---
artifact_kind: optimizer-card
artifact_name: <e.g., "nash-ga-v2" | "nsga-ii-coalition-v1" | "mcts-lats-v1">
version: <semver>
owner: <team>
last_reviewed: <YYYY-MM-DD>
next_review: <YYYY-MM-DD>
optimizer_kind:
  <genetic | multi-objective | mcts | cp-sat | bayesian | psro | coalition>
related_phase_179_tasks:
  [
    <179.4.2.1 | 179.4.2.2 | 179.4.2.3 | 179.4.2.4 | 179.4.2.5 | 179.4.2.6 |
    179.4.2.7>,
  ]
---

# Optimizer Card — <artifact_name> v<version>

## 1. Identity

- **Optimizer name:** <name>
- **Algorithm family:** <Nash-product GA / NSGA-II / MAP-Elites / MCTS / LATS /
  CP-SAT / MILP / Bayesian / PSRO / coalition-core>
- **Implementation source:** <library / repo / paper citation>
- **Intended use:**
  <which Concordia candidate-search scenarios this optimizer serves best>

## 2. Scoring rule

- **Utility aggregation:** <Nash product / utilitarian sum / max-min /
  Kalai-Smorodinsky / multi-objective Pareto>
- **Fairness metrics reported:** <which metrics are emitted from this optimizer>
- **Handling of hard constraints:** <how §179.3.3.2 hard constraints are
  separated from soft preferences>
- **Handling of uncertainty:**
  <how posterior means and credible intervals feed the optimizer>

## 3. Inputs

- **Candidate space:** <parameterization, mutators, clause library>
- **Initial population / seeds:** <from §179.4.1.1 sources>
- **Time / iteration budget:** <wall-clock, iteration cap, early-stop criteria>
- **Randomness:** <seed policy, reproducibility posture>

## 4. Outputs

- **Accepted candidate contract:** <JSON schema; point to `@concordia/contracts`
  types>
- **Pareto-frontier diagnostics:** <what is reported for explanation>
- **Uncertainty propagation:**
  <how optimizer output carries uncertainty forward>

## 5. Evaluation

| Metric                          | Value | Evaluator card | Date  |
| ------------------------------- | ----- | -------------- | ----- |
| Pareto-front coverage           | <...> | <...>          | <...> |
| Diversity preservation          | <...> | <...>          | <...> |
| Fairness (Nash / KS / max-min)  | <...> | <...>          | <...> |
| Regret vs oracle                | <...> | <...>          | <...> |
| Specification-gaming resistance | <...> | <...>          | <...> |
| Runtime (p50 / p95 / p99)       | <...> | <...>          | <...> |
| Cost per run (USD)              | <...> | <...>          | <...> |

## 6. Known limitations

- <convergence issues, mode-collapse, sensitivity to hyperparameters, etc.>

## 7. Guardrails

- **Candidate safety gate:** <§179.4.1.5 feasibility / legality / safety filter
  applied before scoring>
- **Clause static validation:** <§179.2.3.4 — mandatory before optimizer entry>
- **Redline separation:** <hard constraints never laundered into utility>

## 8. References

- <link to implementation code, tuning notebooks, academic paper the optimizer implements>