feat(state): scope card evidence and policy memory by reported run identity

AGENTS.md

@@ -6,6 +6,7 @@
 - `policy/combat.py` and `policy/selection.py`: combat and selection proposals.
 - `policy/context.py`: shared `Decision`, pending actions, and session-owned `PolicyContext`.
 - `run.py`: observe–decide–act loop, captures, and session attribution.
+- `run_state.py`: reported run identity and combat-pile provenance; see `docs/RUN_STATE.md`.
 - `sts2.py`: local game HTTP client. `jev.py`: TypeSafe model client and gates.
 - `migrate.py`: dataset migration and integrity checks.
 - `CONTEXT.md`: domain terms. `docs/POLICY.md`: policy boundaries. `docs/DATASET.md`: dataset schema and limits.
@@ -35,7 +36,8 @@ Prefer integration/end-to-end checks; keep only essential regression tests and u
 
 ## Safety and scope
 - Default to offline tests. Ask before live game actions, model calls, or evaluation batches.
-- `run.py --dry-run` sends no action POSTs or deck-cache writes; state reads, model calls, and logs still run.
+- `run.py --dry-run` sends no action POSTs; state/identity reads, model calls, and logs still run.
+- The runner ignores legacy `deck.json`. Combat-pile evidence is session-local and is not the persistent deck.
 - Never print or commit credentials. Do not read secret files for development checks.
 - Do not rebuild `dataset/`, change captures, or edit `vendor/` unless the task requires it.
 - `migrate.py --verify` rebuilds data; use `--check-only` for routine verification.

CONTEXT.md

@@ -38,6 +38,13 @@ An accepted request to change a card's selection status. It does not prove that
 **Screen flow**:
 Related game screens that form one interaction, such as opening and skipping a card reward.
 
+**Reported run identity**:
+The run identifier supplied by the game service. It identifies a reported attempt, not a seed or an atomic state snapshot.
+
+**Combat-pile snapshot**:
+The cards visible in combat piles at one observation. It can include temporary cards and omit unavailable cards.
+_Avoid_: Persistent deck, complete deck
+
 **Run deck**:
 The persistent collection of cards owned during a run, distinct from temporary combat cards and combat piles.
 _Avoid_: Treating all visible combat cards as the persistent deck.

docs/POLICY.md

@@ -4,6 +4,7 @@
 
 ```text
 run.py                 observe, execute, report results, record the session
+run_state.py           reported run identity and scoped combat-pile evidence
 brain.py               entry point, dispatch, navigation, shops, minigames
 policy/
     __init__.py        package marker; no registration or initialization
@@ -13,8 +14,9 @@ policy/
 facts.py               observation parsing and combat calculations
 ```
 
-`brain.decide()` remains the policy entry point. The runner owns one
-`PolicyContext` per session and reports action results to it.
+`brain.decide()` remains the policy entry point. The runner owns a `RunContext`,
+which holds policy memory and replaces it on reported run changes. The runner
+reports action results through that context.
 `brain.Decision` and `brain.PolicyContext` remain available as direct imports
 of the shared types, so the runner interface does not change.
 
@@ -47,8 +49,8 @@ These are plain modules, not a plugin framework or class hierarchy. Navigation,
 shops, and minigames stay together until a further split helps development.
 
 The extraction does not change game decisions, confidence gates, fallbacks,
-state reconciliation, or recording. Run identity and persistent-deck provenance
-remain separate work.
+state reconciliation, or recording. The subsequent
+[run-state pass](RUN_STATE.md) adds reported identity and card-evidence provenance.
 
 ## Extraction verification
 

docs/RUN_STATE.md

@@ -0,0 +1,117 @@
+# Run identity and card-evidence provenance
+
+## What is available
+
+The vendored mod exposes `current_run.run_id` through `GET /api/v1/compendium`.
+Its format includes save scope, profile, and start time. The seed describes
+content; it is not the identity of an attempt.
+
+The live player serializer exposes hand, draw, discard, and exhaust piles during
+combat. It does not expose the persistent run deck. The compendium exposes a
+save path, but the runner does not follow that path or read the save file.
+
+Sources checked:
+
+- `vendor/STS2MCP/McpMod.Compendium.cs`: `BuildCurrentRunContext`.
+- `vendor/STS2MCP/McpMod.StateBuilder.cs`: `BuildPlayerState`.
+
+## Ownership and identity
+
+`run.py` owns one `run_state.RunContext` per invocation. The context owns the
+current `PolicyContext` and optional combat-pile evidence. The context performs
+no network or file operations.
+
+The runner requests identity after each stable in-run observation, before
+proposing an action. Menu, game-over, unknown, and overlay observations do not
+trigger this request. Dry-run still performs identity reads.
+
+This deliberately uses the existing full compendium endpoint. It adds one local
+HTTP read per stable in-run step. A smaller identity endpoint would reduce cost,
+but this pass does not modify the mod or add a polling cache with stale identities.
+Live latency has not been measured.
+
+A usable identity requires `is_in_progress: true` and a nonempty string `run_id`.
+Missing or malformed identity remains unknown. The runner never substitutes a
+seed, character name, or floor number as an identity.
+
+- A different reported run ID resets policy memory, duplicate-action guards,
+  rejection counts, and card evidence.
+- `--stop-on-run-end` stops before acting in the replacement run, with reason
+  `run_changed`. This reports a run boundary, not victory or defeat.
+- Observed menu/game-over transitions clear run evidence and active-run policy memory.
+- Unknown states and overlays preserve pending-action guards.
+- Missing identity or a transport error discards card evidence but retains action
+  guards. Erasing accepted toggles during a metadata outage could toggle them again.
+- The last known ID is retained privately across a metadata outage, so recovery
+  with a different ID can still reset state. Decision rows use null during the outage.
+
+### Identity is reported, not atomic
+
+The mod derives identity from save metadata. The observation and compendium are
+separate requests. They are not an atomic snapshot, and save metadata can lag.
+The runner cannot prove that both responses describe exactly the same instant.
+A live observation-level run ID would provide a stronger association.
+
+Do not infer exact attribution or game-action completion from this identity alone.
+Externally switching runs while the bot acts remains a coordination risk.
+
+## Card evidence
+
+The runner no longer reads or writes `deck.json`. Existing files remain untouched.
+No card evidence survives a runner restart. Unscoped legacy input to
+`facts.deck_context` becomes `unknown`.
+
+A combat-pile snapshot is retained only when a reported run ID is available.
+Its provenance includes:
+
+- `source: combat_piles`;
+- reported `run_id`;
+- `observed_step` within the current session;
+- observed act and floor;
+- `persistent_deck: false`;
+- which pile lists were present;
+- freshness: `observed_combat_piles` or `historical_combat_piles`.
+
+The counts retain visible temporary cards, including Status cards. Filtering out
+Status cards would not reconstruct the persistent deck: generated Attack cards,
+temporary upgrades, and unavailable card data would still remain problems.
+Pile-list presence does not establish complete coverage.
+
+Evidence can carry into immediate reward decisions in the same reported room.
+Opening a card reward or skipping it preserves that limited evidence. Other
+accepted non-combat actions discard it because they can change cards or leave the
+room. Rejected requests do not establish such a change.
+
+A room change, missing room coordinates, identity failure, new reported run ID,
+or observed run exit discards evidence. The policy receives `unknown` when no
+scoped evidence remains. This reduces available synergy context intentionally;
+it is safer than presenting stale combat piles as the current persistent deck.
+
+An exact persistent-deck source is still unavailable. This pass labels that
+limitation; it does not implement a complete deck tracker or a save-file parser.
+
+## Minimal trace additions
+
+Decision rows now include reported `run_id`, `run_seed`, and available
+`deck_provenance`. The provenance describes available macro-policy evidence;
+it does not mean every handler used it. Combat policy uses the combat observation directly.
+
+Identity changes and identity transport failures have separate trace events.
+Full observation/action linkage, unique capture names, session identity, and
+session finalization remain recording work.
+
+## Validation
+
+- 279 existing-script assertions passed: 68 facts, 110 policy, and 101 runner.
+  This includes two upgrade-selection assertions added concurrently.
+- Runner integration flows cover same-seed run replacement, single-run stopping,
+  metadata outages, temporary cards, provenance, mutation invalidation, room
+  boundaries, unknown rooms, and restart/cache isolation.
+- Eighteen temporary whole-process scenarios passed with real HTTP clients and
+  loopback fixtures. Five exercise the new identity/provenance behavior.
+- The offline audit replayed 346 stored observations without policy exceptions.
+- Dataset integrity passed for 37 runs, 1,052 decisions, and 346 observations.
+- Python compilation and shell syntax checks passed.
+
+No live game actions, paid model calls, dependencies, vendor edits, or dataset
+rewrites were required. Temporary probes were not added to the repository.

docs/research/13-policy-state.md

@@ -107,7 +107,7 @@ Combat and selection policy have since been extracted without changing execution
 behavior. See [the current policy structure](../POLICY.md). No plugin framework
 or class hierarchy was added.
 
-The next state work is run identity and deck provenance. Recording should then
-link observations, proposals, action attempts, results, and reconciliation.
+[Run identity and card-evidence provenance](../RUN_STATE.md) have since been added.
+Recording should next link observations, proposals, action attempts, results, and reconciliation.
 Session finalization, capture-name collisions, and actual policy-gate metadata
 remain separate work.

facts.py

@@ -538,7 +538,7 @@ class CombatFacts:
             f"block_available={self.max_block_available} survives_with_cards={self.survives_with_cards}",
             f"potions={[p.get('name') for p in self.potions]}",
             f"piles: draw={self.draw_pile_count} discard={self.discard_pile_count} exhaust={self.exhaust_pile_count}",
-            f"deck: {self.deck_counts}",
+            f"visible combat piles: {self.deck_counts}",
         ]
         for e in self.enemies:
             lines.append(
@@ -620,22 +620,12 @@ def enemy_status_names(enemy: EnemyFact) -> list:
 
 
 def deck_summary(player: dict) -> dict:
-    """
-    Stable aggregates over ALL FOUR PILES, for macro decisions.
+    """Aggregate visible combat piles, not the persistent run deck.
 
-    The reward, shop and map states do NOT expose the deck, so this travels with
-    the composition snapshot taken during combat. It is deliberately a
-    *supplement* to the name->count map, never a replacement: card identities
-    are the signal for synergy, redundancy and upgrades, and only counts can
-    carry them. These aggregates are the part that is arithmetic, so it is
-    computed here and never asked of the model.
-
-    Every field is computed over the whole deck and is stable across snapshots
-    of the same deck. There is deliberately no cost aggregate: the state does
-    not expose a card's BASE cost, and a temporary in-combat cost modifier
-    applies to the copy in hand. Measured over 600 captures, one Strike read
-    `cost: "0"` in hand while the same Strike read `cost: "1"` in the draw pile,
-    so an average cost would differ between two snapshots of an identical deck.
+    Generated cards, Status cards, temporary upgrades, and missing pile data
+    limit this evidence. These counts can change during combat. Cost is omitted
+    because temporary modifiers make displayed costs unsuitable for a base-cost
+    aggregate. The historical function name is retained for existing callers.
     """
     names: list[str] = []
     attacks = blocks = other = upgraded = 0
@@ -668,22 +658,19 @@ def deck_summary(player: dict) -> dict:
 
 
 def deck_context(deck: Any) -> Any:
-    """
-    What a macro question receives as deck context: the full name->count map
-    (card identities, which carry the synergy and redundancy signal) plus the
-    stable aggregates. Tolerates a legacy flat name->count snapshot.
-    """
+    """Supply card evidence with its limits; unscoped legacy data is unknown."""
     if not deck:
         return "unknown"
     if isinstance(deck, dict) and "counts" in deck:
         counts = deck.get("counts") or {}
         if not counts:
             return "unknown"
-        return {"cards": counts, **({"summary": deck["summary"]}
-                                    if deck.get("summary") else {})}
-    if isinstance(deck, dict):
-        return {"cards": deck}
-    return deck
+        provenance = deck.get("provenance")
+        if not isinstance(provenance, dict) or not provenance.get("run_id"):
+            return "unknown"
+        return {"cards": counts, "provenance": provenance,
+                **({"summary": deck["summary"]} if deck.get("summary") else {})}
+    return "unknown"
 
 
 # --------------------------------------------------------------------------

run.py

@@ -28,8 +28,7 @@ import brain
 import facts as F
 import sts2
 from jev import JevClient, JevError, answer_record
-
-DECK_FILE = pathlib.Path("deck.json")
+from run_state import RunContext
 
 # Where the game writes its own run records. This is the OUTCOME source: it
 # carries win/loss, the killer, the seed and the final deck. Override with
@@ -163,28 +162,6 @@ def decision_record(step: int, state_type: str, run_state, decision,
     }
 
 
-# Known starting decks, used only until the first combat exposes the real one.
-# Source: the character_select state, which lists starting_deck per character.
-STARTING_DECKS: dict[str, dict[str, int]] = {
-    "The Ironclad": {"Strike": 5, "Defend": 4, "Bash": 1},
-    "The Silent": {"Strike": 5, "Defend": 5, "Neutralize": 1, "Survivor": 1},
-    "The Regent": {"Strike": 4, "Defend": 4, "Falling Star": 1, "Venerate": 1},
-}
-
-
-def load_deck() -> dict | None:
-    if DECK_FILE.exists():
-        try:
-            return json.loads(DECK_FILE.read_text())
-        except (json.JSONDecodeError, OSError):
-            return None
-    return None
-
-
-def save_deck(deck: dict) -> None:
-    DECK_FILE.write_text(json.dumps(deck, indent=2, sort_keys=True))
-
-
 def observe() -> dict:
     return sts2.state()
 
@@ -243,8 +220,8 @@ def main() -> int:
     ap = argparse.ArgumentParser()
     ap.add_argument("--steps", type=int, default=20)
     ap.add_argument("--dry-run", action="store_true",
-                    help="send no action POSTs and do not update deck.json; "
-                         "state reads, model calls, and capture logs still run")
+                    help="send no action POSTs; "
+                         "state/identity reads, model calls, and capture logs still run")
     ap.add_argument("--no-jev", action="store_true")
     ap.add_argument("--pause", type=float, default=0.6,
                     help="seconds to wait after each action")
@@ -349,12 +326,9 @@ def main() -> int:
 
     atexit.register(write_session_row)
 
-    # The card_reward state does not expose the deck, but combat states expose
-    # all four piles. Snapshot composition during combat and persist it so it
-    # survives a restart.
-    deck_snapshot: dict | None = load_deck()
-    if deck_snapshot:
-        print(f"deck snapshot loaded: {deck_snapshot}")
+    # Never load the old unscoped deck.json. Visible combat piles are limited
+    # evidence, not a persistent deck, and remain local to this session/run.
+    run_context = RunContext()
     waits = 0
     rejected = 0
     last_sig: str | None = None
@@ -365,7 +339,6 @@ def main() -> int:
     duplicate_waits = 0
     jev_errors = 0
     saw_a_run = False
-    policy_context = brain.PolicyContext()
     pending_action = None
     pending_since = time.monotonic()
 
@@ -377,6 +350,28 @@ def main() -> int:
             return finish(1, "state_error")
 
         st = obs.get("state_type")
+        previous_id = run_context.run_id
+        previous_known_id = run_context.last_known_id
+        identity = None
+        if st not in ("menu", "game_over", "unknown", "overlay"):
+            try:
+                identity = sts2.compendium().get("current_run")
+            except sts2.Sts2Error as exc:
+                trace({"step": step, "event": "run_identity_error", "error": str(exc)[:160]})
+        run_context.observe(obs, identity)
+        if run_context.run_id != previous_id:
+            trace({"step": step, "event": "run_identity", "run_id": run_context.run_id,
+                   "seed": run_context.seed, "source": "compendium.current_run"})
+        run_changed = (previous_known_id is not None and run_context.run_id is not None
+                       and previous_known_id != run_context.run_id)
+        if run_changed:
+            if args.stop_on_run_end:
+                return finish(0, "run_changed")
+            last_sig = last_action_key = None
+            same_state = rejected = duplicate_waits = 0
+            same_state_since = time.monotonic()
+            last_ok = True
+        policy_context = run_context.policy
 
         # One session = one run, when asked. Otherwise the bot dies, returns to
         # the menu and starts a fresh run inside the same session, so a single
@@ -446,13 +441,7 @@ def main() -> int:
             print(f"[{step:03d}] COMBAT  {f.describe().splitlines()[0]}")
             for line in f.describe().splitlines()[1:]:
                 print(f"        {line}")
-            # The snapshot carries the name->count map (card identities, which
-            # are the synergy signal) plus stable all-pile aggregates. Only the
-            # counts decide whether the deck actually changed.
-            if f.deck_counts and f.deck_counts != (deck_snapshot or {}).get("counts"):
-                deck_snapshot = {"counts": f.deck_counts, "summary": f.deck_summary}
-                if not args.dry_run:
-                    save_deck(deck_snapshot)
+            run_context.capture_combat(obs, f, step)
 
         decision = None
         decide_error = None
@@ -461,7 +450,7 @@ def main() -> int:
             # must not inherit the previous step's answers in the log.
             client.last = None
         try:
-            decision = brain.decide(obs, client, deck_snapshot, context=policy_context)
+            decision = brain.decide(obs, client, run_context.deck, context=policy_context)
             # A procedural action or animation wait does not establish model
             # recovery. Only a successful model response resets the budget.
             if client is not None and client.last is not None:
@@ -481,7 +470,7 @@ def main() -> int:
             # the same failing request -- measured, a DNS blip re-raised out of
             # the "fallback" and killed the session.
             try:
-                decision = brain.decide(obs, None, deck_snapshot, context=policy_context)
+                decision = brain.decide(obs, None, run_context.deck, context=policy_context)
             except Exception as inner:  # noqa: BLE001
                 trace({"step": step, "event": "fallback_error", "error": str(inner)[:160]})
                 print(f"[{step:03d}] fallback also failed: {inner}")
@@ -539,7 +528,10 @@ def main() -> int:
 
         stats[decision.source] = stats.get(decision.source, 0) + 1
         print(f"[{step:03d}] DECIDE  {decision}")
-        trace(decision_record(step, st, obs.get("run"), decision, decide_error, client))
+        record = decision_record(step, st, obs.get("run"), decision, decide_error, client)
+        record.update(run_id=run_context.run_id, run_seed=run_context.seed,
+                      deck_provenance=(run_context.deck or {}).get("provenance"))
+        trace(record)
         last_action_key = action_key
 
         if args.dry_run:
@@ -553,7 +545,7 @@ def main() -> int:
                    "action": decision.action, "error": str(exc)[:200]})
             return finish(1, "action_error")
 
-        policy_context.record_result(obs, decision, accepted=result.ok)
+        run_context.record_result(obs, decision, accepted=result.ok)
         if not result.ok:
             print(f"[{step:03d}] action rejected: {result.message}")
             trace({"step": step, "event": "action_rejected",

run_state.py

@@ -0,0 +1,87 @@
+"""Run-scoped policy memory and explicitly limited card evidence. No file or network I/O."""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+import facts as F
+from policy.context import Decision, PolicyContext
+
+
+@dataclass
+class RunContext:
+    policy: PolicyContext = field(default_factory=PolicyContext)
+    run_id: str | None = None
+    seed: str | None = None
+    last_known_id: str | None = None
+    active: bool = False
+    deck: dict | None = None
+
+    def observe(self, obs: dict, current_run: dict | None) -> None:
+        """Bind evidence to reported identity, never to a seed or a local save path.
+
+        Missing metadata discards card evidence, but retains action guards.
+        Clearing accepted toggles on a transient metadata failure is unsafe.
+        A different known run ID or an observed run exit resets those guards.
+        """
+        st = obs.get("state_type")
+        if st in ("unknown", "overlay"):
+            return
+        if st in ("menu", "game_over"):
+            if self.active:
+                self.policy = PolicyContext()
+            self.run_id = self.seed = self.last_known_id = None
+            self.deck = None
+            self.active = False
+            return
+        self.active = True
+        identity = current_run if isinstance(current_run, dict) else {}
+        run_id = identity.get("run_id")
+        if identity.get("is_in_progress") is not True or not isinstance(run_id, str) or not run_id.strip():
+            self.run_id = self.seed = None
+            self.deck = None
+            return
+        if self.last_known_id is not None and run_id != self.last_known_id:
+            self.policy = PolicyContext()
+            self.deck = None
+        self.run_id = self.last_known_id = run_id
+        self.seed = identity.get("seed") if isinstance(identity.get("seed"), str) else None
+        if self.deck is not None:
+            self.deck["provenance"]["freshness"] = "historical_combat_piles"
+            # Room changes make this weak evidence even less useful. Do not
+            # carry it into a later room as a substitute for the run deck.
+            location = {k: (obs.get("run") or {}).get(k) for k in ("act", "floor")}
+            if any(value is None for value in location.values()) or location != self.deck["provenance"]["location"]:
+                self.deck = None
+
+    def capture_combat(self, obs: dict, facts: F.CombatFacts, step: int) -> None:
+        """Visible piles include temporary cards and can omit unseen cards."""
+        if self.run_id is None or not facts.deck_counts:
+            self.deck = None
+            return
+        self.deck = {
+            "counts": facts.deck_counts,
+            "summary": facts.deck_summary,
+            "provenance": {
+                "source": "combat_piles",
+                "run_id": self.run_id,
+                "observed_step": step,
+                "location": {k: (obs.get("run") or {}).get(k) for k in ("act", "floor")},
+                "persistent_deck": False,
+                "pile_lists_present": [name for name in ("hand", "draw_pile", "discard_pile", "exhaust_pile")
+                                       if isinstance((obs.get("player") or {}).get(name), list)],
+                "freshness": "observed_combat_piles",
+            },
+        }
+
+    def record_result(self, obs: dict, decision: Decision, *, accepted: bool) -> None:
+        self.policy.record_result(obs, decision, accepted=accepted)
+        if not accepted or obs.get("state_type") in ("monster", "elite", "boss"):
+            return
+        # Keep limited evidence while opening/skipping a card reward. Other
+        # accepted non-combat actions can change the deck or leave this room.
+        card_reward = decision.action == "claim_reward" and any(
+            item.get("index") == decision.params.get("index") and item.get("type") == "card"
+            for item in (obs.get("rewards") or {}).get("items", [])
+        )
+        if not card_reward and decision.action != "skip_card_reward":
+            self.deck = None

test_run.py

@@ -22,6 +22,7 @@ import contextlib
 import io
 import itertools
 import json
+import os
 import pathlib
 import sys
 import tempfile
@@ -55,12 +56,14 @@ class StubClient:
     def __init__(self, noul: float = 0.9):
         self.noul = noul
         self.calls = 0
+        self.requests = []
 
     def __repr__(self) -> str:
         return "StubClient()"
 
     def ask(self, state, questions, model=None):
         self.calls += 1
+        self.requests.append((state, questions))
         answers = {}
         for qid, q in questions.items():
             if q.get("type") == "choice":
@@ -87,12 +90,14 @@ class FakeSts2:
 
     BASE = "offline://game"
 
-    def __init__(self, states, *, action_ok=True, action_error=None):
+    def __init__(self, states, *, action_ok=True, action_error=None, identities=None):
         self.states = states
         self.i = 0
         self.actions = []
         self.action_ok = action_ok
         self.action_error = action_error
+        self.identities = identities or [{"is_in_progress": True, "run_id": "fixture:A", "seed": "same-seed"}]
+        self.identity_reads = 0
 
     def is_up(self) -> bool:
         return True
@@ -104,6 +109,13 @@ class FakeSts2:
             raise state
         return state
 
+    def compendium(self):
+        identity = self.identities[min(self.identity_reads, len(self.identities) - 1)]
+        self.identity_reads += 1
+        if isinstance(identity, Exception):
+            raise identity
+        return {"current_run": identity}
+
     def act(self, *a, **k):
         self.actions.append((a, k))
         if self.action_error:
@@ -121,8 +133,11 @@ def invoke(fake, *flags, client=None, clock=None, decide=None):
         stack.enter_context(patch.object(run, "sts2", fake))
         stack.enter_context(patch.object(run, "JevClient", return_value=client or StubClient()))
         stack.enter_context(patch.object(run, "history_snapshot", return_value=set()))
-        stack.enter_context(patch.object(run, "load_deck", return_value=None))
-        save = stack.enter_context(patch.object(run, "save_deck"))
+        stack.callback(os.chdir, os.getcwd())
+        os.chdir(directory)
+        legacy = pathlib.Path(directory) / "deck.json"
+        legacy_text = '{"counts": {"STALE CARD": 999}}'
+        legacy.write_text(legacy_text)
         stack.enter_context(patch.object(run.atexit, "register", side_effect=callbacks.append))
         stack.enter_context(patch.object(run.time, "sleep"))
         stack.enter_context(patch.object(sys, "argv", ["run.py", "--steps", "10", "--pause", "0",
@@ -139,7 +154,7 @@ def invoke(fake, *flags, client=None, clock=None, decide=None):
             path = capdir / name
             return [json.loads(line) for line in path.read_text().splitlines()] if path.exists() else []
         return types.SimpleNamespace(rc=rc, rows=rows("decisions.jsonl"),
-                                     sessions=rows("sessions.jsonl"), saved=save.call_count,
+                                     sessions=rows("sessions.jsonl"), saved=int(not legacy.exists() or legacy.read_text() != legacy_text),
                                      output=output.getvalue())
 
 
@@ -390,6 +405,58 @@ check("pending purchase times out despite unrelated observation changes",
       (result.rc, result.sessions[0]["stop_reason"], len(fake.actions)),
       (1, "pending_action_timeout", 1))
 
+print()
+print("=== 9. run identity and card-evidence provenance ===")
+run_a = {"is_in_progress": True, "run_id": "fixture:A", "seed": "same-seed"}
+run_b = dict(run_a, run_id="fixture:B")
+fake = FakeSts2([menu, grid], identities=[run_a, run_b])
+result = invoke(fake, "--no-jev", "--steps", "2")
+check("different run IDs reset selection and duplicate guards despite identical seeds/screens",
+      fake.actions, [(("select_card",), {"index": 0})] * 2)
+check("decision rows carry reported run identity",
+      [r["run_id"] for r in result.rows if r["event"] == "decide"], ["fixture:A", "fixture:B"])
+fake = FakeSts2([menu, grid], identities=[run_a, run_b])
+result = invoke(fake, "--no-jev", "--steps", "2", "--stop-on-run-end")
+check("single-run mode never acts in the replacement run",
+      (result.rc, result.sessions[0]["stop_reason"], len(fake.actions)), (0, "run_changed", 1))
+
+fake = FakeSts2([menu, grid, grid, confirmable],
+                identities=[run_a, FakeSts2.Sts2Error("metadata unavailable"), run_a])
+result = invoke(fake, "--no-jev", "--steps", "3")
+check("metadata failure preserves accepted toggles instead of selecting them again",
+      fake.actions, [(("select_card",), {"index": 0}), (("select_card",), {"index": 1}),
+                     (("confirm_selection",), {})])
+check("metadata failure is traced without inventing identity",
+      [r["run_id"] for r in result.rows if r["event"] == "decide"], ["fixture:A", None, "fixture:A"])
+
+reward_here = dict(card_reward, run=combat["run"])
+with_status = dict(combat, player=dict(combat["player"], hand=combat["player"]["hand"] + [
+    {"index": 1, "name": "Wound", "type": "Status", "cost": "1", "description": "Unplayable.", "can_play": False}]))
+client = StubClient(.1)  # Declines skipping; selects a card, then invalidates old evidence.
+result = invoke(FakeSts2([menu, with_status, reward_here, reward_here]), "--steps", "3", client=client)
+evidence = [state["deck_composition"] for state, _ in client.requests if "deck_composition" in state]
+check("temporary Status cards remain labeled combat evidence, not a persistent deck",
+      (evidence[0]["cards"].get("Wound"), evidence[0]["provenance"]["persistent_deck"],
+       evidence[0]["provenance"]["source"]), (1, False, "combat_piles"))
+check("carried card evidence records run, step, age, and exposed piles",
+      {k: evidence[0]["provenance"][k] for k in ("run_id", "observed_step", "freshness", "pile_lists_present")},
+      {"run_id": "fixture:A", "observed_step": 1, "freshness": "historical_combat_piles", "pile_lists_present": ["hand"]})
+check("accepted card addition invalidates old composition", evidence[1], "unknown")
+check("legacy deck.json stays untouched even during an executing session", result.saved, 0)
+
+for label, states, identities in [
+    ("another run", [menu, combat, reward_here], [run_a, run_b]),
+    ("missing identity", [menu, combat, reward_here], [None]),
+    ("identity read failure", [menu, combat, reward_here], [run_a, FakeSts2.Sts2Error("unavailable")]),
+    ("another room", [menu, combat, card_reward], [run_a]),
+    ("unknown room", [menu, combat, dict(reward_here, run=None)], [run_a]),
+    ("fresh session with a legacy cache", [menu, reward_here], [run_a]),
+]:
+    client = StubClient()
+    result = invoke(FakeSts2(states, identities=identities), "--steps", str(len(states) - 1), client=client)
+    evidence = [state["deck_composition"] for state, _ in client.requests if "deck_composition" in state]
+    check(label + " receives unknown deck context", evidence[-1], "unknown")
+
 print()
 print(f"=== {PASS} passed, {FAIL} failed ===")
 sys.exit(1 if FAIL else 0)