged head. The preceding code-fix head also passed PRIMARY's full gate and all 26 focused tests passed in independent review.
• PRIMARY independently verified exact-copy acceptance, nine altered-copy rejections, three altered-default-buffer rejections, one byte read per control source, and the meaningful old-baseline/empty-input defect. Default serialized output remains byte-for-byte identical (201,939 bytes; SHA-256 db100e926bfcf6331c303087846c4d51117a4411c6afdf58877aada5dbf14245).
• All frozen scientific inputs are preserved. Earlier verification of the 185 predecessor recommendations and 32 predecessor assumptions carries forward; the fresh fix commits change none of those files. Historical oracle scalars were source-checked, not remeasured with licensed tools.
• Linux and macOS CI: both passed on this exact head. Final read-only review is complete, all in-scope findings are fixed, and CLEAN merge status was verified before the user-authorized squash merge.

Closes #911, closes #917, closes #918, closes #919, closes #920, closes #921, closes #922, closes #926, closes #927, closes #928, closes #929, closes #930, closes #931, closes #932, closes #933, closes #934, closes #935, closes #936, closes #937, closes #938, closes #939, closes #940, closes #941, closes #942, closes #943, closes #944, closes #945, closes #946, closes #947, closes #948, closes #949, closes #950, closes #951.

Existing follow-ups #760, #765, #790, #799, #872, #873 and #876 remain open and outside this PR. No new follow-up was added by the fresh review.

The dated audit checkpoints preserve review history; the final review and CI state are recorded here.

[realmarcin]

Final acceptance at e8769e6366736196a42909445f0beeffae6ef598: PRIMARY verified the accepted fixes and exact committed side effects, the fresh read-only re-review, and same-head validation. No unresolved in-scope product finding remains. The PR is unmerged; merge remains the user's decision.

The required locked Python 3.12.11 benchmark gate passed in 130.76 seconds on a clean, unchanged head. Linux and macOS CI both passed: https://github.com/CultureBotAI/protstruct_review/actions/runs/41915447226. The final #949 matrix and #934 reflection-audit prose repairs are verified.

All 33 findings are recorded: 30 fixed in this PR (#917–922 and #926–949), and #923/#924/#925 rejected and closed not planned. The fixed issues remain open until merge; the PR description has the 31 closing references including parent #911. Existing follow-ups #760/#765/#790/#799/#872/#873/#876 remain open and outside this PR.

Review scope is explicit: newly inspected source is distinguished from reviews carried forward on unchanged blobs. PRIMARY checked the final review's actual read/command receipts. Task009's false source/execution claims and the failed scratch suite are excluded from acceptance evidence. Current matrix applicability was checked directly; committed wrong-task matrix-mutation coverage remains a documented limitation.

The secondary is idle. PRIMARY remains on main, the original untracked handoff is uncommitted, and no orchestration files were committed.

[realmarcin]

Fresh-context adversarial review completed at e8769e6366736196a42909445f0beeffae6ef598, against 6bd461b5e9f005646431cd373e97ffad3aa5d80d. The reviewer received no prior reports, fix history or issue list; all15 PR files were covered. PRIMARY independently verified its three findings before filing or dispatching fixes.

• P2: distinguish Gemmi superposition from experimental-frame placement validation #950 — P2, in scope: the new guide promotes free Gemmi superposition into experimental-frame placement validation. A20Å rigid translation is erased by the fitted RMSD. Correct the capability/integration boundary.
• P2: bind fixed control conclusions to their admitted source bytes #951 — P2, in scope: the new control reader accepts changed inputs but retains fixed scientific conclusions. Empty inputs produce zero-target cohorts with22-target prose. Bind fixed conclusions to admitted source bytes and add a meaningful altered-input regression.
• P2: 1SAR case study invents Servalcat free-flag and R-factor mechanisms #934 — P3 duplicate residual, in scope: case91 attaches723 to a historical PHENIX evaluator whose effective selection is not retained. Preserve the count as archive/header context.

No new follow-up was triaged from this review; no P0/P1 or user-scientific-judgment blocker was found. Existing follow-ups remain separate. Fixes, PRIMARY QC, a fresh fix-diff review and final-head Linux/macOS SUCCESS plus CLEAN are required before the specifically authorized squash merge. No merge has occurred.

[realmarcin]

Fix commit 2f52b15e60be0161811a1a07e71ee43ef0a9e972 is local and not pushed yet. PRIMARY verified #951 source admission with nine altered-copy probes, three altered-default-buffer probes, exact-copy equivalence and one byte read per source. Old baseline succeeds before reproducing the empty-control defect; the new default generated assessment remains byte-for-byte identical. The #934 denominator wording now preserves the historical evaluator-selection gap.

Two in-scope P3 prose residuals are recorded on existing issues: #950 option names and #926 audit chronology/status. No new finding issue was needed. A separate fresh-context read-only review of the five-file fix diff and PRIMARY's locked QC are running; final push, CI and merge await acceptance.

[realmarcin]

Final acceptance at d039ebd8786e7b90d1e76197a418ce352fec6548: no unresolved in-scope finding. Independent full-diff review and fresh fix-diff re-reviews are complete. PRIMARY verified the source-admission counterexamples, unchanged default output, retained evidence and exact final prose corrections. Locked Python 3.12 QC passed on the clean final head.

Both validate (ubuntu-latest) and validate (macos-latest) are SUCCESS in run41990664847; GitHub reports CLEAN. The user explicitly authorized PR916-only squash merge after these gates. Proceeding with --squash --match-head-commit d039ebd8786e7b90d1e76197a418ce352fec6548, followed by squash-tree equality, exact issue-closure checks and authorized branch/worktree cleanup. The seven listed follow-ups remain open.

[realmarcin]

PR #916 is squash-merged as 0320d1a2b95301076e59056d2ab390a5a844a305. Final reviewed head d039ebd8786e7b90d1e76197a418ce352fec6548 and the squash have the same tree: 50ae34b59fa2935fac3e8abad18bceede93415d5.

Both final-head validate checks passed in run41990664847, and merge status was CLEAN immediately before the head-pinned squash. PRIMARY also ran the locked Python 3.12 gate on main after fast-forwarding: exit 0 in 129.98 seconds, stable clean tracked state.

Fresh review filed and fixed #950/#951, with deduplicated residuals addressed on #934/#926. All 33 intended issues are verified CLOSED/COMPLETED: #911, #917–#922 and #926–#951. GitHub completed every closure automatically; no manual closure was needed. Rejected #923–#925 remain CLOSED/NOT_PLANNED. Follow-ups #760, #765, #790, #799, #872, #873 and #876 remain OPEN; the complete remaining queue is seven issues and zero PRs.

The clean, fully pushed protstruct_review-wt-001 was removed after tree verification. Local and remote fix-911-prosmart-rama-z branches are deleted and independently verified absent. Main equals the squash/origin main. The untracked handoff prompt, excluded orchestration records and all three other-session scratch-worktree records were preserved. The secondary is idle. No additional project was started.