Validation

scafld validate <task-id> checks the Markdown spec shape before execution:

• spec_version is 2.0
• task_id is stable and machine-safe
• lifecycle status is valid
• harden status is valid
• phase ids are unique
• criteria ids are unique
• executable criteria use a known expected kind

Acceptance Criteria

Executable criteria are Markdown checklist items with structured child lines:

Acceptance:
- [ ] `ac1_1` test: Unit tests pass.
  - Command: `go test ./...`
  - Expected kind: `exit_code_zero`

Supported expected kinds:

• exit_code_zero: command exits 0.
• exit_code_nonzero: command exits non-zero.
• no_matches: command produces no output or returns a no-match status.
• browser_evidence: browser command exits 0 and emits structured evidence.

Criteria without a known expected kind are rejected before execution. This is intentional: free-form Expected: prose is documentation, not an executable contract.

browser_evidence criteria must be typed as browser. The command writes one JSON evidence object to stdout with url, viewport, console_errors, network_errors, and at least one screenshot, trace, video, or artifact path. Playwright is supported by convention, not dependency: if a Playwright-shaped browser command fails because Playwright or browser binaries are missing, scafld adds install guidance to the failed criterion reason.

Build-Time Evaluation

After a phase is open and implemented, scafld build runs that phase's command criteria and evaluates the result against Expected kind. The full command output is captured in the session/diagnostic surface; the spec keeps only concise projected evidence.

Validation profiles are agent guidance, not hidden runtime gates. Runtime success is proven by explicit acceptance criteria in the spec, review dossier validation, and the complete gate's review authority checks.

Manual Evidence

Manual work belongs in prose, harden rounds, review findings, or session notes produced by commands. The executable gate remains explicit command evidence.