diff --git a/docs/architecture/legacy-engine-constraints.md b/docs/architecture/legacy-engine-constraints.md
index 1bfd409..365924f 100644
--- a/docs/architecture/legacy-engine-constraints.md
+++ b/docs/architecture/legacy-engine-constraints.md
@@ -47,6 +47,117 @@ Constraints are organized by domain:
 
 Each constraint sets a failure reason string used by UI and logs.
 
+
+## Rewrite Constraint Contract (Draft)
+
+This section proposes a shared contract for the rewrite so both the turn daemon
+(in-memory) and API server (DB-backed) can evaluate constraints consistently.
+
+### Goals
+
+- Single source of truth for constraint logic.
+- Support full evaluation in daemon and precheck in API.
+- Explicit data requirements for batching and caching.
+- Deterministic reasons for deny vs unknown outcomes.
+
+### Types (TypeScript sketch)
+
+```ts
+export type ConstraintResult =
+    | { kind: 'allow' }
+    | { kind: 'deny'; reason: string; code?: string }
+    | { kind: 'unknown'; missing: RequirementKey[] };
+
+export type RequirementKey =
+    | { kind: 'general'; id: number }
+    | { kind: 'city'; id: number }
+    | { kind: 'nation'; id: number }
+    | { kind: 'destGeneral'; id: number }
+    | { kind: 'destCity'; id: number }
+    | { kind: 'destNation'; id: number }
+    | { kind: 'arg'; key: string }
+    | { kind: 'env'; key: string };
+
+export interface ConstraintContext {
+    actorId: number;
+    cityId?: number;
+    nationId?: number;
+    destGeneralId?: number;
+    destCityId?: number;
+    destNationId?: number;
+    args: Record<string, unknown>;
+    env: Record<string, unknown>;
+    mode: 'full' | 'precheck';
+}
+
+export interface StateView {
+    has(req: RequirementKey): boolean;
+    get(req: RequirementKey): unknown | null;
+}
+
+export interface Constraint {
+    name: string;
+    requires(ctx: ConstraintContext): RequirementKey[];
+    test(ctx: ConstraintContext, view: StateView): ConstraintResult;
+}
+```
+
+### Evaluation Flow
+
+- `ConstraintPlanner` collects requirements across constraints.
+- `StateView` loads those requirements (daemon: in-memory, API: DB).
+- `test()` returns:
+  - `allow` if constraint passes.
+  - `deny` with a stable reason/code for UI.
+  - `unknown` if required data is missing and `mode === 'precheck'`.
+
+```ts
+function evaluateConstraints(
+    constraints: Constraint[],
+    ctx: ConstraintContext,
+    view: StateView,
+): ConstraintResult {
+    for (const constraint of constraints) {
+        const missing = constraint.requires(ctx).filter((req) => !view.has(req));
+        if (missing.length && ctx.mode === 'precheck') {
+            return { kind: 'unknown', missing };
+        }
+        const result = constraint.test(ctx, view);
+        if (result.kind !== 'allow') {
+            return result;
+        }
+    }
+    return { kind: 'allow' };
+}
+```
+
+### StateView Selection Boundary
+
+The split between in-memory and DB-backed evaluation happens outside the
+constraint logic. A factory or loader chooses the `StateView` implementation
+based on the execution environment:
+
+- Turn daemon -> `InMemoryStateView` with a full in-memory snapshot.
+- API server -> `DbStateView` (or `ProjectedStateView`) that fetches only the
+  required fields from the DB or precomputed projections.
+
+This keeps constraints pure and deterministic, while the infrastructure layer
+decides how to satisfy `requires()` in each runtime.
+
+### Mapping from Legacy Flags
+
+- `REQ_GENERAL` -> `{ kind: 'general', id: actorId }`
+- `REQ_CITY` -> `{ kind: 'city', id: cityId }`
+- `REQ_NATION` -> `{ kind: 'nation', id: nationId }`
+- `REQ_DEST_*` -> respective `dest` key
+- `REQ_ARG` -> `{ kind: 'arg', key: <argName> }`
+- `env` dependencies (for example `turnterm`, `year`) -> `{ kind: 'env', key: 'turnterm' }`
+
+### Data Projection Suggestion
+
+- API prechecks can rely on a small read model (for example `general_summary`)
+  updated by the turn daemon; `StateView` selects the source per requirement.
+
 ## Open Questions / Follow-ups
 
 - Some constraints rely on `env` values (`turnterm`, `year`, etc.); document
diff --git a/docs/architecture/todo.md b/docs/architecture/todo.md
index aa9ac3e..3eedfa1 100644
--- a/docs/architecture/todo.md
+++ b/docs/architecture/todo.md
@@ -8,6 +8,7 @@ Move items into the main docs once they are finalized.
 - Turn daemon scheduling details and preemption rules
 - Turn daemon vs API server priority policy under load
 - In-memory state lifecycle and DBMS flush checkpoints
+- [AI suggestion] Define a constraint evaluation contract with explicit data requirements and a tri-state result (allow/deny/unknown) to support DB-backed prechecks vs in-memory full checks.
 - Recovery behavior after partial flush or crash
 - Observability: metrics, logs, and alerts for turn processing
 - [AI suggestion] Define a stable in-memory AI state contract (snapshot + delta invalidation rules) aligned with `GeneralAI` inputs.