# Polymyth Coherence V5.1.2 application protocol

Use this protocol with `Polymyth_Coherence_Assessment_Instrument_V5.1.2.xlsx` to begin a new, evidence-led assessment. The workbook is blank: it contains `TEMPLATE-001`, forty unchecked template QA rows, formulas, research references, and synthetic regression tests, but no completed case audit and no expected case verdict.

This protocol is model-neutral. An AI may assist with a case, but it cannot autonomously certify the result. Automated output alone cannot establish a repair ruling, a Deep verdict, or Full qualification.

## Required material

Before assessment begins, provide:

- the exact work, version, cut, edition, or released artifact;
- the declared corpus and canon, including exclusions;
- the medium and organizing model;
- the intended claim and audit depth;
- primary material or exact primary locators;
- relevant tradition-specific evidential and relational conditions;
- permissions where they are constitutive;
- required adaptations and any remainder the workbook cannot responsibly carry.

Do not invent missing primary material, evidence, permissions, reviewer identities, or locators.

## File boundary

Treat these workbook sheets as assessment criteria, governance, research reference, or instrument tests. Do not rewrite them to make a case pass:

- `START HERE`
- `GOVERNING RULES`
- `INTERNAL`
- `MEZO`
- `EXTERNAL`
- `CROSS-LAYER`
- `BOUNDARY CASES`
- `DEBATES`
- `SOURCES`
- `RULE-SOURCE MAP`
- `COVERAGE`
- `CODEBOOK`
- `CHANGELOG`
- `AUDIT PROTOCOL`
- `QA MASTER`
- `REGRESSION TESTS`
- `V5 INTEGRITY`

`BOUNDARY CASES`, `DEBATES`, and `REGRESSION TESTS` explain or test the method. They are not answers for the new case.

Enter the new case only in the application sheets:

- `AUDIT INDEX`
- `SCENE-EVENT LEDGER`
- `STATE-TIME-SPACE`
- `RULE-CAPABILITY`
- `KNOWLEDGE-DECISION`
- `CAUSAL-TRANSITIONS`
- `FINDINGS-REPAIRS`
- `RED-TEAM QA`
- `MEZO RELATIONS`
- `EXTERNAL CLAIMS`
- `SCOPE-CORPUS`
- `PRODUCTION DEFECTS`
- `ORGANIZING FORM`
- `TRUTH-DIRECTION`
- `SECOND-ORDER`
- `INTERPRETATIONS`
- `QA EVIDENCE`
- `SALIENCE PROFILE`

Preserve formula cells, validation lists, table headers, QA IDs, rule IDs, regression fixtures, and the workbook structure. Write only to the intended case-input fields. If the AI cannot edit the workbook safely, return proposed entries under the exact sheet and column headings for a person to review and enter.

Never use `Polymyth_Coherence_Worked_Applications.xlsx` or a prior case result as an answer key. A prior application may be compared only when that comparison is itself declared, bounded, and evidenced.

## Assessment sequence

### 1. Decide protocol fit before auditing

Record one P0 classification:

- `Applicable`
- `Applicable with Adaptation`
- `Outside Protocol Scope`

For `Applicable with Adaptation`, state every adaptation and the remainder it cannot carry. `Outside Protocol Scope` is neither Fail nor Underdetermined and receives no Polymyth verdict. Analysis may continue only as a clearly bounded external account.

### 2. Declare depth and intended claim

Use the P23 boundary:

- Screening may locate candidates, possible defects, and research needs. It cannot certify absence, Full qualification, clean-list inclusion, or a population-wide claim.
- `Standard` may support a broader bounded analysis, but it remains noncertifying and cannot establish Full or clean claims.
- `Deep` is required for Full or clean claims. A smaller-scope result never propagates automatically to a larger work, series, tradition, or population.

### 3. Create a new case

In `AUDIT INDEX` and `SCOPE-CORPUS`, replace or copy the template with a unique Case ID. Declare the exact artifact, version, medium, corpus, exclusions, organizing model, depth, auditor, independent reviewer, dates, and intended qualified conclusion.

Do not present `TEMPLATE-001` as a real case.

### 4. Reconstruct the primary artifact

Use exact primary locators. Complete the applicable scene/event, state, capability, knowledge/decision, causal, noncausal organizing-form, and production ledgers. Record gaps and uncertainty instead of filling them with plausibility.

### 5. Record findings and repairs

Enter every candidate defect in `FINDINGS-REPAIRS` or `PRODUCTION DEFECTS`. Keep hard consistency and substantive coherence distinct. Grade the strongest proposed repair by the workbook's R0–R5/X ladder. A merely possible repair cannot defeat a finding.

### 6. Audit all three layers

- Internal: test the bounded released artifact on its own commitments. Internal is the absolute gate.
- Mezo: record at least one evidenced, typed, substantive relation and every declared load-bearing relation.
- External: test every substantive outside-frame claim with claim-appropriate evidence.

Mezo, External, reception, salience, truth-direction, or second-order meaning cannot repair an Internal failure.

### 7. Complete QA and independent review

Answer Q-001 through Q-040 exactly once for the case in `RED-TEAM QA`, with linked evidence. A genuinely different independent reviewer must reconstruct the material evidence, test the strongest allegation and repair, and rule by evidence. The auditor cannot serve as their own independent reviewer.

Unresolved material disagreement remains visible and blocks a clean result.

### 8. Report only the result the evidence supports

Report:

- Case ID and exact scope;
- protocol-fit classification and adaptations;
- audit depth and intended claim;
- primary evidence coverage and limitations;
- Internal, Mezo, and External results separately;
- unresolved findings, reviewer disagreements, and blocker codes;
- the qualified, scope-bound conclusion;
- salience and truth-direction only as separate post-coherence judgments.

Do not convert the method into a composite numerical score.

## Automation boundary

Under G-052, an AI may:

- extract and normalize locators;
- propose ledger entries;
- perform deterministic completeness checks;
- flag candidate contradictions, gaps, or missing links;
- organize evidence for human review.

An AI acting alone may not:

- establish a repair ruling;
- issue a Deep verdict;
- grant Full qualification;
- fabricate an independent reviewer;
- treat its own second pass as independent review;
- infer a desired answer from methodological examples or prior applications.

A human adjudicator must verify the evidence and make the required qualitative rulings. A genuinely different independent reviewer remains mandatory. Every automated task requires task-specific validation.

## Full-qualification gate

Full qualification remains blocked unless all of the following are present:

- protocol fit is declared and permits the assessment;
- audit depth is Deep;
- primary material is obtained and coverage is complete for the declared scope;
- `V5 INTEGRITY` returns `PASS`;
- Q-001 through Q-040 are complete and valid;
- the auditor and independent reviewer are different people;
- a qualified conclusion is present;
- Internal passes;
- at least one substantive Mezo relation passes and no declared load-bearing relation fails;
- at least one substantive External claim is supported and no applicable External blocker remains;
- no unresolved material finding, evidence gap, production contradiction, or invalid link remains.

If a gate is missing, report the blocker. Do not silently upgrade Screening, partial coverage, or automated proposals into certification.

## Release identity

- Instrument version: `V5.1.2`
- Instrument SHA-256: `0624c76e0ae1351dcfe6a9d2cabf6e4e581820b63a0fc40e5ad755bbd1d93550`
- Formula count: 85,768
- Formula signature: `94a9cd749e53b71c056c2a41c68d27cdb9a344eaf9bb3865b37a27f2b58684be`
- Completed case audits included: 0