E2e validate

End-to-End Validation Orchestrator

Scope

Primary entry point for executing a deepest-plan artifact. Routes to platform-specific validation via detected platform. Applies to any project — detects stack and selects strategy automatically.

Iron Rule

If the real system doesn't work, FIX the real system. NEVER create mocks, stubs, test doubles, or test files. ALWAYS validate through the interfaces real users hit. ALWAYS capture evidence. ALWAYS review evidence. ALWAYS write verdicts.

Evidence you don't READ is evidence you don't HAVE.

See no-mocking-validation-gates for the full block list.

Platform Detection

First match wins.

Priority	Signals	Platform	Skill
1	.xcodeproj, Package.swift	ios	ios-validation
2	Cargo.toml [[bin]], go.mod + main.go, package.json "bin"	cli	cli-validation
3	Backend routes without frontend templates	api	api-validation
4	Frontend framework without backend routes	web	web-validation
5	Frontend AND backend in same project	fullstack	fullstack-validation
6	None of the above	generic	adaptive (use functional-validation directly)

Command Routing

Command flag	Behavior
(none) / --run	Full pipeline: preflight → plan → execute → verdicts → report
--plan	Generate validation-plan.md only (no execution)
--execute	Execute an existing plan; no re-planning
--fix	Auto-fix failures, re-run (3-strike protocol); see error-recovery
--audit	Read-only; no code changes
--report	Export a human-readable report
--ci	Non-interactive; exit codes; no approval gates

Validation Order

Always bottom-up for multi-layer systems:

Data Layer → Backend Business Logic → API Endpoints → UI / CLI

Higher layers depend on lower layers. Start at the data layer to find causes, not symptoms.

Pipeline

1. preflight            → e2e-evidence/preflight-report.md (must be CLEAR)
2. create-validation-plan → e2e-evidence/validation-plan.md (if missing or --plan)
3. Human review         → skip in --ci
4. Execute per platform → invoke platform skill
                          each journey produces e2e-evidence/j{N}-*
5. gate-validation-discipline → verdict per criterion
6. Report assembly      → e2e-evidence/report.md
7. On FAIL:
     --fix  → error-recovery 3-strike loop
     --audit → skip fix, record findings only

Success Criteria (a run is "done" only when ALL true)

1. Platform detected; every journey identified
2. PASS criteria defined for every journey BEFORE execution
3. Real system built AND running
4. Every journey exercised through real interfaces
5. Evidence captured AND read (content described, not just "exists")
6. Evidence matched to criteria with PASS/FAIL verdicts written
7. Failures diagnosed with root cause
8. Report saved to e2e-evidence/report.md
9. Zero unreviewed evidence files

Rules

1. Never skip platform detection — wrong platform = wrong validation
2. Never execute without a plan — PASS criteria must exist first
3. Never claim PASS without cited evidence — see gate-validation-discipline
4. Always validate bottom-up for fullstack

Security Policy

Evidence directories may contain screenshots of authenticated screens or API responses with tokens. Store in e2e-evidence/ (gitignored). Never commit evidence containing credentials or PII to public repos.