# Reading the frozen Aligners workbook

Edition **5.9**; current correction build `MG-RL-13.0-20260926-RELEASE-I`. The workbook remains a frozen worked-run example, not a general operational selector. Thirty-three existing string cells (17 primary edits and 16 matching snapshot records) and two print-area scope attributes changed. No formula, numeric input, cached formula result, sheet order, style or geometry changed. The manifest binds the corrected bytes.

## Point difference, normalized comparison and final verdict
`RLS!A35` labels the raw difference as `ΔRLS (raw)`. `Config!C22` names the normalized SignedGap criterion. `RLS!F67` identifies only a nominal demonstration comparison; it does not declare final decisive selection. The final posture is still `REFUSE_DETERMINISTIC_SELECTION`, with authority choice separately recorded. Required uncertainty/dependence variants cannot be ignored.

## Assessed zero is not missing data
`Start_Here!B7` explicitly stipulates synthetic assessed zeros. A has seven input-bearing cells and B six, leaving 42 and 43 no-row cells respectively. None of those stipulated fixture cells is silently missing. All 49 active cells retain the declared demonstration uncertainty. A zero point impact does not imply zero uncertainty. The exact-zero convention clarified in Canon Appendix P/R.19 concerns their separate Method B synthetic proxies, not this workbook's uncertainty basis. Do not inject new phantom assumptions into a replay of this fixture.

## Runtime records and detectors
`Audit_Flags` contains frozen worked records. Rows 29–31 are Agent-owned runtime flags: `RUNTIME_BLOCK`, `RUNTIME_BLOCK`, `RUNTIME_INVALID`. These are not additions to the PCC severity vocabulary. The explanatory note at A39 makes ownership explicit. Actual input/integrity detectors are elsewhere in the workbook and in the external verifier.

## Two rights orderings
`Parameters!A3:B12` is an anchor/presentation order, not emergency priority. The emergency priority is in `CANON!D29:D37`, read with the rights labels in column A and Canon §7.5.1: LIFE, BODY, ECOL, LBTY, NEED, DIGN, PROC, INFO. Neither existing display or threshold changed.

## Verification
Run `Reference/Workbook_Verifier.py` against the separately trusted manifest, and `Reference/full_formula_replay.py` for cache-independent calculation. Formula presence alone is not formula identity. The complete 20,212 formula and cache records have the same numerical fingerprint as the supplied baseline.

No native I recalculation/save/reopen was executed. G's named native receipt remains source-build evidence, not recertification of changed bytes. The full formula/cache fingerprint is unchanged and the current cache-independent replay is separate evidence. See the current Release_I_Verification report.

See Feedback_Clarifications.md for the independent zero-cell/epsilon explanation and the distinction between current typed snapshots and historical layout values.

## Current local-integrity boundary and errata

The `Build_Integrity!A2` instruction explicitly limits local guards to retained inputs/literals, selected formula presence and RLS reconciliation. Verdict-display replacement is not included in the summed guard; adding dependent displays there would create a circular check. `RLS!B40` can therefore be overwritten without that local banner changing. The external `Workbook_Verifier` rejects any byte difference, including this mutation, relative to the trusted release manifest.

`Sensitivity_Analysis!E13` is now `ΔRLS (raw)`, not normalized Gap. The seven previously unclassified emissions are labeled TOOL_LOCAL in their existing source-reference cells; dashboard counts group stored records by severity, not canonical registry implementation coverage. Print-area local scopes now point to Build_Snapshot and Build_Integrity. See the Release I adjudication and exact workbook record.