| name | final-exp |
| description | Internal Harness instruction source for final-exp. Route through visible Harness aliases or hook contracts instead of invoking directly. |
Final Exp
References
Read these first:
../../../.agents/references/workflow-guide.md
../../../.agents/references/context-layering-policy.md
../../../.agents/references/contract-gating-rule.md
../../../.agents/references/language-policy.md
../../../.agents/references/documentation-evidence-rule.md
../../../.agents/references/documentation-style.md
./references/experiment-matrix.md
../../../iteration_log.json
../../../PROJECT_STATE.json
../../../docs/10_contract/Project_Contract.md if it exists
../../../docs/10_contract/Evaluation_Contract.md if it exists
../../../docs/10_contract/Baseline_Contract.md if it exists
../../../docs/10_contract/Claim_Boundary.md if it exists
When To Use
Use this skill for WF11 when the user wants the final validation matrix after the main approach is stable.
Required Work
- Read the best iteration, available iteration reports, and the contract set.
Dynamic-context projects must cite the current Project Contract, Evaluation
Contract, and Claim Boundary, or record the fallback evidence/policy reason
when operating under an Automation Policy.
- Design canonical ablations that isolate major component contributions.
- Define hyperparameter search, robustness tests, and cross-dataset evaluation.
- Estimate compute budget and execution order.
- Ensure the matrix respects the Evaluation Contract and Claim Boundary. Record
Claim Delta Evidence when the matrix narrows, removes, or changes a claim
implication. In legacy or standard projects without dynamic contracts,
report the fallback evidence source instead of treating missing contracts as
approval.
- Before metric-bearing final experiment evaluation, create or verify
pre_eval_commit, or record pre_eval_commit_NOT_CHANGED.
- Write
docs/Final_Experiment_Matrix.md using the canonical template.
- Update
PROJECT_STATE.json when appropriate.
Durable Docs Render
After stable Markdown outputs for this skill are finalized, invoke $docs-site or report docs_site_boundary_report. Do not render after temporary draft edits; Markdown remains the source of truth.
Codex Adaptation
- Treat natural-language requests as the canonical
$final-exp flow.
- Preserve the original emphasis on ablation isolation, reviewer expectations, and compute budgeting.
- Keep the canonical output path and state updates.
- Include evidence sources for selected iterations, component definitions, metrics, and compute estimates.
- Report a Gate ledger for dynamic-context readiness, protocol drift, experiment-matrix writes, and workflow-state checks. Mark skipped checks
NOT_RUN with the reason.
- Treat template wording as structure-only; localize headings and narrative text according to
../../../.agents/references/language-policy.md unless a field is explicitly English-only.
Execution Rule
Follow the local planning prompt, experiment-matrix template, and language policy rather than replacing them with a generic experiment checklist.
Durable Docs Render
After stable Markdown is finalized, invoke $docs-site or report
docs_site_boundary_report / docs_site_render_or_NOT_RUN. Do not render for
temporary drafts.