# Result Lens Methodology

Status: current v0.74 methodology note. This document explains result categories only. It does not change evidence rows, Bayes factors, priors, dependency logic, hypothesis IDs, evidence IDs, source trails, or stage flow.

For the broader strict-logic defense of hypotheses, evidence rows, Bayes factors, dependency caps, transfer weights, confidence ceilings, and live objections, see `docs/methodology_strict_logic_defense.md`.

## Why The Layers Must Stay Separate

The Signal is not trying to make the smallest number sound humble or the largest number sound impressive. Both habits can mislead. A lower result is not automatically more fair. A higher result is not automatically inflated. Fairness means each constraint must correspond to a real uncertainty, dependency, source limitation, or rival pressure.

That is why the result labels matter. They tell the reader which constraints have been applied and why.

## Six Distinctions

1. Row-level Bayes factors are the local likelihood judgments attached to individual evidence rows. They ask how a particular datum leans under named hypotheses. They are not final conclusions.

2. Dependency and cluster caps are anti-double-counting controls. They keep closely related rows from being counted as though each were wholly independent. These caps are justified when rows share sources, mechanisms, family structure, or explanatory overlap.

3. Stage-transfer weights govern how earlier stage evidence conditions later stages. They do not copy a whole posterior forward as fresh evidence. They let upstream pressure matter while keeping later claims responsible for their own direct evidence.

4. Audit and guardrail confidence ceilings are conservative stress-test controls. They are not row-level dependency caps. If a ceiling is retained, it must be labeled as a guardrail or sensitivity unless a separate argument justifies treating it as the main fair calculation.

5. Sensitivity diagnostics preserve alternate readings for audit comparison. A sensitivity result is useful because it shows how much a choice matters. It is not automatically the official result.

6. Full-strength diagnostic views show what happens when the staged model is allowed to carry full cumulative force under stated assumptions. They are upper-bound stress views, not proof, certainty, salvation, or the official audit result.

## Current v0.74 Result Outputs

- Strict Audit Calculation: the primary hostile/fair audit lens. It uses skeptical priors, rival-positive pressure, dependency and cluster caps, source limitations, and no arbitrary global confidence ceiling.
- Audit Guardrail 75% Ceiling Stress-Test: the former 75% capped audit output retained visibly as a conservative stress-test. It is useful for inspection, but it is not the main fair audit posterior.
- Guided Staged Path Result: the fair intended staged calculation. It uses dependency caps, transfer weights, and rival pressure, with no final confidence ceiling.
- Guided 90% Carry-Cap Sensitivity: the older guided result with a 90% carried-prior ceiling retained for comparison.
- Full Signal Diagnostic / Uncapped Stage-5 View: full-strength upper-bound stress view.

## Audit Cap Semantics

The 75% audit ceiling is not a dependency cap. It is a global conditioned-prior ceiling applied after dependency-adjusted upstream signals have already been generated and before direct downstream evidence is added. In the current release it suppresses the Resurrection conditioned prior from the raw audit-conditioned value to 75%.

That does not make the ceiling dishonest. It does mean it should be named honestly. It is a conservative audit guardrail, not the main strict audit calculation and not the fair staged path.

## Guided Path Semantics

Guided Staged Path should represent the intended staged calculation: dependency-capped evidence, governed stage-transfer weights, and rival pressure preserved. It should not add an arbitrary final confidence ceiling unless that ceiling is separately justified.

The old 90% Stage-5 carried-prior cap is therefore retained as a sensitivity result, not as the main Guided Staged Path.

## Full Diagnostic Semantics

The Full Signal Diagnostic is different from "Guided Path without a cap." It also removes the guided transfer-weight discipline and lets each stage carry at full stage strength. That is why it belongs as an upper-bound stress view. It is useful, but it is not the strict audit result, the 75% guardrail stress-test, or the guided staged path.

## Audit Rule

When challenging a result, name the layer being challenged:

- the row-level Bayes factor,
- the dependency or cluster cap,
- the stage-transfer weight,
- the audit ceiling,
- a sensitivity diagnostic,
- or the full-strength diagnostic view.

Then name the exact row, cluster, source limitation, rival pressure, or stage assumption that justifies the change. If no such reason can be named, the verdict should remain: unclear pending named row or cluster audit.