@serfab/cadre-core 1.5.0 → 1.7.0

package/src/peer-join-backfill.ts

@@ -1,647 +1,709 @@
-import debug from 'debug';
-import type { Connection, Libp2p, PeerId } from '@libp2p/interface';
-import type { ActionId, IPeerNetwork } from '@optimystic/db-core';
-import { BlockTransferClient, type BlockCommitProof, type IRawStorage } from '@optimystic/db-p2p';
-import { peerJoinPushBudget } from './link-budget.js';
-
-// Peer-join whole-store block catch-up, shared by the STRAND networks and the CONTROL
-// network. On `connection:open` (debounced) it pushes every committed, materialized block
-// in the local raw store to the new peer, so a machine that joined after blocks were
-// committed still ends up physically holding them. Optimystic has no cohort-join catch-up
-// of its own; a block committed while its writer was alone has exactly one holder forever
-// without this (named collection-header blocks are written once, at collection creation,
-// and their revision never moves again — so no later commit ever carries them anywhere).
-//
-// NOTE: this copies the WHOLE local store to every newly connected peer, which is right
-// while a network is one party's handful of machines (see docs/architecture.md →
-// Replication cluster size — the control network replicates to the whole party by intent,
-// and "On a strand, the machine count is not a party count"). If these meshes ever get
-// large, filter the pushed set by FRET cohort responsibility for each block id instead of
-// pushing everything.
-//
-// NOTE: this lives in cadre-core only because `../optimystic` is read-only from this repo.
-// If Optimystic ever grows a cohort-join catch-up of its own, delete this module rather
-// than running two.
-//
-// Membership — the one place the two networks differ, expressed as the optional
-// `authorizePeer` dep:
-//
-// - STRAND networks pass none. Anything connected on a strand's own libp2p network already
-//   receives cohort replicas of new commits, so pushing older blocks to it exposes nothing
-//   new; adding a gate would diverge from what ordinary replication already does there. A
-//   peer that does NOT speak the strand's block-transfer protocol (a bare circuit relay or
-//   bootstrap node the strand node also connects to) cannot receive anything: the protocol
-//   id is namespaced per strand, so the dial fails and the push is dropped.
-// - The CONTROL network MUST pass one, because that argument does not carry over: its
-//   inbound connection gate deliberately admits non-members in several states (an
-//   un-enrolled node taking its seed, an open enrollment window, an outstanding
-//   invitation, configured bootstrap/relay peers — see docs/architecture.md, the
-//   control-network inbound connection gate). Pushing the whole control store to such a
-//   peer would hand a stranger the party's entire membership, addresses and strand list.
-//   `CadreNode` passes `isAuthorizedMember`; the gate is consulted at PUSH time, not at
-//   schedule time, and fails closed on a thrown check. A denied run is not memoized, so
-//   the peer is retried on its next `connection:open` — or sooner, via
-//   `scheduleConnectedPeers()` on a membership change (the production join order is
-//   connect-then-authorize, so the denial at dial time is the expected first pass).
-//
-// Both per-push deadlines are DERIVED from the declared link round trip
-// (`link-budget.ts`), not fixed milliseconds. That is not tuning — it is what makes this
-// module work at all over a relay. Opening a relayed connection costs a fixed number of
-// exchanges, so the 3000 ms this catch-up used to allow its dial could never finish one
-// above 375 ms of one-way link delay: the catch-up existed so that a machine which joined
-// after blocks were committed physically ends up holding them, and through a relay on any
-// link slow enough to matter it had never once managed to. The reproduction is
-// `packages/integration-tests/src/scenarios/relayed-dial-cost-by-latency.integration.ts`.
-//
-// It does NOT work at every speed. Above roughly 1250 ms one-way, two libp2p budgets that
-// sereus cannot reach abandon the connection before any deadline here is consulted, and the
-// listener's one makes that failure look like an absent peer rather than a timeout — see
-// `link-budget.ts` ("The ceiling this does NOT lift") and
-// `tickets/blocked/how-slow-a-relayed-link-does-sereus-carry`.
-//
-// RETRY, and why it backs off. A run whose PUSH FAILED — the transport threw, which is what
-// a dial or response deadline expiring looks like here — re-arms on a doubling backoff
-// (`retryBackoffMs` to `maxRetryBackoffMs`), and a `connection:open` arriving while that wait
-// is outstanding is DROPPED rather than collapsing it back to the debounce. Both halves are
-// load-bearing:
-//
-// - Without the re-arm, a transient push failure over a stable connection left the peer
-//   partially copied until its next reconnect (read repair still covered reads meanwhile).
-// - Without dropping churn inside the wait, a peer that cannot be reached at all is
-//   re-dialled on every `connection:open` forever. That is not hypothetical: the 2026-09-26
-//   relayed reproduction shows one peer re-opening a connection about every 14.5 s for a
-//   200-second run, each event starting a catch-up whose dial could not possibly finish.
-//   The connection kept re-appearing because Optimystic's own block-transfer push path
-//   budgets its dial at `transferTimeoutMs ?? 30000` — the one dial budget in the stack
-//   above the measured relayed setup cost — so it succeeded where this one could not.
-//
-// A NON-CLEAN run is NOT on its own enough to re-arm, and the distinction is what keeps the
-// backoff from becoming a worse problem than the one it fixes. `clean` also goes false for
-// outcomes no amount of retrying can change: the membership gate DENIED the peer; this
-// network's raw storage implements no `listBlockIds`, so the catch-up is inert; or the
-// receiver reported blocks in `missing`, which it does per block for a payload it cannot
-// parse and for a revision whose retained commit proof this node does not hold (a push
-// carrying no proof is refused outright by a receiver running the default
-// `requirePushCertificate: true`, and an unretained proof is the ordinary case — see
-// {@link Chunk.proofs}). Re-arming on those would re-push the WHOLE store to that peer every
-// `maxRetryBackoffMs` for as long as the node runs, and report the failure below as a link
-// budget problem when nothing about the link is wrong. All three still leave the peer
-// un-memoized, so its next `connection:open` retries — the behaviour that predates the
-// backoff, and the right one for a verdict rather than a timeout. Denial in particular is
-// re-driven on purpose by `scheduleConnectedPeers()` the moment the membership commit lands,
-// because the control network's join order is connect-then-authorize.
-//
-// A re-arm is also skipped when the peer is no longer CONNECTED. This module's trigger is
-// `connection:open`, so a peer that went away already has one: re-arming instead would leave
-// a machine dialing a peer it cannot see once a minute for the rest of its uptime.
-//
-// After `PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES` consecutive failures one `console.warn`
-// names the peer and the budget, because otherwise a machine on a too-slow link says
-// nothing at all: the connection simply never appears and only a DEBUG log mentions a dial
-// timeout.
-//
-// NOTE: a failing dial to a non-speaking peer costs one dial timeout per attempt, and is
-// never memoized (only clean runs are). Bounded by the backoff above plus the
-// unreachable-peer bail in `runCatchUp` — one dial per attempt, not one per chunk. If a node
-// ever holds many such connections, or the enumeration ahead of the first chunk gets
-// expensive, pre-check `libp2p.peerStore` for this network's block-transfer protocol before
-// enumerating.
-//
-// NOTE: a peer that disconnects INSIDE a backoff wait still costs the one attempt that wait
-// was already armed for — the connectivity check is made when the re-arm is decided, not when
-// the timer fires, and this module subscribes to `connection:open` only. One dial, not a
-// recurring one; if a node ever holds many transient peers, subscribe to `connection:close`
-// and clear the pending timer there.
-
-const log = debug('sereus:cadre:peer-join-backfill');
-
-/**
- * The block-transfer protocol's hard cap on one length-prefixed message. Mirrors
- * `MAX_BLOCK_MESSAGE_BYTES` in `@optimystic/db-p2p`'s `protocol-limits.ts` (8 MiB), which
- * that package does not re-export from its index — keep the two in sync if upstream ever
- * changes it or starts exporting it.
- */
-export const MAX_BLOCK_MESSAGE_BYTES = 8 * 1024 * 1024;
-
-/**
- * Consecutive failed catch-up runs against ONE peer before the module says so on
- * `console.warn` rather than only in its DEBUG log. At the default backoff that is reached
- * about 35 seconds in (5 s + 10 s + 20 s), which is late enough to have ruled out a blip and
- * early enough to be the first thing an operator sees about a machine that is not catching up.
- */
-export const PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES = 3;
-
-/** Tuning for the per-peer catch-up. Every field optional; defaults in {@link DEFAULT_PEER_JOIN_BACKFILL}. */
-export interface PeerJoinBackfillConfig {
-  /** Default true. False disables the catch-up entirely (the pre-existing behaviour). */
-  enabled?: boolean;
-  /** Settle time after a connection opens before catching that peer up, ms. Default 1000. */
-  debounceMs?: number;
-  /** Ceiling on blocks copied in one catch-up. Default 10_000. Reaching it is LOGGED, never silent. */
-  maxBlocks?: number;
-  /** Soft byte budget per push message. Default 1 MiB. Protocol hard cap is {@link MAX_BLOCK_MESSAGE_BYTES} (8 MiB). */
-  maxChunkBytes?: number;
-  /** Max blocks per push message. Default 64. */
-  maxChunkBlocks?: number;
-  /**
-   * Per-push dial deadline, ms. Default {@link peerJoinPushBudget}'s `dialTimeoutMs` — four link
-   * round trips at the declared link, 8000 ms as shipped. NOT a fixed number: a relayed dial
-   * costs a fixed number of exchanges, so a host on a slower link moves this (and every other
-   * cadre dial budget) by declaring `NetworkConfig.linkRoundTripMs`. Naming it here still wins
-   * over the derived value — `link-budget.ts` has the counts and the measurement.
-   */
-  dialTimeoutMs?: number;
-  /**
-   * Per-push response deadline, ms. Default {@link peerJoinPushBudget}'s `responseTimeoutMs` —
-   * two link round trips at the declared link plus a transfer allowance for the chunk's own
-   * bytes, 10_000 ms as shipped. Derived differently from {@link dialTimeoutMs} because it
-   * bounds a data transfer over a connection that is already open, not a dial.
-   */
-  responseTimeoutMs?: number;
-  /**
-   * First wait before re-running a catch-up whose last run did not complete cleanly, ms.
-   * Default 5000. Doubles per consecutive failure up to {@link maxRetryBackoffMs}.
-   */
-  retryBackoffMs?: number;
-  /** Ceiling on that doubling wait, ms. Default 60_000. */
-  maxRetryBackoffMs?: number;
-}
-
-/** The resolved defaults every {@link PeerJoinBackfill} starts from. */
-export const DEFAULT_PEER_JOIN_BACKFILL: Required<PeerJoinBackfillConfig> = {
-  enabled: true,
-  debounceMs: 1000,
-  maxBlocks: 10_000,
-  maxChunkBytes: 1024 * 1024,
-  maxChunkBlocks: 64,
-  // Derived at the DEFAULT declared link, for a `PeerJoinBackfill` built without a host's
-  // declaration (this module's own tests, an embedder driving it directly). The two production
-  // construction sites pass `peerJoinPushBudget(network?.linkRoundTripMs)` so a host that
-  // declares a slower link moves both — see `link-budget.ts`.
-  ...peerJoinPushBudget(),
-  retryBackoffMs: 5000,
-  maxRetryBackoffMs: 60_000
-};
-
-/**
- * The one capability the catch-up needs from a transfer client. Structural (rather than
- * the concrete `BlockTransferClient`) so unit tests can capture pushes without dialing
- * libp2p — see {@link PeerJoinBackfillDeps.createPushClient}.
- */
-export type PeerJoinBackfillPushClient = Pick<BlockTransferClient, 'pushBlocks'>;
-
-export interface PeerJoinBackfillDeps {
-  /** Log tag naming which network this catch-up serves (a strand id, or `control-<partyId>`). */
-  label: string;
-  /** The libp2p node of the network being caught up — source of connection events and peer ids. */
-  libp2p: Libp2p;
-  /** `node.keyNetwork`, the IPeerNetwork BlockTransferClient dials through. */
-  peerNetwork: IPeerNetwork;
-  /** This network's own raw block store — the same instance handed to the libp2p node. */
-  storage: IRawStorage;
-  /** Must equal the prefix the receiver registered its handler under: `/optimystic/<networkName>`. */
-  protocolPrefix: string;
-  /**
-   * Membership gate, consulted at PUSH time (fails closed on a throw). Absent = push to
-   * every connected peer — correct for strand networks, NEVER for the control network;
-   * see the module comment for why the strand no-gate argument does not carry over.
-   */
-  authorizePeer?: (peerId: string) => Promise<boolean>;
-  /**
-   * Test seam: build the per-peer push client. Defaults to a real
-   * `BlockTransferClient` over {@link peerNetwork} + {@link protocolPrefix}.
-   */
-  createPushClient?: (peerId: PeerId) => PeerJoinBackfillPushClient;
-}
-
-/** What one peer's catch-up actually did. Returned for tests and logged at the end of each run. */
-export interface PeerJoinBackfillResult {
-  /** Blocks offered: had a committed `latest` AND materialized content locally. */
-  offered: number;
-  /** Blocks the remote reported it persisted. */
-  accepted: number;
-  /** Block ids the remote reported in `missing` (parse or persist failure on its side). */
-  rejected: string[];
-  /** Skipped: metadata has no `latest` (pending-only — not yet a durability claim here). */
-  uncommitted: number;
-  /** Skipped: `latest` exists but no materialized block is stored for that actionId. */
-  unmaterialized: number;
-  /** Not attempted because `maxBlocks` was reached. */
-  capped: number;
-  /** Skipped: a single block whose wire size alone exceeds {@link MAX_BLOCK_MESSAGE_BYTES}. */
-  oversized: string[];
-  /** True when the `authorizePeer` gate refused (or threw) — nothing was pushed, peer not memoized. */
-  denied: boolean;
-}
-
-function emptyResult(): PeerJoinBackfillResult {
-  return { offered: 0, accepted: 0, rejected: [], uncommitted: 0, unmaterialized: 0, capped: 0, oversized: [], denied: false };
-}
-
-/** One in-flight push message being accumulated. */
-interface Chunk {
-  ids: string[];
-  buffers: Uint8Array[];
-  meta: Record<string, { rev: number; actionId: ActionId }>;
-  /**
-   * The cohort commit proof retained for each block's pushed revision, where one exists.
-   *
-   * A receiver running the default `requirePushCertificate: true` REJECTS a block pushed
-   * without one, so this is what makes peer-join catch-up land at all — a push carrying only
-   * `meta` is the legacy shape and is refused. Sparse deliberately: a block whose proof was
-   * never retained is still offered with its meta, which is the only thing that lets a
-   * receiver migrating with the flag off land the replica at this node's revision rather than
-   * a fabricated rev 1. Upstream's rule is the other direction and holds here by construction:
-   * a proof is never attached WITHOUT its meta, because both are written from the same
-   * `latest` in one place.
-   */
-  proofs: Record<string, BlockCommitProof>;
-  bytes: number;
-}
-
-/**
- * Base64 wire size of a raw buffer — `pushBlocks` base64-encodes each block into the JSON
- * request, so the protocol cap must be judged against the encoded size, not the raw bytes.
- */
-function base64WireBytes(rawBytes: number): number {
-  return Math.ceil(rawBytes / 3) * 4;
-}
-
-/**
- * Copies every block in one network's own raw store to each peer its libp2p node connects
- * to, so a peer that joined the network after blocks were committed still ends up holding
- * them physically (the receiver persists each push via `saveReplicatedBlock`, which is
- * monotonic and idempotent — crossing pushes from both ends cannot regress a revision).
- *
- * Best-effort throughout: nothing here throws into a libp2p event handler or into a
- * runtime bring-up; a failed chunk is logged and the run continues. A peer is marked
- * fully caught up ONLY after a run with no thrown chunk and an empty `missing` list, so
- * the next `connection:open` from a peer whose catch-up failed retries it.
- */
-export class PeerJoinBackfill {
-  private readonly config: Required<PeerJoinBackfillConfig>;
-  private readonly createPushClient: (peerId: PeerId) => PeerJoinBackfillPushClient;
-  private started = false;
-  private stopped = false;
-  /** Peers fully caught up this runtime (never retried until the runtime is rebuilt). */
-  private readonly done = new Set<string>();
-  /** Peers with a catch-up currently running (suppresses concurrent duplicates). */
-  private readonly inFlight = new Set<string>();
-  /**
-   * Peers whose (re-)schedule arrived while their OWN run was in flight, replayed once
-   * that run finishes. Load-bearing for the gated path: the gate's authorization check is
-   * a control-database read, so the window between it and the run's end is wide enough for
-   * the very membership commit that would have authorized the peer to land inside it —
-   * dropping that schedule (rather than deferring it) leaves the denied peer waiting for a
-   * reconnect, which is exactly what the re-arm exists to avoid.
-   */
-  private readonly rearmAfterFlight = new Set<string>();
-  /** Pending per-peer timers — the debounce, or a backoff re-arm. Cleared on stop. */
-  private readonly timers = new Map<string, ReturnType<typeof setTimeout>>();
-  /** Consecutive non-clean runs per peer, cleared when one finishes cleanly. */
-  private readonly failures = new Map<string, number>();
-  /**
-   * Epoch ms before which a peer's catch-up must not be re-run, set alongside a backoff timer
-   * that will run it. Its presence is what makes `connection:open` churn free: a schedule
-   * arriving inside the wait is dropped, rather than collapsing the backoff to the debounce.
-   */
-  private readonly retryAfter = new Map<string, number>();
-  /** Peers already reported on `console.warn`; reset when a run finally lands cleanly. */
-  private readonly warned = new Set<string>();
-  private readonly onConnectionOpen: (evt: CustomEvent<Connection>) => void;
-  private loggedNoListBlockIds = false;
-
-  constructor(private readonly deps: PeerJoinBackfillDeps, config?: PeerJoinBackfillConfig) {
-    this.config = { ...DEFAULT_PEER_JOIN_BACKFILL, ...config };
-    this.createPushClient = deps.createPushClient
-      ?? ((peerId) => new BlockTransferClient(peerId, deps.peerNetwork, deps.protocolPrefix));
-    this.onConnectionOpen = (evt) => this.schedulePeer(evt.detail.remotePeer);
-  }
-
-  /** Subscribe to connection:open AND schedule a catch-up for peers already connected. */
-  start(): void {
-    if (this.started || this.stopped || !this.config.enabled) return;
-    this.started = true;
-    this.deps.libp2p.addEventListener('connection:open', this.onConnectionOpen);
-    // A runtime rebuilt over live connections (resumeStrand) never sees their
-    // connection:open, so walk what is already connected once.
-    const scheduled = this.scheduleConnectedPeers();
-    log('[%s] started (%d peer(s) already connected)', this.deps.label, scheduled);
-  }
-
-  /** Unsubscribe, clear timers; in-flight runs observe the stopped flag and bail. */
-  stop(): void {
-    if (this.stopped) return;
-    this.stopped = true;
-    if (this.started) {
-      this.deps.libp2p.removeEventListener('connection:open', this.onConnectionOpen);
-    }
-    for (const timer of this.timers.values()) {
-      clearTimeout(timer);
-    }
-    this.timers.clear();
-    this.rearmAfterFlight.clear();
-    this.retryAfter.clear();
-    log('[%s] stopped', this.deps.label);
-  }
-
-  /**
-   * (Re-)schedule a debounced catch-up for every currently-connected peer not yet caught
-   * up. Idempotent and cheap (caught-up peers are skipped before any timer is set, and a
-   * peer whose run is in flight is deferred to the end of that run rather than dropped).
-   * Driven by {@link start}, and — on a gated network — by the embedder whenever
-   * membership changes, so a peer whose first pass was denied (connected before it was
-   * authorized: the production join order) is retried without waiting for a reconnect.
-   * Returns how many distinct peers were considered (for the start() log) — a peer holding
-   * several connections is one peer, and is scheduled once.
-   */
-  scheduleConnectedPeers(): number {
-    if (this.stopped) return 0;
-    const peers = new Set<string>();
-    for (const connection of this.deps.libp2p.getConnections()) {
-      const key = connection.remotePeer.toString();
-      if (peers.has(key)) continue;
-      peers.add(key);
-      this.schedulePeer(connection.remotePeer);
-    }
-    return peers.size;
-  }
-
-  /** Debounced entry point for connection churn: one run per peer per settle window. */
-  private schedulePeer(peerId: PeerId): void {
-    const key = peerId.toString();
-    if (this.stopped || this.done.has(key)) return;
-    if (this.inFlight.has(key)) {
-      // Defer rather than drop: the running pass may already be past its gate check (or
-      // past the block this schedule was meant to carry), so replay it when that pass ends.
-      this.rearmAfterFlight.add(key);
-      return;
-    }
-    // A peer whose last run failed is already re-armed on a backoff timer that will run it, so
-    // DROP this schedule instead of shortening the wait — otherwise a peer that cannot be
-    // reached is re-dialled on every connection:open forever (see the module comment's retry
-    // paragraph). A denied run never gets here: it does not set a backoff.
-    const retryAt = this.retryAfter.get(key);
-    if (retryAt !== undefined && Date.now() < retryAt) return;
-    const existing = this.timers.get(key);
-    if (existing) clearTimeout(existing);
-    this.timers.set(key, setTimeout(() => {
-      this.timers.delete(key);
-      void this.catchUpPeer(peerId);
-    }, this.config.debounceMs));
-  }
-
-  /**
-   * Run one peer's catch-up now, bypassing the debounce. Never rejects. Returns an
-   * all-zero result without pushing when the peer is already caught up, already being
-   * caught up, or this backfill is stopped.
-   */
-  async catchUpPeer(peerId: PeerId): Promise<PeerJoinBackfillResult> {
-    const key = peerId.toString();
-    if (this.stopped || this.done.has(key) || this.inFlight.has(key)) {
-      return emptyResult();
-    }
-    this.inFlight.add(key);
-    try {
-      const { result, clean, pushFailed } = await this.runCatchUp(peerId);
-      // NOTE: a run that hit `maxBlocks` is still "clean" and still memoizes the peer, so
-      // the tail past the ceiling never reaches it. Deliberate: enumeration is not
-      // resumable, so not memoizing would re-push the same prefix on every reconnect
-      // without ever advancing. Loud in the log below (capped > 0). If a store can
-      // realistically exceed maxBlocks, the fix is a resumable cursor, not either policy.
-      if (clean && !this.stopped) {
-        this.done.add(key);
-        this.failures.delete(key);
-        this.retryAfter.delete(key);
-        this.warned.delete(key);
-      } else if (pushFailed && !this.stopped && this.started && this.isConnected(peerId)) {
-        // `pushFailed`, not `!clean`: a denial, an inert store and a receiver's per-block
-        // rejection are all verdicts a retry cannot change, and re-arming on them would
-        // re-push the whole store to that peer forever. See the module comment.
-        //
-        // Gated on `started` too: the re-arm exists to REPLACE a connection:open-driven retry,
-        // so a caller driving `catchUpPeer` by hand against a backfill that was never started
-        // owns its own retry policy and must not be left holding a background timer.
-        this.scheduleRetryWithBackoff(peerId);
-      }
-      log('[%s] catch-up peer=%s offered=%d accepted=%d rejected=%d uncommitted=%d unmaterialized=%d capped=%d oversized=%d denied=%s done=%s',
-        this.deps.label, key, result.offered, result.accepted, result.rejected.length,
-        result.uncommitted, result.unmaterialized, result.capped, result.oversized.length, result.denied, clean);
-      return result;
-    } catch (error) {
-      // runCatchUp already contains a per-chunk catch; this guards the enumeration and
-      // metadata reads too — a backfill fault must never surface through a libp2p event.
-      log('[%s] catch-up peer=%s failed: %o', this.deps.label, key, error);
-      return emptyResult();
-    } finally {
-      this.inFlight.delete(key);
-      // Replay a schedule that arrived mid-run. `schedulePeer` re-checks `done`, so a run
-      // that finished clean re-arms nothing; a denied or failed one gets its retry.
-      if (this.rearmAfterFlight.delete(key)) {
-        this.schedulePeer(peerId);
-      }
-    }
-  }
-
-  /** Whether this network's libp2p node still holds a connection to that peer. */
-  private isConnected(peerId: PeerId): boolean {
-    return this.deps.libp2p.getConnections(peerId).length > 0;
-  }
-
-  /**
-   * Re-arm one peer's catch-up after a run whose push failed, on a wait that doubles
-   * per consecutive failure up to `maxRetryBackoffMs`. Sets {@link retryAfter} alongside the
-   * timer, which is what makes `connection:open` churn inside the wait free.
-   */
-  private scheduleRetryWithBackoff(peerId: PeerId): void {
-    const key = peerId.toString();
-    const failures = (this.failures.get(key) ?? 0) + 1;
-    this.failures.set(key, failures);
-    const delayMs = Math.min(
-      this.config.retryBackoffMs * 2 ** (failures - 1),
-      this.config.maxRetryBackoffMs
-    );
-    this.retryAfter.set(key, Date.now() + delayMs);
-    const existing = this.timers.get(key);
-    if (existing) clearTimeout(existing);
-    this.timers.set(key, setTimeout(() => {
-      this.timers.delete(key);
-      this.retryAfter.delete(key);
-      void this.catchUpPeer(peerId);
-    }, delayMs));
-    log('[%s] catch-up peer=%s failed %d time(s) in a row; retrying in %dms',
-      this.deps.label, key, failures, delayMs);
-    this.warnPersistentFailure(key, failures);
-  }
-
-  /**
-   * Say ONCE per peer, outside the DEBUG log, that its catch-up is not landing. Without this a
-   * machine on a link too slow for a relayed dial produces no statement that anything is wrong:
-   * the peer never appears to hold the blocks, and the only trace is a DEBUG line naming a dial
-   * timeout. Reset when a run finally lands cleanly, so a peer that recovers can report again.
-   */
-  private warnPersistentFailure(key: string, failures: number): void {
-    if (failures < PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES || this.warned.has(key)) return;
-    this.warned.add(key);
-    console.warn(
-      `[cadre:${this.deps.label}] peer-join block catch-up to peer ${key} has failed ${failures} times in a row `
-      + `(dial budget ${this.config.dialTimeoutMs}ms, response budget ${this.config.responseTimeoutMs}ms). `
-      + 'That peer may not be holding blocks committed before it joined. If it is reachable only through a relay, '
-      + 'these budgets are derived from network.linkRoundTripMs (see link-budget.ts); above about a 2.5-second '
-      + 'round trip no relayed connection can be established at all, whatever they are set to.'
-    );
-  }
-
-  /**
-   * The actual copy. `clean` = every chunk pushed and the remote persisted every block.
-   * `pushFailed` = at least one push THREW, which is the only non-clean outcome a retry can
-   * change; the others are verdicts (denied, inert store, blocks the receiver refused).
-   */
-  private async runCatchUp(peerId: PeerId): Promise<{ result: PeerJoinBackfillResult; clean: boolean; pushFailed: boolean }> {
-    const result = emptyResult();
-    const { storage } = this.deps;
-
-    // The membership gate, judged at push time (never at schedule time — authorization can
-    // change during the debounce window in either direction). Fails CLOSED on a throw: a
-    // gate that cannot answer must not leak the store. A denied run is not clean, so the
-    // peer is retried later rather than memoized as caught up.
-    if (this.deps.authorizePeer) {
-      let authorized = false;
-      try {
-        authorized = await this.deps.authorizePeer(peerId.toString());
-      } catch (error) {
-        log('[%s] authorizePeer(%s) threw — treating as denied: %o', this.deps.label, peerId.toString(), error);
-      }
-      if (!authorized) {
-        result.denied = true;
-        return { result, clean: false, pushFailed: false };
-      }
-    }
-
-    if (!storage.listBlockIds) {
-      if (!this.loggedNoListBlockIds) {
-        this.loggedNoListBlockIds = true;
-        log('[%s] raw storage does not implement listBlockIds(); backfill is inert', this.deps.label);
-      }
-      return { result, clean: false, pushFailed: false };
-    }
-
-    const client = this.createPushClient(peerId);
-    const encoder = new TextEncoder();
-    let chunk: Chunk = { ids: [], buffers: [], meta: {}, proofs: {}, bytes: 0 };
-    let chunkFailed = false;
-    let anyChunkDelivered = false;
-    /**
-     * A peer that has answered no push at all, and failed one, is not reachable on this
-     * protocol — abandon rather than spend a dial timeout per remaining chunk. Once ONE
-     * push has been answered the peer demonstrably speaks it, so a later failure is a
-     * transient blip and the rest of the store is still worth pushing.
-     */
-    const peerUnreachable = (): boolean => chunkFailed && !anyChunkDelivered;
-
-    const flush = async (): Promise<void> => {
-      if (chunk.ids.length === 0) return;
-      const { ids, buffers, meta, proofs } = chunk;
-      chunk = { ids: [], buffers: [], meta: {}, proofs: {}, bytes: 0 };
-      try {
-        // ONE certification value rather than two arguments, so a caller cannot pair a proof
-        // for one revision with meta for another — both were written from the same `latest`.
-        const response = await client.pushBlocks(ids, buffers, 'replication',
-          { blockMeta: meta, blockProofs: proofs }, {
-          dialTimeoutMs: this.config.dialTimeoutMs,
-          responseTimeoutMs: this.config.responseTimeoutMs
-        });
-        anyChunkDelivered = true;
-        const missing = new Set(response.missing);
-        for (const id of ids) {
-          if (missing.has(id)) {
-            result.rejected.push(id);
-          } else {
-            result.accepted += 1;
-          }
-        }
-        if (response.missing.length > 0) {
-          log('[%s] peer=%s rejected %d block(s): %o', this.deps.label, peerId.toString(), response.missing.length, response.missing);
-        }
-      } catch (error) {
-        chunkFailed = true;
-        log('[%s] push to peer=%s failed for %d block(s): %s', this.deps.label, peerId.toString(), ids.length, (error as Error).message);
-      }
-    };
-
-    for await (const blockId of storage.listBlockIds()) {
-      if (this.stopped) break;
-      if (result.offered >= this.config.maxBlocks) {
-        // Past the ceiling: count what is left (id enumeration only — no more reads or
-        // pushes) so the cap is loud in the end-of-run log rather than silent.
-        result.capped += 1;
-        continue;
-      }
-
-      const metadata = await storage.getMetadata(blockId);
-      const latest = metadata?.latest;
-      if (!latest) {
-        result.uncommitted += 1;
-        continue;
-      }
-      // NOTE: a block whose latest revision is a DELETE materializes to nothing (a
-      // tombstone), so it is skipped here — a peer that joins after such a delete never
-      // receives the tombstone from this path and relies on read repair for it. Fine
-      // while nothing deletes whole blocks pre-join; if the whole-store coverage gate in
-      // strand-membership-closed-strand-e2e ever reports a tombstone residue, teach this
-      // to push the promoted delete transform instead of skipping.
-      const block = await storage.getMaterializedBlock(blockId, latest.actionId);
-      if (!block) {
-        result.unmaterialized += 1;
-        log('[%s] block=%s has latest rev %d but no materialized content; skipped', this.deps.label, blockId, latest.rev);
-        continue;
-      }
-
-      const buffer = encoder.encode(JSON.stringify(block));
-      // NOTE: MAX_BLOCK_MESSAGE_BYTES caps the whole framed request, not one block, and
-      // this test ignores the request envelope (ids array, blockMeta, JSON punctuation).
-      // A lone block within a few KiB of the cap therefore still ships and is rejected by
-      // the receiver's length-prefix decoder — a permanent chunk failure for that peer. No
-      // block comes close today; if one ever can, subtract a measured envelope allowance
-      // here rather than raising the cap.
-      if (base64WireBytes(buffer.length) >= MAX_BLOCK_MESSAGE_BYTES) {
-        result.oversized.push(blockId);
-        log('[%s] block=%s is %d bytes (%d on the wire) — exceeds the %d-byte protocol cap alone; skipped',
-          this.deps.label, blockId, buffer.length, base64WireBytes(buffer.length), MAX_BLOCK_MESSAGE_BYTES);
-        continue;
-      }
-
-      // Flush before adding when this block would overflow either budget. A single block
-      // larger than maxChunkBytes still ships — alone, in its own chunk.
-      if (chunk.ids.length > 0
-        && (chunk.ids.length + 1 > this.config.maxChunkBlocks || chunk.bytes + buffer.length > this.config.maxChunkBytes)) {
-        await flush();
-        if (this.stopped || peerUnreachable()) break;
-      }
-      chunk.ids.push(blockId);
-      chunk.buffers.push(buffer);
-      chunk.meta[blockId] = { rev: latest.rev, actionId: latest.actionId };
-      // Read the proof for EXACTLY the revision being pushed, from the same `latest` the meta
-      // above was written from. Absent is normal (nothing retained a proof for that revision);
-      // the block still ships with its meta and the receiver decides.
-      const proof = await storage.getBlockProof(blockId, latest.rev);
-      if (proof) {
-        chunk.proofs[blockId] = proof;
-      }
-      chunk.bytes += buffer.length;
-      result.offered += 1;
-    }
-
-    if (!this.stopped && !peerUnreachable()) {
-      await flush();
-    }
-
-    if (result.capped > 0) {
-      log('[%s] peer=%s catch-up CAPPED at %d blocks; %d block id(s) not attempted',
-        this.deps.label, peerId.toString(), this.config.maxBlocks, result.capped);
-    }
-
-    const clean = !chunkFailed && result.rejected.length === 0;
-    return { result, clean, pushFailed: chunkFailed };
-  }
-}
+import debug from 'debug';
+import type { IdentifyResult, Libp2p, PeerId } from '@libp2p/interface';
+import type { ActionId, IPeerNetwork } from '@optimystic/db-core';
+import { BlockTransferClient, buildBlockTransferProtocol, type BlockCommitProof, type IRawStorage } from '@optimystic/db-p2p';
+import { peerJoinPushBudget } from './link-budget.js';
+
+// Peer-join whole-store block catch-up, shared by the STRAND networks and the CONTROL
+// network. On a new peer's `peer:identify` result — libp2p's identify protocol runs once per
+// connection and reports the protocols the remote supports — naming this network's own
+// block-transfer protocol, it schedules a debounced push of every committed, materialized
+// block in the local raw store to that peer, so a machine that joined after blocks were
+// committed still ends up physically holding them. Optimystic has no cohort-join catch-up
+// of its own; a block committed while its writer was alone has exactly one holder forever
+// without this (named collection-header blocks are written once, at collection creation,
+// and their revision never moves again — so no later commit ever carries them anywhere).
+//
+// A peer whose identify does not name the protocol — a bare circuit relay, a bootstrap node,
+// or (on the control network) a stranger the inbound gate admits without membership — is
+// never scheduled at all: it cannot receive anything, since the protocol id is namespaced per
+// network and the dial would simply fail. Before this filter existed, such a peer WAS
+// scheduled, failed every dial, and after `PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES` produced a
+// `console.warn` naming it — gotchoices/sereus#18 is two relay-only parties restarting and
+// printing exactly that warning about their own relay, forever, on every reconnect. See
+// {@link speaksBlockTransfer}.
+//
+// NOTE: this copies the WHOLE local store to every newly connected peer, which is right
+// while a network is one party's handful of machines (see docs/architecture.md →
+// Replication cluster size — the control network replicates to the whole party by intent,
+// and "On a strand, the machine count is not a party count"). If these meshes ever get
+// large, filter the pushed set by FRET cohort responsibility for each block id instead of
+// pushing everything.
+//
+// NOTE: this lives in cadre-core only because `../optimystic` is read-only from this repo.
+// If Optimystic ever grows a cohort-join catch-up of its own, delete this module rather
+// than running two.
+//
+// Membership — the one place the two networks differ, expressed as the optional
+// `authorizePeer` dep:
+//
+// - STRAND networks pass none. Anything that speaks the strand's own block-transfer protocol
+//   already receives cohort replicas of new commits, so pushing older blocks to it exposes
+//   nothing new; adding a gate would diverge from what ordinary replication already does
+//   there.
+// - The CONTROL network MUST pass one, because that argument does not carry over: its
+//   inbound connection gate deliberately admits non-members in several states (an
+//   un-enrolled node taking its seed, an open enrollment window, an outstanding
+//   invitation, configured bootstrap/relay peers — see docs/architecture.md, the
+//   control-network inbound connection gate). Pushing the whole control store to such a
+//   peer would hand a stranger the party's entire membership, addresses and strand list.
+//   `CadreNode` passes `isAuthorizedMember`; the gate is consulted at PUSH time, not at
+//   schedule time, and fails closed on a thrown check. A denied run is not memoized, so
+//   the peer is retried on its next `peer:identify` (a reconnect re-runs identify) — or
+//   sooner, via `scheduleConnectedPeers()` on a membership change (the production join order is
+//   connect-then-authorize, so the denial at dial time is the expected first pass).
+//
+// Both per-push deadlines are DERIVED from the declared link round trip
+// (`link-budget.ts`), not fixed milliseconds. That is not tuning — it is what makes this
+// module work at all over a relay. Opening a relayed connection costs a fixed number of
+// exchanges, so the 3000 ms this catch-up used to allow its dial could never finish one
+// above 375 ms of one-way link delay: the catch-up existed so that a machine which joined
+// after blocks were committed physically ends up holding them, and through a relay on any
+// link slow enough to matter it had never once managed to. The reproduction is
+// `packages/integration-tests/src/scenarios/relayed-dial-cost-by-latency.integration.ts`.
+//
+// It does NOT work at every speed. Above the declared link, libp2p's own dial and
+// inbound-upgrade limits — derived from the same declaration — abandon the connection before
+// any deadline here is consulted, and the listener's one makes that failure look like an absent
+// peer rather than a timeout. That is also what happens when the machine being caught up
+// declared a FASTER link than this one, since its limit is the listener's. See
+// `link-budget.ts` ("libp2p's own two limits", "What still fails at the supported link").
+//
+// RETRY, and why it backs off. A run whose PUSH FAILED — the transport threw, which is what
+// a dial or response deadline expiring looks like here — re-arms on a doubling backoff
+// (`retryBackoffMs` to `maxRetryBackoffMs`), and a `peer:identify` arriving while that wait
+// is outstanding is DROPPED rather than collapsing it back to the debounce. Both halves are
+// load-bearing:
+//
+// - Without the re-arm, a transient push failure over a stable connection left the peer
+//   partially copied until its next reconnect (read repair still covered reads meanwhile).
+// - Without dropping churn inside the wait, a peer that cannot be reached at all is
+//   re-dialled on every `peer:identify` forever. That is not hypothetical: the 2026-09-26
+//   relayed reproduction shows one peer re-opening a connection about every 14.5 s for a
+//   200-second run, each event starting a catch-up whose dial could not possibly finish.
+//   The connection kept re-appearing because Optimystic's own block-transfer push path
+//   budgets its dial at `transferTimeoutMs ?? 30000` — the one dial budget in the stack
+//   above the measured relayed setup cost — so it succeeded where this one could not.
+//
+// A NON-CLEAN run is NOT on its own enough to re-arm, and the distinction is what keeps the
+// backoff from becoming a worse problem than the one it fixes. `clean` also goes false for
+// outcomes no amount of retrying can change: the membership gate DENIED the peer; this
+// network's raw storage implements no `listBlockIds`, so the catch-up is inert; or the
+// receiver reported blocks in `missing`, which it does per block for a payload it cannot
+// parse and for a revision whose retained commit proof this node does not hold (a push
+// carrying no proof is refused outright by a receiver running the default
+// `requirePushCertificate: true`, and an unretained proof is the ordinary case — see
+// {@link Chunk.proofs}). Re-arming on those would re-push the WHOLE store to that peer every
+// `maxRetryBackoffMs` for as long as the node runs, and report the failure below as a link
+// budget problem when nothing about the link is wrong. All three still leave the peer
+// un-memoized, so its next `peer:identify` retries — the behaviour that predates the
+// backoff, and the right one for a verdict rather than a timeout. Denial in particular is
+// re-driven on purpose by `scheduleConnectedPeers()` the moment the membership commit lands,
+// because the control network's join order is connect-then-authorize.
+//
+// A re-arm is also skipped when the peer is no longer CONNECTED. This module's trigger is
+// `peer:identify`, so a peer that went away already has one: re-arming instead would leave
+// a machine dialing a peer it cannot see once a minute for the rest of its uptime.
+//
+// After `PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES` consecutive failures one `console.warn`
+// names the peer and the budget, because otherwise a machine on a too-slow link says
+// nothing at all: the connection simply never appears and only a DEBUG log mentions a dial
+// timeout.
+//
+// NOTE: the protocol check happens once — either when a connection's `peer:identify` fires,
+// or, for a peer already connected when this instance starts, when `scheduleConnectedPeers()`
+// reads its protocols back from the peer store — and is never repeated once a peer is
+// scheduled. A peer that stops speaking the protocol after being scheduled (should that ever
+// happen) is not de-scheduled; `runCatchUp`'s own unreachable-peer bail is what would catch
+// that case instead. `peer:identify` also fires on every identify-push from a connected peer
+// (address or protocol changes), which only re-enters the debounce — cheap, and dropped
+// outright for caught-up peers and peers inside a backoff wait.
+//
+// NOTE: a same-network peer whose identify never completes on a connection (the identify
+// stream times out over a slow relayed link, say) is not caught up until a later identify
+// succeeds — a reconnect, an identify-push, or on the control network a membership-change
+// re-drive. Accepted: every db-p2p node runs identify, and read repair still covers reads
+// meanwhile; if relayed links ever routinely outlast identify's timeout, raise that timeout
+// in the node builder rather than scheduling unidentified peers here.
+//
+// NOTE: a peer that disconnects INSIDE a backoff wait still costs the one attempt that wait
+// was already armed for — the connectivity check is made when the re-arm is decided, not when
+// the timer fires, and this module subscribes to `peer:identify` only. One dial, not a
+// recurring one; if a node ever holds many transient peers, subscribe to `connection:close`
+// and clear the pending timer there.
+
+const log = debug('sereus:cadre:peer-join-backfill');
+
+/**
+ * The block-transfer protocol's hard cap on one length-prefixed message. Mirrors
+ * `MAX_BLOCK_MESSAGE_BYTES` in `@optimystic/db-p2p`'s `protocol-limits.ts` (8 MiB), which
+ * that package does not re-export from its index — keep the two in sync if upstream ever
+ * changes it or starts exporting it.
+ */
+export const MAX_BLOCK_MESSAGE_BYTES = 8 * 1024 * 1024;
+
+/**
+ * Whether an identified peer's protocol list names this network's own block-transfer
+ * protocol — the one gate that decides whether {@link PeerJoinBackfill} schedules a peer at
+ * all. A peer that lacks it (a bare circuit relay, a bootstrap node, a stranger the control
+ * network's inbound gate admitted without membership) cannot receive a push no matter how
+ * long it is dialed, since the protocol id is namespaced per network. Exported because a
+ * later strand peer-address ticket's identify-driven observation needs the identical test.
+ */
+export function speaksBlockTransfer(protocols: readonly string[], protocolPrefix: string): boolean {
+  return protocols.includes(buildBlockTransferProtocol(protocolPrefix));
+}
+
+/**
+ * Consecutive failed catch-up runs against ONE peer before the module says so on
+ * `console.warn` rather than only in its DEBUG log. At the default backoff that is reached
+ * about 35 seconds in (5 s + 10 s + 20 s), which is late enough to have ruled out a blip and
+ * early enough to be the first thing an operator sees about a machine that is not catching up.
+ */
+export const PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES = 3;
+
+/** Tuning for the per-peer catch-up. Every field optional; defaults in {@link DEFAULT_PEER_JOIN_BACKFILL}. */
+export interface PeerJoinBackfillConfig {
+  /** Default true. False disables the catch-up entirely (the pre-existing behaviour). */
+  enabled?: boolean;
+  /** Settle time after a connection opens before catching that peer up, ms. Default 1000. */
+  debounceMs?: number;
+  /** Ceiling on blocks copied in one catch-up. Default 10_000. Reaching it is LOGGED, never silent. */
+  maxBlocks?: number;
+  /** Soft byte budget per push message. Default 1 MiB. Protocol hard cap is {@link MAX_BLOCK_MESSAGE_BYTES} (8 MiB). */
+  maxChunkBytes?: number;
+  /** Max blocks per push message. Default 64. */
+  maxChunkBlocks?: number;
+  /**
+   * Per-push dial deadline, ms. Default {@link peerJoinPushBudget}'s `dialTimeoutMs` — four link
+   * round trips at the declared link, 14 000 ms as shipped. NOT a fixed number: a relayed dial
+   * costs a fixed number of exchanges, so a host on a slower link moves this (and every other
+   * cadre dial budget) by declaring `NetworkConfig.linkRoundTripMs`. Naming it here still wins
+   * over the derived value — `link-budget.ts` has the counts and the measurement.
+   */
+  dialTimeoutMs?: number;
+  /**
+   * Per-push response deadline, ms. Default {@link peerJoinPushBudget}'s `responseTimeoutMs` —
+   * two link round trips at the declared link plus a transfer allowance for the chunk's own
+   * bytes, 13 000 ms as shipped. Derived differently from {@link dialTimeoutMs} because it
+   * bounds a data transfer over a connection that is already open, not a dial.
+   */
+  responseTimeoutMs?: number;
+  /**
+   * First wait before re-running a catch-up whose last run did not complete cleanly, ms.
+   * Default 5000. Doubles per consecutive failure up to {@link maxRetryBackoffMs}.
+   */
+  retryBackoffMs?: number;
+  /** Ceiling on that doubling wait, ms. Default 60_000. */
+  maxRetryBackoffMs?: number;
+}
+
+/** The resolved defaults every {@link PeerJoinBackfill} starts from. */
+export const DEFAULT_PEER_JOIN_BACKFILL: Required<PeerJoinBackfillConfig> = {
+  enabled: true,
+  debounceMs: 1000,
+  maxBlocks: 10_000,
+  maxChunkBytes: 1024 * 1024,
+  maxChunkBlocks: 64,
+  // Derived at the DEFAULT declared link, for a `PeerJoinBackfill` built without a host's
+  // declaration (this module's own tests, an embedder driving it directly). The two production
+  // construction sites pass `peerJoinPushBudget(network?.linkRoundTripMs)` so a host that
+  // declares a slower link moves both — see `link-budget.ts`.
+  ...peerJoinPushBudget(),
+  retryBackoffMs: 5000,
+  maxRetryBackoffMs: 60_000
+};
+
+/**
+ * The one capability the catch-up needs from a transfer client. Structural (rather than
+ * the concrete `BlockTransferClient`) so unit tests can capture pushes without dialing
+ * libp2p — see {@link PeerJoinBackfillDeps.createPushClient}.
+ */
+export type PeerJoinBackfillPushClient = Pick<BlockTransferClient, 'pushBlocks'>;
+
+export interface PeerJoinBackfillDeps {
+  /** Log tag naming which network this catch-up serves (a strand id, or `control-<partyId>`). */
+  label: string;
+  /** The libp2p node of the network being caught up — source of connection events and peer ids. */
+  libp2p: Libp2p;
+  /** `node.keyNetwork`, the IPeerNetwork BlockTransferClient dials through. */
+  peerNetwork: IPeerNetwork;
+  /** This network's own raw block store — the same instance handed to the libp2p node. */
+  storage: IRawStorage;
+  /** Must equal the prefix the receiver registered its handler under: `/optimystic/<networkName>`. */
+  protocolPrefix: string;
+  /**
+   * Membership gate, consulted at PUSH time (fails closed on a throw). Absent = push to
+   * every connected peer — correct for strand networks, NEVER for the control network;
+   * see the module comment for why the strand no-gate argument does not carry over.
+   */
+  authorizePeer?: (peerId: string) => Promise<boolean>;
+  /**
+   * Test seam: build the per-peer push client. Defaults to a real
+   * `BlockTransferClient` over {@link peerNetwork} + {@link protocolPrefix}.
+   */
+  createPushClient?: (peerId: PeerId) => PeerJoinBackfillPushClient;
+}
+
+/** What one peer's catch-up actually did. Returned for tests and logged at the end of each run. */
+export interface PeerJoinBackfillResult {
+  /** Blocks offered: had a committed `latest` AND materialized content locally. */
+  offered: number;
+  /** Blocks the remote reported it persisted. */
+  accepted: number;
+  /** Block ids the remote reported in `missing` (parse or persist failure on its side). */
+  rejected: string[];
+  /** Skipped: metadata has no `latest` (pending-only — not yet a durability claim here). */
+  uncommitted: number;
+  /** Skipped: `latest` exists but no materialized block is stored for that actionId. */
+  unmaterialized: number;
+  /** Not attempted because `maxBlocks` was reached. */
+  capped: number;
+  /** Skipped: a single block whose wire size alone exceeds {@link MAX_BLOCK_MESSAGE_BYTES}. */
+  oversized: string[];
+  /** True when the `authorizePeer` gate refused (or threw) — nothing was pushed, peer not memoized. */
+  denied: boolean;
+}
+
+function emptyResult(): PeerJoinBackfillResult {
+  return { offered: 0, accepted: 0, rejected: [], uncommitted: 0, unmaterialized: 0, capped: 0, oversized: [], denied: false };
+}
+
+/** One in-flight push message being accumulated. */
+interface Chunk {
+  ids: string[];
+  buffers: Uint8Array[];
+  meta: Record<string, { rev: number; actionId: ActionId }>;
+  /**
+   * The cohort commit proof retained for each block's pushed revision, where one exists.
+   *
+   * A receiver running the default `requirePushCertificate: true` REJECTS a block pushed
+   * without one, so this is what makes peer-join catch-up land at all — a push carrying only
+   * `meta` is the legacy shape and is refused. Sparse deliberately: a block whose proof was
+   * never retained is still offered with its meta, which is the only thing that lets a
+   * receiver migrating with the flag off land the replica at this node's revision rather than
+   * a fabricated rev 1. Upstream's rule is the other direction and holds here by construction:
+   * a proof is never attached WITHOUT its meta, because both are written from the same
+   * `latest` in one place.
+   */
+  proofs: Record<string, BlockCommitProof>;
+  bytes: number;
+}
+
+/**
+ * Base64 wire size of a raw buffer — `pushBlocks` base64-encodes each block into the JSON
+ * request, so the protocol cap must be judged against the encoded size, not the raw bytes.
+ */
+function base64WireBytes(rawBytes: number): number {
+  return Math.ceil(rawBytes / 3) * 4;
+}
+
+/**
+ * Copies every block in one network's own raw store to each peer its libp2p node connects
+ * to, so a peer that joined the network after blocks were committed still ends up holding
+ * them physically (the receiver persists each push via `saveReplicatedBlock`, which is
+ * monotonic and idempotent — crossing pushes from both ends cannot regress a revision).
+ *
+ * Best-effort throughout: nothing here throws into a libp2p event handler or into a
+ * runtime bring-up; a failed chunk is logged and the run continues. A peer is marked
+ * fully caught up ONLY after a run with no thrown chunk and an empty `missing` list, so
+ * the next `peer:identify` from a peer whose catch-up failed retries it.
+ */
+export class PeerJoinBackfill {
+  private readonly config: Required<PeerJoinBackfillConfig>;
+  private readonly createPushClient: (peerId: PeerId) => PeerJoinBackfillPushClient;
+  private started = false;
+  private stopped = false;
+  /** Peers fully caught up this runtime (never retried until the runtime is rebuilt). */
+  private readonly done = new Set<string>();
+  /** Peers with a catch-up currently running (suppresses concurrent duplicates). */
+  private readonly inFlight = new Set<string>();
+  /**
+   * Peers whose (re-)schedule arrived while their OWN run was in flight, replayed once
+   * that run finishes. Load-bearing for the gated path: the gate's authorization check is
+   * a control-database read, so the window between it and the run's end is wide enough for
+   * the very membership commit that would have authorized the peer to land inside it —
+   * dropping that schedule (rather than deferring it) leaves the denied peer waiting for a
+   * reconnect, which is exactly what the re-arm exists to avoid.
+   */
+  private readonly rearmAfterFlight = new Set<string>();
+  /** Pending per-peer timers — the debounce, or a backoff re-arm. Cleared on stop. */
+  private readonly timers = new Map<string, ReturnType<typeof setTimeout>>();
+  /** Consecutive non-clean runs per peer, cleared when one finishes cleanly. */
+  private readonly failures = new Map<string, number>();
+  /**
+   * Epoch ms before which a peer's catch-up must not be re-run, set alongside a backoff timer
+   * that will run it. Its presence is what makes `peer:identify` churn free: a schedule
+   * arriving inside the wait is dropped, rather than collapsing the backoff to the debounce.
+   */
+  private readonly retryAfter = new Map<string, number>();
+  /** Peers already reported on `console.warn`; reset when a run finally lands cleanly. */
+  private readonly warned = new Set<string>();
+  private readonly onPeerIdentify: (evt: CustomEvent<IdentifyResult>) => void;
+  private loggedNoListBlockIds = false;
+
+  constructor(private readonly deps: PeerJoinBackfillDeps, config?: PeerJoinBackfillConfig) {
+    this.config = { ...DEFAULT_PEER_JOIN_BACKFILL, ...config };
+    this.createPushClient = deps.createPushClient
+      ?? ((peerId) => new BlockTransferClient(peerId, deps.peerNetwork, deps.protocolPrefix));
+    this.onPeerIdentify = (evt) => {
+      const { peerId, protocols } = evt.detail;
+      if (speaksBlockTransfer(protocols, this.deps.protocolPrefix)) {
+        this.schedulePeer(peerId);
+      }
+    };
+  }
+
+  /**
+   * Subscribe to `peer:identify` AND schedule a catch-up for already-connected peers whose
+   * stored protocols already name this network's block-transfer protocol — a peer that does
+   * not speak it (a bare circuit relay, a bootstrap node) is never scheduled at all. See
+   * {@link speaksBlockTransfer} and the module comment.
+   */
+  start(): void {
+    if (this.started || this.stopped || !this.config.enabled) return;
+    this.started = true;
+    this.deps.libp2p.addEventListener('peer:identify', this.onPeerIdentify);
+    // A runtime rebuilt over live connections (resumeStrand) never sees their peer:identify —
+    // it already fired before this instance existed, but its result is persisted in the peer
+    // store — so walk what is already connected once.
+    void this.scheduleConnectedPeers().then((scheduled) => {
+      log('[%s] started (%d peer(s) already connected)', this.deps.label, scheduled);
+    });
+  }
+
+  /** Unsubscribe, clear timers; in-flight runs observe the stopped flag and bail. */
+  stop(): void {
+    if (this.stopped) return;
+    this.stopped = true;
+    if (this.started) {
+      this.deps.libp2p.removeEventListener('peer:identify', this.onPeerIdentify);
+    }
+    for (const timer of this.timers.values()) {
+      clearTimeout(timer);
+    }
+    this.timers.clear();
+    this.rearmAfterFlight.clear();
+    this.retryAfter.clear();
+    log('[%s] stopped', this.deps.label);
+  }
+
+  /**
+   * (Re-)schedule a debounced catch-up for every currently-connected peer whose protocols, as
+   * already stored in the libp2p peer store, name this network's block-transfer protocol. A
+   * peer not yet in the peer store — its identify has not finished — is left to the
+   * `peer:identify` handler rather than dialed blind; one whose stored protocols are known
+   * and lack ours is skipped outright. Idempotent and cheap otherwise (caught-up peers are
+   * skipped before any timer is set, and a peer whose run is in flight is deferred to the end
+   * of that run rather than dropped). Driven by {@link start}, and — on a gated network — by
+   * the embedder whenever membership changes, so a peer whose first pass was denied (connected
+   * before it was authorized: the production join order) is retried without waiting for a
+   * reconnect. Returns how many distinct peers were considered (for the start() log) — a peer
+   * holding several connections is one peer, and is scheduled once.
+   */
+  async scheduleConnectedPeers(): Promise<number> {
+    if (this.stopped) return 0;
+    const peers = new Map<string, PeerId>();
+    for (const connection of this.deps.libp2p.getConnections()) {
+      peers.set(connection.remotePeer.toString(), connection.remotePeer);
+    }
+    await Promise.all([...peers.values()].map((peerId) => this.scheduleIfKnownToSpeak(peerId)));
+    return peers.size;
+  }
+
+  /** Schedules `peerId` only if the peer store already knows it speaks this network's protocol. */
+  private async scheduleIfKnownToSpeak(peerId: PeerId): Promise<void> {
+    let protocols: string[];
+    try {
+      ({ protocols } = await this.deps.libp2p.peerStore.get(peerId));
+    } catch (error) {
+      // NotFoundError: identify has not finished. Leave it to the `peer:identify` handler
+      // rather than dialing blind. Anything else is unexpected — log it and skip the peer.
+      if ((error as Error).name !== 'NotFoundError') {
+        log('[%s] peer store read for %s failed — not scheduling: %o', this.deps.label, peerId.toString(), error);
+      }
+      return;
+    }
+    if (speaksBlockTransfer(protocols, this.deps.protocolPrefix)) {
+      this.schedulePeer(peerId);
+    }
+  }
+
+  /** Debounced entry point for connection churn: one run per peer per settle window. */
+  private schedulePeer(peerId: PeerId): void {
+    const key = peerId.toString();
+    if (this.stopped || this.done.has(key)) return;
+    if (this.inFlight.has(key)) {
+      // Defer rather than drop: the running pass may already be past its gate check (or
+      // past the block this schedule was meant to carry), so replay it when that pass ends.
+      this.rearmAfterFlight.add(key);
+      return;
+    }
+    // A peer whose last run failed is already re-armed on a backoff timer that will run it, so
+    // DROP this schedule instead of shortening the wait — otherwise a peer that cannot be
+    // reached is re-dialled on every peer:identify forever (see the module comment's retry
+    // paragraph). A denied run never gets here: it does not set a backoff.
+    const retryAt = this.retryAfter.get(key);
+    if (retryAt !== undefined && Date.now() < retryAt) return;
+    const existing = this.timers.get(key);
+    if (existing) clearTimeout(existing);
+    this.timers.set(key, setTimeout(() => {
+      this.timers.delete(key);
+      void this.catchUpPeer(peerId);
+    }, this.config.debounceMs));
+  }
+
+  /**
+   * Run one peer's catch-up now, bypassing the debounce. Never rejects. Returns an
+   * all-zero result without pushing when the peer is already caught up, already being
+   * caught up, or this backfill is stopped.
+   */
+  async catchUpPeer(peerId: PeerId): Promise<PeerJoinBackfillResult> {
+    const key = peerId.toString();
+    if (this.stopped || this.done.has(key) || this.inFlight.has(key)) {
+      return emptyResult();
+    }
+    this.inFlight.add(key);
+    try {
+      const { result, clean, pushFailed } = await this.runCatchUp(peerId);
+      // NOTE: a run that hit `maxBlocks` is still "clean" and still memoizes the peer, so
+      // the tail past the ceiling never reaches it. Deliberate: enumeration is not
+      // resumable, so not memoizing would re-push the same prefix on every reconnect
+      // without ever advancing. Loud in the log below (capped > 0). If a store can
+      // realistically exceed maxBlocks, the fix is a resumable cursor, not either policy.
+      if (clean && !this.stopped) {
+        this.done.add(key);
+        this.failures.delete(key);
+        this.retryAfter.delete(key);
+        this.warned.delete(key);
+      } else if (pushFailed && !this.stopped && this.started && this.isConnected(peerId)) {
+        // `pushFailed`, not `!clean`: a denial, an inert store and a receiver's per-block
+        // rejection are all verdicts a retry cannot change, and re-arming on them would
+        // re-push the whole store to that peer forever. See the module comment.
+        //
+        // Gated on `started` too: the re-arm exists to REPLACE a peer:identify-driven retry,
+        // so a caller driving `catchUpPeer` by hand against a backfill that was never started
+        // owns its own retry policy and must not be left holding a background timer.
+        this.scheduleRetryWithBackoff(peerId);
+      }
+      log('[%s] catch-up peer=%s offered=%d accepted=%d rejected=%d uncommitted=%d unmaterialized=%d capped=%d oversized=%d denied=%s done=%s',
+        this.deps.label, key, result.offered, result.accepted, result.rejected.length,
+        result.uncommitted, result.unmaterialized, result.capped, result.oversized.length, result.denied, clean);
+      return result;
+    } catch (error) {
+      // runCatchUp already contains a per-chunk catch; this guards the enumeration and
+      // metadata reads too — a backfill fault must never surface through a libp2p event.
+      log('[%s] catch-up peer=%s failed: %o', this.deps.label, key, error);
+      return emptyResult();
+    } finally {
+      this.inFlight.delete(key);
+      // Replay a schedule that arrived mid-run. `schedulePeer` re-checks `done`, so a run
+      // that finished clean re-arms nothing; a denied or failed one gets its retry.
+      if (this.rearmAfterFlight.delete(key)) {
+        this.schedulePeer(peerId);
+      }
+    }
+  }
+
+  /** Whether this network's libp2p node still holds a connection to that peer. */
+  private isConnected(peerId: PeerId): boolean {
+    return this.deps.libp2p.getConnections(peerId).length > 0;
+  }
+
+  /**
+   * Re-arm one peer's catch-up after a run whose push failed, on a wait that doubles
+   * per consecutive failure up to `maxRetryBackoffMs`. Sets {@link retryAfter} alongside the
+   * timer, which is what makes `peer:identify` churn inside the wait free.
+   */
+  private scheduleRetryWithBackoff(peerId: PeerId): void {
+    const key = peerId.toString();
+    const failures = (this.failures.get(key) ?? 0) + 1;
+    this.failures.set(key, failures);
+    const delayMs = Math.min(
+      this.config.retryBackoffMs * 2 ** (failures - 1),
+      this.config.maxRetryBackoffMs
+    );
+    this.retryAfter.set(key, Date.now() + delayMs);
+    const existing = this.timers.get(key);
+    if (existing) clearTimeout(existing);
+    this.timers.set(key, setTimeout(() => {
+      this.timers.delete(key);
+      this.retryAfter.delete(key);
+      void this.catchUpPeer(peerId);
+    }, delayMs));
+    log('[%s] catch-up peer=%s failed %d time(s) in a row; retrying in %dms',
+      this.deps.label, key, failures, delayMs);
+    this.warnPersistentFailure(key, failures);
+  }
+
+  /**
+   * Say ONCE per peer, outside the DEBUG log, that its catch-up is not landing. Without this a
+   * machine on a link too slow for a relayed dial produces no statement that anything is wrong:
+   * the peer never appears to hold the blocks, and the only trace is a DEBUG line naming a dial
+   * timeout. Reset when a run finally lands cleanly, so a peer that recovers can report again.
+   */
+  private warnPersistentFailure(key: string, failures: number): void {
+    if (failures < PEER_JOIN_BACKFILL_WARN_AFTER_FAILURES || this.warned.has(key)) return;
+    this.warned.add(key);
+    console.warn(
+      `[cadre:${this.deps.label}] peer-join block catch-up to peer ${key} has failed ${failures} times in a row `
+      + `(dial budget ${this.config.dialTimeoutMs}ms, response budget ${this.config.responseTimeoutMs}ms). `
+      + 'That peer may not be holding blocks committed before it joined. If it is reachable only through a relay, '
+      + 'these budgets and libp2p\'s own connection limits are derived from network.linkRoundTripMs (see link-budget.ts); '
+      + 'a link slower than that declaration, or a peer that declared a faster one, cannot open a relayed connection at all.'
+    );
+  }
+
+  /**
+   * The actual copy. `clean` = every chunk pushed and the remote persisted every block.
+   * `pushFailed` = at least one push THREW, which is the only non-clean outcome a retry can
+   * change; the others are verdicts (denied, inert store, blocks the receiver refused).
+   */
+  private async runCatchUp(peerId: PeerId): Promise<{ result: PeerJoinBackfillResult; clean: boolean; pushFailed: boolean }> {
+    const result = emptyResult();
+    const { storage } = this.deps;
+
+    // The membership gate, judged at push time (never at schedule time — authorization can
+    // change during the debounce window in either direction). Fails CLOSED on a throw: a
+    // gate that cannot answer must not leak the store. A denied run is not clean, so the
+    // peer is retried later rather than memoized as caught up.
+    if (this.deps.authorizePeer) {
+      let authorized = false;
+      try {
+        authorized = await this.deps.authorizePeer(peerId.toString());
+      } catch (error) {
+        log('[%s] authorizePeer(%s) threw — treating as denied: %o', this.deps.label, peerId.toString(), error);
+      }
+      if (!authorized) {
+        result.denied = true;
+        return { result, clean: false, pushFailed: false };
+      }
+    }
+
+    if (!storage.listBlockIds) {
+      if (!this.loggedNoListBlockIds) {
+        this.loggedNoListBlockIds = true;
+        log('[%s] raw storage does not implement listBlockIds(); backfill is inert', this.deps.label);
+      }
+      return { result, clean: false, pushFailed: false };
+    }
+
+    const client = this.createPushClient(peerId);
+    const encoder = new TextEncoder();
+    let chunk: Chunk = { ids: [], buffers: [], meta: {}, proofs: {}, bytes: 0 };
+    let chunkFailed = false;
+    let anyChunkDelivered = false;
+    /**
+     * A peer that has answered no push at all, and failed one, is not reachable on this
+     * protocol — abandon rather than spend a dial timeout per remaining chunk. Once ONE
+     * push has been answered the peer demonstrably speaks it, so a later failure is a
+     * transient blip and the rest of the store is still worth pushing.
+     */
+    const peerUnreachable = (): boolean => chunkFailed && !anyChunkDelivered;
+
+    const flush = async (): Promise<void> => {
+      if (chunk.ids.length === 0) return;
+      const { ids, buffers, meta, proofs } = chunk;
+      chunk = { ids: [], buffers: [], meta: {}, proofs: {}, bytes: 0 };
+      try {
+        // ONE certification value rather than two arguments, so a caller cannot pair a proof
+        // for one revision with meta for another — both were written from the same `latest`.
+        const response = await client.pushBlocks(ids, buffers, 'replication',
+          { blockMeta: meta, blockProofs: proofs }, {
+          dialTimeoutMs: this.config.dialTimeoutMs,
+          responseTimeoutMs: this.config.responseTimeoutMs
+        });
+        anyChunkDelivered = true;
+        const missing = new Set(response.missing);
+        for (const id of ids) {
+          if (missing.has(id)) {
+            result.rejected.push(id);
+          } else {
+            result.accepted += 1;
+          }
+        }
+        if (response.missing.length > 0) {
+          log('[%s] peer=%s rejected %d block(s): %o', this.deps.label, peerId.toString(), response.missing.length, response.missing);
+        }
+      } catch (error) {
+        chunkFailed = true;
+        log('[%s] push to peer=%s failed for %d block(s): %s', this.deps.label, peerId.toString(), ids.length, (error as Error).message);
+      }
+    };
+
+    for await (const blockId of storage.listBlockIds()) {
+      if (this.stopped) break;
+      if (result.offered >= this.config.maxBlocks) {
+        // Past the ceiling: count what is left (id enumeration only — no more reads or
+        // pushes) so the cap is loud in the end-of-run log rather than silent.
+        result.capped += 1;
+        continue;
+      }
+
+      const metadata = await storage.getMetadata(blockId);
+      const latest = metadata?.latest;
+      if (!latest) {
+        result.uncommitted += 1;
+        continue;
+      }
+      // NOTE: a block whose latest revision is a DELETE materializes to nothing (a
+      // tombstone), so it is skipped here — a peer that joins after such a delete never
+      // receives the tombstone from this path and relies on read repair for it. Fine
+      // while nothing deletes whole blocks pre-join; if the whole-store coverage gate in
+      // strand-membership-closed-strand-e2e ever reports a tombstone residue, teach this
+      // to push the promoted delete transform instead of skipping.
+      const block = await storage.getMaterializedBlock(blockId, latest.actionId);
+      if (!block) {
+        result.unmaterialized += 1;
+        log('[%s] block=%s has latest rev %d but no materialized content; skipped', this.deps.label, blockId, latest.rev);
+        continue;
+      }
+
+      const buffer = encoder.encode(JSON.stringify(block));
+      // NOTE: MAX_BLOCK_MESSAGE_BYTES caps the whole framed request, not one block, and
+      // this test ignores the request envelope (ids array, blockMeta, JSON punctuation).
+      // A lone block within a few KiB of the cap therefore still ships and is rejected by
+      // the receiver's length-prefix decoder — a permanent chunk failure for that peer. No
+      // block comes close today; if one ever can, subtract a measured envelope allowance
+      // here rather than raising the cap.
+      if (base64WireBytes(buffer.length) >= MAX_BLOCK_MESSAGE_BYTES) {
+        result.oversized.push(blockId);
+        log('[%s] block=%s is %d bytes (%d on the wire) — exceeds the %d-byte protocol cap alone; skipped',
+          this.deps.label, blockId, buffer.length, base64WireBytes(buffer.length), MAX_BLOCK_MESSAGE_BYTES);
+        continue;
+      }
+
+      // Flush before adding when this block would overflow either budget. A single block
+      // larger than maxChunkBytes still ships — alone, in its own chunk.
+      if (chunk.ids.length > 0
+        && (chunk.ids.length + 1 > this.config.maxChunkBlocks || chunk.bytes + buffer.length > this.config.maxChunkBytes)) {
+        await flush();
+        if (this.stopped || peerUnreachable()) break;
+      }
+      chunk.ids.push(blockId);
+      chunk.buffers.push(buffer);
+      chunk.meta[blockId] = { rev: latest.rev, actionId: latest.actionId };
+      // Read the proof for EXACTLY the revision being pushed, from the same `latest` the meta
+      // above was written from. Absent is normal (nothing retained a proof for that revision);
+      // the block still ships with its meta and the receiver decides.
+      const proof = await storage.getBlockProof(blockId, latest.rev);
+      if (proof) {
+        chunk.proofs[blockId] = proof;
+      }
+      chunk.bytes += buffer.length;
+      result.offered += 1;
+    }
+
+    if (!this.stopped && !peerUnreachable()) {
+      await flush();
+    }
+
+    if (result.capped > 0) {
+      log('[%s] peer=%s catch-up CAPPED at %d blocks; %d block id(s) not attempted',
+        this.deps.label, peerId.toString(), this.config.maxBlocks, result.capped);
+    }
+
+    const clean = !chunkFailed && result.rejected.length === 0;
+    return { result, clean, pushFailed: chunkFailed };
+  }
+}