validation-methodology

name	validation-methodology
description	Reusable validation and review methodology for staged validation, failure triage, and prosecution-depth setup. Use when running validation ladders, triaging failures, or executing adversarial review passes. DO NOT USE FOR: CE Gate orchestration, specialist dispatch ownership, or step execution flow (keep those in Code-Conductor).

Validation Methodology

Reusable entryway for staged validation and extracted review-reconciliation references.

When to Use

• When an agent needs a graduated validation sequence from cheap automated checks to independent review
• When validation failures need a consistent triage rule before re-delegation
• When a review cycle needs the shared review-reconciliation contract without re-reading Code-Conductor
• When an agent needs reusable review heuristics but must keep specialist routing, express-lane ownership, and CE Gate ownership in the agent

Purpose

Run validation in increasing-cost tiers, classify failures before rework, and point callers at the canonical review-reconciliation references. The owning agent still decides who speaks next, when CE Gate runs, and how accepted fixes are routed.

Validation Ladder

Validation runs in this graduated 4-tier order:

1. Tier 1 — Build & Validate: run quick-validate commands, lint/typecheck, and the full test suite together when practical so all failures are visible before fixing. Prefer lint/typecheck before tests when the project supports it. For migration-type issues, also run the migration completeness scan required by the owning workflow. Projects with slow test suites can override in copilot-instructions.md by adding <!-- slow-test-suite: true -->, which splits Tier 1 into: (a) quick-validate, lint/typecheck, and targeted tests, then (b) the full test suite.
2. Tier 2 — Structural validation: run project architecture or structural validation commands defined by the repository.
3. Tier 3 — Strength validation: run configured coverage, robustness, or deeper confidence checks defined by the repository.
4. Tier 4 — Independent review lane: load references/review-reconciliation.md after the lower tiers pass. The owning agent keeps any CE Gate orchestration, post-fix routing, and specialist-dispatch authority that follows this tier.

Do not skip ahead when an earlier tier fails. Resolve failures at the current tier before advancing.

Failure Triage Rule

When a validation tier fails, classify first, then route:

• code defect → route to the implementation specialist with failing evidence
• test defect → route to the testing specialist with failure analysis
• harness/env defect → route to the responsible tooling or environment path
• rc-divergence → run one dedicated correction cycle outside the main convergence budget: dispatch implementation first with the divergent acceptance-criteria items, re-run Tier 1, re-evaluate all acceptance-criteria items, then dispatch testing only if divergence persists and the assertions must be re-derived from the Requirement Contract rather than the corrected implementation

Always include the failure evidence, attempted diagnosis, and the next action in the handoff prompt. Avoid blind retries.

Composite References

• references/review-reconciliation.md: pre-review gate, prosecution-depth setup and use, fixed 3-pass review mechanics, non-GitHub review mode, improvement-first rule, AC cross-check, R4 batch dispatch, and the R2 post-fix review loop
• references/review-state-persistence.md: review-state persistence shape, lookup precedence, completion gate, and resume evidence rules
• references/post-judgment-routing.md: post-judgment prosecution-depth re-activation detection and the paired routing index
• ../code-review-intake/references/express-lane.md: canonical R6 express-lane gate, scope restriction, and Tier 1 re-validation requirement

Review Heuristics At A Glance

• Complete the full prosecution to defense to judgment cycle before fix routing.
• Avoid blind retries; carry forward the failure evidence and current hypothesis with each re-delegation.
• Preserve the change-type classification and any prosecution-depth exclusions in every prosecution prompt.
• Keep merged-ledger deduplication evidence-based: same evidence at the same file and line collapses; materially different evidence remains additive.

Boundaries

This skill does not decide:

• which agent speaks next
• when a step begins or ends
• how accepted findings are routed to specialists
• whether express-lane routing is available or how it dispatches
• when CE Gate orchestration or PR creation runs
• when review-mode entry, post-fix prosecution entry, or post-judgment follow-up occurs
• whether durable handoff state is written
• whether calibration or other write-back side effects occur

Keep those decisions in the owning agent.

Gotchas

Trigger	Gotcha	Fix
Copying specialist routing, CE Gate ownership, or step-order decisions into this skill	The skill stops being reusable methodology and starts stealing orchestration authority from the agent	Keep routing, step execution flow, durable handoff ownership, and CE Gate orchestration in the owning agent; move only reusable validation and review method text here