Merge pull request 'feat(sidecar): carry the event plane (Phase 11b)' (#37) from feature/events-p11b-leases-participation into edge

Reviewed-on: #37

.gitea/workflows/release.yml

@@ -149,6 +149,39 @@ jobs:
             fi
           fi
 
+          # ── Orphan sweep ────────────────────────────────────────────────
+          #
+          # The check above is VERSION-SCOPED: it only ever asks about the one
+          # version this run computed. That is enough to recover an orphan on
+          # the very next run, and useless afterwards — once any releasable
+          # commit lands, the next run computes a NEW version, never looks at
+          # the old tag again, and the orphan becomes permanent and silent.
+          #
+          # servuo-plugins v0.1.0 is the proof, and the proof is pointed: the
+          # commit that ADDED the recovery above was itself typed
+          # `fix(release): ... recover the orphaned v0.1.0 tag`, so it bumped to
+          # v0.1.1 — and the run that introduced the recovery stepped straight
+          # past the tag it was written to rescue. That tag is still orphaned.
+          #
+          # So every v* tag is checked, and anything missing a release is
+          # WARNED about. Deliberately not recovered: publishing an old version
+          # would mean building today's tree and shipping it under a tag whose
+          # tree it is not, which is worse than the inconsistency it fixes.
+          # A human decides whether to recover or drop it.
+          #
+          # Never fails the run. A sweep that can break a good release is a
+          # sweep someone will delete.
+          ORPHANS=""
+          for T in $(git tag -l 'v*' --sort=-v:refname); do
+            T_HTTP="$(curl -s -o /dev/null -w '%{http_code}' \
+              -H "Authorization: token $(printf '%s' "${REGISTRY_TOKEN:-}" | tr -d '\r\n')" \
+              "https://${GITEA_HOST}/api/v1/repos/${REPO}/releases/tags/${T}" || echo 000)"
+            [ "$T_HTTP" = "404" ] && ORPHANS="${ORPHANS} ${T}"
+          done
+          if [ -n "${ORPHANS}" ]; then
+            echo "::warning::Tags with no release:${ORPHANS} — a run failed after tagging. Publish or delete them; this job will not do either."
+          fi
+
           # Changelog range. A recovery run has nothing after the tag, so
           # summarize what the tag itself contains rather than emitting an empty
           # list: the range that produced it, i.e. previous-tag..this-tag.
@@ -342,18 +375,75 @@ jobs:
           # corrupt the Authorization header.
           CI_TOKEN="$(printf '%s' "${REGISTRY_TOKEN}" | tr -d '\r\n')"
 
-          REL_ID="$(curl -sSf -X POST "${API}/releases" \
+          PAYLOAD="$(jq -n --arg tag "$TAG" --arg body "$BODY" \
-            -H "Authorization: token ${CI_TOKEN}" \
+                       '{tag_name:$tag, name:$tag, body:$body, draft:false, prerelease:false}')"
-            -H "Content-Type: application/json" \
+
-            -d "$(jq -n --arg tag "$TAG" --arg body "$BODY" \
+          # This POST is the step that orphaned tag v0.1.1 (run 75): it landed one
-                   '{tag_name:$tag, name:$tag, body:$body, draft:false, prerelease:false}')" \
+          # second after the tag push and Gitea answered 500, having not finished
-            | jq -r '.id')"
+          # processing the pushed tag. Re-running the workflow published the same
+          # four assets untouched, so the failure was a race, not a bad request.
+          #
+          # Two things went wrong there, and both are fixed here.
+          #
+          # 1. `curl -sSf` prints NO response body on an error status, so all the
+          #    log carried was "curl: (22) ... error: 500" and the cause had to be
+          #    inferred from timestamps. Capture the body and print it.
+          # 2. Nothing retried, so a transient 5xx became a permanent orphan tag.
+          #    The plan step CAN recover one, but only on a run that reaches it --
+          #    and a later push with no releasable commits stands down before it
+          #    gets there, so in practice the tag sits until a human notices.
+          #
+          # 4xx is deliberately NOT retried: a bad token or a malformed body does
+          # not improve by being sent again, and retrying only turns a clear
+          # failure into a slow one.
+          REL_ID=""
+          for attempt in 1 2 3 4 5; do
+            HTTP="$(curl -s -o /tmp/rel.json -w '%{http_code}' -X POST "${API}/releases" \
+              -H "Authorization: token ${CI_TOKEN}" \
+              -H "Content-Type: application/json" \
+              -d "${PAYLOAD}" || echo 000)"
+
+            if [ "$HTTP" = "201" ] || [ "$HTTP" = "200" ]; then
+              REL_ID="$(jq -r '.id' /tmp/rel.json)"
+              break
+            fi
+
+            echo "::warning::POST /releases attempt ${attempt} returned HTTP ${HTTP}"
+            echo "--- response body ---"
+            cat /tmp/rel.json || true
+            echo
+            echo "---------------------"
+
+            case "$HTTP" in
+              4*) echo "::error::HTTP ${HTTP} is a client error - not retrying."; exit 1 ;;
+            esac
+
+            if [ "$attempt" = 5 ]; then
+              echo "::error::POST /releases still failing after 5 attempts. Tag ${TAG} is pushed but has no release."
+              echo "::error::Re-run this workflow - the plan step detects the orphan tag and republishes it."
+              exit 1
+            fi
+            sleep $(( attempt * 5 ))
+          done
+
+          if [ -z "$REL_ID" ] || [ "$REL_ID" = "null" ]; then
+            echo "::error::Release created but no id came back; refusing to upload assets blind."
+            exit 1
+          fi
           echo "Created release ${TAG} (id=${REL_ID})"
 
           for f in "${BIN}-linux-x86_64" "${BIN}-linux-aarch64" "${BIN}-windows-x86_64.exe" SHA256SUMS; do
-            curl -sSf -X POST "${API}/releases/${REL_ID}/assets?name=${f}" \
+            # Same treatment. An upload that fails quietly leaves a release whose
+            # SHA256SUMS does not cover every binary it advertises, which is worse
+            # than no release at all -- that file IS the trust anchor.
+            HTTP="$(curl -s -o /tmp/asset.json -w '%{http_code}' -X POST "${API}/releases/${REL_ID}/assets?name=${f}" \
               -H "Authorization: token ${CI_TOKEN}" \
-              -F "attachment=@dist/${f}" >/dev/null
+              -F "attachment=@dist/${f}" || echo 000)"
+            if [ "$HTTP" != "201" ] && [ "$HTTP" != "200" ]; then
+              echo "::error::uploading ${f} returned HTTP ${HTTP}"
+              cat /tmp/asset.json || true
+              exit 1
+            fi
             echo "  uploaded ${f}"
           done
 

sidecar/src/main.rs

@@ -52,7 +52,46 @@ use tracing_subscriber::EnvFilter;
 /// kinds are new, `GET /guilds` grows a `roster` key, and nothing existing changed shape. This is the
 /// first bump that also needed a **store migration** (`guilds.members`), because it is the first to
 /// add a column to a table that already exists rather than a whole new table; see `store::migrate`.
-pub const PROTOCOL_VERSION: u32 = 4;
+///
+/// v5 (Protocol 5): three enrichments that are additive in the same way again, bumped together
+/// rather than one at a time because a protocol bump is not cheap here — it costs a sidecar
+/// release, a republished bundle and an operator update on every shard, so a field left out costs
+/// a whole second round of that rather than a follow-up commit. They are:
+///
+///   * `house.decay` gains `ownerName` and a decay SCHEDULE — `nextStage`, `decayPeriodSec`,
+///     `dynamicDecay`, and `estimatedCollapse` only where it is exactly knowable (at IDOC under
+///     dynamic decay; at any stage under static decay, which has no randomness to wait out).
+///   * `vendor.listing` gains `ownerAcct` — without which the frame names an owner nobody can
+///     resolve to a person — and a `fees` object carrying the charge, the funds, the pay interval
+///     and the resolved `dismissalAt`.
+///   * `account.login.result` is a NEW kind: the verdict of a login, which the pre-existing
+///     `account.login.attempt` structurally cannot carry (its EventSink fires before the auth
+///     decision is made).
+///
+/// **No store migration this time**, unlike v4. Every frame is persisted whole and the board tables
+/// index only the columns they already had, so the new fields ride inside the stored JSON and the
+/// new kind lands in `events` like any other. That is the dumb-forwarder property doing its job:
+/// the sidecar defines no schema for a frame's contents and so needs no change when they grow.
+///
+/// v6 (Protocol 6): the first bump that is about a GUARANTEE rather than about data, and the first
+/// the sidecar mostly gets for free. Two things:
+///
+///   * **`idempotencyKey` on inbound commands.** A command that carries one is executed by the shard
+///     at most once; a repeat is answered with the original reply rather than re-run. That is what
+///     makes a world-writing verb retryable at all — until now a lost acknowledgement was
+///     indistinguishable from a command that never applied, so the website had to declare every
+///     write un-retryable and accept losing one rather than risk doubling it. The sidecar's part is
+///     to CARRY the key (it rides in the command body, which every write endpoint already passes
+///     through verbatim) and to understand the one new answer the shard can now give: `bridge.busy`,
+///     meaning a command under that key is still in flight. See `web::respond`.
+///   * **`champ.boss.killed` is a new kind**: a champion's defeat, with the damage table only the
+///     shard ever sees. It was previously inferable from `champ.update` going `bossUp` true then
+///     false alongside a nearby `mob.killed`, which is fragile and says nothing about who did the
+///     work. It lands in `events` and on the feed like any other kind, with no code here at all —
+///     the dumb-forwarder property again.
+///
+/// **No store migration.** Nothing gains a column; the new kind is persisted whole like every other.
+pub const PROTOCOL_VERSION: u32 = 6;
 
 // Not `#[tokio::main]`: on Windows the SCM dispatcher takes over this thread and starts the runtime
 // itself, on its own thread, once the service actually begins. The runtime is built by whichever

sidecar/src/web.rs

@@ -72,6 +72,22 @@ pub async fn serve(addr: &str, state: AppState) -> anyhow::Result<()> {
         .route("/admin/ban", post(admin_ban))
         .route("/admin/unban", post(admin_unban))
         .route("/admin/broadcast", post(admin_broadcast))
+        // The event plane (protocol 6, EVENTS_PLAN.md Phase 11b). Leases are a live config value
+        // the website holds for a bounded time; the shard restores baseline when the deadline
+        // passes whether or not anyone asks it to. GET lists the whole catalog with current values,
+        // which is the one read both `read()` and `inForce()` on the website's side are served by.
+        .route("/lease", get(lease_list).post(lease_apply))
+        .route("/lease/release", post(lease_release))
+        // The run-scoped participation ledger. `snapshot` is a POST despite being a read: it
+        // carries the caller's `idempotencyKey`, and on a well-attended run the shard walks its
+        // members across ticks rather than in one inbound call -- so a repeat arriving mid-walk is
+        // answered `bridge.busy`, and a read that can be refused as a repeat is not a GET.
+        .route("/participation", post(participation_open))
+        .route(
+            "/participation/:run_id/snapshot",
+            post(participation_snapshot),
+        )
+        .route("/participation/:run_id/close", post(participation_close))
         // Help-page (support) queue: snapshot the open queue, respond to / close a page.
         .route("/pages", get(pages_list))
         .route("/pages/:id/respond", post(page_respond))
@@ -246,13 +262,31 @@ fn constant_time_eq(a: &[u8], b: &[u8]) -> bool {
 
 // ---- shared reply handling ----
 
+/// Protocol 6. `bridge.busy` says a command carrying this `idempotencyKey` is already in flight on
+/// the shard: nothing was run, and the caller should come back.
+///
+/// It maps to **425 Too Early**, which is what that status is for — a server unwilling to risk
+/// processing a request that might be a replay. The obvious alternative, 409, is already the
+/// protocol-version gate's answer, and those two want opposite dispositions from a client: a version
+/// mismatch is a deployment fault nobody should retry, and a busy shard is a retry that should
+/// succeed on its own. Sharing a status would have made the difference readable only by inspecting
+/// the body, which is exactly how a retry loop ends up hiding a mismatched deployment.
+///
+/// It is checked BEFORE the `.error` suffix test in each responder below, and it is deliberately not
+/// spelled `bridge.busy.error`: nothing is wrong. The work is happening.
+const BUSY_KIND: &str = "bridge.busy";
+const BUSY_STATUS: StatusCode = StatusCode::TOO_EARLY;
+
 /// Turns an RPC result into an HTTP response. A `bridge.error` reply from the shard becomes a 4xx;
-/// a real reply is returned as-is; transport failures map to 503/504.
+/// a `bridge.busy` reply becomes a 425; a real reply is returned as-is; transport failures map to
+/// 503/504.
 fn respond(result: Result<Value, RpcError>) -> (StatusCode, Json<Value>) {
     match result {
         Ok(value) => {
             let kind = value.get("kind").and_then(|k| k.as_str()).unwrap_or("");
-            if kind == "bridge.error" || kind.ends_with(".error") {
+            if kind == BUSY_KIND {
+                (BUSY_STATUS, Json(value))
+            } else if kind == "bridge.error" || kind.ends_with(".error") {
                 let reason = value
                     .get("reason")
                     .and_then(|r| r.as_str())
@@ -286,7 +320,9 @@ fn respond_admin(result: Result<Value, RpcError>) -> (StatusCode, Json<Value>) {
     match result {
         Ok(value) => {
             let kind = value.get("kind").and_then(|k| k.as_str()).unwrap_or("");
-            if kind == "admin.error" {
+            if kind == BUSY_KIND {
+                (BUSY_STATUS, Json(value))
+            } else if kind == "admin.error" {
                 let reason = value
                     .get("reason")
                     .and_then(|r| r.as_str())
@@ -317,6 +353,56 @@ fn respond_admin(result: Result<Value, RpcError>) -> (StatusCode, Json<Value>) {
     }
 }
 
+/// Like `respond`, but for the event plane: leases and the participation ledger.
+///
+/// Two mappings are the point of it existing rather than reusing `respond`.
+///
+/// **`lease.drifted` is a 200.** The shard was asked to compare and set, it compared, and it
+/// refused to overwrite somebody's deliberate change -- that is the mechanism working, not a
+/// failure, and `cleanup.js` on the website treats `drifted` as a distinct successful outcome
+/// rather than an error. It is also why this is not a 409: 409 is the protocol-version gate's, and
+/// a version mismatch and a drifted lease want opposite dispositions from a caller. The same
+/// argument protocol 6 made for `bridge.busy` being a 425.
+///
+/// **The event plane being switched off is a 403**, not the 400 the generic responder's
+/// reason-sniffing would produce. `Bridge.EventsEnabled` is an operator's deliberate refusal to let
+/// the website change the world on a schedule, and telling the website it sent a bad request would
+/// send an administrator hunting a bug in a step that is written correctly.
+fn respond_event(result: Result<Value, RpcError>) -> (StatusCode, Json<Value>) {
+    match result {
+        Ok(value) => {
+            let kind = value.get("kind").and_then(|k| k.as_str()).unwrap_or("");
+            if kind == BUSY_KIND {
+                (BUSY_STATUS, Json(value))
+            } else if kind.ends_with(".error") {
+                let reason = value
+                    .get("reason")
+                    .and_then(|r| r.as_str())
+                    .unwrap_or("request rejected");
+                let code = if reason.contains("disabled") {
+                    StatusCode::FORBIDDEN
+                } else if reason.contains("no lease is offered") || reason.contains("not counting")
+                {
+                    StatusCode::NOT_FOUND
+                } else {
+                    StatusCode::BAD_REQUEST
+                };
+                (code, Json(value))
+            } else {
+                (StatusCode::OK, Json(value))
+            }
+        }
+        Err(RpcError::NoShard) => (
+            StatusCode::SERVICE_UNAVAILABLE,
+            Json(json!({"error": "shard not connected"})),
+        ),
+        Err(RpcError::Timeout) => (
+            StatusCode::GATEWAY_TIMEOUT,
+            Json(json!({"error": "shard did not reply in time"})),
+        ),
+    }
+}
+
 /// Like `respond`, but for the account-provisioning plane. Maps an `account.error` reply to a
 /// status by its reason: a name clash is a 409, the per-IP cap is a 429, a disabled/protected/
 /// refused action is a 403, an unknown target or "not linked" is a 404, anything else a 400.
@@ -324,7 +410,9 @@ fn respond_account(result: Result<Value, RpcError>) -> (StatusCode, Json<Value>)
     match result {
         Ok(value) => {
             let kind = value.get("kind").and_then(|k| k.as_str()).unwrap_or("");
-            if kind == "account.error" {
+            if kind == BUSY_KIND {
+                (BUSY_STATUS, Json(value))
+            } else if kind == "account.error" {
                 let reason = value
                     .get("reason")
                     .and_then(|r| r.as_str())
@@ -454,6 +542,16 @@ async fn link_delete(
 /// Forwards a staff moderation command to the shard, correlated on a fresh reqId. Injects `kind`
 /// and `reqId`, requiring the caller-supplied `actor` up front (the shard enforces it too). The
 /// body's remaining fields (account/serial/durationSec/reason/text/hue) pass straight through.
+///
+/// **Protocol 6: `idempotencyKey` is one of those remaining fields**, and passing it through is the
+/// whole of the sidecar's part in the guarantee. It is worth stating rather than leaving to the
+/// word "remaining", because a later refactor that narrowed this to a known field list would quietly
+/// turn every retried world write back into a possible duplicate, and nothing here would fail.
+///
+/// The key belongs to the CALLER's unit of work — the website's event step — so the sidecar neither
+/// generates one nor validates it. Note also that `reqId` is regenerated on every call: a retry
+/// carries the same idempotency key under a NEW correlation id, which is exactly why the shard
+/// re-stamps a replayed reply rather than echoing the id the first attempt used.
 async fn admin_call(st: &AppState, kind: &str, body: Value) -> (StatusCode, Json<Value>) {
     let mut obj = match body {
         Value::Object(m) => m,
@@ -504,6 +602,123 @@ async fn admin_broadcast(State(st): State<AppState>, Json(body): Json<Value>) ->
     admin_call(&st, "admin.broadcast", body).await
 }
 
+// ---- event plane handlers (protocol 6, Phase 11b) ----
+
+/// Forwards an event-plane command to the shard, correlated on a fresh reqId.
+///
+/// Deliberately NOT `admin_call`: that one requires an `actor`, because every verb behind it is a
+/// staff member pressing a button and the shard's audit trail has to name them. An event verb's
+/// author is a RUN, which the body already carries as `runId` -- and demanding an actor here would
+/// have the runner inventing a human name for something no human is doing.
+///
+/// Everything else about it is the same, and the `idempotencyKey` passthrough matters for the same
+/// reason it does there: the key is one of the body's remaining fields, and a refactor that
+/// narrowed this to a known field list would silently make every retried lease a possible duplicate.
+async fn event_call(st: &AppState, kind: &str, body: Value) -> (StatusCode, Json<Value>) {
+    let mut obj = match body {
+        Value::Object(m) => m,
+        Value::Null => serde_json::Map::new(),
+        _ => {
+            return (
+                StatusCode::BAD_REQUEST,
+                Json(json!({"error": "body must be a JSON object"})),
+            )
+        }
+    };
+
+    let req_id = st.rpc.next_req_id();
+    obj.insert("kind".to_string(), json!(kind));
+    obj.insert("reqId".to_string(), json!(req_id));
+
+    respond_event(st.rpc.call(&st.shard, Value::Object(obj), &req_id).await)
+}
+
+/// Every lease this shard offers, with what each is worth right now and what is holding it.
+///
+/// One read answers both questions the website asks about a lease: `read()` wants the current value
+/// before it applies anything, and `inForce()` wants to know whether the shard still has a record
+/// of the hold. Splitting them would be two round trips for one key.
+///
+/// **`held` means "the shard still has a record of this lease", not "the value is still
+/// overridden".** A lease whose deadline has already fired stays listed, with `expired: true`,
+/// until teardown collects its verdict -- otherwise a reconcile in that window would report it gone
+/// and the website would write off a correctly-working backstop as an orphaned resource.
+async fn lease_list(State(st): State<AppState>) -> impl IntoResponse {
+    event_call(&st, "lease.list", Value::Null).await
+}
+
+/// Body: {"key":"...","value":"...","holdMs":<ms>,"untilMs":<opt>,"runId":<opt>,"idempotencyKey":<opt>}.
+///
+/// **`holdMs` is authoritative and `untilMs` is carried for display.** An absolute deadline computed
+/// on the website and honoured on the shard is a deadline measured against two clocks, and a shard
+/// running ten minutes fast would restore a ten-minute lease the moment it took it. A duration is
+/// immune to that; the absolute time is still worth sending so a console can say when the hold ends.
+///
+/// Values cross as TEXT whatever the lease's declared type, because JSON would otherwise decide for
+/// us: `1200` and `1200.0` are one number to a parser and two strings to a compare-and-set.
+async fn lease_apply(State(st): State<AppState>, Json(body): Json<Value>) -> impl IntoResponse {
+    event_call(&st, "lease.apply", body).await
+}
+
+/// Body: {"key":"...","expected":"...","baseline":"...","idempotencyKey":<opt>}.
+///
+/// `expected` is what the event applied and `baseline` is what to put back, both out of the
+/// website's ledger rather than the shard's memory -- so a release still works after a reconnect,
+/// and a shard that has forgotten the lease entirely (a restart, which reverts every lease anyway)
+/// can answer honestly instead of refusing.
+///
+/// A mismatch comes back `lease.drifted` with a **200**: see `respond_event`.
+async fn lease_release(State(st): State<AppState>, Json(body): Json<Value>) -> impl IntoResponse {
+    event_call(&st, "lease.release", body).await
+}
+
+/// Body: {"runId":"...","map":"Felucca","x":N,"y":N,"radius":N,"holdMs":<opt>}.
+///
+/// Declares where a run happens and starts counting who is there. The area is a map, a point and a
+/// radius rather than a region name, because protocol 6's own live walk established that the most
+/// specific region containing an event is routinely anonymous.
+async fn participation_open(
+    State(st): State<AppState>,
+    Json(body): Json<Value>,
+) -> impl IntoResponse {
+    event_call(&st, "participation.open", body).await
+}
+
+/// Body: {"idempotencyKey":<opt>}. Answers the run's tally, best-effort resolved to accounts.
+///
+/// A POST for a read, and the reason is worth keeping: on a well-attended run the shard walks its
+/// members in chunks across Core ticks rather than handing the whole resolve to one inbound call,
+/// so the handler completes after its call returned and a repeat arriving in between is answered
+/// `bridge.busy`. A read that can legitimately be refused as a repeat in flight is not a GET.
+async fn participation_snapshot(
+    State(st): State<AppState>,
+    Path(run_id): Path<String>,
+    Json(body): Json<Value>,
+) -> impl IntoResponse {
+    let mut obj = match body {
+        Value::Object(m) => m,
+        _ => serde_json::Map::new(),
+    };
+    obj.insert("runId".to_string(), json!(run_id));
+    event_call(&st, "participation.snapshot", Value::Object(obj)).await
+}
+
+/// Body: {"idempotencyKey":<opt>}. Stops counting; the tally stays readable through the shard's
+/// grace window, because closing an event and collecting its results are two steps and either can
+/// be retried.
+async fn participation_close(
+    State(st): State<AppState>,
+    Path(run_id): Path<String>,
+    Json(body): Json<Value>,
+) -> impl IntoResponse {
+    let mut obj = match body {
+        Value::Object(m) => m,
+        _ => serde_json::Map::new(),
+    };
+    obj.insert("runId".to_string(), json!(run_id));
+    event_call(&st, "participation.close", Value::Object(obj)).await
+}
+
 // ---- help-page queue handlers ----
 
 /// The open help-page queue, correlated on reqId. Returns a pages.list.
@@ -978,3 +1193,153 @@ async fn ws_client(mut socket: WebSocket, state: AppState) {
 
     info!("ws client disconnected");
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn reply(kind: &str) -> Result<Value, RpcError> {
+        Ok(json!({"t": 1, "kind": kind, "reqId": "r-9"}))
+    }
+
+    /// Protocol 6. Every responder must recognise `bridge.busy`, because every write plane can be
+    /// retried: the staff plane, the account plane and the plain command plane all reach handlers
+    /// that a keyed retry can arrive at. A responder that missed it would return 200 with a body
+    /// saying nothing happened, which is the worst of the three possible answers.
+    #[test]
+    fn busy_maps_to_425_on_every_plane() {
+        assert_eq!(respond(reply("bridge.busy")).0, StatusCode::TOO_EARLY);
+        assert_eq!(respond_admin(reply("bridge.busy")).0, StatusCode::TOO_EARLY);
+        assert_eq!(
+            respond_account(reply("bridge.busy")).0,
+            StatusCode::TOO_EARLY
+        );
+        assert_eq!(respond_event(reply("bridge.busy")).0, StatusCode::TOO_EARLY);
+    }
+
+    /// The event plane is the FIRST place `bridge.busy` is reachable on a live shard rather than
+    /// only in a unit test: `participation.snapshot` walks a well-attended run's members across
+    /// Core ticks, so it completes after its inbound call returned and a repeat can genuinely land
+    /// mid-flight. 11a built the door and had nothing to walk through it.
+    #[test]
+    fn a_drifted_lease_is_a_200_not_a_409() {
+        let value =
+            json!({"kind": "lease.drifted", "key": "PlayerCaps.SkillCap", "current": "1300"});
+        let (status, body) = respond_event(Ok(value));
+
+        // The shard was asked to compare and set, it compared, and it declined to overwrite
+        // somebody's deliberate change. That is the mechanism working; the website records
+        // `drifted` as a distinct successful outcome rather than an error.
+        assert_eq!(status, StatusCode::OK);
+        assert_eq!(body.0.get("current").and_then(|v| v.as_str()), Some("1300"));
+
+        // And explicitly not the version gate's status, for the reason 425 is not either: a
+        // mismatched deployment and a moved value want opposite dispositions from a caller.
+        assert_ne!(status, StatusCode::CONFLICT);
+    }
+
+    /// The event plane being switched off is an operator's refusal, not a malformed request. A 400
+    /// would send an administrator hunting a bug in a step that is written correctly.
+    #[test]
+    fn the_event_gate_being_off_is_a_403() {
+        assert_eq!(
+            respond_event(Ok(json!({
+                "kind": "lease.error",
+                "reason": "the event plane is disabled on this shard (Bridge.EventsEnabled)"
+            })))
+            .0,
+            StatusCode::FORBIDDEN
+        );
+        assert_eq!(
+            respond_event(Ok(json!({
+                "kind": "participation.error",
+                "reason": "the event plane is disabled on this shard (Bridge.EventsEnabled)"
+            })))
+            .0,
+            StatusCode::FORBIDDEN
+        );
+    }
+
+    /// An unknown lease key and an unknown run are not-founds; anything else the shard refuses is a
+    /// bad request. The catalog is short and a typo in a step is the likely cause of both.
+    #[test]
+    fn unknown_lease_and_run_are_404s() {
+        assert_eq!(
+            respond_event(Ok(json!({
+                "kind": "lease.error",
+                "reason": "no lease is offered for key 'Loot.MaxProps'"
+            })))
+            .0,
+            StatusCode::NOT_FOUND
+        );
+        assert_eq!(
+            respond_event(Ok(json!({
+                "kind": "participation.error",
+                "reason": "this shard is not counting run '42'"
+            })))
+            .0,
+            StatusCode::NOT_FOUND
+        );
+        assert_eq!(
+            respond_event(Ok(json!({
+                "kind": "lease.error",
+                "reason": "a lease needs a positive holdMs"
+            })))
+            .0,
+            StatusCode::BAD_REQUEST
+        );
+    }
+
+    /// A lease taken, a tally answered: an ordinary success carries straight through.
+    #[test]
+    fn event_successes_are_200s() {
+        assert_eq!(respond_event(reply("lease.ok")).0, StatusCode::OK);
+        assert_eq!(respond_event(reply("lease.list.ok")).0, StatusCode::OK);
+        assert_eq!(respond_event(reply("participation.ok")).0, StatusCode::OK);
+        assert_eq!(
+            respond_event(reply("participation.snapshot.ok")).0,
+            StatusCode::OK
+        );
+    }
+
+    /// 425 must not collide with the protocol-version gate's 409: a mismatch is a deployment fault
+    /// nobody should retry, a busy shard is a retry that will succeed. Same-status would make the
+    /// two readable only by inspecting the body.
+    #[test]
+    fn busy_is_not_the_version_gates_status() {
+        assert_ne!(BUSY_STATUS, StatusCode::CONFLICT);
+    }
+
+    /// A replayed reply is an ordinary success. The shard marks it `replayed: true` for the log, and
+    /// the caller must be able to treat it exactly as it would have treated the answer it lost.
+    #[test]
+    fn a_replayed_reply_is_still_a_200() {
+        let value = json!({"t": 1, "kind": "admin.ok", "reqId": "r-9", "replayed": true});
+        let (status, body) = respond_admin(Ok(value));
+        assert_eq!(status, StatusCode::OK);
+        assert_eq!(body.0.get("replayed").and_then(|v| v.as_bool()), Some(true));
+    }
+
+    /// The error mapping the busy arm is threaded in front of must be untouched by it.
+    #[test]
+    fn errors_still_map_as_before() {
+        assert_eq!(
+            respond(Ok(
+                json!({"kind": "bridge.error", "reason": "unknown account"})
+            ))
+            .0,
+            StatusCode::NOT_FOUND
+        );
+        assert_eq!(
+            respond_admin(Ok(json!({"kind": "admin.error", "reason": "protected"}))).0,
+            StatusCode::FORBIDDEN
+        );
+        assert_eq!(
+            respond_account(Ok(
+                json!({"kind": "account.error", "reason": "already exists"})
+            ))
+            .0,
+            StatusCode::CONFLICT
+        );
+    }
+}