# Contract: POLICY-CONSOLE-23-001 — Console findings/export & simulation surfaces

**Status:** Draft → Proposed (2025-12-02)

**Scope**
- Provide deterministic, tenant-scoped APIs from Policy Engine to StellaOps Console for findings browse/export and simulation/explain experiences.
- Replace legacy ad-hoc Console queries with cursor-based, RBAC-aware endpoints that expose provenance and aggregation hints.
- Keep all responses deterministic (stable ordering, explicit timestamps, no wall-clock/default time windows).

## Versioning & Compatibility
- `schemaVersion`: `console-policy-23-001` (bumped on breaking changes).
- Media type: `application/vnd.stellaops.console-policy-23-001+json` (clients MUST send `Accept` and SHOULD send `Content-Type`).
- Backward-compatible additions follow additive fields; ordering and cursor format remain stable.

## Authentication & RBAC
- Required scopes: `policy:read`, `effective:read`, `explain:read` (all tenant-scoped).
- Optional `findings:export` to enable NDJSON bulk export.
- All endpoints require `X-Tenant-Id`; server enforces tenant filter and rejects cross-tenant cursor reuse.

## Determinism Rules
- Ordering: `policyVersion DESC`, `artifactDigest ASC`, `purl ASC`, `ruleId ASC`, `findingId ASC`.
- Cursor: opaque, URL-safe base64 of the last tuple above; contains `policyVersion|artifactDigest|purl|ruleId|findingId` plus `schemaVersion`. No server clocks in cursors.
- Timestamps: clients MUST provide `evaluationTimestamp` or `timeWindowStart/End`; server never injects `DateTime.UtcNow` defaults.
- Randomness/network access disallowed; sampling ratios must be provided by the client or policy config.

## Endpoints

### 1) List findings (paged)
- **GET** `/policy/console/findings`
- **Query params**
  - `cursor` (string, optional)
  - `limit` (int, 1–500, default 100)
  - `severityBand[]` (enum: critical|high|medium|low|unknown)
  - `ruleId[]`, `policyId`, `policyVersion`
  - `artifactDigest[]`, `purl[]`, `namespace[]`
  - `advisoryId[]`, `vexStatement[]`
  - `state[]` (open|waived|fixed|not_applicable)
  - `timeWindowStart`, `timeWindowEnd` (ISO-8601, optional)
  - `sort` (one of `default`, `severity_desc`, `artifact`, `rule`); default respects deterministic tuple above.
- **Response**
  ```json
  {
    "schemaVersion": "console-policy-23-001",
    "items": [
      {
        "findingId": "ulid",
        "policyVersion": "2025.11.24",
        "artifactDigest": "sha256:...",
        "purl": "pkg:maven/org.example/foo@1.2.3",
        "ruleId": "RULE-1234",
        "severity": "high",
        "state": "open",
        "explainSummary": {
          "hitRules": ["RULE-1234"],
          "traceSampleId": "ulid",
          "rationale": ["package matches advisory CVE-2025-1234"]
        },
        "provenance": {
          "evaluationTimestamp": "2025-11-28T00:00:00Z",
          "effectiveFindingHash": "be...",
          "source": "materialized"
        }
      }
    ],
    "cursor": { "next": "b64...", "prev": "b64..." },
    "aggregates": {
      "countsBySeverity": {"critical": 1, "high": 5, "medium": 12, "low": 3, "unknown": 0},
      "countsByRule": [{"ruleId": "RULE-1234", "count": 4}],
      "countsByPolicyVersion": [{"policyVersion": "2025.11.24", "count": 25}]
    }
  }
  ```

### 2) Finding explain trace (summary)
- **GET** `/policy/console/findings/{findingId}/explain`
- Returns deterministic trace summary for UI drawer (no full trace fan-out): hit rules, key facts, sampled trace token, policyVersion, evaluationTimestamp, hashes.
- Optional `format` (`json` default, `markdown` for UI preview); output ordering stable.

### 3) Simulation/export diff (used by POLICY-CONSOLE-23-002)
- **POST** `/policy/console/simulations/diff`
- **Body**
  ```json
  {
    "baselinePolicyVersion": "2025.11.24",
    "candidatePolicyVersion": "2025.12.02",
    "artifactScope": [{"artifactDigest": "sha256:..."}],
    "budget": {"maxFindings": 2000, "maxExplainSamples": 50},
    "filters": {"severityBand": ["high","critical"]}
  }
  ```
- **Response**
  ```json
  {
    "schemaVersion": "console-policy-23-001",
    "summary": {
      "before": {"total": 120, "severity": {"critical":4,"high":30,"medium":60,"low":26}},
      "after": {"total": 98, "severity": {"critical":3,"high":22,"medium":55,"low":18}},
      "delta": {"added":12,"removed":34,"regressed":2}
    },
    "ruleImpact": [
      {"ruleId":"RULE-1234","added":3,"removed":10,"severityShift":{"high→medium":6}},
      {"ruleId":"RULE-2000","added":1,"removed":0}
    ],
    "samples": {
      "explain": ["trace-token-1","trace-token-2"],
      "findings": ["finding-ulid-1","finding-ulid-2"]
    },
    "provenance": {
      "baselinePolicyVersion": "2025.11.24",
      "candidatePolicyVersion": "2025.12.02",
      "evaluationTimestamp": "2025-12-02T00:00:00Z"
    }
  }
  ```
- Ordering of ruleImpact array: `ruleId ASC`; samples ordered by hash.

### 4) Bulk export (NDJSON)
- **POST** `/policy/console/findings/export`
- Body accepts same filters as list endpoint plus `format` (`ndjson` only) and `maxRows` (hard cap 50k).
- Response streams NDJSON of finding records in deterministic ordering with content hashes.

## Error Model
- 400 with machine-readable code (`invalid_filter`, `unsupported_schemaVersion`, `budget_exceeded`).
- 401/403 for auth/scope failures; 409 when `schemaVersion` mismatch.
- 429 when budget limits tripped; include `retryAfterSeconds` but never implicit sleep in server.

## Non-Goals
- No mutable state or approvals exposed here; status transitions remain in Console backend via existing endpoints.
- No live wall-clock filtering; clients must pass explicit windows.

## Testing Hooks
- Provide `X-Dry-Run: true` to validate filters and budgets without executing evaluation.
- `X-Debug-Sampling: <0..1>` allowed in non-production tenants only; otherwise rejected.

## Implementation Notes
- Reuse batch evaluation pipeline for simulation diff; reuse materialized `effective_finding_*` collections for listing/export.
- Enforce deterministic `evaluationTimestamp` supplied by caller; reject missing timestamp when `baselinePolicyVersion != candidatePolicyVersion`.
- All aggregates computed in-memory over deterministically ordered result sets; no sampling unless explicitly requested.