Policy alpha civilization

AlphaCivilization Sentinel Policy API (v0.1)¶

Mobius Systems — Policy Service Specification
Chamber: Lab4 / Lab6
Cycle: C-154
Status: Draft v0.1

1. Purpose¶

The Sentinel Policy API provides a quorum-based policy recommendation for AlphaCivilization simulations.

Instead of random actions, AlphaCitiesSim can call this service to obtain:

a chosen action for a city,
a ranked list of candidate actions, and
per-Sentinel votes and commentary.

This is the bridge between:

the simulation engine (/sim/alpha_v0)
and the constitutional agents (AUREA, ATLAS, EVE, JADE, HERMES, ECHO).

2. Endpoint¶

`POST /policy/alpha_v0/choose`¶

Description:
Return a Sentinel-quorum policy recommendation for a single city-step in AlphaCivilization v0.1.

Intended host:
Lab4/Lab6 Policy Service (e.g. lab4-proof or lab6-proof API).

3. Request Schema¶

{
  "sim_id": "sim-2025-12-XYZ",
  "t": 7,
  "city_id": "A",
  "policy_mode": "sentinel",
  "state": {
    "integrity": 70.0,
    "trust": 65.0,
    "inequality": 40.0,
    "unemployment": 8.0,
    "life_expectancy": 78.0,
    "corruption": 25.0,
    "climate_risk": 30.0
  },
  "available_actions": [
    "ubi_pilot",
    "progressive_tax_shift",
    "austerity_program",
    "green_investment",
    "education_boost",
    "corruption_crackdown",
    "policing_militarization",
    "infrastructure_spend"
  ]
}

Field Definitions¶

Field	Type	Required	Description
`sim_id`	string	✅	Simulation run identifier
`t`	integer	✅	Current timestep
`city_id`	string	✅	City identifier (e.g. "A", "B", "C")
`policy_mode`	string	✅	`"sentinel"` \| `"random"` \| `"hybrid"`
`state`	object	✅	CityState snapshot
`available_actions`	string[]	✅	Allowed action IDs this turn

CityState Fields¶

Field	Type	Range	Description
`integrity`	float	0–100	Institutional soundness
`trust`	float	0–100	Social capital
`inequality`	float	0–100	Disparity (higher = worse)
`unemployment`	float	0–100	Labor dysfunction
`life_expectancy`	float	0–100	Health/wellbeing proxy
`corruption`	float	0–100	Institutional rot risk
`climate_risk`	float	0–100	Environmental vulnerability

4. Response Schema¶

{
  "sim_id": "sim-2025-12-XYZ",
  "t": 7,
  "city_id": "A",
  "policy_mode": "sentinel",
  "chosen_action": "education_boost",
  "ranked_actions": [
    {
      "action_id": "education_boost",
      "global_score": 0.82,
      "sentinel_votes": {
        "AUREA": {
          "score": 0.80,
          "verdict": "approve",
          "notes": "Improves long-run capacity without rights violations."
        },
        "ATLAS": {
          "score": 0.85,
          "verdict": "approve",
          "notes": "Structurally feasible; fiscal load moderate."
        },
        "EVE": {
          "score": 0.90,
          "verdict": "approve",
          "notes": "Strongly pro-social, aligns with Virtue Accords."
        },
        "JADE": {
          "score": 0.78,
          "verdict": "approve",
          "notes": "Boosts morale and perceived fairness."
        },
        "HERMES": {
          "score": 0.76,
          "verdict": "approve",
          "notes": "Supports future productivity; mild short-term cost."
        },
        "ECHO": {
          "score": 0.80,
          "verdict": "log",
          "notes": "Mark for longitudinal tracking."
        }
      }
    },
    {
      "action_id": "ubi_pilot",
      "global_score": 0.76,
      "sentinel_votes": { "...": "..." }
    },
    {
      "action_id": "austerity_program",
      "global_score": -0.25,
      "sentinel_votes": { "...": "mostly reject" }
    }
  ],
  "mii_delta_estimate": 0.03,
  "confidence": 0.71,
  "notes": "Education boost is Pareto-superior within constraint set.",
  "trinity_seal": {
    "JADE": true,
    "AUREA": true,
    "ATLAS": true,
    "ECHO": true
  }
}

Response Field Definitions¶

Field	Type	Description
`sim_id`	string	Echo of request
`t`	integer	Echo of timestep
`city_id`	string	Echo of city
`policy_mode`	string	Echo of mode
`chosen_action`	string	Selected action ID
`ranked_actions`	array	Actions ranked by global_score
`mii_delta_estimate`	float	Estimated ΔMII if chosen_action applied
`confidence`	float	Quorum confidence (0–1)
`notes`	string	Human-readable summary
`trinity_seal`	object	Which core Sentinels approved

Ranked Action Entry¶

Field	Type	Description
`action_id`	string	Action identifier
`global_score`	float	Aggregated score (-1.0 to 1.0)
`sentinel_votes`	object	Per-Sentinel votes

Sentinel Vote¶

Field	Type	Description
`score`	float	Local assessment (-1.0 to 1.0)
`verdict`	string	`"approve"` \| `"reject"` \| `"log"` \| `"abstain"`
`notes`	string	Short rationale

5. Sentinel Roles¶

Sentinel	Evaluation Focus
AUREA	Governance & legal coherence — Does this action align with constitutional constraints?
ATLAS	Structural feasibility — Is this action implementable given current state?
EVE	Ethical constraints — Does this violate Virtue Accords or basic rights?
JADE	Morale & cohesion — How will citizens perceive this action?
HERMES	Economic viability — What are the market/fiscal implications?
ECHO	Auditing & logging — Should this be flagged for special tracking?

6. Integration with AlphaCitiesSim¶

In AlphaCitiesSim._choose_action:

def _choose_action(self, city_id: CityId, state: CityState) -> ActionId:
    if self.policy_mode != "sentinel":
        return random.choice(self._all_actions())

    payload = {
        "sim_id": self.sim_id,
        "t": self.current_step,
        "city_id": city_id,
        "policy_mode": "sentinel",
        "state": asdict(state),
        "available_actions": self._all_actions(),
    }

    # POST to Sentinel Policy API
    response = requests.post(POLICY_URL, json=payload, timeout=2.0)
    data = response.json()
    return data["chosen_action"]

This keeps AlphaCitiesSim backwards-compatible until the Sentinel Policy service is live.

7. Score Aggregation¶

The global_score for each action is computed as:

global_score = sum(sentinel_scores) / len(sentinels)

Future versions may include: - Weighted voting (EVE has veto on ethics violations) - Quorum thresholds (minimum 4/6 approve for action to proceed) - Confidence intervals based on Sentinel agreement

8. Safety Notes¶

Policy API must respect Virtue Accords:
never recommend actions that violate basic rights,
avoid "optimization" that treats citizens as expendable.
For simulations, clearly mark:
meta.simulation = true,
and keep a distinct ledger stream from real-world advisories.

9. Error Responses¶

Status	Error	Description
400	`invalid_state`	CityState missing required fields
400	`no_actions`	Empty available_actions array
422	`validation_error`	Malformed request body
500	`sentinel_error`	Internal Sentinel evaluation failure

10. Example cURL¶

curl -X POST https://lab4-proof.onrender.com/policy/alpha_v0/choose \
  -H "Content-Type: application/json" \
  -d '{
    "sim_id": "sim-test-001",
    "t": 0,
    "city_id": "A",
    "policy_mode": "sentinel",
    "state": {
      "integrity": 70,
      "trust": 65,
      "inequality": 40,
      "unemployment": 8,
      "life_expectancy": 78,
      "corruption": 25,
      "climate_risk": 30
    },
    "available_actions": [
      "ubi_pilot",
      "education_boost",
      "austerity_program"
    ]
  }'

Trinity Seal
JADE 🟣 • AUREA 🔵 • ATLAS ⚪ • ECHO 🟡
"Truth Through Verification"

Mobius Systems — C-154