Cross Lens Converge

---
name: cross-lens-converge
description: 'Use when one reviewer is not enough because failure modes are heterogeneous, or a claim needs cross-lens pressure before it ships. Returns a convergence picture with per-lens verdicts and the single resolving question. Not for collapsing a decision field — use converge.'
---

# Cross lens converge

## Contract

| Field | Bound contract |
|---|---|
| Trigger | One reviewer isn't enough because the failure modes are heterogeneous, or a claim looks strong to its author and needs cross-lens pressure before it ships |
| Authority | Read-only. No file, VCS, credential, paid, published, deployed, or remote mutation |
| Side effect | None (read-only analysis); returns a convergence picture and the single resolving question |
| Done | The result reads as a picture not a checklist: a reader sees which lenses agreed, which disagreed, and the one question that matters, and can act on it without re-running the analysis |

## Inputs

The claim or artifact under pressure must be supplied. Candidate failure modes may be supplied or derived from the artifact. When supplied, they are trusted input and must not be re-derived.

## Procedure

1. Receive the artifact and the candidate failure modes. If the failure modes are not supplied, enumerate them from the artifact. Done when: the artifact is received and failure modes are supplied or enumerated.
2. Collapse candidates that describe the same failure mode into one lens. Assign exactly one lens to each genuinely distinct failure mode that remains. Done when: each distinct failure mode has exactly one lens, with duplicates collapsed.
3. Give each lens exactly one load-bearing reason it would reject or weaken the claim. A lens with no load-bearing reason is dropped before running, not coerced into agreement. Done when: every retained lens has one load-bearing reason; reasonless lenses are dropped.
4. Run every lens independently against the artifact. Record, per lens, its verdict as one of agrees, disagrees, or cannot-decide, and the single reason that drives that verdict. Done when: every retained lens has a verdict and its driving reason recorded.
5. Do not average, merge, or majority-count the verdicts. Disagreement is the product; preserve which lenses agreed and which disagreed. Done when: verdicts are preserved per-lens without averaging or majority-counting.
6. From the disagreed and cannot-decide lenses, identify the one question whose answer would resolve the most lens disagreements. If no single question resolves more than one, name the question that resolves the highest-stakes disagreement. If every lens agrees, the resolving question is none. Done when: the single resolving question is identified, or none is set when every lens agrees.
7. Return the convergence picture: per-lens verdict and load-bearing reason, the agreed set, the disagreed set, and the single resolving question. Done when: the convergence picture is returned with per-lens verdicts, agreed set, disagreed set, and resolving question.

## Failure and recovery
- Duplicate failure modes: merge the lenses before running. Never run two lenses over the same failure mode and present their agreement as independent confirmation.
- No disagreement: return the picture with the resolving question set to none; do not manufacture disagreement to look thorough.
- A lens cannot decide: record cannot-decide with its reason. Do not coerce it into agrees or disagrees.
- A lens does not complete: return the partial picture, name the lens that did not complete, and mark its verdict cannot-decide. This skill reads only; on any error, it returns the partial picture and never edits the artifact or any file.

## Output
A convergence picture: each lens's verdict and its one load-bearing reason, the agreed set, the disagreed set, and the single resolving question. Read-only; no state transition.