toolroll 0.9.52 → 0.9.53

package/dist/operate.js

@@ -42,7 +42,7 @@ import { packageVersion, startMonitoring, targetOf } from "./monitoring.js";
 import { origin, readMonitoring } from "./monitoring-settings.js";
 import { logEvent } from "./log.js";
 import { FLOW_EVERY_MS, flowHousekeeping, moveCards, moveCardsAfter } from "./flow-cadence.js";
-import { readHooksBase } from "./flow-triggers.js";
+import { readHooksBase, scheduleFromWords } from "./flow-triggers.js";
 import { sendTeammateSummaries } from "./teammate-admin.js";
 import { runRequestedUndos, sendTeammateWeeklies } from "./teammate-week.js";
 import { refreshConnections } from "./mcp-connect.js";
@@ -83,10 +83,10 @@ import { taskReviewBrief, renderReviewBrief } from "./task-review-brief.js";
  *
  * **Nothing ever prompts.** There is no terminal on the other end at 3am.
  */
-import { homedir, hostname, tmpdir } from "node:os";
+import { homedir, hostname } from "node:os";
 import { basename, dirname, join, resolve } from "node:path";
 import { releaseStalledUpdate, waitingUpdate } from "./toolroll-update.js";
-import { openStore, openStoreNoMigrate, CONCURRENT_WRITER_WAIT_MS, isDatabaseBusy, databasePath, BUILT_IN, DEFAULT_ACTOR, parseCapabilityKey, verifiedAuthor, contestantProfileOf, isDigestTime, RESULT_SCREENSHOTS, } from "./store.js";
+import { openStore, openStoreNoMigrate, CONCURRENT_WRITER_WAIT_MS, isDatabaseBusy, databasePath, BUILT_IN, DEFAULT_ACTOR, parseCapabilityKey, verifiedAuthor, isDigestTime, RESULT_SCREENSHOTS, } from "./store.js";
 import { createHash, randomBytes, randomInt, randomUUID } from "node:crypto";
 import { authorizePlanUnderMode } from "./plan-auto.js";
 import { ghDispatchAdapter, mirrorTaskId, syncPass } from "./sync.js";
@@ -94,8 +94,8 @@ import { LIVE_FILE_BOUND, LIVE_RETAIN_MS, sweepLiveLogs } from "./live.js";
 import { configPath, addRepos, removeRepos, updateRepos, loadRepos, loadProjectRegistry, updateProjectRegistry } from "./repos.js";
 import { deleteProject, holdingsWords, projectHoldings, projectRunning } from "./project-delete.js";
 import { pushPass } from "./push.js";
-import { chmodSync, closeSync, constants as fsConstants, existsSync, fstatSync, fsyncSync, openSync, readFileSync, readSync, realpathSync, renameSync, unlinkSync, writeSync, writeFileSync, mkdirSync } from "node:fs";
-import { BRANCH_PREFIX, envTwins, envValue, existingOrFirst, namedPath, taskBranches } from "./names.js";
+import { chmodSync, closeSync, constants as fsConstants, existsSync, fstatSync, fsyncSync, openSync, readFileSync, readSync, realpathSync, renameSync, unlinkSync, writeSync, writeFileSync } from "node:fs";
+import { BRANCH_PREFIX, envTwins, envValue, existingOrFirst, taskBranches } from "./names.js";
 import { claimActor, currentActor, parseLeadToken, withActor } from "./actor.js";
 import { CLI_PASSWORD_SOURCE, withPasswordSource } from "./sign-in-guard.js";
 import { admissionRecorded, admitProject, commandLineActor } from "./project-admission.js";
@@ -166,17 +166,15 @@ import { tally, spendLine } from "./summary.js";
 import { bodyHashOf, describePublicationGrant, observeChecks, publicationBody, publishPass, sweepMerges, } from "./publish.js";
 import { checkPublishing, publishingOf, savePublishing, followPullRequests, newestPullRequestOf } from "./pull-request-flow.js";
 import { starterOf, startersFor, switchOnStarter } from "./flow-starters.js";
-import { acquire, acquireFallback, acquireIfReady, completeFenced, finalizeFailureFenced, finalizeMalformedFenced, finalizeParkFenced, finalizePlanFenced, finalizePlanFailureFenced, finalizeScoutFenced, finalizeScoutFailureFenced, heartbeat, release, reap, currentClaim, DEFAULT_LEASE_MS, SYNC_MAX_AGE_MS, acquireContinuation, } from "./claim.js";
+import { acquire, acquireIfReady, heartbeat, release, reap, currentClaim, DEFAULT_LEASE_MS, SYNC_MAX_AGE_MS, finalize, } from "./claim.js";
 import { disposeBuildOutcome, holdStaleApproval, regateTask } from "./dispose.js";
-import { attendedLivenessState } from "./liveness.js";
-import { HeldSessionCoordinator, sweepHeldOrphans } from "./held.js";
 import { proposeGrant, describeGrant, describeWithheld, permits, MUTATION_CLASSES, DEFAULT_MUTATIONS, } from "./grant.js";
 import { builtIn, guarded } from "./backend.js";
 import { register, authenticate, heartbeat as heartbeatRunner, isAlive, recoverDead, recoveredAnything, acquireWatchLeaseAuthed, heartbeatWatchLeaseAuthed, addRunnerReposAuthed, removeRunnerRepoAuthed, registerRunnerIfIdle, retireRunnerIfCurrent, normalizeRunnerName, validRunnerName, canonicalRepos, RUNNER_NAME_MAX, RUNNER_CAPACITY_MAX, parseRunnerCapacity, } from "./runner.js";
 import { mintCoordinator, revokeCoordinator, listCoordinators } from "./coordinator.js";
 import { serveMcp } from "./mcp.js";
 import { createInterface } from "node:readline";
-import { propose, approve, addApprover, authenticateAccount, authenticateApprover as passwordApprover, describeScope, approvalOf, hashToken as hashApproverToken, profileFromJson, fileAndSealUnderMode, modeFilingCoverage, acceptanceLinesToInput, parseAcceptanceCriteria, splitAcceptanceRubric, rubricIsPlaceholder, isCommitSha } from "./scope.js";
+import { propose, approve, addApprover, authenticateAccount, authenticateApprover as passwordApprover, describeScope, approvalOf, hashToken as hashApproverToken, fileAndSealUnderMode, modeFilingCoverage, acceptanceLinesToInput, parseAcceptanceCriteria, splitAcceptanceRubric, rubricIsPlaceholder, isCommitSha } from "./scope.js";
 import { presetTerms, modeTermsJson, modeDigestOf, modeTermsFromJson, modeWords, MODE_MAX_DAYS } from "./modes.js";
 import { WorktreePool } from "./worktree.js";
 import { requestTaskStop, resumeTaskStop, taskControlOf } from "./task-control.js";
@@ -186,25 +184,23 @@ import { readExecutorOf } from "./read-executor.js";
 import { parseWorkView } from "./workspace-ui.js";
 import { taskWorkSummaryOf } from "./work-summary.js";
 import { assignmentOf, assignmentBrief, syncAssignmentHandoffs } from "./assignment.js";
-import { approveRoutine, describeRoutine, fireRoutine, refreshRoutineAgents, routineAgentsState, routineDigestOf, validateRoutineTerms, ROUTINE_NAME, } from "./routine.js";
-import { fileTaskProposal, fileRoutineProposal, validateTaskText } from "./proposal.js";
+import { createScheduledFlow, describeSchedule, parseSchedule } from "./flow-schedule.js";
+import { fileTaskProposal, validateTaskText } from "./proposal.js";
 import { TEMPLATES, templateByName } from "./templates.js";
-import { planTournament, planComparison, contestNoun, jointApprovalDigest, admitContest, crossReadyBarrier, finalizeContestant, recoverContests, maybeAggregate as contestMaybeAggregate, sweepContestCleanup, escalateOverdueContests } from "./contest.js";
 import { isDirectChatProvider, isSubscriptionChatProvider, priceOf, PRICED_MODELS } from "./converse.js";
-import { resolvePhaseAgent, resolveScopeProfile, resolveScopeChain, resolveRouteCandidates, routeOfTask, INSTALLATION_SCOPE } from "./agentconfig.js";
-import { isRiskLevel, isTaskSize, legOf, projectRoute, riskConsequence, routeDigestOf, routeWords, RISK_LEVELS, TASK_SIZES, PHASES as ROUTE_PHASES } from "./phase-routing.js";
+import { resolvePhaseAgent, resolveScopeProfile, resolveRouteCandidates, routeOfTask, INSTALLATION_SCOPE } from "./agentconfig.js";
+import { isTaskSize, legOf, projectRoute, routeDigestOf, routeWords, TASK_SIZES, PHASES as ROUTE_PHASES } from "./phase-routing.js";
 import { observeProviderReadiness, reportProviderReadinessAuthed } from "./runner.js";
 import { parseDemoUrl, projectDemoUrl, saveProjectDemo } from "./project-demo.js";
 import { effectiveConcurrency, maySlotTake, parseProjectConcurrency, PROJECT_CONCURRENCY_DEFAULT, ProjectPasses, projectConcurrency, saveProjectConcurrency, savedProjectConcurrency } from "./project-concurrency.js";
 import { activeWebhookTargets, webhookPass, LEGACY_WEBHOOK_WARNING, loadWebhookTargets, effectivePrimary, isMessagingChannel, loadConsoleUrl, phoneOrigin, saveConsoleUrl, savePrimary } from "./webhooks.js";
 import { auditOf, inspectionOf, isProviderId, MONEY_CAPABILITIES, PROVIDER_IDS, validModelId, validateSpec, ALL_CREDENTIAL_ENV } from "./provider.js";
 import { attestProvider, attestationOf, versionInRange } from "./attest.js";
-import { recognizesEligible } from "./exhaustion.js";
 import { build, proveApprovedProfile, } from "./builder.js";
 import { plan as planTask } from "./planner.js";
 import { attachTmux, elapsedWords, openInTmux, PEEK_TAIL_LINES, runPeek, snapshotLiveRuns } from "./peek-cli.js";
 import { scout as scoutTask } from "./scout.js";
-import { profileDigestOf, chainDigestOf, entryDigestOf } from "./scope.js";
+import { profileDigestOf } from "./scope.js";
 import { PROVIDER_KEY_ENV, SUBSCRIPTION_CAPABLE, clearProviderKey, keyStatus, readAuthMode, readAuthModeStrict, readProviderKey, saveProviderKey, setAuthMode, verifyProviderKey, verdictWords } from "./keys.js";
 import { run, terminateLiveProviders, run as execRun } from "./exec.js";
 import { containmentNotice, containmentStatus, currentContainment, describeContainment, resolveContainment } from "./containment.js";
@@ -359,14 +355,8 @@ External trackers — build what a tracker nominates, under local approvals
   toolroll approver list
   toolroll task scope <id> --goal <what success is>
       [--not <text>] [--touches a,b] [--budget-usd <n>]
-      [--race provider:model[,provider:model…]] [--race-count 2..4]
-      [--race-per-usd <n>] [--race-total-usd <n>]
-      [--compare provider:model[,provider:model…]]  (labeled comparison — no dollar caps; needs a lane no budget can bound)
-                                        a tournament races 2-4 agents on the
-                                        task; you compare and pick one
   toolroll task approve <id>         the yes — interactive, or
-      --yes --digest <d> --as <you> --token <t> for scripts; a tournament
-      approves both documents with one yes, on the joint fingerprint
+      --yes --digest <d> --as <you> --token <t> for scripts
   toolroll task requeue <id> --as <you> --token <t>
                                         exit a stall: incidents resolved,
                                         strikes cleared, queued again
@@ -374,8 +364,7 @@ External trackers — build what a tracker nominates, under local approvals
                                         run the approved check again on the
                                         last attempt's exact commit — a new
                                         attempt, no agent, saved result
-  toolroll config set budgets [--build-usd <n>] [--race-per-usd <n>]
-      [--race-total-usd <n>] [--race-agents 2..4] --as <you> --token <t>
+  toolroll config set budgets [--build-usd <n>] --as <you> --token <t>
                                         spend defaults new filings pre-fill
                                         from; config clear budgets resets
 
@@ -443,35 +432,19 @@ Capabilities — what the work needs, recorded and probed, never valued
                                         plan before building: an agent reads
                                         the repo, asks you questions, and
                                         proposes a scope you approve
-  toolroll task route <id> [--risk routine|elevated|high]
-      [--size small|medium|large] [--risky yes|no]
+  toolroll task route <id> [--size small|medium|large] [--risky yes|no]
       [--phase plan|build|repair --provider <p> [--model <m>] | --clear-phase <phase>]
       --as <you> --token <t>            which agent plans, builds, and repairs
                                         this task, with the reason for
-                                        each; declare its risk or override a
+                                        each; set its size or override a
                                         phase — approval seals the route
 
-Routines — tasks that fire on a schedule, each instance isolated
-  toolroll template list             common routines, shipped
-  toolroll template show <name>      the full prefill + what to edit
+Templates — common work, shipped
+  toolroll template list | show <name>
   toolroll template apply <name> --repo <path> [--file]
-      previews the exact filing; --file files it UNAPPROVED through the
-      same door as a manual filing — a template carries no authority
-
-  toolroll routine add <name> --repo <path> --goal <text>
-      --schedule every:<min>|daily:<HH:MM>[@Zone]|weekly:<0-6>:<HH:MM>[@Zone] (UTC by default)
-      [--not <text>] [--touches a,b] [--require kind:name,…] [--ceiling <usd>]
-      [--budget-usd <n>]                    what each firing may spend
-  toolroll routine approve <name>    the step-up: approving means each
-                                        firing builds WITHOUT asking, inside
-                                        exactly the stated terms; editing any
-                                        term voids the approval
-  toolroll routine list | show <name>
-  toolroll routine refresh <name>    re-resolve the agents it freezes from
-                                        today's configuration; approve again
-                                        afterwards — nothing fires until then
-  toolroll routine pause|resume <name>
-  toolroll routine run-now <name> --as <you> --token <t>
+      previews the exact work; --file makes it UNAPPROVED — a task, or a
+      scheduled flow whose schedule starts paused (turn it on with
+      \`flows trigger resume\`; each run waits for approval)
 
 Flows — processes cards move through; the console's rules
   toolroll flows list [--repo <path>] | show <id>
@@ -521,9 +494,10 @@ Agents — which provider and model each phase runs on
                                         the build it mends.
   toolroll config set <phase> --tier strong --provider <p> --model <m>
       [--repo <path>] --as <you> --token <t>
-                                        the STRONG agent high-risk, strict,
-                                        screenshot-proof, and automerge
-                                        routes reach for; never inferred
+                                        the STRONG agent large or risky,
+                                        strict, screenshot-proof, and
+                                        automerge routes reach for; never
+                                        inferred
   toolroll config set build --tier light --provider <p> --model <m>
       [--repo <path>] --as <you> --token <t>
                                         the fast agent small changes build
@@ -561,8 +535,8 @@ Agents — which provider and model each phase runs on
                                         approve the fast check Quick runs; verify show
                                         suggests one from the project's scripts
   Pass flags still win for one pass: --provider/--model,
-  --plan-provider/--plan-model, --repair-model. A routine instance is
-  pinned at fire time and ignores all of them.
+  --plan-provider/--plan-model, --repair-model. A task a schedule files
+  under an approval moved from a routine is pinned and ignores all of them.
   toolroll peek [<run-id>] [--tmux]  watch live agents: one pane per
                                         open run — stage, clock, and what
                                         the agent is saying; digits focus,
@@ -630,8 +604,6 @@ export const TASK_ACTIONS = [
 export const PUBLISH_ACTIONS = ["setup", "grant", "revoke", "status", "unblock", "rearm", "merge", "refire"];
 export const CONFIG_ACTIONS = ["show", "set", "clear"];
 export const APPROVER_ACTIONS = ["list", "add"];
-export const ROUTINE_ACTIONS = ["list", "add", "show", "approve", "refresh", "pause", "resume", "run-now"];
-export const CONTEST_ACTIONS = ["show", "exclude"];
 export const PEOPLE_ACTIONS = ["list", "invite", "projects", "revoke"];
 export const KEYS_ACTIONS = ["status", "set", "clear", "verify", "auth"];
 /**
@@ -644,7 +616,7 @@ export const OPERATE_VALUE_FLAGS = new Set([
     "allow", "selector", "paths", "credentials", "repo", "token", "capacity",
     "goal", "not", "touches", "acceptance", "candidate", "by", "digest", "as", "branch", "pool", "base", "model", "turns",
     "max", "cap", "probe", "kind", "expires", "cmd", "since", "repair-model",
-    "choose", "note", "max-open-decisions", "max-held-sessions", "name", "days", "publication", "auto-approve", "review-auto", "entries", "port", "host", "allow-host",
+    "choose", "note", "max-open-decisions", "name", "days", "publication", "auto-approve", "review-auto", "port", "host", "allow-host",
     "for", "tick-every", "bridge-every", "reconcile-every", "incarnation",
     "say", "ceiling-usd", "cap-usd",
     "token-file", "bin", "poll", "github", "remote", "head-prefix", "password",
@@ -659,8 +631,8 @@ export const OPERATE_VALUE_FLAGS = new Set([
     "project-root", "schedule", "ceiling", "require",
     "provider", "plan-model", "plan-provider", "public-url", "editor",
     "command", "timeout-seconds", "setup-digest", "stop-grace", "title", "name", "every", "lines",
-    "label", "reviewers", "limit", "role", "key-file", "weekly-usd", "daily-turns", "per-hour", "token-file", "race", "compare", "race-per-usd", "race-total-usd", "race-count", "race-agents", "budget-usd", "build-usd", "sync-max-age", "merge-method",
-    "phase", "risk", "tier", "clear-phase", "size", "risky",
+    "label", "reviewers", "limit", "role", "key-file", "weekly-usd", "daily-turns", "per-hour", "token-file", "budget-usd", "build-usd", "sync-max-age", "merge-method",
+    "phase", "tier", "clear-phase", "size", "risky",
     "run", "containment", "agent",
     // onboard: the starter flows to switch on.
     "starter",
@@ -679,7 +651,7 @@ export const OPERATE_VALUE_FLAGS = new Set([
     "token-name",
 ]);
 export const OPERATE_BOOLEAN_FLAGS = new Set([
-    "json", "yes", "all", "brief", "local", "history", "latest-watch", "dry-run", "file", "allow-paid-fallback",
+    "json", "yes", "all", "brief", "local", "history", "latest-watch", "dry-run", "file",
     "clear", "follow", "ready", "all-tasks", "inbound-only", "help", "undo", "anyone", "allow-dispatch", "allow-merge", "merge-delete-branch",
     "no-open", "remove", "no-verify", "no-follow", "end", "report", "off", "tmux",
     "self-heal", "plan-auto", "chat-approve", "full-access", "repair-auto", "review-retry-auto", "no-local",
@@ -898,7 +870,6 @@ async function dispatchOn(command, positional, flags, repoList, base, options, e
             ...(options.integrationIo === undefined ? {} : { integrationIo: options.integrationIo }),
             ...(options.shouldStop === undefined ? {} : { shouldStop: options.shouldStop }),
             ...(options.mateSeams === undefined ? {} : { mateSeams: options.mateSeams }),
-            ...(options.heldCoordinator === undefined ? {} : { heldCoordinator: options.heldCoordinator }),
             ...(options.waitSleep === undefined ? {} : { waitSleep: options.waitSleep }),
             ...(options.releaseIo === undefined ? {} : { releaseIo: options.releaseIo }),
             ...(options.installBin === undefined ? {} : { installBin: options.installBin }),
@@ -1048,13 +1019,11 @@ function telegramCanDeliver(context, token) {
 }
 /** Ordinary paired text talks to the shared assistant: the same evidence
  * root the console reads results from, the same membership harness seam
- * the CLI's chat uses, and this process's held-session supervisor for a
- * confirmed stop. The pass, the follower and the watch all wire it. */
+ * the CLI's chat uses. The pass, the follower and the watch all wire it. */
 function telegramConversation(context, options = {}) {
     return {
         evidenceRoot: context.evidenceRoot,
         ...(context.mateSeams?.subscriptionRunner === undefined ? {} : { subscriptionRunner: context.mateSeams.subscriptionRunner }),
-        ...(context.heldCoordinator === undefined ? {} : { held: context.heldCoordinator }),
         // Re-read on every card and every `/task`: the same console-url chat
         // links use, held to an https origin, and — inside `up`, where this
         // process also serves the console — equal to that console's own
@@ -1208,8 +1177,6 @@ async function dispatch(command, positional, flags, context) {
             return decideCommand(positional, flags, context);
         case "incident":
             return incidentCommand(positional, flags, context);
-        case "routine":
-            return routineCommand(positional, flags, context);
         case "flows":
             return flowsCommand(positional, flags, context);
         case "config":
@@ -1245,8 +1212,6 @@ async function dispatch(command, positional, flags, context) {
             return providersCommand(flags, context);
         case "template":
             return templateCommand(positional, flags, context);
-        case "contest":
-            return contestCommand(positional, flags, context);
         case "webhook":
             return webhookCommand(positional, flags, context);
         case "sync":
@@ -1465,14 +1430,6 @@ function claimCommand(positional, flags, context) {
                 detail: result.detail,
             });
         }
-        if (result.reason === "attended-held") {
-            return fail(write, json, "claim", "attended-held", `an attended session holds this task for ${result.runner}`, EXIT.refused, {
-                runner: result.runner,
-            });
-        }
-        if (result.reason === "attended-only") {
-            return fail(write, json, "claim", "attended-only", "this task runs only while its operator watches — it needs a live attended authorization or a real approval", EXIT.refused);
-        }
         if (result.reason === "mode-ended") {
             return fail(write, json, "claim", "mode-ended", result.message, EXIT.refused);
         }
@@ -2065,11 +2022,9 @@ async function buildCommand(positional, flags, context) {
     // the same proof the tick's admission wears, refused in words here.
     // Every road stamps at insert (v48 integrity): a routed row's sealed
     // build leg, a pre-routing row's sealed profile, the bare word on a task
-    // with no scope — and a chain approval's base custody rides the same
-    // insert (the standalone road never resumes a parked chain tail; the
-    // tick's proven transfer does that).
+    // with no scope.
     const standaloneModel = text(flags, "model");
-    const authority = store.routeAuthorityFor(ref.id, "builder", null, { provider: "claude", model: standaloneModel ?? null });
+    const authority = store.routeAuthorityFor(ref.id, "builder", { provider: "claude", model: standaloneModel ?? null });
     if (authority !== null && !authority.ok) {
         await worktrees.release(leased.worktree.path, now);
         return fail(write, json, "build", "admission-refused", `${id}: ${authority.problem}`, EXIT.refused);
@@ -2099,7 +2054,6 @@ async function buildCommand(positional, flags, context) {
             worktree: leased.worktree.path,
             ...(authority === null ? {} : { provider: authority.stamp.provider, route: authority.stamp }),
             ...(standaloneModel === undefined ? {} : { model: standaloneModel }),
-            ...(store.approvedChainOf(id) === null ? {} : { custody: { kind: "base" } }),
             now,
         });
     }
@@ -2187,12 +2141,6 @@ async function buildCommand(positional, flags, context) {
         "Nothing has been pushed. Look at the branch before it goes anywhere.",
     ]);
 }
-// ---- the unattended pass --------------------------------------------------
-/** An attended authorization whose signed head the leased worktree no
- * longer matches (final authority closure): thrown BEFORE admission so no
- * run opens and no attempt is spent, reported as its own skip reason. */
-class StaleAuthorization extends Error {
-}
 /** How a worker's flow pass reaches outside: the tests' seams, or the real commands, network and folders. */
 function flowIoOf(context, pool, git, base) {
     const dir = dirname(context.databaseFile);
@@ -2270,7 +2218,7 @@ async function tickCommand(flags, context) {
     const repo = repoFrom(flags);
     // THE PRE-I/O MEMBERSHIP CHECK (MCP spec v6, round-4 finding 1): the
     // pass proves its canonical repo is in the runner's BOUND list before
-    // any git access, probe, or routine fires. The --repo flag stops being
+    // any git access, probe, or schedule fires. The --repo flag stops being
     // authority on every runner road, not only at the claim.
     if (!auth.runner.repos.includes(canonicalProject(repo) ?? resolve(repo))) {
         return fail(write, json, "tick", "unauthorized-repo", `${runner} is not bound to ${repo} — \`runner bind\` adds it`, EXIT.refused);
@@ -2308,29 +2256,9 @@ async function tickCommand(flags, context) {
     // in. Statuses land in the store; the gate inside the claim transaction is
     // what acts on them.
     await probeRepo(store, repo, runner, clock());
-    // Standing orders fire before the ready set is read, so a fresh instance
-    // joins THIS pass. dueRoutines only nominates; every proof — approval
-    // digest, pause, due, single-flight, budget — is re-made inside
-    // fireRoutine's own transaction, and skipped slots ledger and page
-    // themselves there. This loop just reports.
-    const routines = [];
-    for (const routine of buildsOnly ? [] : store.dueRoutines(repo, clock())) {
-        // The stop fence (audit IV-1): a signal that landed mid-pass stops
-        // every further admission — a routine not yet fired stays unfired.
-        if (context.shouldStop?.() === true || context.shouldPauseAdmission?.() === true)
-            break;
-        const outcome = fireRoutine(store, routine.id, clock());
-        routines.push(outcome.ok
-            ? { routine: routine.name, outcome: "fired", taskId: outcome.taskId }
-            : {
-                routine: routine.name,
-                outcome: outcome.reason,
-                ...(outcome.detail === undefined ? {} : { detail: outcome.detail }),
-            });
-    }
-    // Flows move before the ready set is read, like routines: a card entering
-    // a build or research zone files its task now, and that task joins THIS
-    // pass. The engine is model-free; approvals the task needs still apply.
+    // Flows move before the ready set is read: a card entering
+    // a build or research zone files its task now (as does a schedule's
+    // standing order, v115), and that task joins THIS pass. The engine is model-free; approvals the task needs still apply.
     // Triggers first: a schedule, GitHub, Linear or another flow may start
     // cards, which then move in the same pass. Checking an outside service is
     // `gh` or one HTTPS request, never a model, and only when it is due.
@@ -2372,7 +2300,7 @@ async function tickCommand(flags, context) {
     // other builds may hold the worker for minutes.
     let settled = 0;
     const settleFinished = () => {
-        const finished = dispatched.slice(settled).filter(one => one.outcome !== "skipped" && one.outcome !== "contest").map(one => one.id);
+        const finished = dispatched.slice(settled).filter(one => one.outcome !== "skipped").map(one => one.id);
         settled = dispatched.length;
         if (finished.length === 0)
             return;
@@ -2381,24 +2309,10 @@ async function tickCommand(flags, context) {
         flowPass.filed.push(...after.filed);
         flowPass.problems.push(...after.problems);
     };
-    // Tournament housekeeping before the ordinary pass (stage 4): interrupted
-    // races recover by CAS, and an ANSWERED question re-admits its parked
-    // agent — fresh claim, fresh slot, remaining budget only, the SAME
-    // verified checkout on the SAME runner (finding 29's custody rule).
-    if (!buildsOnly)
-        recoverContests(store, clock());
     // Expired ceremony nonces are litter with a bound (round-3 finding 30):
     // the mint refuses past 50 open per approver, so the sweep keeps the
     // ceiling meaningful rather than letting dead rows consume it.
     store.sweepCeremonyNonces(clock());
-    // Decided tournaments give their checkouts back (stage 6) — this runner's
-    // custody only; a checkout that will not release cleanly is flagged and
-    // paged, never force-cleaned. Undecided ones escalate once at 14 days.
-    if (!buildsOnly) {
-        await sweepContestCleanup(store, path => worktrees.release(path, clock()), runner, clock());
-        escalateOverdueContests(store, clock());
-    }
-    const resumed = [];
     // v105: monthly budgets, read when first needed and again after any build (its spend counts): a used-up hard-stop
     // budget holds back new work billed to an API key, on every road below.
     let budgetGate = null;
@@ -2406,13 +2320,10 @@ async function tickCommand(flags, context) {
     const budgetWords = (hold) => budgetHoldWords(hold, monthOf(clock()).name);
     // Sprint 8: the organisation policy, asked before any claim on every road below: a provider or model it doesn't
     // allow never starts (the task waits, saying which rule and where to change it); terms above its permission ceiling
-    // start and run lowered (build() lowers them and says so), except an attended session's, signed at exactly those.
+    // start and run lowered (build() lowers them and says so).
     const policyHold = (agents) => {
         for (const agent of agents) {
             if ("profile" in agent) {
-                const refused = agent.attended === true ? store.attendedPolicyRefusal(agent.profile) : null;
-                if (refused !== null)
-                    return refused;
                 const verdict = store.runPolicy(agent.profile);
                 if (!verdict.ok)
                     return verdict.message;
@@ -2426,297 +2337,12 @@ async function tickCommand(flags, context) {
         }
         return null;
     };
-    for (const waiting of buildsOnly ? [] : store.contestsInStates(["decision-wait"])) {
-        if (context.shouldStop?.() === true || context.shouldPauseAdmission?.() === true)
-            break;
-        // D1 belt-and-braces (external dispatch, finding 41): a mirror and a
-        // contest should never coexist; if one ever does, its race resumes
-        // NOTHING — no claim, no run, no worktree, no spend.
-        const waitingTaskId = store.externalIdFor(waiting.taskRef);
-        if (waitingTaskId !== null && store.mirrorByTask(waitingTaskId) !== null) {
-            resumed.push({ id: waitingTaskId, outcome: "skipped", reason: "external-race" });
-            continue;
-        }
-        // THE ANSWERED BATCH IS MARKED ACTIVE FIRST (Codex slice-B finding 2):
-        // resuming lanes one at a time let the FIRST finisher aggregate the
-        // contest while later answered lanes were still 'parked' — excluded
-        // from active, their decisions no longer open — stranding them in a
-        // contest that had already moved on. 'ready' counts as active, so
-        // aggregation waits for the whole batch. Any lane that bails before
-        // its build reverts to 'parked' so the next pass retries it.
-        const batch = [];
-        for (const racer of store.contestants(waiting.id).filter(one => one.state === "parked")) {
-            if (store.answeredDecisionForContestant(racer.id) === null)
-                continue;
-            if (store.casContestantState(racer.id, ["parked"], "ready", racer.generation)) {
-                batch.push({ racer });
-            }
-        }
-        for (const { racer } of batch) {
-            const backToParked = () => {
-                const current = store.getContestant(racer.id);
-                if (current !== null)
-                    store.casContestantState(racer.id, ["ready"], "parked", current.generation);
-            };
-            if (context.shouldStop?.() === true || context.shouldPauseAdmission?.() === true) {
-                backToParked();
-                continue;
-            }
-            const custody = racer.custody === null ? null : JSON.parse(racer.custody);
-            const taskId = store.externalIdFor(waiting.taskRef);
-            if (custody === null || custody.runner !== runner || taskId === null) {
-                backToParked();
-                continue;
-            }
-            const lanePolicy = policyHold([racer.profile == null ? { provider: racer.provider, model: racer.model } : { profile: racer.profile }]);
-            if (lanePolicy !== null) {
-                backToParked();
-                resumed.push({ id: taskId, outcome: "skipped", reason: "policy", detail: lanePolicy });
-                continue;
-            }
-            const laneBudget = budgetHold(waiting.taskRef, store.agentsFor([racer.provider]));
-            if (laneBudget.over !== null) {
-                backToParked();
-                resumed.push({ id: taskId, outcome: "skipped", reason: "budget", detail: budgetWords(laneBudget) });
-                continue;
-            }
-            // The sign-in pause holds a resumed lane too: it waits, parked, instead of failing.
-            const laneGate = signInGate(store, [racer.provider], clock());
-            if (laneGate.waiting !== null) {
-                backToParked();
-                resumed.push({ id: taskId, outcome: "skipped", reason: "signed-out", detail: signInWords(laneGate.waiting) });
-                continue;
-            }
-            // Legacy comparison lanes carried an absolute clock. New profiles use
-            // a progress watchdog and may remain alive indefinitely while useful
-            // work is observable; only legacy approvals retain their cumulative
-            // wall-clock contract.
-            const remaining = waiting.kind === "comparison" ? null : racer.budgetMicrousd - racer.accountedMicrousd;
-            if (remaining !== null && remaining <= 0) {
-                laneGate.giveBack();
-                const current = store.getContestant(racer.id);
-                if (current !== null)
-                    store.casContestantState(racer.id, ["ready"], "stopped", current.generation);
-                contestMaybeAggregate(store, waiting.id, clock());
-                resumed.push({ id: taskId, outcome: "skipped", reason: "over-ceiling" });
-                continue;
-            }
-            if (waiting.kind === "comparison") {
-                const laneProfile = racer.profile ?? contestantProfileOf(racer.provider, racer.model, racer.repairModel);
-                const clockCapMs = 3 * laneProfile.timeoutSeconds * 1000;
-                if (laneProfile.timeoutKind !== "idle" && store.contestantCumulativeMs(racer.id) >= clockCapMs) {
-                    laneGate.giveBack();
-                    const current = store.getContestant(racer.id);
-                    if (current !== null)
-                        store.casContestantState(racer.id, ["ready"], "stopped", current.generation);
-                    contestMaybeAggregate(store, waiting.id, clock());
-                    resumed.push({ id: taskId, outcome: "skipped", reason: "over-ceiling" });
-                    continue;
-                }
-            }
-            const reclaimed = acquire(store, waiting.taskRef, runner, { now: clock(), token, ttlMs: leaseTtlMs });
-            if (!reclaimed.ok) {
-                laneGate.giveBack();
-                backToParked();
-                continue;
-            }
-            const freshContest = store.getContest(waiting.id);
-            if (freshContest === null || !store.casContestState(waiting.id, ["decision-wait", "racing"], "racing", freshContest.generation)) {
-                release(store, reclaimed.claim.leaseId, clock());
-                laneGate.giveBack();
-                backToParked();
-                continue;
-            }
-            store.stampContestLease(waiting.id, reclaimed.claim.leaseId, runner, text(flags, "incarnation") ?? null);
-            const leased = await worktrees.lease({ repo, branch: racer.branch, runner, taskRef: waiting.taskRef, now: clock() });
-            const headCheck = leased.ok ? await git("git", ["rev-parse", "HEAD"], { cwd: leased.worktree.path }) : null;
-            if (!leased.ok || (custody.head !== null && headCheck !== null && headCheck.stdout.trim() !== custody.head)) {
-                // The tree cannot be proved to be the one the agent left — stop the
-                // agent rather than cold-starting against a different history.
-                if (leased.ok)
-                    await worktrees.release(leased.worktree.path, clock());
-                const current = store.getContestant(racer.id);
-                if (current !== null)
-                    store.casContestantState(racer.id, ["ready"], "stopped", current.generation);
-                store.setContestantCleanup(racer.id, "attention");
-                contestMaybeAggregate(store, waiting.id, clock());
-                release(store, reclaimed.claim.leaseId, clock());
-                resumed.push({ id: taskId, outcome: "failed", reason: "contest-custody" });
-                continue;
-            }
-            const [resumeSlot] = store.reserveExecutionSlots(runner, 1, clock());
-            const parkedRun = racer.activeRun;
-            // A contest lane spends under its race-approved profile and says so
-            // at insert (v48 integrity): the lane's proven profile digest, the
-            // exact pair it will spend as — the store's own answer, re-proved
-            // and bound to the lane inside the admission.
-            const laneStamp = store.laneAuthorityFor(racer.id);
-            if (laneStamp === null) {
-                await worktrees.release(leased.worktree.path, clock());
-                release(store, reclaimed.claim.leaseId, clock());
-                backToParked();
-                resumed.push({ id: taskId, outcome: "failed", reason: "admission-refused", detail: `contestant ${racer.id} is gone` });
-                continue;
-            }
-            // THE LANE ADMISSION (atomic authority closure): the resume opens on
-            // the lane under the contest's live custody — this lease, this
-            // runner, this watch incarnation, stamped on the contest just above
-            // — presenting the lane's stored sealed profile; value-shaped.
-            const admittedResume = store.admitContestLane({
-                taskRef: waiting.taskRef,
-                leaseId: reclaimed.claim.leaseId,
-                runner,
-                incarnation: text(flags, "incarnation") ?? null,
-                branch: racer.branch,
-                worktree: leased.worktree.path,
-                provider: racer.provider,
-                model: racer.model,
-                contestant: racer.id,
-                ...(parkedRun === null ? {} : { parentRun: parkedRun }),
-                now: clock(),
-                route: laneStamp,
-            });
-            if (!admittedResume.ok) {
-                await worktrees.release(leased.worktree.path, clock());
-                release(store, reclaimed.claim.leaseId, clock());
-                backToParked();
-                resumed.push({ id: taskId, outcome: "failed", reason: "admission-refused", detail: admittedResume.problem });
-                continue;
-            }
-            const resumeRun = admittedResume.runId;
-            // The lane's pointer moved to the resume INSIDE its admission (raw
-            // authority repair): from the parked attempt it continues, proved
-            // there — no release-then-claim window exists any more.
-            const afterClaim = store.getContestant(racer.id);
-            if (afterClaim !== null)
-                store.casContestantState(racer.id, ["ready"], "building", afterClaim.generation);
-            budgetGate = null; // this build's spend counts toward the next check
-            const resumeResult = await build(store, {
-                taskId,
-                taskRef: waiting.taskRef,
-                runner,
-                leaseId: reclaimed.claim.leaseId,
-                runnerToken: token,
-                runId: resumeRun,
-                evidenceRoot: context.evidenceRoot,
-                worktree: leased.worktree.path,
-                branch: racer.branch,
-                now: clock(),
-                clock,
-                provider: racer.provider,
-                // v24: the contestant's OWN sealed profile is the authority — the
-                // proof holds the lane to it (model, limits, permissions), so no
-                // flag-shaped overrides ride along.
-                contestProfile: racer.profile ?? contestantProfileOf(racer.provider, racer.model, racer.repairModel),
-                ...(remaining === null ? {} : { maxBudgetUsd: Math.min(remaining, laneBudget.remainingMicrousd ?? Infinity) / 1_000_000 }),
-                onProviderSpawn: (pid) => {
-                    worktrees.recordProviderOccupancy(leased.worktree.path, runner, pid, leased.worktree.leaseEpoch);
-                    if (resumeSlot !== undefined) {
-                        const facts = { run: resumeRun, contestant: racer.id, incarnation: text(flags, "incarnation") ?? null, processGroup: pid };
-                        if (!store.markSlotRunning(resumeSlot, facts, clock()) && !store.refreshSlotProcess(resumeSlot, facts))
-                            throw new Error("the provider's execution slot no longer belongs to this attempt");
-                    }
-                },
-                ...(context.agentRunner === undefined ? {} : { agent: context.agentRunner }),
-                ...(context.gitRunner === undefined ? {} : { git: context.gitRunner }),
-                ...(context.shouldStop === undefined ? {} : { shouldStop: context.shouldStop }),
-            });
-            const resumedRunRow = store.getRun(resumeRun);
-            const resumeMeasured = resumedRunRow === null || resumedRunRow.providerStartedAt === null
-                ? 0
-                : resumedRunRow.costUsd !== null
-                    ? Math.round(resumedRunRow.costUsd * 1_000_000)
-                    : null;
-            let resumedOutcome = "failed";
-            let resumedCommitted = false;
-            if (resumeResult.ok && resumeResult.parked !== undefined) {
-                resumedOutcome = "parked";
-                const asked = resumeResult.parked.decision;
-                const racerDecision = store.saveDecision({
-                    run: resumeRun,
-                    contestant: racer.id,
-                    urgency: asked.urgency,
-                    recap: asked.recap,
-                    question: asked.question,
-                    options: asked.options,
-                    recommendation: asked.recommendation,
-                    ...(asked.assignee === null ? {} : { assignee: asked.assignee }),
-                    ...(asked.deadline === null ? {} : { deadline: asked.deadline }),
-                }, clock());
-                // A racing agent's question pages like any other (arc 3 finding 21):
-                // aggregation stays quiet ASSUMING this row already spoke.
-                store.enqueueNotification({
-                    source: { run: resumeRun },
-                    dedupeKey: `decision:${racerDecision}`,
-                    kind: "decision",
-                    subject: `${taskId} parked a decision (${contestNoun(waiting.kind)} agent)`,
-                    body: `\`toolroll decide ${racerDecision}\``,
-                    pushClass: "decision",
-                    link: `/d/${racerDecision}`,
-                }, clock());
-            }
-            else if (resumeResult.ok) {
-                resumedOutcome = "built";
-                resumedCommitted = resumeResult.committed;
-            }
-            else if (resumeResult.reason === "stopped") {
-                resumedOutcome = "stopped";
-            }
-            if (resumedRunRow !== null && resumedRunRow.outcome === null) {
-                // The reason rides the resumed run too (Codex slice-B finding 7):
-                // both settlement paths, one honesty.
-                const resumeReason = !resumeResult.ok ? resumeResult.reason : undefined;
-                store.finishRun(resumeRun, {
-                    outcome: resumedOutcome === "built" && !resumedCommitted ? "no-change" : resumedOutcome === "stopped" ? "failed" : resumedOutcome === "parked" ? "parked" : resumedOutcome,
-                    committed: resumedCommitted,
-                    ...(resumeReason === undefined ? {} : { reason: resumeReason }),
-                    now: clock(),
-                });
-            }
-            if (resumedOutcome === "parked") {
-                // Custody refreshes on EVERY park (Codex slice-B finding 9): a
-                // re-parked lane whose custody still named the pre-resume head
-                // would falsely stop as contest-custody on its next answer.
-                const headNow = await git("git", ["rev-parse", "HEAD"], { cwd: leased.worktree.path });
-                const dirtyNow = await git("git", ["status", "--porcelain"], { cwd: leased.worktree.path });
-                store.setContestantCustody(racer.id, JSON.stringify({
-                    branch: racer.branch,
-                    head: headNow.code === 0 ? headNow.stdout.trim() : null,
-                    runner,
-                    dirty: dirtyNow.stdout.trim() !== "",
-                    at: clock().toISOString(),
-                }));
-            }
-            await worktrees.release(leased.worktree.path, clock());
-            const resumedFinal = finalizeContestant(store, {
-                contestId: waiting.id,
-                contestantId: racer.id,
-                runId: resumeRun,
-                outcome: resumedOutcome,
-                measuredMicrousd: resumeMeasured,
-                slotId: resumeSlot ?? null,
-            }, clock());
-            resumed.push({ id: taskId, outcome: "contest", reason: resumedFinal.aggregated ?? "racing" });
-        }
-    }
-    // THE CHAIN RECONCILER (E3d, review finding 3): a crash between a run's
-    // disposition and its cycle resolution leaves an OPEN cycle with a
-    // concluded tail — resolve each through the SAME resolver the disposition
-    // uses, before any admission can re-tag or race it. Advancing here lands
-    // the cycle in pending-admission for THIS pass's chain admission below.
-    // One shared piece with the fault tests (F+G review, finding 5).
-    if (!buildsOnly)
-        store.reconcileStrandedChains(repo, clock());
     // The DISPATCH view of the queue (queue columns, v19): this runner's own
     // reserved work first, then the shared queue; work reserved for other
     // workers is absent. The claim primitive re-proves the reservation.
     const ready = store.listReady(clock(), runner);
     const considered = ready.length;
-    const dispatched = [...resumed];
-    // Expired attended authorizations close durably each pass (round-6
-    // finding 8): the partial unique frees, and the claim gates stop
-    // honoring corpses.
-    store.sweepExpiredAuthorizations(clock());
+    const dispatched = [];
     // The stale-scan (dispatch v3, finding 20): mirrors the courtesy filter
     // kept out of ready are REPORTED here, typed, with a paged episode —
     // an undispatakable tracker item is a 9am fact, not a silent absence.
@@ -2748,15 +2374,9 @@ async function tickCommand(flags, context) {
         settleFinished();
         untakenTrial?.giveBack();
         untakenTrial = null;
-        // The build budget governs UNATTENDED admissions (round-1 finding 4):
-        // once it is spent, the pass keeps SCANNING for attended
-        // authorizations — operator-invoked sessions launch regardless —
-        // while declining every further ordinary admission.
-        if (built >= max) {
-            const openAuth = store.openAuthorizationFor(ref.id);
-            if (openAuth === null || openAuth.runner !== runner || openAuth.attemptRun !== null)
-                continue;
-        }
+        // The build budget: once it is spent, no further admission this pass.
+        if (built >= max)
+            continue;
         // The stop fence (audit IV-1): checked before every claim. The build
         // already in flight finishes under its own bounds; nothing NEW is
         // admitted once the operator has said stop.
@@ -2768,91 +2388,17 @@ async function tickCommand(flags, context) {
             dispatched.push({ id, outcome: "skipped", reason: "other-repo" });
             continue;
         }
-        // EVERY live fallback cycle defers the ordinary road (Codex E3d review,
-        // finding 5) — a pending admission belongs to the chain pass below, and
-        // an open cycle's custody moves only through proven roads, never an
-        // in-passing re-tag. The ONE exception: an open cycle whose tail PARKED
-        // — the paused lineage — proceeds, and the new run takes custody through
-        // the proven parked-resume transfer after it is created.
-        const liveCycle = store.fallbackCycleFor(ref.id);
-        const parkedChainTail = liveCycle !== null && liveCycle.state === "open" && liveCycle.tailRun !== null
-            ? store.getRun(liveCycle.tailRun)
-            : null;
-        if (liveCycle !== null && !(parkedChainTail !== null && parkedChainTail.outcome === "parked")) {
-            dispatched.push({ id, outcome: "skipped", reason: "fallback-active" });
-            continue;
-        }
         // A plan the operator asked for dispatches a PLANNER — the one
         // legitimate spend on a task with no approved scope. Everything else
-        // unapproved is a person's pending decision: skip, not refuse — EXCEPT
-        // the attended road (Phase 2, v6 W1): a live attended authorization
-        // naming THIS runner is authority for one watched attempt, and the
-        // skip for everything short of that is its own typed word, never the
-        // generic `unapproved` the round-5 review caught masking it.
+        // unapproved is a person's pending decision: skip, not refuse.
         const scopeApproved = scopeApprovedForDispatch(store, ref.id, clock());
         const wantsPlan = ref.plan === "requested" && !scopeApproved;
         // A report task with an approved scope dispatches a SCOUT (mate arc
         // §10): the same approval, the planner's read-only road, a report back.
         const wantsScout = ref.deliverable === "report" && scopeApproved;
-        let attendedDispatch = null;
         if (!wantsPlan && !scopeApproved) {
-            const open = store.openAuthorizationFor(ref.id);
-            const watching = open === null
-                ? null
-                : attendedLivenessState(open.lastBeatAt === null ? null : Date.parse(open.lastBeatAt), clock().getTime(), Date.parse(open.absoluteExpiry));
-            if (open !== null &&
-                open.runner === runner &&
-                (watching === "live" || watching === "grace") &&
-                open.attemptRun === null &&
-                context.heldCoordinator !== undefined) {
-                // v28: sessions are unbounded by default; an operator-set cap
-                // skips FURTHER launches in words. The gauge is durable custody
-                // rows, never the in-process map — a restarted up with orphans
-                // pending must count them.
-                if (context.maxHeldSessions !== undefined && store.openHeldSessionCount(runner) >= context.maxHeldSessions) {
-                    dispatched.push({
-                        id,
-                        outcome: "skipped",
-                        reason: "session-cap",
-                        detail: `this machine holds ${store.openHeldSessionCount(runner)} of ${context.maxHeldSessions} attended sessions — end one, or raise --max-held-sessions`,
-                    });
-                    continue;
-                }
-                attendedDispatch = open;
-            }
-            else if (open !== null) {
-                dispatched.push({ id, outcome: "skipped", reason: "attended-only" });
-                continue;
-            }
-            else {
-                dispatched.push({ id, outcome: "skipped", reason: "unapproved" });
-                continue;
-            }
-        }
-        // TOURNAMENT TERMS FIRST (foundations finding 8's reorder): an approved
-        // race's admission is governed by its own fingerprinted terms, and pass
-        // flags must not be able to shape it — so the terms are discovered
-        // before any flag-shaped resolution runs, and a raced task's build
-        // resolution ignores the flags outright.
-        const racedAhead = wantsPlan || wantsScout || attendedDispatch !== null ? null : store.activeTournamentTerms(ref.id);
-        // The attended spec comes from the authorization's PINNED terms — the
-        // courtesy half of the proof; the coordinator's transaction re-proves
-        // byte-for-byte at the actual HEAD (v6 W1).
-        let attendedSpec = null;
-        if (attendedDispatch !== null) {
-            try {
-                const terms = JSON.parse(attendedDispatch.termsJson);
-                const pinned = profileFromJson(typeof terms.profileJson === "string" ? terms.profileJson : null);
-                if (pinned !== null)
-                    attendedSpec = { provider: pinned.provider, model: pinned.model, digest: profileDigestOf(pinned), profile: pinned };
-            }
-            catch {
-                attendedSpec = null;
-            }
-            if (attendedSpec === null) {
-                dispatched.push({ id, outcome: "skipped", reason: "attended-only", detail: "the authorization's pinned profile cannot be read" });
-                continue;
-            }
+            dispatched.push({ id, outcome: "skipped", reason: "unapproved" });
+            continue;
         }
         // The phase agent, resolved BEFORE anything is claimed and snapshotted
         // into the run: pin > flags > project > installation > default. Planner
@@ -2860,14 +2406,14 @@ async function tickCommand(flags, context) {
         // when no plan-specific flag is given.
         // THE ROUTE (v47): the sealed route when the approval stands, else the
         // working proposed route, else — with no scope yet — a live
-        // recommendation from the task's own risk, overrides, and pins; a row
+        // recommendation from the task's own size, overrides, and pins; a row
         // proven to predate routing is the LEGACY road (its sealed profile,
         // then flags and configuration); a routed row whose route cannot be
         // read FAILS CLOSED here, in words. Dispatch resolves each phase from
         // ITS leg: the plan leg for a planner (pass flags may restate it, never
         // contradict it), the SEALED build leg for every routed approval
         // (mutable configuration cannot reroute an approved build).
-        const taskRoute = attendedSpec !== null || racedAhead !== null ? null : routeOfTask(store, id, ref, clock());
+        const taskRoute = routeOfTask(store, id, ref, clock());
         if (taskRoute !== null && taskRoute.kind === "unreadable") {
             dispatched.push({ id, outcome: "skipped", reason: "agent-config", detail: taskRoute.problem });
             continue;
@@ -2877,9 +2423,8 @@ async function tickCommand(flags, context) {
         // authority closure): a planner on a filed, unapproved scope runs
         // under the WORKING route only as the whole scope proves — exact raw
         // terms (a proposed-via marker this code never writes), a resolved
-        // profile, a whole fallback chain (an unresolved `[]` is none), route
-        // parity with the signed risk, the digest, and the live auth mode
-        // agreeing with a chain's pinned base mode. One disagreement skips the
+        // profile, route parity with the signed risk, the digest, and the live
+        // auth mode. One disagreement skips the
         // task in words before any claim; the admission and the spawn ask the
         // same question again.
         if (wantsPlan && route !== null && route.source === "proposed") {
@@ -2891,21 +2436,6 @@ async function tickCommand(flags, context) {
         }
         const planLeg = route === null ? null : legOf(route.route, "plan");
         const sealedBuildLeg = route !== null && route.source === "approved" ? legOf(route.route, "build") : null;
-        // A parked chain tail PAST the base (v48 authority repair): its successor resumes the
-        // entry's custody and spends as that entry — the exact approved pair
-        // under `fallback` provenance — never as the sealed build leg. An
-        // entry the approved chain no longer carries under the tail's digest
-        // is refused here, in words, before any claim moves.
-        let parkedEntry = null;
-        if (parkedChainTail !== null && parkedChainTail.chainIndex != null && parkedChainTail.chainIndex > 0 && !wantsPlan && !wantsScout && attendedSpec === null && racedAhead === null) {
-            const chain = store.approvedChainOf(id);
-            const entry = chain === null ? undefined : chain[parkedChainTail.chainIndex];
-            if (entry === undefined || entryDigestOf(entry) !== parkedChainTail.entryDigest) {
-                dispatched.push({ id, outcome: "skipped", reason: "stale-approval", detail: `${id}: the parked attempt #${parkedChainTail.id} is bound to fallback entry ${parkedChainTail.chainIndex}, which the approved chain no longer carries — nothing resumes it` });
-                continue;
-            }
-            parkedEntry = { index: parkedChainTail.chainIndex, provider: entry.profile.provider, model: entry.profile.model };
-        }
         const planFlagged = planProvider !== undefined || planModel !== undefined || providerFlag !== undefined || model !== undefined;
         if (wantsPlan && planLeg !== null && planFlagged) {
             // Pass flags cannot contradict the task's plan leg: a flag that names
@@ -2923,48 +2453,40 @@ async function tickCommand(flags, context) {
                 continue;
             }
         }
-        const resolution = attendedSpec !== null
-            ? null
-            : wantsPlan
-                ? planLeg !== null
-                    ? resolvePhaseAgent(store, "plan", repo, { provider: planLeg.provider, model: planLeg.model })
-                    : resolvePhaseAgent(store, "plan", repo, {
-                        // The legacy road (P2/C7 precedence): the task's plan PIN
-                        // beats every flag and config row; flags beat config.
-                        provider: ref.planProvider ?? planProvider ?? providerFlag,
-                        model: ref.planProvider !== null ? (ref.planModel ?? undefined) : (planModel ?? model),
-                    })
-                : racedAhead !== null
-                    ? resolvePhaseAgent(store, "build", repo, {}, ref)
-                    : parkedEntry !== null
-                        ? resolvePhaseAgent(store, "build", repo, { provider: parkedEntry.provider, model: parkedEntry.model })
-                        : sealedBuildLeg !== null
-                            ? resolvePhaseAgent(store, "build", repo, { provider: sealedBuildLeg.provider, model: sealedBuildLeg.model }, ref)
-                            : resolvePhaseAgent(store, "build", repo, { provider: providerFlag, model }, ref);
-        if (resolution !== null && !resolution.ok) {
+        const resolution = wantsPlan
+            ? planLeg !== null
+                ? resolvePhaseAgent(store, "plan", repo, { provider: planLeg.provider, model: planLeg.model })
+                : resolvePhaseAgent(store, "plan", repo, {
+                    // The legacy road (P2/C7 precedence): the task's plan PIN
+                    // beats every flag and config row; flags beat config.
+                    provider: ref.planProvider ?? planProvider ?? providerFlag,
+                    model: ref.planProvider !== null ? (ref.planModel ?? undefined) : (planModel ?? model),
+                })
+            : sealedBuildLeg !== null
+                ? resolvePhaseAgent(store, "build", repo, { provider: sealedBuildLeg.provider, model: sealedBuildLeg.model }, ref)
+                : resolvePhaseAgent(store, "build", repo, { provider: providerFlag, model }, ref);
+        if (!resolution.ok) {
             dispatched.push({ id, outcome: "skipped", reason: "agent-config", detail: resolution.problem });
             continue;
         }
-        const spec = resolution === null ? attendedSpec : resolution.spec;
+        const spec = resolution.spec;
         // THE AUTH MODE, strictly, before any claim or row (atomic authority
         // closure): a present mode file for the provider this pass would spend
         // as that says neither word is a stated problem — the same reader the
         // filing, the seal, and the spawn use — and the task is skipped in
         // words with nothing opened, never dispatched to be refused later.
-        const modeProviders = racedAhead !== null ? racedAhead.agents.filter(agent => isProviderId(agent.provider)).map(agent => agent.provider) : [spec.provider];
-        const brokenMode = [...new Set(modeProviders)].map(one => readAuthModeStrict(one)).find(one => !one.ok);
-        if (brokenMode !== undefined && !brokenMode.ok) {
+        const brokenMode = readAuthModeStrict(spec.provider);
+        if (!brokenMode.ok) {
             dispatched.push({ id, outcome: "skipped", reason: "auth-mode", detail: brokenMode.problem });
             continue;
         }
         // THE SIGN-IN PAUSE: a provider whose sign-in stopped working takes no
-        // new work — no claim, no run, never a substitute (a lapsed login must not
-        // reach a paid fallback) — until a run or check on it works again or a
+        // new work — no claim, no run, never a substitute — until a run or check on it works again or a
         // person resumes it. Other providers keep working.
         // One task every AUTH_TRIAL_MS goes ahead as the trial: a sign-in check
         // can say "logged in" for a session that cannot refresh, so a real run
         // decides — it lifts the pause, or fails into the same incident.
-        const gate = signInGate(store, modeProviders, clock(), ref.id);
+        const gate = signInGate(store, [spec.provider], clock(), ref.id);
         if (gate.waiting !== null) {
             dispatched.push({ id, outcome: "skipped", reason: "signed-out", detail: signInWords(gate.waiting) });
             continue;
@@ -2972,38 +2494,27 @@ async function tickCommand(flags, context) {
         // A trial whose task is skipped before its claim succeeds is given back
         // at the next task (or after the pass), so the next pass may take it.
         untakenTrial = gate;
-        // The leg is the authority: what resolved must BE the leg, exactly —
-        // the sealed build leg, or the parked fallback entry's own pair.
-        const governingLeg = wantsPlan ? planLeg : parkedEntry !== null ? { provider: parkedEntry.provider, model: parkedEntry.model, chosen: "fallback" } : sealedBuildLeg;
+        // The leg is the authority: what resolved must BE the leg, exactly.
+        const governingLeg = wantsPlan ? planLeg : sealedBuildLeg;
         if (governingLeg !== null && (spec.provider !== governingLeg.provider || spec.model !== governingLeg.model)) {
             dispatched.push({ id, outcome: "skipped", reason: "agent-config", detail: `${id}: the ${wantsPlan ? "plan" : "build"} leg names ${governingLeg.provider} · ${governingLeg.model} but resolution produced ${spec.provider} · ${spec.model ?? "(no model)"} — nothing substitutes` });
             continue;
         }
         // Route provenance for the run this pass opens (v47): PRESENTED to the
         // admission transaction, from the leg that governs it — the store
-        // dictates nothing (v48 authority repair). A parked fallback entry's successor presents
-        // `fallback` under the sealed route — there is no chain-only digest
-        // (raw authority repair): with no sealed route it presents nothing,
-        // and the admission refuses in words.
-        // Every road presents at insert (v48 integrity): an attended session
-        // its pinned profile; a pre-routing row its sealed profile (or the
-        // bare word on a task with no scope) — the store's own answer, so the
+        // dictates nothing (v48 authority repair).
+        // Every road presents at insert (v48 integrity): a pre-routing row its
+        // sealed profile (or the bare word on a task with no scope) — the store's own answer, so the
         // admission proves exactly what the row holds.
         const routeStamp = (phase) => {
-            if (attendedSpec !== null && phase === "build") {
-                return { routeDigest: `profile:${attendedSpec.digest}`, phase, provider: spec.provider, model: spec.model, chosen: "legacy" };
-            }
             // A task with NO scope holds no filed route (final authority
             // closure): its planner presents the bare word `legacy` for the pair
             // the live recommendation resolved — the recommendation chose the
             // agent, but nothing anybody filed is the authority it spends under.
             if (governingLeg === null || route === null || (phase === "plan" && route.source === "live")) {
-                const legacy = store.routeAuthorityFor(ref.id, phase === "plan" ? "planner" : "builder", null, { provider: spec.provider, model: spec.model });
+                const legacy = store.routeAuthorityFor(ref.id, phase === "plan" ? "planner" : "builder", { provider: spec.provider, model: spec.model });
                 return legacy !== null && legacy.ok ? legacy.stamp : null;
             }
-            if (parkedEntry !== null && phase === "build") {
-                return { routeDigest: routeDigestOf(route.route), phase, provider: spec.provider, model: spec.model, chosen: "fallback" };
-            }
             return { routeDigest: routeDigestOf(route.route), phase, provider: spec.provider, model: spec.model, chosen: governingLeg.chosen };
         };
         // THE LEGACY STALE CHECK, before any claim or row (v48 integrity): a
@@ -3012,7 +2523,7 @@ async function tickCommand(flags, context) {
         // must BE that pair. A disagreement used to open a row and refuse it
         // at the dispatch proof; now nothing opens — the task is held under
         // the same backoff the approval door lifts, and paged once.
-        if (!wantsPlan && !wantsScout && attendedSpec === null && racedAhead === null && route === null) {
+        if (!wantsPlan && !wantsScout && route === null) {
             const legacyStamp = routeStamp("build");
             if (legacyStamp !== null && (legacyStamp.provider !== spec.provider || legacyStamp.model !== spec.model)) {
                 const message = `${id}: the sealed profile builds on ${legacyStamp.provider} · ${legacyStamp.model ?? "(no model)"} but today's routing resolved ${spec.provider} · ${spec.model ?? "(no model)"} — nothing runs on it (stale-approval)`;
@@ -3024,30 +2535,16 @@ async function tickCommand(flags, context) {
         // THE READINESS HALT (v47): a provider THIS runner has reported
         // unavailable never claims and never spends — no substitution, no
         // second-best; the skip names the observation and the ways out.
-        // Unknown readiness passes (every existing gate still applies); a
-        // raced task halts if any lane's provider is unavailable, because a
-        // subset is a different contest than the one signed. The ONE road
-        // onward is an EXPLICITLY APPROVED fallback chain: when the sealed
-        // build leg's provider is unavailable, the task moves to the approved
-        // next entry — that one and no other — and the chain admission pass
-        // below runs it after re-proving everything; otherwise it fails closed.
+        // Unknown readiness passes (every existing gate still applies).
         {
-            const providersToProve = racedAhead !== null ? racedAhead.agents.map(agent => agent.provider) : [spec.provider];
-            const unavailable = providersToProve
-                .map(candidate => store.runnerReadinessOf(runner, candidate))
-                .find(seen => seen !== null && seen.state === "unavailable");
-            if (unavailable !== undefined && unavailable !== null) {
+            const unavailable = store.runnerReadinessOf(runner, spec.provider);
+            if (unavailable !== null && unavailable.state === "unavailable") {
                 const observed = `${unavailable.provider} is reported unavailable on ${runner} (${unavailable.reason}; observed ${unavailable.observedAt})`;
-                const onward = !wantsPlan && !wantsScout && racedAhead === null && attendedSpec === null && sealedBuildLeg !== null
-                    ? store.skipUnavailablePrimary(ref.id, id, repo, runner, clock())
-                    : null;
                 dispatched.push({
                     id,
                     outcome: "skipped",
                     reason: "provider-unavailable",
-                    detail: onward !== null && onward.ok
-                        ? `${observed} — moving to the approved fallback ${onward.next.provider} · ${onward.next.model}, the only substitution the approval allows; it is admitted next`
-                        : `${observed} — nothing substitutes for a routed provider${onward === null || onward.reason === "no-chain" ? "" : ` (${onward.detail})`}: override the phase with \`task route\` or restore the provider and report readiness again`,
+                    detail: `${observed} — nothing substitutes for a routed provider: override the phase with \`task route\` or restore the provider and report readiness again`,
                 });
                 continue;
             }
@@ -3057,23 +2554,14 @@ async function tickCommand(flags, context) {
         // worktree, no wake churn. The provider source is the AUTHORITATIVE
         // one per road: the sealed approval snapshot for ordinary builds (the
         // same snapshot the dispatch proof enforces), the plan resolver for
-        // planner runs. Attended work is claude-only and excluded; tournament
-        // contestants cannot be tier-2 today (the money gate refuses them at
-        // filing). A missing or malformed snapshot is NOT skipped here — the
+        // planner runs. A missing
+        // or malformed snapshot is NOT skipped here — the
         // existing approval refusals own that road, and attestation must
         // never mask them. The gateway re-checks before spawn; this skip only
         // keeps the normal road cheap.
-        const skipProviders = attendedSpec !== null
-            ? []
-            : racedAhead !== null
-                ? // The contest road (slice B): every lane's provider, so an
-                    // out-of-range attested lane skips the WHOLE contest before
-                    // any claim — running a subset is a different contest than
-                    // the one the operator signed.
-                    racedAhead.agents.filter(agent => isProviderId(agent.provider)).map(agent => agent.provider)
-                : wantsPlan
-                    ? [spec.provider]
-                    : [store.getScope(id)?.approvedProfile?.provider].filter((one) => one !== null && one !== undefined);
+        const skipProviders = wantsPlan
+            ? [spec.provider]
+            : [store.getScope(id)?.approvedProfile?.provider].filter((one) => one !== null && one !== undefined);
         let unattestedLane = null;
         for (const candidate of new Set(skipProviders)) {
             if (attestationOf(candidate) === null)
@@ -3094,10 +2582,9 @@ async function tickCommand(flags, context) {
         }
         // Sprint 8: the organisation policy before the claim (build() looks again before spawn, and lowers).
         {
-            const sealed = wantsPlan || wantsScout ? null : store.approvedChainOf(id)?.[0]?.profile ?? store.getScope(id)?.approvedProfile ?? null;
-            const refused = racedAhead !== null ? policyHold(racedAhead.agents.map(agent => ({ provider: agent.provider, model: agent.model })))
-                : policyHold([attendedSpec !== null ? { profile: attendedSpec.profile, attended: true } : sealed !== null ? { profile: sealed }
-                        : { provider: spec.provider, model: spec.model, ...(wantsPlan ? { session: "planning" } : wantsScout ? { session: "scouting" } : {}) }]);
+            const sealed = wantsPlan || wantsScout ? null : store.getScope(id)?.approvedProfile ?? null;
+            const refused = policyHold([sealed !== null ? { profile: sealed }
+                    : { provider: spec.provider, model: spec.model, ...(wantsPlan ? { session: "planning" } : wantsScout ? { session: "scouting" } : {}) }]);
             if (refused !== null) {
                 dispatched.push({ id, outcome: "skipped", reason: "policy", detail: refused });
                 continue;
@@ -3106,18 +2593,16 @@ async function tickCommand(flags, context) {
         // v105: a monthly budget that stops new API work, used up: the task waits in the queue until the month turns or the
         // budget is raised (the 100% alert went out; the task page and the spend page say why).
         // Budgets are dollars: work that runs only on subscriptions passes (see Store.budgetGate).
-        // A chain's base entry says how it bills (a pinned key); otherwise each provider as it bills now.
-        const chainBase = wantsPlan || wantsScout ? undefined : store.approvedChainOf(id)?.[0];
-        const budgeted = budgetHold(ref.id, chainBase !== undefined ? [{ provider: chainBase.profile.provider, billing: chainBase.authMode }] : store.agentsFor([spec.provider, ...skipProviders]));
+        // Each provider as it bills now.
+        const budgeted = budgetHold(ref.id, store.agentsFor([spec.provider, ...skipProviders]));
         if (budgeted.over !== null) {
             dispatched.push({ id, outcome: "skipped", reason: "budget", detail: budgetWords(budgeted) });
             continue;
         }
         // THE DAILY RAIL (modes chain D4): a live mode's run cap reserves at
         // admission — atomic, so two watch loops cannot both slip under it.
-        // Attended sessions and raced tasks reserve on their own roads below;
-        // an ordinary/planner start reserves ONE here. No mode = no-op.
-        if (attendedDispatch === null && racedAhead === null && repo !== null) {
+        // An ordinary/planner start reserves ONE here. No mode = no-op.
+        if (repo !== null) {
             const railed = store.reserveModeRail(repo, 1, clock());
             if (!railed.ok) {
                 dispatched.push({ id, outcome: "skipped", reason: railed.rail, detail: railed.detail });
@@ -3174,240 +2659,6 @@ async function tickCommand(flags, context) {
             continue;
         }
         const lease = claimed.claim.leaseId;
-        // A tournament rides this claim (stage 3b): approved race terms send N
-        // agents instead of one builder. Everything after admission either
-        // reaches the ready barrier for ALL agents or interrupts the whole
-        // tournament — a partial race is never dispatched (finding 19).
-        const raceTerms = racedAhead;
-        if (raceTerms !== null && raceTerms.approvedDigest === raceTerms.raceDigest) {
-            const admittedKind = raceTerms.kind;
-            // The rail reserves every lane at once (D4) — a tournament is N
-            // starts from one filing. Refused before any skeleton exists.
-            if (repo !== null) {
-                const railedRace = store.reserveModeRail(repo, raceTerms.n, clock());
-                if (!railedRace.ok) {
-                    dispatched.push({ id, outcome: "skipped", reason: railedRace.rail, detail: railedRace.detail });
-                    continue;
-                }
-            }
-            const scopeRow = store.getScope(id);
-            const admitted = admitContest(store, {
-                taskId: id,
-                taskRef: ref.id,
-                runner,
-                leaseId: lease,
-                incarnation: text(flags, "incarnation") ?? null,
-                scopeDigest: scopeRow?.digest ?? "",
-                scopeApproved: scopeRow !== null && scopeApprovedForDispatch(store, ref.id, clock()),
-                // 'tasks' capacity mode keeps the claim-counted contract; the
-                // slot ledger records regardless (finding 26).
-                capacity: null,
-                quotaBlocked: (provider, model) => store.quotaState(runner, provider, model, clock())?.state ?? null,
-            }, clock());
-            if (!admitted.ok) {
-                release(store, lease, clock());
-                dispatched.push({ id, outcome: "skipped", reason: admitted.reason });
-                continue;
-            }
-            const interrupt = async (why, leasedPaths) => {
-                const fresh = store.getContest(admitted.contestId);
-                if (fresh !== null)
-                    store.casContestState(admitted.contestId, ["dispatching", "racing"], "interrupted", fresh.generation);
-                store.releaseSlotsForContest(admitted.contestId, clock());
-                for (const path of leasedPaths)
-                    await worktrees.release(path, clock());
-                release(store, lease, clock());
-                dispatched.push({ id, outcome: "failed", reason: why });
-                broke++;
-            };
-            const baseRead = await git("git", ["rev-parse", "HEAD"], { cwd: repo });
-            if (baseRead.code !== 0) {
-                await interrupt("contest-base", []);
-                continue;
-            }
-            const baseSha = baseRead.stdout.trim();
-            store.stampContestDispatch(admitted.contestId, baseSha, store.liveWorktreeSetup(repo)?.digest ?? null);
-            const agents = store.contestants(admitted.contestId);
-            const prepared = [];
-            let prepFailed = false;
-            for (const [index, agent] of agents.entries()) {
-                const leased = await worktrees.lease({ repo, branch: agent.branch, runner, taskRef: ref.id, now: clock(), base: baseSha });
-                if (!leased.ok) {
-                    prepFailed = true;
-                    break;
-                }
-                store.setContestantWorktree(agent.id, leased.worktree.path);
-                // The lane's provenance rides its insert (v48 integrity): the
-                // race-approved profile it will be proved against, exactly — the
-                // store's own answer, re-proved and bound to the lane there.
-                const laneStamp = store.laneAuthorityFor(agent.id);
-                if (laneStamp === null) {
-                    prepFailed = true;
-                    break;
-                }
-                // THE LANE ADMISSION (atomic authority closure): under the custody
-                // admitContest stamped — this lease, runner, and incarnation.
-                const admittedLane = store.admitContestLane({
-                    taskRef: ref.id,
-                    leaseId: lease,
-                    runner,
-                    incarnation: text(flags, "incarnation") ?? null,
-                    branch: agent.branch,
-                    worktree: leased.worktree.path,
-                    provider: agent.provider,
-                    model: agent.model,
-                    contestant: agent.id,
-                    now: clock(),
-                    route: laneStamp,
-                });
-                if (!admittedLane.ok) {
-                    prepFailed = true;
-                    break;
-                }
-                const contestantRun = admittedLane.runId;
-                // The lane's pointer was bound inside the insert above (raw
-                // authority repair); nothing claims it after the fact.
-                prepared.push({
-                    contestantId: agent.id,
-                    slotId: admitted.slotIds[index] ?? -1,
-                    runId: contestantRun,
-                    worktree: leased.worktree.path,
-                    branch: agent.branch,
-                    leaseEpoch: leased.worktree.leaseEpoch ?? null,
-                });
-            }
-            const freshContest = store.getContest(admitted.contestId);
-            if (prepFailed || freshContest === null || !crossReadyBarrier(store, freshContest, admitted.contestantIds)) {
-                await interrupt("contest-admission", prepared.map(one => one.worktree));
-                continue;
-            }
-            // Every agent is READY and nothing has spawned: cross into racing and
-            // spend. The builds run concurrently; the stop fence stops them all.
-            budgetGate = null; // the race's spend counts toward the next check
-            // v105: the lanes run at once, so they share what's left of a monthly budget.
-            const laneShare = budgeted.remainingMicrousd === null ? null : Math.max(1, Math.floor(budgeted.remainingMicrousd / Math.max(1, prepared.length)));
-            const settled = await Promise.allSettled(prepared.map(async (entry) => {
-                const agent = store.getContestant(entry.contestantId);
-                if (agent === null)
-                    throw new Error("contestant vanished");
-                store.casContestantState(entry.contestantId, ["ready"], "building", agent.generation);
-                return build(store, {
-                    taskId: id,
-                    taskRef: ref.id,
-                    runner,
-                    leaseId: lease,
-                    runId: entry.runId,
-                    evidenceRoot: context.evidenceRoot,
-                    worktree: entry.worktree,
-                    branch: entry.branch,
-                    now: clock(),
-                    clock,
-                    provider: agent.provider,
-                    contestProfile: contestantProfileOf(agent.provider, agent.model, agent.repairModel),
-                    // A comparison lane has no dollar cap — the sealed clock is the
-                    // bound; only race lanes carry the harness stop (E1).
-                    ...(agent.budgetMicrousd > 0 ? { maxBudgetUsd: Math.min(agent.budgetMicrousd, laneShare ?? Infinity) / 1_000_000 } : {}),
-                    onProviderSpawn: pid => {
-                        worktrees.recordProviderOccupancy(entry.worktree, runner, pid, entry.leaseEpoch);
-                        const facts = { run: entry.runId, contestant: entry.contestantId, incarnation: text(flags, "incarnation") ?? null, processGroup: pid };
-                        if (!store.markSlotRunning(entry.slotId, facts, clock()) && !store.refreshSlotProcess(entry.slotId, facts))
-                            throw new Error("the provider's execution slot no longer belongs to this attempt");
-                    },
-                    ...(context.agentRunner === undefined ? {} : { agent: context.agentRunner }),
-                    ...(context.gitRunner === undefined ? {} : { git: context.gitRunner }),
-                    ...(context.shouldStop === undefined ? {} : { shouldStop: context.shouldStop }),
-                });
-            }));
-            let lastAggregate = null;
-            for (const [index, entry] of prepared.entries()) {
-                const outcome = settled[index];
-                const run = store.getRun(entry.runId);
-                const measured = run === null || run.providerStartedAt === null
-                    ? 0
-                    : run.costUsd !== null
-                        ? Math.round(run.costUsd * 1_000_000)
-                        : null;
-                let contestantOutcome = "failed";
-                let committed = false;
-                if (outcome !== undefined && outcome.status === "fulfilled") {
-                    const result = outcome.value;
-                    if (result.ok && result.parked !== undefined) {
-                        contestantOutcome = "parked";
-                        // The question still reaches the operator, tagged with its agent;
-                        // decision-wait mechanics land in stage 4 — the card works today.
-                        const asked = result.parked.decision;
-                        const contestantDecision = store.saveDecision({
-                            run: entry.runId,
-                            contestant: entry.contestantId,
-                            urgency: asked.urgency,
-                            recap: asked.recap,
-                            question: asked.question,
-                            options: asked.options,
-                            recommendation: asked.recommendation,
-                            ...(asked.assignee === null ? {} : { assignee: asked.assignee }),
-                            ...(asked.deadline === null ? {} : { deadline: asked.deadline }),
-                        }, clock());
-                        store.enqueueNotification({
-                            source: { run: entry.runId },
-                            dedupeKey: `decision:${contestantDecision}`,
-                            kind: "decision",
-                            subject: `${id} parked a decision (${contestNoun(admittedKind)} agent)`,
-                            body: `\`toolroll decide ${contestantDecision}\``,
-                            pushClass: "decision",
-                            link: `/d/${contestantDecision}`,
-                        }, clock());
-                    }
-                    else if (result.ok) {
-                        contestantOutcome = "built";
-                        committed = result.committed;
-                    }
-                    else if (result.reason === "stopped") {
-                        contestantOutcome = "stopped";
-                    }
-                }
-                if (run !== null && run.outcome === null) {
-                    // The reason rides the run (slice B, E2 — kind-agnostic): "lane 3
-                    // failed" with no words when a binary drifted out of its attested
-                    // range is exactly the silence the attested runtime rules out.
-                    const laneReason = outcome !== undefined && outcome.status === "fulfilled" && !outcome.value.ok
-                        ? outcome.value.reason
-                        : undefined;
-                    store.finishRun(entry.runId, {
-                        outcome: contestantOutcome === "built" && !committed ? "no-change" : contestantOutcome === "stopped" ? "failed" : contestantOutcome === "parked" ? "parked" : contestantOutcome,
-                        committed,
-                        ...(laneReason === undefined ? {} : { reason: laneReason }),
-                        now: clock(),
-                    });
-                }
-                if (contestantOutcome === "parked") {
-                    // Custody (round-3 finding 29): who owns this checkout while the
-                    // question waits, and what exact state it was left in — the
-                    // resume verifies all of it before trusting the tree again.
-                    const headNow = await git("git", ["rev-parse", "HEAD"], { cwd: entry.worktree });
-                    const dirtyNow = await git("git", ["status", "--porcelain"], { cwd: entry.worktree });
-                    store.setContestantCustody(entry.contestantId, JSON.stringify({
-                        branch: entry.branch,
-                        head: headNow.code === 0 ? headNow.stdout.trim() : null,
-                        runner,
-                        dirty: dirtyNow.stdout.trim() !== "",
-                        at: clock().toISOString(),
-                    }));
-                }
-                await worktrees.release(entry.worktree, clock());
-                const final = finalizeContestant(store, {
-                    contestId: admitted.contestId,
-                    contestantId: entry.contestantId,
-                    runId: entry.runId,
-                    outcome: contestantOutcome,
-                    measuredMicrousd: measured,
-                    slotId: entry.slotId >= 0 ? entry.slotId : null,
-                }, clock());
-                if (final.aggregated !== null)
-                    lastAggregate = final.aggregated;
-            }
-            dispatched.push({ id, outcome: "contest", reason: lastAggregate ?? "racing" });
-            continue;
-        }
         if (wantsPlan) {
             // THE FILED REQUEST, FIRST (contract handoff, task 1): everything the
             // operator filed — scope, rubric, terms, revision brief, earlier
@@ -3480,11 +2731,11 @@ async function tickCommand(flags, context) {
             }
             catch (error) {
                 await worktrees.release(planLeased.worktree.path, clock());
-                const unrecorded = finalizePlanFailureFenced(store, {
-                    leaseId: lease,
+                const unrecorded = finalize(store, lease, {
+                    kind: "plan-failure",
                     runId: planRunId,
                     taskId: id,
-                    kind: "failure",
+                    failure: "failure",
                     message: `the planner's source could not be recorded: ${error instanceof Error ? error.message : String(error)}`,
                     now: clock(),
                 });
@@ -3520,8 +2771,8 @@ async function tickCommand(flags, context) {
             });
             await worktrees.release(planLeased.worktree.path, clock());
             if (outcome.ok && "parked" in outcome) {
-                const sealed = finalizeParkFenced(store, {
-                    leaseId: lease,
+                const sealed = finalize(store, lease, {
+                    kind: "park",
                     runId: planRunId,
                     taskId: id,
                     decision: outcome.parked.decision,
@@ -3544,8 +2795,8 @@ async function tickCommand(flags, context) {
                 continue;
             }
             if (outcome.ok) {
-                const sealed = finalizePlanFenced(store, {
-                    leaseId: lease,
+                const sealed = finalize(store, lease, {
+                    kind: "plan",
                     runId: planRunId,
                     taskId: id,
                     plan: outcome.drafted.plan,
@@ -3578,11 +2829,11 @@ async function tickCommand(flags, context) {
                 }
                 continue;
             }
-            const sealedFailure = finalizePlanFailureFenced(store, {
-                leaseId: lease,
+            const sealedFailure = finalize(store, lease, {
+                kind: "plan-failure",
                 runId: planRunId,
                 taskId: id,
-                kind: outcome.kind,
+                failure: outcome.kind,
                 ...(outcome.kind === "malformed" ? { malformed: outcome.reason === "malformed-decision" ? "decision" : "plan" } : {}),
                 message: outcome.message,
                 now: clock(),
@@ -3608,7 +2859,7 @@ async function tickCommand(flags, context) {
             const scopeRow = store.getScope(id);
             // Approvals bind exact routing for a scout exactly as for a build:
             // the pinned profile is proved BEFORE the workspace is leased.
-            const proof = proveApprovedProfile(scopeRow, null, {
+            const proof = proveApprovedProfile(scopeRow, {
                 provider: spec.provider,
                 model: spec.model ?? undefined,
                 maxTurns: undefined,
@@ -3687,8 +2938,8 @@ async function tickCommand(flags, context) {
             // never silently kept; `toolroll worktrees` lists it.
             const leftover = discarded.ok ? {} : { detail: `checkout kept: ${discarded.message}` };
             if (scouted.ok && "parked" in scouted) {
-                const sealed = finalizeParkFenced(store, {
-                    leaseId: lease,
+                const sealed = finalize(store, lease, {
+                    kind: "park",
                     runId: scoutRunId,
                     taskId: id,
                     decision: scouted.parked.decision,
@@ -3706,8 +2957,8 @@ async function tickCommand(flags, context) {
                 continue;
             }
             if (scouted.ok) {
-                const sealed = finalizeScoutFenced(store, {
-                    leaseId: lease,
+                const sealed = finalize(store, lease, {
+                    kind: "scout",
                     runId: scoutRunId,
                     taskId: id,
                     report: scouted.reported.report,
@@ -3727,11 +2978,11 @@ async function tickCommand(flags, context) {
                 }
                 continue;
             }
-            const sealedFailure = finalizeScoutFailureFenced(store, {
-                leaseId: lease,
+            const sealedFailure = finalize(store, lease, {
+                kind: "scout-failure",
                 runId: scoutRunId,
                 taskId: id,
-                kind: scouted.kind,
+                failure: scouted.kind,
                 ...(scouted.kind === "malformed" ? { malformed: scouted.reason === "malformed-decision" ? "decision" : "report" } : {}),
                 message: scouted.message,
                 now: clock(),
@@ -3793,102 +3044,12 @@ async function tickCommand(flags, context) {
         // a row with no outcome — an attempt that vanished, visible by morning.
         // Its route provenance (v47) is written in the same admission
         // transaction; build() then refuses to spend as anything else.
-        // The chain custody for this run (E3b/E3d, atomic since the v48 authority repair): a parked
-        // chain tail hands custody to this successor through the PROVEN resume
-        // transfer, otherwise a chain approval opens its fresh cycle bound to
-        // this run — both proved and written IN the run's own insert. A
-        // single-profile approval — every task until an operator configures a
-        // fallback chain — binds nothing, so this is inert by default. A
-        // binding that cannot be proved, or a stamp the task's authority does
-        // not admit, rolls the insert back: no row, no claim kept, said why.
+        // A stamp the task's authority does not admit rolls the insert back:
+        // no row, no claim kept, said why.
         const buildStamp = routeStamp("build");
         let runId;
         try {
-            if (parkedEntry !== null && parkedChainTail !== null && liveCycle !== null && buildStamp !== null) {
-                // A parked FALLBACK tail's successor is admitted by the one
-                // fallback road (v48 integrity): every fact the tail's binding
-                // states is presented and re-proved — cycle, index, digest, auth
-                // mode, provider, exact model, repair binding, the sealed-profile
-                // mirror, the approved chain — with the parked run as the live
-                // tail, before any row exists.
-                const chain = store.approvedChainOf(id);
-                const entry = chain === null ? undefined : chain[parkedEntry.index];
-                const mirror = store.getScope(id)?.approvedProfile ?? null;
-                if (chain === null || entry === undefined || mirror === null || parkedChainTail.entryDigest == null || parkedChainTail.authMode == null || spec.model === null) {
-                    throw new Error(`${id}: the parked attempt #${parkedChainTail.id}'s fallback binding cannot be restated against the approved chain — nothing resumes it`);
-                }
-                const admitted = store.admitFallback({
-                    kind: "resume",
-                    parkedRun: parkedChainTail.id,
-                    cycleId: liveCycle.id,
-                    expectCursor: parkedEntry.index,
-                    expectTail: parkedChainTail.id,
-                    entryDigest: parkedChainTail.entryDigest,
-                    authMode: parkedChainTail.authMode,
-                    repairModel: entry.profile.repairModel === "inherit" ? entry.profile.model : entry.profile.repairModel,
-                    approved: { chainDigest: chainDigestOf(chain), profile: mirror },
-                    run: {
-                        taskRef: ref.id, leaseId: lease, runner, branch, worktree: leased.worktree.path, provider: spec.provider, model: spec.model,
-                        // The recovered draft's lineage rides the insert (raw
-                        // authority repair) — never a later stamp.
-                        ...(leased.resumedFromRun === undefined ? {} : { recoveredFrom: leased.resumedFromRun }),
-                    },
-                    route: buildStamp,
-                }, clock());
-                if (!admitted.ok)
-                    throw new Error(admitted.problem);
-                runId = admitted.runId;
-            }
-            else if (attendedDispatch !== null && attendedSpec !== null) {
-                // THE ATTENDED ADMISSION (atomic authority closure): the one
-                // watched attempt opens under exactly this authorization — named
-                // by id, runner, and generation — and the insert consumes its
-                // attempt and binds the row to it in one transaction.
-                if (buildStamp === null)
-                    throw new Error(`${id}: the attended authorization's pinned profile presents no build authority`);
-                // THE SIGNED HEAD, BEFORE THE ATTEMPT IS SPENT (final authority
-                // closure): the authorization signed the exact commit the watched
-                // attempt would start from. The leased worktree's HEAD is read
-                // here and held to it BEFORE admission — a head that moved opens
-                // no run and spends no attempt; the authorization closes in the
-                // refusal's words so the operator sees why and may authorize
-                // again at today's head. The coordinator's final proof re-reads
-                // the same term against the captured base revision.
-                let signedHead = null;
-                try {
-                    const terms = JSON.parse(attendedDispatch.termsJson);
-                    signedHead = typeof terms.head === "string" && terms.head !== "" ? terms.head : null;
-                }
-                catch {
-                    signedHead = null;
-                }
-                if (signedHead === null) {
-                    store.closeAuthorization(attendedDispatch.id, "refused:stale-authorization", clock());
-                    throw new StaleAuthorization(`${id}: the attended authorization ${attendedDispatch.id} signs no readable head — nothing opens under it`);
-                }
-                const headRead = await git("git", ["rev-parse", "HEAD"], { cwd: leased.worktree.path });
-                const headNow = headRead.code === 0 ? headRead.stdout.trim() : "";
-                if (headNow !== signedHead) {
-                    store.closeAuthorization(attendedDispatch.id, "refused:stale-authorization", clock());
-                    throw new StaleAuthorization(`${id}: the head moved since the attended authorization ${attendedDispatch.id} was signed (${signedHead.slice(0, 12)} → ${headNow === "" ? "unreadable" : headNow.slice(0, 12)}) — no run opened, no attempt spent; authorize it again at today's head`);
-                }
-                const admittedAttended = store.admitAttended({
-                    taskRef: ref.id,
-                    leaseId: lease,
-                    runner,
-                    branch,
-                    worktree: leased.worktree.path,
-                    provider: spec.provider,
-                    ...(spec.model === null ? {} : { model: spec.model }),
-                    authorization: { id: attendedDispatch.id, runner: attendedDispatch.runner, generation: attendedDispatch.runnerGeneration },
-                    now: clock(),
-                    route: buildStamp,
-                });
-                if (!admittedAttended.ok)
-                    throw new Error(admittedAttended.problem);
-                runId = admittedAttended.runId;
-            }
-            else if (leased.resumedFromRun !== undefined) {
+            if (leased.resumedFromRun !== undefined) {
                 // THE RECOVERED-DRAFT ADMISSION (atomic authority closure): the
                 // fresh attempt inherits the interrupted attempt's draft, proved
                 // this task's own interrupted builder in this very worktree.
@@ -3905,7 +3066,6 @@ async function tickCommand(flags, context) {
                     recoveredFrom: leased.resumedFromRun,
                     now: clock(),
                     route: buildStamp,
-                    custody: parkedChainTail !== null && liveCycle !== null ? { kind: "resume", parkedRun: parkedChainTail.id } : { kind: "base" },
                 });
                 if (!admittedRecovered.ok)
                     throw new Error(admittedRecovered.problem);
@@ -3922,14 +3082,13 @@ async function tickCommand(flags, context) {
                     ...(spec.model === null ? {} : { model: spec.model }),
                     now: clock(),
                     ...(buildStamp === null ? {} : { route: buildStamp }),
-                    custody: parkedChainTail !== null && liveCycle !== null ? { kind: "resume", parkedRun: parkedChainTail.id } : { kind: "base" },
                 });
             }
         }
         catch (error) {
             await worktrees.release(leased.worktree.path, clock());
             release(store, lease, clock());
-            dispatched.push({ id, outcome: "skipped", reason: error instanceof StaleAuthorization ? "stale-authorization" : "admission-refused", detail: error instanceof Error ? error.message : String(error) });
+            dispatched.push({ id, outcome: "skipped", reason: "admission-refused", detail: error instanceof Error ? error.message : String(error) });
             continue;
         }
         if (leased.resumedFromRun !== undefined) {
@@ -3945,11 +3104,9 @@ async function tickCommand(flags, context) {
         const budgetLeft = budgeted.remainingMicrousd !== null && MONEY_CAPABILITIES[spec.provider].nativeDollarCapFlag !== null ? Math.max(1, budgeted.remainingMicrousd) : null;
         const capMicrousd = [scopeBudget, backstop, budgetLeft].reduce((least, one) => one === null ? least : least === null ? one : Math.min(least, one), null);
         if (capMicrousd !== null && MONEY_CAPABILITIES[spec.provider].nativeDollarCapFlag === null) {
-            // The run row and any chain cycle already exist — FINISH and RESOLVE
-            // them (E3d verify, R6): a refused-but-open run would defer a chain
-            // task forever and read as a vanished attempt everywhere else.
+            // The run row already exists — FINISH it: a refused-but-open run
+            // would read as a vanished attempt everywhere else.
             store.finishRun(runId, { outcome: "refused", reason: "budget-unenforceable", now: clock() });
-            store.resolveChainOnRunEnd(ref.id, id, repo, runId, clock());
             release(store, lease, clock());
             await worktrees.release(leased.worktree.path, clock());
             dispatched.push({ id, outcome: "skipped", reason: "budget-unenforceable" });
@@ -3988,29 +3145,7 @@ async function tickCommand(flags, context) {
             ...(context.shouldStop === undefined ? {} : { shouldStop: context.shouldStop }),
             ...(leased.resumedFromRun === undefined ? {} : { recoveredDraftRun: leased.resumedFromRun }),
             ...(leased.recoveryKind === undefined ? {} : { recoveredDraftKind: leased.recoveryKind }),
-            ...(attendedDispatch === null || context.heldCoordinator === undefined
-                ? {}
-                : {
-                    attended: {
-                        authorization: attendedDispatch,
-                        coordinator: context.heldCoordinator,
-                        upIncarnation: context.upIncarnation ?? "unknown",
-                        socketDir: context.heldSocketDir ?? tmpdir(),
-                        releaseWorktree: async (path) => worktrees.release(path, clock()),
-                        dispose: { repo, origin: ref.origin, provider: spec.provider, model: spec.model },
-                        ...(context.heldStarter === undefined ? {} : { starter: context.heldStarter }),
-                        ...(context.heldGraceMs === undefined ? {} : { graceMs: context.heldGraceMs }),
-                        ...(context.maxHeldSessions === undefined ? {} : { maxHeldSessions: context.maxHeldSessions }),
-                    },
-                }),
         });
-        // THE HELD HANDOFF (Phase 2, v2 S0d): ownership transferred — the
-        // coordinator owns run, lease, and worktree; this pass releases and
-        // settles NOTHING and moves on. Nonblocking is the whole point.
-        if (result.ok && result.parked === undefined && result.held === true) {
-            dispatched.push({ id, outcome: "held", branch, worktree: leased.worktree.path });
-            continue;
-        }
         // Handed back either way; a tree with somebody's work in it comes back
         // unverified rather than cleaned, same as `build`.
         await worktrees.release(leased.worktree.path, clock());
@@ -4034,13 +3169,6 @@ async function tickCommand(flags, context) {
             evidenceRoot: context.evidenceRoot,
             clock,
         }, result);
-        // The chain step (E3c/E3d): EVERY concluded run resolves its cycle
-        // through the one resolver — success closes, an ordinary end closes, a
-        // parked tail stays open for repair, and a recognized eligible
-        // exhaustion advances to pending-admission (fail-closed at every gate,
-        // the live grant re-proved in its own transaction). Inert unless the
-        // task filed under an explicit chain — no cycle, fast no-op.
-        store.resolveChainOnRunEnd(ref.id, id, repo, runId, clock());
         switch (disposition.kind) {
             case "parked":
                 dispatched.push({ id, outcome: "parked", reason: `decision:${disposition.decisionId}`, worktree: leased.worktree.path });
@@ -4062,453 +3190,51 @@ async function tickCommand(flags, context) {
                 dispatched.push({ id, outcome: "failed", reason: "fenced", branch, worktree: leased.worktree.path });
                 broke++;
                 break;
-            case "skipped":
-                dispatched.push({ id, outcome: "skipped", reason: disposition.reason });
-                break;
-            case "fenced":
-                dispatched.push({ id, outcome: "failed", reason: "fenced", worktree: leased.worktree.path });
-                broke++;
-                break;
-            case "malformed":
-                dispatched.push({ id, outcome: "failed", reason: disposition.sealed ? "malformed-decision" : "fenced", worktree: leased.worktree.path });
-                broke++;
-                break;
-            case "failed":
-                dispatched.push({
-                    id,
-                    outcome: "failed",
-                    reason: disposition.sealed
-                        ? disposition.disposition === "auth-expired"
-                            ? "auth-expired — requeued, sign-in needed"
-                            : `${disposition.failureClass}${disposition.disposition === "backoff" ? ` — retry ${disposition.strikes}/3` : disposition.disposition === "stalled" ? " — stalled" : ""}`
-                        : "fenced",
-                    ...(result.ok ? {} : { detail: result.message.slice(0, 200) }),
-                    worktree: leased.worktree.path,
-                });
-                broke++;
-                break;
-            case "recorded":
-                // The standalone-only arm; unreachable under the tick policy.
-                break;
-            case "stopped":
-                // An operator's stop (v52): interrupted with its work preserved —
-                // not a break, not a strike; the task waits under the stop's hold.
-                dispatched.push({ id, outcome: "stopped", reason: `stop:${disposition.stopRun}`, branch, worktree: leased.worktree.path });
-                break;
-            case "handed-back":
-                // The service is stopping: the work is kept and the task requeued
-                // to resume on the next pass — not a break, not a strike.
-                dispatched.push({ id, outcome: "stopped", reason: "service-stop", detail: result.ok ? "" : result.message.slice(0, 200), branch, worktree: leased.worktree.path });
-                break;
-            case "invariant":
-                dispatched.push({ id, outcome: "failed", reason: disposition.reason });
-                broke++;
-                break;
-        }
-    }
-    untakenTrial?.giveBack();
-    // THE CHAIN ADMISSION PASS (E3d): cycles a recognized exhaustion advanced
-    // to pending-admission dispatch their NEXT approved entry here — the ONLY
-    // road that runs a fallback entry. Every authority is re-derived inside
-    // admitNextChainEntry (approved chain standing + digest match + LIVE
-    // paid-fallback grant + the single-use pending edge); this loop carries
-    // only claim, worktree, and rail, and its run then re-proves the
-    // chain-entry dispatch proof inside build() before any money moves.
-    for (const pending of buildsOnly ? [] : store.pendingChainAdmissions(repo)) {
-        settleFinished();
-        if (context.shouldStop?.() === true || context.shouldPauseAdmission?.() === true)
-            break;
-        if (built >= max)
-            break;
-        // v105: a fallback billed to an API key spends dollars: a used-up budget holds it like any new work.
-        const chainEntry = store.approvedChainOf(pending.taskId)?.[pending.cursor];
-        const chainAgents = chainEntry === undefined ? [] : [{ provider: chainEntry.profile.provider, billing: chainEntry.authMode }];
-        // Sprint 8: a fallback entry the organisation policy doesn't allow never starts.
-        const chainPolicy = chainEntry === undefined ? null : policyHold([{ profile: chainEntry.profile }]);
-        if (chainPolicy !== null) {
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: "policy", detail: chainPolicy });
-            continue;
-        }
-        const chainBudget = budgetHold(pending.taskRef, chainAgents);
-        if (chainBudget.over !== null) {
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: "budget", detail: budgetWords(chainBudget) });
-            continue;
-        }
-        const railed = store.reserveModeRail(repo, 1, clock());
-        if (!railed.ok) {
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: railed.rail, detail: railed.detail });
-            continue;
-        }
-        // A peek proves the entry EXISTS and carries an enforceable budget
-        // BEFORE any claim or run row exists (review findings 4/6); the
-        // admission below re-derives the entry as authority in its transaction.
-        const peek = store.approvedChainOf(pending.taskId)?.[pending.cursor];
-        if (peek === undefined) {
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: "fallback-unadmittable" });
-            continue;
-        }
-        const scopeBudget = store.getScope(pending.taskId)?.budgetMicrousd ?? null;
-        const backstop = store.getSpendDefaults()?.buildPerRunMicrousd ?? null;
-        const capMicrousd = scopeBudget === null ? backstop : backstop === null ? scopeBudget : Math.min(scopeBudget, backstop);
-        if (capMicrousd !== null && MONEY_CAPABILITIES[peek.profile.provider].nativeDollarCapFlag === null) {
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: "budget-unenforceable" });
-            continue;
-        }
-        // The sign-in pause holds a fallback entry too: the admission stays pending until its provider works.
-        const entryGate = signInGate(store, [peek.profile.provider], clock(), pending.taskRef);
-        if (entryGate.waiting !== null) {
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: "signed-out", detail: signInWords(entryGate.waiting) });
-            continue;
-        }
-        // The FALLBACK claim (finding 4): every acquireIfReady gate — task
-        // state, non-backoff holds, blockers, approved scope + mode belt,
-        // capability, capacity, and quota keyed by the PINNED credential —
-        // with only the predecessor's backoff exempted.
-        const claimed = acquireFallback(store, pending.taskRef, runner, {
-            now: clock(),
-            token,
-            ttlMs: leaseTtlMs,
-            repo,
-            provider: peek.profile.provider,
-            model: peek.profile.model,
-            authMode: peek.authMode,
-            // The watch incarnation rides the claim (F+G review, finding 3): a
-            // daemon that dies after this claim — admitted or not — must be
-            // recoverable by its successor's incarnation takeover, exactly like
-            // the ordinary road.
-            ...(text(flags, "incarnation") === undefined ? {} : { incarnation: text(flags, "incarnation") }),
-            ...(text(flags, "max-open-decisions") === undefined
-                ? {}
-                : { maxOpenDecisions: Number(text(flags, "max-open-decisions")) }),
-        });
-        if (!claimed.ok) {
-            entryGate.giveBack();
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: claimed.reason });
-            continue;
-        }
-        const lease = claimed.claim.leaseId;
-        const branch = await existingOrFirst(taskBranches(pending.taskId), async (one) => (await git("git", ["rev-parse", "--verify", "--quiet", `refs/heads/${one}`], { cwd: repo })).code === 0);
-        const exists = await git("git", ["rev-parse", "--verify", "--quiet", `refs/heads/${branch}`], { cwd: repo });
-        const leased = await worktrees.lease({
-            repo,
-            branch,
-            runner,
-            taskRef: pending.taskRef,
-            now: clock(),
-            ...(exists.code === 0 ? {} : { base }),
-            reclaim: { evidenceRoot: context.evidenceRoot },
-        });
-        if (!leased.ok) {
-            release(store, lease, clock());
-            dispatched.push({ id: pending.taskId, outcome: "failed", reason: leased.reason });
-            broke++;
-            continue;
-        }
-        const admitted = store.admitNextChainEntry(pending.cycleId, { leaseId: lease, runner, branch, worktree: leased.worktree.path, ...(leased.resumedFromRun === undefined ? {} : { recoveredFrom: leased.resumedFromRun }) }, clock());
-        if (!admitted.ok) {
-            await worktrees.release(leased.worktree.path, clock());
-            release(store, lease, clock());
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: `fallback-${admitted.reason}`, ...(admitted.detail === undefined ? {} : { detail: admitted.detail }) });
-            continue;
-        }
-        if (leased.resumedFromRun !== undefined) {
-            store.addRunNote(admitted.runId, "Toolroll", `Recovered the ${leased.recoveryKind === "completed" ? "completed source draft" : "work-in-progress draft"} from interrupted attempt #${leased.resumedFromRun}. This fresh attempt is reviewing and verifying it; the safety patch is retained.`, clock());
-        }
-        // The effective cap, RE-DERIVED after admission (E3d verify, R6): the
-        // pre-claim value is a survey; a backstop set while the claim and
-        // worktree awaits ran must govern the spend that actually happens. The
-        // belt then re-proves capability against the ADMITTED provider — and on
-        // refusal the run is FINISHED and its cycle resolved, never abandoned.
-        const scopeBudgetNow = store.getScope(pending.taskId)?.budgetMicrousd ?? null;
-        const backstopNow = store.getSpendDefaults()?.buildPerRunMicrousd ?? null;
-        const budgetNow = chainEntry?.authMode === "api-key" ? budgetHold(pending.taskRef, chainAgents).remainingMicrousd : null;
-        const capNow = [scopeBudgetNow, backstopNow, budgetNow === null ? null : Math.max(1, budgetNow)]
-            .reduce((least, one) => one === null ? least : least === null ? one : Math.min(least, one), null);
-        if (capNow !== null && MONEY_CAPABILITIES[admitted.provider].nativeDollarCapFlag === null) {
-            store.finishRun(admitted.runId, { outcome: "refused", reason: "budget-unenforceable", now: clock() });
-            store.resolveChainOnRunEnd(pending.taskRef, admitted.taskId, repo, admitted.runId, clock());
-            release(store, lease, clock());
-            await worktrees.release(leased.worktree.path, clock());
-            dispatched.push({ id: pending.taskId, outcome: "skipped", reason: "budget-unenforceable" });
-            continue;
-        }
-        budgetGate = null; // this build's spend counts toward the next check
-        const result = await build(store, {
-            taskId: pending.taskId,
-            taskRef: pending.taskRef,
-            runner,
-            leaseId: lease,
-            runnerToken: token,
-            runId: admitted.runId,
-            evidenceRoot: context.evidenceRoot,
-            worktree: leased.worktree.path,
-            branch,
-            now: clock(),
-            clock,
-            ...(capNow === null ? {} : { maxBudgetUsd: capNow / 1_000_000 }),
-            onProviderSpawn: pid => {
-                worktrees.recordProviderOccupancy(leased.worktree.path, runner, pid, leased.worktree.leaseEpoch);
-            },
-            provider: admitted.provider,
-            model: admitted.model,
-            ...(context.agentRunner === undefined ? {} : { agent: context.agentRunner }),
-            ...(context.gitRunner === undefined ? {} : { git: context.gitRunner }),
-            ...(context.shouldStop === undefined ? {} : { shouldStop: context.shouldStop }),
-            ...(leased.resumedFromRun === undefined ? {} : { recoveredDraftRun: leased.resumedFromRun }),
-            ...(leased.recoveryKind === undefined ? {} : { recoveredDraftKind: leased.recoveryKind }),
-        });
-        await worktrees.release(leased.worktree.path, clock());
-        const disposition = disposeBuildOutcome({
-            store,
-            policy: "tick",
-            leaseId: lease,
-            runId: admitted.runId,
-            taskId: pending.taskId,
-            taskRef: pending.taskRef,
-            runner,
-            repo,
-            branch,
-            origin: "ours",
-            provider: admitted.provider,
-            model: admitted.model,
-            worktreePath: leased.worktree.path,
-            evidenceRoot: context.evidenceRoot,
-            clock,
-        }, result);
-        // The same one resolver: a fallback entry that itself exhausts advances
-        // again; one that succeeds or ordinarily ends closes its cycle.
-        store.resolveChainOnRunEnd(pending.taskRef, pending.taskId, repo, admitted.runId, clock());
-        switch (disposition.kind) {
-            case "built":
-                dispatched.push({ id: pending.taskId, outcome: "built", committed: disposition.committed, branch, worktree: leased.worktree.path });
-                built++;
-                break;
-            case "stopped":
-                dispatched.push({ id: pending.taskId, outcome: "stopped", reason: `stop:${disposition.stopRun}`, branch, worktree: leased.worktree.path });
-                break;
-            case "handed-back":
-                dispatched.push({ id: pending.taskId, outcome: "stopped", reason: "service-stop", branch, worktree: leased.worktree.path });
-                break;
-            case "parked":
-                dispatched.push({ id: pending.taskId, outcome: "parked", reason: `decision:${disposition.decisionId}`, worktree: leased.worktree.path });
-                parked++;
-                break;
-            case "skipped":
-                dispatched.push({ id: pending.taskId, outcome: "skipped", reason: disposition.reason });
-                break;
-            default:
-                dispatched.push({ id: pending.taskId, outcome: "failed", reason: "fallback-attempt", worktree: leased.worktree.path });
-                broke++;
-                break;
-        }
-    }
-    // THE CONTINUATION PASS (Phase 2E, A4): open continuation authorizations
-    // named to this runner dispatch here — the finished parent task never
-    // re-enters the queue; the authorization is the claimable unit (v3 R7).
-    // Only a co-located coordinator can hold the session, and everything
-    // else (liveness, one attempt, the final proof at the parent's exact
-    // head) is re-proved on the way in.
-    if (context.heldCoordinator !== undefined && !buildsOnly) {
-        for (const continuation of store.openContinuationAuthorizations(runner)) {
-            if (context.shouldStop?.() === true || context.shouldPauseAdmission?.() === true)
-                break;
-            const watching = attendedLivenessState(continuation.lastBeatAt === null ? null : Date.parse(continuation.lastBeatAt), clock().getTime(), Date.parse(continuation.absoluteExpiry));
-            if (watching !== "live" && watching !== "grace")
-                continue;
-            const parent = continuation.parentRun === null ? null : store.getRun(continuation.parentRun);
-            const parentRef = parent === null ? null : store.refForId(parent.taskRef);
-            if (parent === null || parentRef === null)
-                continue;
-            // A continuation continues WORK: a branchless parent (the reviewer
-            // role, v29) has no workspace to lease and can never be continued —
-            // and this guard is what keeps "null" out of `git worktree add`.
-            if (parent.branch === null)
-                continue;
-            const parentBranch = parent.branch;
-            if (parentRef.repo !== null && parentRef.repo !== repo)
-                continue;
-            const taskId = parentRef.externalId;
-            let pinned = null;
-            try {
-                const terms = JSON.parse(continuation.termsJson);
-                const profile = profileFromJson(typeof terms.profileJson === "string" ? terms.profileJson : null);
-                if (profile !== null)
-                    pinned = { provider: profile.provider, model: profile.model, digest: profileDigestOf(profile), profile };
-            }
-            catch {
-                pinned = null;
-            }
-            if (pinned === null) {
-                dispatched.push({ id: taskId, outcome: "skipped", reason: "attended-only", detail: "the continuation's pinned profile cannot be read" });
-                continue;
-            }
-            // Sprint 8: an attended continuation runs at exactly its signed terms, or not at all.
-            const continuationPolicy = policyHold([{ profile: pinned.profile, attended: true }]);
-            if (continuationPolicy !== null) {
-                dispatched.push({ id: taskId, outcome: "skipped", reason: "policy", detail: continuationPolicy });
-                continue;
-            }
-            const continuationBudget = budgetHold(parent.taskRef, store.agentsFor([pinned.provider]));
-            if (continuationBudget.over !== null) {
-                dispatched.push({ id: taskId, outcome: "skipped", reason: "budget", detail: budgetWords(continuationBudget) });
-                continue;
-            }
-            // The sign-in pause holds a continuation too: its authorization stays open and it starts once the sign-in works.
-            const continuationGate = signInGate(store, [pinned.provider], clock(), parent.taskRef);
-            if (continuationGate.waiting !== null) {
-                dispatched.push({ id: taskId, outcome: "skipped", reason: "signed-out", detail: signInWords(continuationGate.waiting) });
-                continue;
-            }
-            const claimed = acquireContinuation(store, continuation, runner, { now: clock(), token, ttlMs: leaseTtlMs });
-            if (!claimed.ok) {
-                continuationGate.giveBack();
-                dispatched.push({ id: taskId, outcome: "skipped", reason: claimed.reason, ...("message" in claimed ? { detail: claimed.message } : {}) });
-                continue;
-            }
-            const lease = claimed.claim.leaseId;
-            // The parent's branch, at the head the terms signed — a moved branch
-            // fails the final proof with words naming the head.
-            const leased = await worktrees.lease({ repo, branch: parentBranch, runner, taskRef: parent.taskRef, now: clock() });
-            if (!leased.ok) {
-                release(store, lease, clock());
-                dispatched.push({ id: taskId, outcome: "failed", reason: leased.reason });
+            case "skipped":
+                dispatched.push({ id, outcome: "skipped", reason: disposition.reason });
+                break;
+            case "fenced":
+                dispatched.push({ id, outcome: "failed", reason: "fenced", worktree: leased.worktree.path });
                 broke++;
-                continue;
-            }
-            // THE SIGNED HEAD, BEFORE THE ATTEMPT IS SPENT (final admission
-            // closure): the continuation's terms signed the exact commit the
-            // parent finished at, and the leased worktree's HEAD is read here
-            // and held to it BEFORE admitAttended — the same order the queue's
-            // attended road keeps. A head the terms do not sign, or one that
-            // moved, opens no run, spends no attempt, invokes no provider: the
-            // worktree and claim are released and the authorization closes in
-            // the refusal's words — one terminal refusal, recorded once — so
-            // the operator authorizes again at today's head.
-            let signedContinuationHead = null;
-            try {
-                const terms = JSON.parse(continuation.termsJson);
-                signedContinuationHead = typeof terms.head === "string" && terms.head !== "" ? terms.head : null;
-            }
-            catch {
-                signedContinuationHead = null;
-            }
-            const continuationHeadRead = signedContinuationHead === null ? null : await git("git", ["rev-parse", "HEAD"], { cwd: leased.worktree.path });
-            if (continuationHeadRead !== null && continuationHeadRead.code !== 0) {
-                // The worktree's HEAD could not be READ — the head did not move,
-                // so nothing terminal is said about the authorization: custody is
-                // released and this tick records the failure; the next one reads
-                // again under the same open authorization.
-                await worktrees.release(leased.worktree.path, clock());
-                release(store, lease, clock());
-                dispatched.push({ id: taskId, outcome: "failed", reason: "head-unreadable", detail: `${taskId}: the leased worktree's HEAD could not be read (${continuationHeadRead.stderr.trim() || `git exited ${continuationHeadRead.code}`}) — the continuation authorization ${continuation.id} stays open; no run opened, no attempt spent` });
+                break;
+            case "malformed":
+                dispatched.push({ id, outcome: "failed", reason: disposition.sealed ? "malformed-decision" : "fenced", worktree: leased.worktree.path });
                 broke++;
-                continue;
-            }
-            const continuationHeadNow = continuationHeadRead === null ? "" : continuationHeadRead.stdout.trim();
-            if (signedContinuationHead === null || continuationHeadNow !== signedContinuationHead) {
-                await worktrees.release(leased.worktree.path, clock());
-                release(store, lease, clock());
-                store.closeAuthorization(continuation.id, "refused:stale-authorization", clock());
+                break;
+            case "failed":
                 dispatched.push({
-                    id: taskId,
-                    outcome: "skipped",
-                    reason: "stale-authorization",
-                    detail: signedContinuationHead === null
-                        ? `${taskId}: the continuation authorization ${continuation.id} signs no readable head — nothing opens under it; no run opened, no attempt spent`
-                        : `${taskId}: the head moved since the continuation authorization ${continuation.id} was signed (${signedContinuationHead.slice(0, 12)} → ${continuationHeadNow === "" ? "empty" : continuationHeadNow.slice(0, 12)}) — no run opened, no attempt spent; authorize it again at today's head`,
+                    id,
+                    outcome: "failed",
+                    reason: disposition.sealed
+                        ? disposition.disposition === "auth-expired"
+                            ? "auth-expired — requeued, sign-in needed"
+                            : `${disposition.failureClass}${disposition.disposition === "backoff" ? ` — retry ${disposition.strikes}/3` : disposition.disposition === "stalled" ? " — stalled" : ""}`
+                        : "fenced",
+                    ...(result.ok ? {} : { detail: result.message.slice(0, 200) }),
+                    worktree: leased.worktree.path,
                 });
-                continue;
-            }
-            // The continuation spends under the authorization's pinned profile
-            // and says so at insert (v48 integrity).
-            // THE ATTENDED ADMISSION (atomic authority closure): the continuation
-            // opens under exactly this authorization — id, runner, generation —
-            // continuing the finished parent it names; the insert consumes the
-            // one attempt and binds the row to it.
-            const admittedContinuation = store.admitAttended({
-                taskRef: parent.taskRef,
-                leaseId: lease,
-                runner,
-                branch: parentBranch,
-                worktree: leased.worktree.path,
-                parentRun: parent.id,
-                provider: pinned.provider,
-                ...(pinned.model === null ? {} : { model: pinned.model }),
-                authorization: { id: continuation.id, runner: continuation.runner, generation: continuation.runnerGeneration },
-                now: clock(),
-                route: { routeDigest: `profile:${pinned.digest}`, phase: "build", provider: pinned.provider, model: pinned.model, chosen: "legacy" },
-            });
-            if (!admittedContinuation.ok) {
-                await worktrees.release(leased.worktree.path, clock());
-                release(store, lease, clock());
-                dispatched.push({ id: taskId, outcome: "skipped", reason: "admission-refused", detail: admittedContinuation.problem });
-                continue;
-            }
-            const runId = admittedContinuation.runId;
-            budgetGate = null; // this build's spend counts toward the next check
-            const result = await build(store, {
-                taskId,
-                taskRef: parent.taskRef,
-                runner,
-                leaseId: lease,
-                runnerToken: token,
-                runId,
-                evidenceRoot: context.evidenceRoot,
-                worktree: leased.worktree.path,
-                branch: parentBranch,
-                now: clock(),
-                clock,
-                provider: pinned.provider,
-                ...(pinned.model === null ? {} : { model: pinned.model }),
-                ...(context.agentRunner === undefined ? {} : { agent: context.agentRunner }),
-                ...(context.gitRunner === undefined ? {} : { git: context.gitRunner }),
-                ...(context.shouldStop === undefined ? {} : { shouldStop: context.shouldStop }),
-                attended: {
-                    authorization: continuation,
-                    coordinator: context.heldCoordinator,
-                    upIncarnation: context.upIncarnation ?? "unknown",
-                    socketDir: context.heldSocketDir ?? tmpdir(),
-                    releaseWorktree: async (path) => worktrees.release(path, clock()),
-                    dispose: { repo, origin: parentRef.origin, provider: pinned.provider, model: pinned.model, policy: "continuation" },
-                    ...(context.heldStarter === undefined ? {} : { starter: context.heldStarter }),
-                    ...(context.heldGraceMs === undefined ? {} : { graceMs: context.heldGraceMs }),
-                    ...(context.maxHeldSessions === undefined ? {} : { maxHeldSessions: context.maxHeldSessions }),
-                },
-            });
-            if (result.ok && result.parked === undefined && result.held === true) {
-                dispatched.push({ id: taskId, outcome: "held", branch: parentBranch, worktree: leased.worktree.path });
-                continue;
-            }
-            // A refusal before the hold (stale proof, spawn failure): record it
-            // through the continuation policy — taskless, always — and release.
-            await worktrees.release(leased.worktree.path, clock());
-            const disposition = disposeBuildOutcome({
-                store,
-                policy: "continuation",
-                leaseId: lease,
-                runId,
-                taskId,
-                taskRef: parent.taskRef,
-                runner,
-                repo,
-                branch: parentBranch,
-                origin: parentRef.origin,
-                provider: pinned.provider,
-                model: pinned.model,
-                worktreePath: leased.worktree.path,
-                evidenceRoot: context.evidenceRoot,
-                clock,
-            }, result);
-            dispatched.push({
-                id: taskId,
-                outcome: disposition.kind === "built" ? "built" : disposition.kind === "stopped" ? "stopped" : "failed",
-                ...(disposition.kind === "stopped" ? { reason: `stop:${disposition.stopRun}` } : result.ok ? {} : { reason: result.reason }),
-                worktree: leased.worktree.path,
-            });
-            if (disposition.kind !== "built" && disposition.kind !== "stopped")
                 broke++;
+                break;
+            case "recorded":
+                // The standalone-only arm; unreachable under the tick policy.
+                break;
+            case "stopped":
+                // An operator's stop (v52): interrupted with its work preserved —
+                // not a break, not a strike; the task waits under the stop's hold.
+                dispatched.push({ id, outcome: "stopped", reason: `stop:${disposition.stopRun}`, branch, worktree: leased.worktree.path });
+                break;
+            case "handed-back":
+                // The service is stopping: the work is kept and the task requeued
+                // to resume on the next pass — not a break, not a strike.
+                dispatched.push({ id, outcome: "stopped", reason: "service-stop", detail: result.ok ? "" : result.message.slice(0, 200), branch, worktree: leased.worktree.path });
+                break;
+            case "invariant":
+                dispatched.push({ id, outcome: "failed", reason: disposition.reason });
+                broke++;
+                break;
         }
     }
+    untakenTrial?.giveBack();
     // The one automatic review each finished build gets when its project's
     // switch is on, before the result is handed on below. Other queued review
     // asks close unrun. Reauthenticated first; only admitted projects.
@@ -4532,11 +3258,6 @@ async function tickCommand(flags, context) {
             ...(triggerPass.added + triggerPass.problems.length === 0 ? {} : { triggers: triggerPass }), ...(stepPass.ran + stepPass.problems.length === 0 ? {} : { steps: stepPass }), ...(replied ? { replies: replyPass } : {}) } };
     const summary = () => {
         const lines = [`Considered ${considered}, built ${built}, parked ${parked}, broke ${broke}.`];
-        for (const entry of routines) {
-            lines.push(entry.outcome === "fired"
-                ? `  routine ${entry.routine.padEnd(16)} fired  ${entry.taskId ?? ""}`.trimEnd()
-                : `  routine ${entry.routine.padEnd(16)} skipped  ${entry.detail ?? entry.outcome}`);
-        }
         for (const entry of dispatched) {
             const detail = entry.outcome === "built"
                 ? entry.committed === true
@@ -4561,21 +3282,19 @@ async function tickCommand(flags, context) {
         return fail(write, json, "tick", "build-failed", `${broke} of ${dispatched.length} dispatched tasks broke`, EXIT.failed, {
             considered,
             dispatched,
-            routines,
         });
     }
-    if (built > 0 || parked > 0 || dispatched.some(one => one.outcome === "planned" || one.outcome === "reported" || one.outcome === "held" || one.outcome === "reviewed" || one.outcome === "not-reviewed")) {
-        return succeed(write, json, "tick", { considered, dispatched, routines, ...flows }, summary);
+    if (built > 0 || parked > 0 || dispatched.some(one => one.outcome === "planned" || one.outcome === "reported" || one.outcome === "reviewed" || one.outcome === "not-reviewed")) {
+        return succeed(write, json, "tick", { considered, dispatched, ...flows }, summary);
     }
     if (considered === 0) {
         return fail(write, json, "tick", "empty", "nothing is ready", EXIT.refused, {
             considered,
             dispatched,
-            routines,
             ...flows,
         });
     }
-    return fail(write, json, "tick", "nothing-dispatched", "everything ready is waiting on a person or held by somebody else", EXIT.refused, { considered, dispatched, routines, ...flows });
+    return fail(write, json, "tick", "nothing-dispatched", "everything ready is waiting on a person or held by somebody else", EXIT.refused, { considered, dispatched, ...flows });
 }
 /**
  * The recovery sweep: everything an unattended stretch may have left behind, in one pass.
@@ -5094,7 +3813,6 @@ async function startConsole(options) {
         ...(options.registryPath === undefined ? {} : { registryPath: options.registryPath }),
         ...(options.upConsole === undefined ? {} : { upConsole: options.upConsole }),
         ...(options.editorLinks === undefined ? {} : { editorLinks: options.editorLinks }),
-        ...(options.attended === undefined ? {} : { attended: options.attended }),
     });
     await new Promise((resolve, reject) => {
         server.once("error", reject);
@@ -5213,9 +3931,6 @@ async function requeueTask(positional, flags, context) {
     if (id === undefined) {
         return fail(write, json, "task requeue", "usage", "`toolroll task requeue <id> --as <you> --token <t>`", EXIT.usage);
     }
-    const racingGuard = refuseWhileRacing(context, "task requeue", id);
-    if (racingGuard !== null)
-        return racingGuard;
     const acting = await askCredentials(flags, context);
     if (acting === null) {
         return fail(write, json, "task requeue", "usage", "requeueing takes `--as <you> --token <t>` — who overrode the stall is recorded, not asserted", EXIT.usage);
@@ -5530,10 +4245,6 @@ async function providersCommand(flags, context) {
         const lastSuccess = store.providerLastSuccess(id);
         const keyPresent = facts.requiresEnv === null ? null : (process.env[facts.requiresEnv] ?? "") !== "";
         const installedVersion = installed ? (version.stdout.trim().split("\n")[0] ?? "") : null;
-        const authMode = readAuthMode(id);
-        const attestation = attestationOf(id);
-        const versionProvenAtSpawn = installedVersion !== null && attestation !== null && versionInRange(installedVersion, attestation);
-        const exhaustionRecognized = versionProvenAtSpawn && recognizesEligible(id, installedVersion, authMode);
         report.push({
             provider: id,
             binary: facts.binary,
@@ -5544,19 +4255,6 @@ async function providersCommand(flags, context) {
             ...(keyPresent === null ? {} : { keyPresent, keyEnv: facts.requiresEnv }),
             measuresCost: facts.measuresCost,
             configuredPhases: configured.get(id) ?? [],
-            fallbackReadiness: {
-                authMode,
-                versionProvenAtSpawn,
-                exhaustionRecognized,
-                automaticSwitchArmed: exhaustionRecognized,
-                reason: exhaustionRecognized
-                    ? "this build can recognize an exhausted credential for the installed provider version"
-                    : attestation === null
-                        ? "this provider does not yet prove its version at spawn, so exhaustion classification fails closed"
-                        : !versionProvenAtSpawn
-                            ? "the installed provider version is not inside this build's attested range"
-                            : "no reviewed exhaustion fixture recognizes this auth mode at the installed provider version",
-            },
             // The audit: facts about the harness, reported before any of them is
             // enforced. What transport we read, whether a session can resume,
             // which init signal exists, what hermetic flag we deliberately do NOT
@@ -5619,14 +4317,10 @@ async function providersCommand(flags, context) {
         if (one["keyEnv"] !== undefined) {
             write(`  ${String(one["keyEnv"])}  ${one["keyPresent"] === true ? "present (not validated — presence is not authorization)" : "ABSENT — runs will fail until the runner exports it"}`);
         }
-        write(`  cost           ${one["measuresCost"] === true ? "measured in dollars per run" : "tokens only — runs land as UNMEASURED; ceilinged routines fail closed on them"}`);
+        write(`  cost           ${one["measuresCost"] === true ? "measured in dollars per run" : "tokens only — runs land as UNMEASURED; schedules with a weekly limit fail closed on them"}`);
         const phases = one["configuredPhases"];
         if (phases.length > 0)
             write(`  configured     ${phases.join(", ")} (installation)`);
-        const fallback = one["fallbackReadiness"];
-        write(`  auto fallback  ${fallback.automaticSwitchArmed
-            ? "armed — a configured and approved chain may advance after proven exhaustion"
-            : `not armed — ${fallback.reason}`}`);
         const audit = one["audit"];
         write(`  transport      ${audit.transport}${audit.initSignal === "none" ? " — no init signal; a failed run cannot say whether the harness came up" : ` — init signal: ${audit.initSignal}`}`);
         write(`  resume         ${audit.resume}`);
@@ -5964,10 +4658,6 @@ async function modeCommand(positional, flags, context) {
     if (terms.planAuto && !terms.autoApproveFiling) {
         return fail(write, json, "mode set", "invalid", "--plan-auto requires automatic filing approval", EXIT.refused);
     }
-    // The paid-fallback grant (R8): NEVER a preset default — only this
-    // explicit flag lets an exhausted subscription switch to another account.
-    if (flag(flags, "allow-paid-fallback"))
-        terms.allowPaidFallback = true;
     // Approving plans and merges from the signer's paired chat: never a preset default, only this flag.
     if (flag(flags, "chat-approve"))
         terms.chatApprove = true;
@@ -6002,9 +4692,8 @@ async function configCommand(positional, flags, context) {
                     : { problem: answer.problem }),
             };
         });
-        const fallback = scope === INSTALLATION_SCOPE ? [] : store.fallbackConfig(scope);
         // The STRONG tier (v47): the named strongest agent per phase, which
-        // high-risk, strict, evidence-sensitive, and publication-sensitive
+        // large, strict, evidence-sensitive, and publication-sensitive
         // routes reach for. Never inferred — absent means "the default, said".
         const candidates = resolveRouteCandidates(store, scope === INSTALLATION_SCOPE ? null : scope);
         const strong = ["plan", "build", "repair", "review"].map(one => ({
@@ -6015,7 +4704,7 @@ async function configCommand(positional, flags, context) {
         // A light planner from before light became build-only is never used: said, with how to clear it.
         const unusedLight = [INSTALLATION_SCOPE, ...(scope === INSTALLATION_SCOPE ? [] : [scope])].filter(one => store.phaseTierConfig(one, "plan", "light") !== null);
         if (json) {
-            write(envelopeJson({ ok: true, command: "config show", installation, project, resolved, fallback, strong, installationStrong: store.listPhaseTierConfig(INSTALLATION_SCOPE), projectStrong: scope === INSTALLATION_SCOPE ? [] : store.listPhaseTierConfig(scope), installationAlso: store.listPhaseTierAlternates(INSTALLATION_SCOPE), projectAlso: scope === INSTALLATION_SCOPE ? [] : store.listPhaseTierAlternates(scope) }));
+            write(envelopeJson({ ok: true, command: "config show", installation, project, resolved, strong, installationStrong: store.listPhaseTierConfig(INSTALLATION_SCOPE), projectStrong: scope === INSTALLATION_SCOPE ? [] : store.listPhaseTierConfig(scope), installationAlso: store.listPhaseTierAlternates(INSTALLATION_SCOPE), projectAlso: scope === INSTALLATION_SCOPE ? [] : store.listPhaseTierAlternates(scope) }));
             return EXIT.ok;
         }
         write(`Effective phase agents${scope === INSTALLATION_SCOPE ? "" : ` for ${scope}`}:`);
@@ -6028,7 +4717,7 @@ async function configCommand(positional, flags, context) {
             }
         }
         write("");
-        write("  strong tier (high-risk, strict, screenshot-proof, and automerge routes reach for these):");
+        write("  strong tier (large or risky, strict, screenshot-proof, and automerge routes reach for these):");
         for (const one of strong) {
             write(`  ${one.phase.padEnd(8)} ${one.strong === null ? "none configured — such routes keep the default above and say so" : `${one.strong.provider}${one.strong.model === null ? " (harness default model)" : ` · ${one.strong.model}`}  [${one.strong.source}]`}`);
         }
@@ -6051,14 +4740,6 @@ async function configCommand(positional, flags, context) {
             }
         }
         write("  repair note: the repair PROVIDER always inherits the build it mends — only its model is configurable.");
-        if (fallback.length > 0) {
-            write("");
-            write("  if the build agent's subscription runs out, NEW approvals bind this fallback chain:");
-            fallback.forEach((one, i) => {
-                write(`    ${i + 1}. ${one.provider} (${one.model}) — ${one.authMode === "subscription" ? "its subscription login" : "your API key"}`);
-            });
-            write("  it fires only when a signed mode allows the paid fallback (`mode set --allow-paid-fallback`).");
-        }
         if (installation.length === 0 && project.length === 0) {
             write("  nothing configured — every phase runs the default (claude).");
             write("  toolroll config set build --provider claude --model sonnet --as <you> --token <t>");
@@ -6081,7 +4762,7 @@ async function configCommand(positional, flags, context) {
             return fail(write, json, `config ${action}`, "unauthenticated", "that is not an approver, or the token does not match", EXIT.refused);
         }
         if (action === "clear") {
-            store.setSpendDefaults({ buildPerRunMicrousd: null, racePerAgentMicrousd: null, raceTotalMicrousd: null }, acting.name, clock());
+            store.setSpendDefaults({ buildPerRunMicrousd: null }, acting.name, clock());
             return succeed(write, json, "config clear", { budgets: null }, () => ["Spend defaults cleared — filings state their own numbers again."]);
         }
         const parseUsd = (flag) => {
@@ -6092,117 +4773,13 @@ async function configCommand(positional, flags, context) {
             return Number.isFinite(value) && value > 0 ? Math.round(value * 1_000_000) : false;
         };
         const build = parseUsd("build-usd");
-        const racePer = parseUsd("race-per-usd");
-        const raceTotal = parseUsd("race-total-usd");
-        // Name the flag that was bad — "budgets are positive dollar amounts"
-        // over four candidates left the caller diffing their own command line.
-        const badFlag = build === false ? "--build-usd" : racePer === false ? "--race-per-usd" : raceTotal === false ? "--race-total-usd" : null;
-        if (badFlag !== null || build === false || racePer === false || raceTotal === false) {
-            return fail(write, json, "config set", "usage", `${badFlag ?? "--build-usd"} is a positive dollar amount`, EXIT.usage);
-        }
-        // The default competing-agent count (operator request): applied only
-        // where a filing names one agent and no explicit count; a race digest
-        // always binds the actual lineup.
-        const agentsGiven = text(flags, "race-agents");
-        const raceAgents = agentsGiven === undefined ? null : Number(agentsGiven);
-        if (raceAgents !== null && (!Number.isInteger(raceAgents) || raceAgents < 2 || raceAgents > 4)) {
-            return fail(write, json, "config set", "usage", "--race-agents is how many agents compete by default: a whole number from 2 to 4", EXIT.usage);
-        }
-        store.setSpendDefaults({ buildPerRunMicrousd: build, racePerAgentMicrousd: racePer, raceTotalMicrousd: raceTotal, raceAgents }, acting.name, clock());
+        if (build === false) {
+            return fail(write, json, "config set", "usage", "--build-usd is a positive dollar amount", EXIT.usage);
+        }
+        store.setSpendDefaults({ buildPerRunMicrousd: build }, acting.name, clock());
         return succeed(write, json, "config set", { budgets: store.getSpendDefaults() }, () => [
             "Spend defaults set. New filings pre-fill from these; every approval still restates its own numbers:",
             ...(build === null ? [] : [`  each ordinary build attempt: $${(build / 1_000_000).toFixed(2)} (also the installation backstop)`]),
-            ...(racePer === null ? [] : [`  each tournament agent: $${(racePer / 1_000_000).toFixed(2)}`]),
-            ...(raceTotal === null ? [] : [`  each tournament total: $${(raceTotal / 1_000_000).toFixed(2)}`]),
-            ...(raceAgents === null ? [] : [`  tournaments race ${raceAgents} agents unless a filing says otherwise`]),
-        ]);
-    }
-    // The FALLBACK CHAIN (Layer F): the ordered entries a repository's NEW
-    // approvals bind after the base agent. Per-repo only — a chain names
-    // credentials for one project's work — and inert on its own: it only
-    // ever fires when a signed mode also allows the paid fallback.
-    if (phase === "fallback") {
-        const acting = await askCredentials(flags, context);
-        if (acting === null) {
-            return fail(write, json, `config ${action}`, "usage", "configuring fallbacks takes `--as <you> --token <t>`", EXIT.usage);
-        }
-        const authedFallback = authenticateApprover(store, acting.name, acting.token);
-        if (!authedFallback.ok) {
-            return fail(write, json, `config ${action}`, "unauthenticated", "that is not an approver, or the token does not match", EXIT.refused);
-        }
-        if (scope === INSTALLATION_SCOPE) {
-            return fail(write, json, `config ${action}`, "usage", "fallbacks are per repository — say --repo <path>", EXIT.usage);
-        }
-        if (action === "clear") {
-            const had = store.clearFallbackConfig(scope);
-            return succeed(write, json, "config clear", { repo: scope, fallback: null }, () => [
-                had
-                    ? `${scope}: fallback chain cleared — new approvals bind the single configured agent again.`
-                    : `${scope}: no fallback chain was configured.`,
-            ]);
-        }
-        const entriesGiven = text(flags, "entries");
-        if (entriesGiven === undefined) {
-            return fail(write, json, "config set", "usage", "`config set fallback --repo <path> --entries provider:model:auth-mode[,…]` — auth-mode is subscription or api-key; 1 to 3 entries", EXIT.usage);
-        }
-        const entries = [];
-        for (const one of entriesGiven.split(",")) {
-            // First and LAST colon split the three parts (F+G review, finding 1):
-            // model ids legitimately carry colons (openrouter's ":free" suffixes),
-            // so the model is everything between the provider and the auth mode.
-            const trimmed = one.trim();
-            const firstColon = trimmed.indexOf(":");
-            const lastColon = trimmed.lastIndexOf(":");
-            if (firstColon === -1 || lastColon === firstColon) {
-                return fail(write, json, "config set", "usage", `"${trimmed}" is not provider:model:auth-mode`, EXIT.usage);
-            }
-            const provider = trimmed.slice(0, firstColon);
-            const model = trimmed.slice(firstColon + 1, lastColon);
-            const authMode = trimmed.slice(lastColon + 1);
-            if (!isProviderId(provider)) {
-                return fail(write, json, "config set", "usage", `unknown provider "${provider}" — one of ${PROVIDER_IDS.join(", ")}`, EXIT.usage);
-            }
-            if (authMode !== "subscription" && authMode !== "api-key") {
-                return fail(write, json, "config set", "usage", `auth-mode is subscription or api-key, not "${authMode}"`, EXIT.usage);
-            }
-            if (authMode === "subscription" && !SUBSCRIPTION_CAPABLE[provider]) {
-                return fail(write, json, "config set", "usage", `${provider} has no subscription login — this entry must use api-key`, EXIT.usage);
-            }
-            // The model rides provider argv: the SAME argv-safety validation every
-            // other sealed model passes (never a leading dash or control bytes).
-            const argvSafe = validateSpec({ provider, model: model === "" ? null : model });
-            if (model === "" || !argvSafe.ok) {
-                return fail(write, json, "config set", "usage", model === "" ? "each entry names an exact model — approvals bind exact routing" : argvSafe.ok ? "invalid entry" : argvSafe.problem, EXIT.usage);
-            }
-            entries.push({ provider, model, authMode });
-        }
-        if (entries.length < 1 || entries.length > 3) {
-            return fail(write, json, "config set", "usage", "1 to 3 fallback entries (the approval binds the whole chain)", EXIT.usage);
-        }
-        // The WHOLE chain must file against the CURRENT base (F+G review,
-        // finding 4): a fallback that duplicates the base — or another entry —
-        // refuses NOW, in words, not at some future filing. Set, prove through
-        // the same resolver filing uses, and restore the old config on refusal.
-        const previous = store.fallbackConfig(scope);
-        store.setFallbackConfig(scope, entries, acting.name, clock());
-        const baseAgent = resolvePhaseAgent(store, "build", scope, {});
-        if (baseAgent.ok) {
-            const proven = resolveScopeChain(store, scope, undefined, {}, readAuthMode(baseAgent.spec.provider));
-            // An UNRESOLVED base (no explicit routing configured yet) cannot be
-            // duplicate-checked — the filing-time resolver holds that line; only
-            // a chain that provably cannot file refuses here.
-            if (!proven.ok && proven.reason !== "base-unresolved") {
-                if (previous.length > 0)
-                    store.setFallbackConfig(scope, previous, acting.name, clock());
-                else
-                    store.clearFallbackConfig(scope);
-                return fail(write, json, "config set", "usage", `this chain cannot file: ${proven.problem}`, EXIT.usage);
-            }
-        }
-        return succeed(write, json, "config set", { repo: scope, fallback: entries }, () => [
-            `${scope}: fallback chain set — NEW approvals bind it; existing approvals are untouched.`,
-            ...entries.map((one, i) => `  ${i + 1}. ${one.provider} (${one.model}) — ${one.authMode === "subscription" ? "its subscription login" : "your API key"}`),
-            "  it fires only when a signed mode allows the paid fallback: `toolroll mode set --allow-paid-fallback …`",
         ]);
     }
     // Chat is its OWN configuration, deliberately not a phase (Codex v3
@@ -6286,7 +4863,7 @@ async function configCommand(positional, flags, context) {
         return fail(write, json, `config ${action}`, authenticated.reason, describeApproveFailure(authenticated.reason, phase), EXIT.refused);
     }
     // The STRONG tier (v47): a second, named row per phase the routing policy
-    // reaches for when risk, quality, evidence, or publication demand it.
+    // reaches for when size, quality, evidence, or publication demand it.
     // Authenticated and audited exactly like the routine row; existing
     // approvals are untouched — a sealed route never re-resolves.
     // The LIGHT tier (v2): the fast agent a small change builds on.
@@ -6380,7 +4957,7 @@ async function configCommand(positional, flags, context) {
         store.setPhaseTierConfig(scope, phase, "strong", providerGiven, modelGiven, acting.name, clock());
         return succeed(write, json, "config set", { scope, phase, tier: "strong", provider: providerGiven, model: modelGiven }, () => [
             `strong ${phase} at ${scope === INSTALLATION_SCOPE ? "the installation" : scope} is ${providerGiven}${modelGiven === null ? "" : ` · ${modelGiven}`}, set by ${acting.name}.`,
-            "  high-risk, strict, screenshot-proof, and automerge routes filed from now on reach for it; sealed routes are untouched.",
+            "  large or risky, strict, screenshot-proof, and automerge routes filed from now on reach for it; sealed routes are untouched.",
             ...(phase === "repair" ? ["  a strong repair agent applies only when its provider is the build's — cross-provider repair does not exist."] : []),
         ]);
     }
@@ -6393,7 +4970,7 @@ async function configCommand(positional, flags, context) {
         warnings.push("OPENROUTER_API_KEY is not present in this environment — runs will fail until the runner exports it.");
     }
     if (providerGiven !== "claude") {
-        warnings.push(`${providerGiven} does not report dollar cost: its runs land as UNMEASURED, and any routine with a cost ceiling fails closed on them by design.`);
+        warnings.push(`${providerGiven} does not report dollar cost: its runs land as UNMEASURED, and any schedule with a weekly limit fails closed on them by design.`);
     }
     return succeed(write, json, "config set", { scope, phase, provider: providerGiven, model: modelGiven, warnings }, () => [
         `${phase} at ${scope === INSTALLATION_SCOPE ? "the installation" : scope} now runs ${providerGiven}${modelGiven === null ? "" : ` · ${modelGiven}`}, set by ${acting.name}.`,
@@ -6964,80 +5541,11 @@ async function intakeCommand(positional, flags, context) {
     ]);
 }
 /**
- * `toolroll contest …` — the tournament from the terminal: `show`
- * for the machine-readable state, `exclude` to stop a racing agent whose
- * question you will not answer (authenticated: it cancels paid-for work
- * and un-sticks the race). The pick itself stays a console ceremony.
- */
-function contestCommand(positional, flags, context) {
-    const { store, write, json, clock } = context;
-    const [action, idGiven, ordinalGiven] = positional;
-    if (action === undefined || !CONTEST_ACTIONS.includes(action)) {
-        return fail(write, json, "contest", "usage", `unknown \`contest ${action ?? ""}\` — try ${CONTEST_ACTIONS.join(", ")}`, EXIT.usage);
-    }
-    if (action === "show") {
-        const contest = store.getContest(Number(idGiven));
-        if (contest === null)
-            return fail(write, json, "contest show", "unknown", "no tournament or comparison with that id", EXIT.refused);
-        const agents = store.contestants(contest.id);
-        if (json) {
-            write(envelopeJson({ ok: true, command: "contest show", contest, agents }));
-            return EXIT.ok;
-        }
-        write(`${contestNoun(contest.kind)} #${contest.id} — ${contest.state}`);
-        for (const racer of agents) {
-            const money = contest.kind === "comparison"
-                ? racer.unknownSpend
-                    ? "spend unmeasured (tokens only)"
-                    : `$${(racer.measuredMicrousd / 1_000_000).toFixed(2)} measured`
-                : `charged $${(racer.accountedMicrousd / 1_000_000).toFixed(2)}${racer.unknownSpend ? " (exact figure unknown — charged the reserved worst case)" : ""}`;
-            write(`  agent ${racer.ordinal}: ${racer.provider} · ${racer.model} — ${racer.state} · ${money}`);
-        }
-        return EXIT.ok;
-    }
-    if (action === "exclude") {
-        return (async () => {
-            const contest = store.getContest(Number(idGiven));
-            const ordinal = Number(ordinalGiven);
-            if (contest === null || !Number.isInteger(ordinal)) {
-                return fail(write, json, "contest exclude", "usage", "`toolroll contest exclude <tournament-id> <agent-number> --as <you> --token <t>`", EXIT.usage);
-            }
-            const acting = await askCredentials(flags, context);
-            if (acting === null) {
-                return fail(write, json, "contest exclude", "usage", "stopping a racing agent takes `--as <you> --token <t>`", EXIT.usage);
-            }
-            const authed = authenticateApprover(store, acting.name, acting.token);
-            if (!authed.ok) {
-                return fail(write, json, "contest exclude", "unauthenticated", "that is not an approver, or the token does not match", EXIT.refused);
-            }
-            const racer = store.contestants(contest.id).find(one => one.ordinal === ordinal);
-            if (racer === undefined || racer.state !== "parked") {
-                return fail(write, json, "contest exclude", "not-waiting", "that agent is not waiting on an answer", EXIT.refused);
-            }
-            const question = store.openDecisionForContestant(racer.id);
-            const moved = store.transact(() => {
-                if (question !== null && !store.excludeDecision(question, acting.name, clock()))
-                    return false;
-                if (!store.casContestantState(racer.id, ["parked"], "stopped", racer.generation))
-                    return false;
-                contestMaybeAggregate(store, contest.id, clock());
-                return true;
-            });
-            if (!moved)
-                return fail(write, json, "contest exclude", "changed", "the tournament moved while you were reading — look again", EXIT.refused);
-            const after = store.getContest(contest.id);
-            return succeed(write, json, "contest exclude", { contest: after }, () => [
-                `Agent ${ordinal} stopped; its question is closed as excluded. The tournament is now ${after?.state ?? "?"}.`,
-            ]);
-        })();
-    }
-    return fail(write, json, "contest", "usage", "`toolroll contest show <id> | exclude <id> <agent-number>`", EXIT.usage);
-}
-/**
- * `toolroll template …` — the shipped library of common standing
- * orders (adoption track, step 2). A template is a pre-filled form:
- * `apply` PREVIEWS by default and files only under `--file`, through the
- * same one door as every manual filing, landing UNAPPROVED. Recipes
+ * `toolroll template …` — the shipped library of common work (adoption
+ * track, step 2). A template is a pre-filled form: `apply` PREVIEWS by
+ * default and files only under `--file`, landing UNAPPROVED — a task through
+ * the same one door as every manual filing, or a scheduled flow whose
+ * schedule starts paused. Recipes
  * (issue-intake, ci-babysitter) display existing ceremonies and cannot be
  * applied — the authority they would need is a separate authenticated act
  * a template must never perform (adoption review, finding 10).
@@ -7054,7 +5562,7 @@ function templateCommand(positional, flags, context) {
             }));
             return EXIT.ok;
         }
-        write("Templates — common routines you edit to fit. Nothing a template");
+        write("Templates — common work you edit to fit. Nothing a template");
         write("files is approved; recipes only show existing ceremonies.");
         write("");
         for (const one of TEMPLATES) {
@@ -7096,8 +5604,8 @@ function templateCommand(positional, flags, context) {
             write(`  title      ${template.title}`);
         }
         else {
-            write(`  files      one routine (cannot fire until you approve its terms)`);
-            write(`  name       ${template.routineName}`);
+            write(`  makes      a scheduled flow (nothing repeats until you turn the schedule on)`);
+            write(`  name       ${template.title}`);
             write(`  schedule   ${template.schedule}`);
         }
         write(`  goal       ${template.goal}`);
@@ -7167,273 +5675,48 @@ function templateCommand(positional, flags, context) {
             ...(filedLink === null ? [] : [`  ${filedLink}`]),
         ]);
     }
-    const routineName = text(flags, "name") ?? template.routineName;
-    const schedule = text(flags, "schedule") ?? template.schedule;
+    const flowName = text(flags, "name") ?? template.title;
+    const scheduleGiven = text(flags, "schedule") ?? template.schedule;
+    const schedule = parseSchedule(scheduleGiven) !== null ? scheduleGiven : scheduleFromWords(scheduleGiven) ?? scheduleGiven;
     const ceilingGiven = text(flags, "ceiling");
     const costCeilingUsd = ceilingGiven === undefined ? template.costCeilingUsd : Number(ceilingGiven);
+    const repo = canonicalProject(repoGiven) ?? resolve(repoGiven);
+    const terms = { goal, outOfScope, touches, requirements: template.requirements, acceptance: template.acceptance, budgetPerRunMicrousd: null, costCeilingUsd };
     if (!flags.has("file")) {
-        const draft = {
-            kind: "routine",
-            name: routineName,
-            repo: canonicalProject(repoGiven) ?? resolve(repoGiven),
-            goal,
-            outOfScope,
-            touches,
-            acceptance: template.acceptance,
-            requirements: template.requirements,
-            schedule,
-            costCeilingUsd,
-            filedVia: `template:${template.name}`,
-        };
+        const draft = { kind: "scheduled flow", name: flowName, repo, schedule, ...terms, filedVia: `template:${template.name}` };
         if (json) {
             write(envelopeJson({ ok: false, command: "template apply", reason: "unconfirmed", draft }));
             return 3;
         }
-        write("Would file, exactly (edit with --name/--goal/--not/--touches/--schedule/--ceiling):");
+        const readable = parseSchedule(schedule);
+        write("Would make this scheduled flow (edit with --name/--goal/--not/--touches/--schedule/--ceiling):");
         write("");
-        write(`  routine  ${draft.name}`);
+        write(`  flow     ${draft.name}`);
         write(`  repo     ${draft.repo}`);
-        write(`  schedule ${draft.schedule}${draft.schedule.startsWith("every:10080") ? "  (weekly)" : ""}`);
+        write(`  schedule ${readable === null ? schedule : describeSchedule(readable)}`);
         write(`  goal     ${draft.goal}`);
         if (draft.outOfScope !== null)
             write(`  not      ${draft.outOfScope}`);
         if (draft.touches.length > 0)
             write(`  touches  ${draft.touches.join(", ")}`);
         if (draft.costCeilingUsd !== null)
-            write(`  ceiling  $${draft.costCeilingUsd}/week`);
+            write(`  ceiling  $${draft.costCeilingUsd} every 7 days`);
         write("");
-        write("Nothing was filed. Re-run with --file to file it — UNAPPROVED either way; it cannot fire until you approve it.");
+        write("Nothing was made. Re-run with --file to make it. Nothing repeats until you turn the schedule on.");
         return 3;
     }
-    const made = fileRoutineProposal(store, {
-        name: routineName,
-        repo: repoGiven,
-        goal,
-        outOfScope,
-        touches,
-        acceptance: template.acceptance,
-        requirements: template.requirements,
-        schedule,
-        costCeilingUsd,
-        filedVia: `template:${template.name}`,
-    }, clock());
+    const made = createScheduledFlow(store, { repo, name: flowName, stem: template.name, schedule, terms, by: context.principal?.account ?? "toolroll", filedBy: context.principal?.account ?? null }, clock());
     if (!made.ok)
-        return fail(write, json, "template apply", made.reason, made.message, made.reason === "duplicate" ? EXIT.refused : EXIT.usage);
-    return succeed(write, json, "template apply", { filed: "routine", id: made.id, approved: false }, () => [
-        `Filed routine ${routineName} from template ${template.name}.`,
+        return fail(write, json, "template apply", "bad-terms", made.message, EXIT.usage);
+    const flowLink = consoleLinkFor(context, `/flows/${made.flow}`);
+    return succeed(write, json, "template apply", { filed: "scheduled flow", flow: made.flow, trigger: made.trigger, on: false, ...(flowLink === null ? {} : { links: { flow: flowLink } }) }, () => [
+        `Made ${flowName}, flow #${made.flow}, from template ${template.name}.`,
         "",
-        "UNAPPROVED — NO AUTHORITY GRANTED. It cannot fire until you approve the standing order:",
-        `  toolroll routine approve ${routineName}`,
+        "Nothing repeats until you turn the schedule on; each run then waits for approval under the project's rules:",
+        `  toolroll flows trigger resume ${made.flow} ${made.trigger}`,
+        ...(flowLink === null ? [] : [`  ${flowLink}`]),
     ]);
 }
-/**
- * `toolroll routine …` — standing orders. Filing one is cheap; the
- * expensive act is the approval, which restates every term including "each
- * firing builds without asking" and takes the approver's credential, same
- * as a scope. Pausing needs no ceremony because stopping spend never does.
- */
-async function routineCommand(positional, flags, context) {
-    const { store, write, json, clock } = context;
-    const [action, name] = positional;
-    if (action !== undefined && !ROUTINE_ACTIONS.includes(action)) {
-        return fail(write, json, "routine", "usage", `unknown \`routine ${action}\` — try ${ROUTINE_ACTIONS.join(", ")}`, EXIT.usage);
-    }
-    if (action === undefined || action === "list") {
-        const repoFilter = text(flags, "repo");
-        const routines = store.listRoutines(repoFilter === undefined ? null : canonicalProject(repoFilter) ?? resolve(repoFilter));
-        if (json) {
-            write(envelopeJson({ ok: true, command: "routine list", routines }));
-            return EXIT.ok;
-        }
-        if (routines.length === 0) {
-            write("No routines. `toolroll routine add <name> --repo <path> --goal <text> --schedule every:60` files one.");
-            return EXIT.ok;
-        }
-        for (const routine of routines) {
-            const approved = routine.approvedAt !== null && routine.approvedDigest === routine.digest;
-            const status = routine.paused ? "paused" : approved ? "live" : "awaiting approval";
-            write(`  ${routine.name.padEnd(20)} ${status.padEnd(18)} ${routine.schedule.padEnd(14)} ${routine.repo}`);
-        }
-        return EXIT.ok;
-    }
-    if (action === "add") {
-        if (name === undefined || !ROUTINE_NAME.test(name)) {
-            return fail(write, json, "routine add", "usage", "a routine's name is lowercase letters, digits, and dashes — it becomes each instance's id", EXIT.usage);
-        }
-        const repoGiven = text(flags, "repo");
-        const goal = text(flags, "goal");
-        const schedule = text(flags, "schedule");
-        if (repoGiven === undefined || goal === undefined || schedule === undefined) {
-            return fail(write, json, "routine add", "usage", "`toolroll routine add <name> --repo <path> --goal <text> --schedule every:<min>|daily:<HH:MM>[@Zone]|weekly:<0-6>:<HH:MM>[@Zone] --acceptance <rubric> [--not <text>] [--touches a,b] [--require kind:name,…] [--ceiling <usd>] [--budget-usd <n>]`", EXIT.usage);
-        }
-        const acceptanceGiven = text(flags, "acceptance");
-        if (acceptanceGiven === undefined) {
-            return fail(write, json, "routine add", "acceptance-required", "a standing order needs at least one signed acceptance criterion — `--acceptance \"<statement>|<evidence,kinds>\"`, `;`-separated for more than one; evidence kinds are check, screenshot, changed-path, manual-review", EXIT.usage);
-        }
-        const ceilingGiven = text(flags, "ceiling");
-        // --budget-usd on a routine caps EACH instance (v16): it becomes the
-        // instance scope's digest-bound budget term, enforced by the same
-        // native-cap plumbing as any other scope budget.
-        const perRunGiven = text(flags, "budget-usd");
-        if (perRunGiven !== undefined && (!Number.isFinite(Number(perRunGiven)) || Number(perRunGiven) <= 0)) {
-            return fail(write, json, "routine add", "bad-budget", "--budget-usd is a positive dollar amount — what each firing may spend", EXIT.usage);
-        }
-        // One filing door for every surface (Codex adoption review, finding 7):
-        // validation, canonicalization, digest, and provenance live in the
-        // service, not here.
-        const created = fileRoutineProposal(store, {
-            name,
-            repo: repoGiven,
-            goal,
-            outOfScope: text(flags, "not") ?? null,
-            touches: (text(flags, "touches") ?? "").split(",").map(one => one.trim()).filter(one => one !== ""),
-            acceptance: acceptanceLinesToInput(splitAcceptanceRubric(acceptanceGiven)),
-            requirements: (text(flags, "require") ?? "").split(",").map(one => one.trim()).filter(one => one !== ""),
-            schedule,
-            costCeilingUsd: ceilingGiven === undefined ? null : Number(ceilingGiven),
-            ...(perRunGiven === undefined ? {} : { budgetPerRunMicrousd: Math.round(Number(perRunGiven) * 1_000_000) }),
-            filedVia: "cli",
-        }, clock());
-        if (!created.ok) {
-            return fail(write, json, "routine add", created.reason, created.message, created.reason === "duplicate" ? EXIT.refused : EXIT.usage);
-        }
-        const routine = store.getRoutine(created.id);
-        return succeed(write, json, "routine add", { routine }, () => [
-            `Filed ${name}. Nothing fires until somebody approves the standing order:`,
-            ...(routine === null ? [] : describeRoutine(routine)),
-            "",
-            `  toolroll routine approve ${name}`,
-        ]);
-    }
-    if (name === undefined) {
-        return fail(write, json, `routine ${action}`, "usage", "which routine? give its name", EXIT.usage);
-    }
-    const routine = store.routineByName(name);
-    if (routine === null) {
-        return fail(write, json, `routine ${action}`, "unknown", `no routine named ${name}`, EXIT.refused);
-    }
-    switch (action) {
-        case "show": {
-            const fires = store.routineFires(routine.id, 14);
-            if (json) {
-                write(envelopeJson({ ok: true, command: "routine show", routine, fires }));
-                return EXIT.ok;
-            }
-            const approved = routine.approvedAt !== null && routine.approvedDigest === routine.digest;
-            write(`${routine.name} — ${routine.paused ? "paused" : approved ? "live" : "awaiting approval"}`);
-            for (const line of describeRoutine(routine))
-                write(line);
-            if (routine.nextFireAt !== null && !routine.paused && approved)
-                write(`  next fire    ${routine.nextFireAt}`);
-            if (fires.length > 0) {
-                write("");
-                write("  recent firings, newest first:");
-                for (const fire of fires) {
-                    const said = fire.outcome === "fired"
-                        ? `${fire.instanceTaskId ?? "instance"}${fire.instanceState === null ? "" : ` (${fire.instanceState})`}${fire.reason === "manual" ? "  (run now)" : ""}`
-                        : `skipped — ${fire.reason ?? ""}`;
-                    write(`    ${fire.scheduledFor.replace(/^manual:/, "")}  ${said}`);
-                }
-            }
-            return EXIT.ok;
-        }
-        case "approve": {
-            let saw = text(flags, "digest");
-            let { name: asWho, token } = credentialsFrom(flags, context);
-            let confirmedAloud = false;
-            if ((!flags.has("yes") || saw === undefined || asWho === undefined || token === undefined) && interactive() && !json) {
-                write(`Approving ${name} makes it a STANDING order:`);
-                write("");
-                for (const line of describeRoutine(routine))
-                    write(line);
-                write("");
-                const agreed = await confirm("Approve exactly this standing order?");
-                if (!agreed) {
-                    write("Nothing approved.");
-                    return EXIT.refused;
-                }
-                saw ??= routine.digest;
-                const acting = await askCredentials(flags, context);
-                if (acting === null)
-                    return fail(write, json, "routine approve", "usage", "approval needs who is agreeing", EXIT.usage);
-                asWho = acting.name;
-                token = acting.token;
-                confirmedAloud = true;
-            }
-            const armed = (flags.has("yes") || confirmedAloud) && saw !== undefined && asWho !== undefined && token !== undefined;
-            if (!armed) {
-                if (json) {
-                    write(envelopeJson({ ok: false, command: "routine approve", reason: "unconfirmed", routine }));
-                    return EXIT.refused;
-                }
-                write(`Would approve this standing order — every firing of it builds without asking:`);
-                write("");
-                for (const line of describeRoutine(routine))
-                    write(line);
-                write("");
-                write("Nothing has been approved. Agree to exactly this with:");
-                write(`  toolroll routine approve ${name} --yes --digest ${routine.digest} --as <you> --token <your password>`);
-                // A preview reached by omitting --yes is the answer "no, not yet" —
-                // exit 3 in both modes, matching the JSON path (round-4 finding 10).
-                return EXIT.refused;
-            }
-            const approved = approveRoutine(store, routine.id, asWho, clock(), saw, token);
-            if (!approved.ok) {
-                return fail(write, json, "routine approve", approved.reason, describeApproveFailure(approved.reason, name), EXIT.refused);
-            }
-            return succeed(write, json, "routine approve", { routine: approved.routine }, () => [
-                `Approved. ${name} fires on its schedule from now on; first at ${approved.routine.nextFireAt}.`,
-                `Pause it any time: toolroll routine pause ${name}`,
-            ]);
-        }
-        case "refresh": {
-            // THE RECOVERY ROAD (v48): re-resolve the agents from today's
-            // configuration and file them as the order's working agents. Nothing
-            // is approved here — the refreshed order waits for the yes.
-            const before = routineAgentsState(routine);
-            const refreshed = refreshRoutineAgents(store, routine.id, clock());
-            if (!refreshed.ok) {
-                return fail(write, json, "routine refresh", refreshed.reason, `${name}: ${refreshed.problem}`, EXIT.refused);
-            }
-            return succeed(write, json, "routine refresh", { routine: refreshed.routine, changed: refreshed.changed, before: before.state }, () => [
-                refreshed.changed
-                    ? `Refreshed ${name}'s agents from today's configuration. Nothing is approved yet — read them and agree:`
-                    : `${name} already names exactly these agents; nothing changed.`,
-                ...describeRoutine(refreshed.routine),
-                "",
-                `  toolroll routine approve ${name}`,
-            ]);
-        }
-        case "pause":
-        case "resume": {
-            store.setRoutinePaused(routine.id, action === "pause", clock());
-            return succeed(write, json, `routine ${action}`, { name }, () => [
-                action === "pause"
-                    ? `${name} is paused — no firing until you resume it. Already-running instances finish.`
-                    : `${name} resumed — the next due slot fires again.`,
-            ]);
-        }
-        case "run-now": {
-            const acting = await askCredentials(flags, context);
-            if (acting === null) {
-                return fail(write, json, "routine run-now", "usage", "run-now takes `--as <you> --token <t>` — it dispatches work that spends", EXIT.usage);
-            }
-            const authenticated = authenticateApprover(store, acting.name, acting.token);
-            if (!authenticated.ok) {
-                return fail(write, json, "routine run-now", authenticated.reason, describeApproveFailure(authenticated.reason, name), EXIT.refused);
-            }
-            const outcome = fireRoutine(store, routine.id, clock(), { manual: true });
-            if (!outcome.ok) {
-                return fail(write, json, "routine run-now", outcome.reason, outcome.detail ?? outcome.reason, EXIT.refused);
-            }
-            return succeed(write, json, "routine run-now", { taskId: outcome.taskId }, () => [
-                `Spawned ${outcome.taskId} — it builds on the next pass. The regular schedule is untouched.`,
-            ]);
-        }
-        default:
-            return fail(write, json, "routine", "usage", "`toolroll routine [add|list|show|approve|refresh|pause|resume|run-now]`", EXIT.usage);
-    }
-}
 /** A credential from a 0600 file, for units that must not carry it inline. */
 function readTokenFile(path) {
     if (path === undefined)
@@ -7761,21 +6044,18 @@ async function runWatchLoop(args) {
     const slackFollower = followSlack({ store, dir: dirname(context.databaseFile), signal: followController.signal,
         readProjects: telegramReadProjects(context), evidenceRoot: context.evidenceRoot,
         ...(context.mateSeams?.subscriptionRunner ? { subscriptionRunner: context.mateSeams.subscriptionRunner } : {}),
-        ...(context.heldCoordinator ? { held: context.heldCoordinator } : {}),
         origin: () => phoneOrigin(process.env, dirname(context.databaseFile), { serverOrigin: text(flags, "public-url") ?? null }),
         notifications: () => effectivePrimary(process.env, dirname(context.databaseFile), loadBotToken(process.env, context.telegramTokenFile) !== null).channel === "slack",
     }).catch(() => progress("watch: Slack stopped. Check Slack settings before reconnecting."));
     const teamsFollower = followTeams({ store, dir: dirname(context.databaseFile), signal: followController.signal,
         readProjects: telegramReadProjects(context), evidenceRoot: context.evidenceRoot,
         ...(context.mateSeams?.subscriptionRunner ? { subscriptionRunner: context.mateSeams.subscriptionRunner } : {}),
-        ...(context.heldCoordinator ? { held: context.heldCoordinator } : {}),
         origin: () => phoneOrigin(process.env, dirname(context.databaseFile), { serverOrigin: text(flags, "public-url") ?? null }),
         notifications: () => effectivePrimary(process.env, dirname(context.databaseFile), loadBotToken(process.env, context.telegramTokenFile) !== null).channel === "teams",
     }).catch(() => progress("watch: Teams stopped. Check Teams settings before reconnecting."));
     const discordFollower = followDiscord({ store, dir: dirname(context.databaseFile), signal: followController.signal,
         readProjects: telegramReadProjects(context), evidenceRoot: context.evidenceRoot,
         ...(context.mateSeams?.subscriptionRunner ? { subscriptionRunner: context.mateSeams.subscriptionRunner } : {}),
-        ...(context.heldCoordinator ? { held: context.heldCoordinator } : {}),
         origin: () => phoneOrigin(process.env, dirname(context.databaseFile), { serverOrigin: text(flags, "public-url") ?? null }),
         notifications: () => effectivePrimary(process.env, dirname(context.databaseFile), loadBotToken(process.env, context.telegramTokenFile) !== null).channel === "discord",
     }).catch(() => progress("watch: Discord stopped. Check Discord settings before reconnecting."));
@@ -8638,7 +6918,6 @@ async function upCommand(flags, context) {
     if ("error" in loadedRegistry) {
         return fail(write, json, "up", "registry", loadedRegistry.error, EXIT.refused);
     }
-    let sweptHeldOrphans = false;
     const gitRun = context.gitRunner ?? ((file, args, opts) => run(file, [...args], { ...(opts?.cwd === undefined ? {} : { cwd: opts.cwd }), timeoutMs: 10_000 }));
     const rootInputs = (text(flags, "project-root") ?? "")
         .split(",")
@@ -8752,18 +7031,6 @@ async function upCommand(flags, context) {
         for (let attempt = 1; attempt <= 5; attempt += 1) {
             const suffix = attempt === 1 ? "" : `-${attempt}`;
             const name = attempt === 1 ? base : `${base.slice(0, RUNNER_NAME_MAX - suffix.length)}${suffix}`;
-            // The held orphan fence runs BEFORE any recovery road (v6 W6): a
-            // crashed predecessor's held session is seized, killed through its
-            // supervisor, and settled — or paged — before registration's
-            // recovery may touch its run or worktree.
-            if (!sweptHeldOrphans) {
-                sweptHeldOrphans = true;
-                const swept = await sweepHeldOrphans(store, `up:${hostname()}:${process.pid}`, clock);
-                if (swept.fenced > 0)
-                    progress(`fenced ${swept.fenced} orphaned held session(s) from a previous up`);
-                if (swept.paged > 0)
-                    progress(`${swept.paged} held session(s) could not be stopped — see the inbox`);
-            }
             // The runner's authority is BOUND to the canonical roots this up
             // serves (MCP spec v6): the claim gate enforces membership, so an
             // unbound registration would deny-all its own dispatches.
@@ -8810,28 +7077,6 @@ async function upCommand(flags, context) {
     finally {
         await new Promise(done => probe.close(() => done()));
     }
-    // 6b. The held-session coordinator (Phase 2): one per up process, shared
-    // by the console and every watch loop through the context. Its socket
-    // directory is deliberately SHORT and flat — sun_path is unforgiving.
-    const heldDir = namedPath(homedir(), ["held"], { dot: true });
-    try {
-        mkdirSync(heldDir, { recursive: true, mode: 0o700 });
-    }
-    catch {
-        // The launch's own path check refuses with words if this failed.
-    }
-    context.heldCoordinator = new HeldSessionCoordinator();
-    context.upIncarnation = randomUUID();
-    context.heldSocketDir = heldDir;
-    // v28: sessions are unbounded unless the operator caps them.
-    const heldCap = text(flags, "max-held-sessions");
-    if (heldCap !== undefined) {
-        const cap = Number(heldCap);
-        if (!Number.isInteger(cap) || cap < 1) {
-            return fail(write, json, "up", "usage", "--max-held-sessions is a whole number of concurrent attended sessions, at least 1", EXIT.usage);
-        }
-        context.maxHeldSessions = cap;
-    }
     // 7. The console, on the just-released port. The tiny window between the
     // probe closing and this bind can lose a race; that failure tears down
     // cleanly below instead of leaving identities half-claimed silently.
@@ -8854,14 +7099,6 @@ async function upCommand(flags, context) {
             currentRepos: () => [...activeRepos],
             admittedRepos: () => [...activeRepos],
             upConsole: true,
-            attended: {
-                runner: runnerName,
-                coordinator: context.heldCoordinator,
-                headOf: async (repo) => {
-                    const answer = await gitRun("git", ["--no-optional-locks", "rev-parse", "HEAD"], { cwd: repo });
-                    return answer.code === 0 ? answer.stdout.trim() : null;
-                },
-            },
             registryPath,
             projectRoots,
             ...(text(flags, "public-url") === undefined ? {} : { publicUrl: text(flags, "public-url") }),
@@ -9203,23 +7440,6 @@ async function upCommand(flags, context) {
         clearTimeout(graceTimer);
     process.removeListener("SIGINT", stop);
     process.removeListener("SIGTERM", stop);
-    // Held sessions fence before the runner retires (v2 S0f): bounded, and
-    // every controller settles conservatively or is paged.
-    if (context.heldCoordinator !== undefined) {
-        const unsettled = await context.heldCoordinator.close();
-        for (const runId of unsettled) {
-            // The shutdown deadline won: say so durably — the orphan sweep of
-            // the NEXT up owns the cleanup, and silence would contradict
-            // "settled conservatively or paged".
-            store.enqueueNotification({
-                source: { run: runId },
-                dedupeKey: `held-shutdown-unsettled:${runId}`,
-                kind: "attended-unsettled",
-                subject: `an attended session did not settle before shutdown (run #${runId})`,
-                body: `The shutdown deadline passed before run #${runId}'s session finished fencing. The next \`toolroll up\` will fence and settle it; its worktree is preserved.`,
-            }, clock());
-        }
-    }
     retireRunnerIfCurrent(store, runnerName, runnerToken, clock());
     await new Promise(done => console_.server.close(() => done()));
     if (fatal !== null) {
@@ -10776,7 +8996,6 @@ function showTask(positional, context) {
         claim: currentClaim(store, ref.id, now),
         scope,
         approval: approvalOf(scope),
-        risk: ref.riskLevel ?? scope?.riskLevel ?? "routine",
         route: taskRouteView(routed, readiness),
         runs: runs.map(one => ({ ...one, route: store.runRoute(one.id) })),
         deliverable: ref.deliverable,
@@ -11174,9 +9393,6 @@ function blockTask(positional, flags, context) {
             return fail(write, json, "task block", "unknown-task", `no task \`${each}\``, EXIT.refused);
         }
     }
-    const racingGuard = refuseWhileRacing(context, "task block", id);
-    if (racingGuard !== null)
-        return racingGuard;
     const result = store.addEdge(id, on, mutationFrom(flags, context.now));
     if (!result.ok)
         return fail(write, json, "task block", "rejected", result.reason, EXIT.refused);
@@ -11213,9 +9429,7 @@ async function steerTask(positional, flags, context) {
             ? `no task \`${id}\``
             : filed.reason === "task-finished"
                 ? `${id} is finished — a note has no next attempt to reach`
-                : filed.reason === "contest-open"
-                    ? "agents are racing on this task — steering waits until the tournament settles"
-                    : (filed.problem ?? "that note will not store");
+                : (filed.problem ?? "that note will not store");
         return fail(write, json, "task steer", filed.reason, detail, EXIT.refused);
     }
     return succeed(write, json, "task steer", { task: id, note: filed.id }, () => [
@@ -11233,9 +9447,6 @@ function unblockTask(positional, flags, context) {
     if (store.getTask(id) === null) {
         return fail(write, json, "task unblock", "unknown-task", `no task \`${id}\``, EXIT.refused);
     }
-    const racingGuard = refuseWhileRacing(context, "task unblock", id);
-    if (racingGuard !== null)
-        return racingGuard;
     const result = store.removeEdge(id, on, mutationFrom(flags, context.now));
     if (!result.ok) {
         return fail(write, json, "task unblock", "not-waiting", `${id} was not waiting on ${on}`, EXIT.refused);
@@ -11271,9 +9482,7 @@ function nextTask(positional, flags, context) {
                 ? `${id} is not queued — only queued work can move up`
                 : moved.reason === "claimed"
                     ? `${id} is being built right now — it needs no place in line`
-                    : moved.reason === "contest-open"
-                        ? "a tournament is running on this task — let it finish, then pick or abandon it from the tournament screen in the console (the task's page links to it)"
-                        : "the queue rank could not be raised any further";
+                    : "the queue rank could not be raised any further";
         return fail(write, json, "task next", moved.reason, message, EXIT.refused);
     }
     return succeed(write, json, "task next", { task: id, priority: moved.priority }, () => [
@@ -11301,13 +9510,11 @@ function assignTask(positional, flags, context) {
                 ? `${id} is not queued — only queued work can be reserved`
                 : moved.reason === "claimed"
                     ? `${id} is being built right now — it needs no reservation`
-                    : moved.reason === "contest-open"
-                        ? "a tournament is running on this task — let it finish, then pick or abandon it from the tournament screen in the console (the task's page links to it)"
-                        : moved.reason === "no-such-worker"
-                            ? `no worker named \`${runner}\` — \`toolroll runner list\` names them`
-                            : moved.reason === "worker-retired"
-                                ? `${runner} is retired — register the name again, or reserve for another worker`
-                                : "the queue did not accept the move";
+                    : moved.reason === "no-such-worker"
+                        ? `no worker named \`${runner}\` — \`toolroll runner list\` names them`
+                        : moved.reason === "worker-retired"
+                            ? `${runner} is retired — register the name again, or reserve for another worker`
+                            : "the queue did not accept the move";
         return fail(write, json, "task assign", moved.reason, message, EXIT.refused);
     }
     return succeed(write, json, "task assign", { task: id, reservedFor: runner ?? null }, () => [
@@ -11338,7 +9545,6 @@ async function reopenTask(positional, flags, context) {
             "not-latched": `${id} was never closed on its tracker — there is nothing to reopen`,
             "not-seen-open": "the tracker has not been SEEN open again since the close — reopen it there, then `toolroll sync`",
             claimed: `${id} is being built right now`,
-            "contest-open": "a tournament is open on this task — decide it first",
             held: "a hold stands — lift it first (`task unhold`, or wait out the timer)",
             "question-open": "an unanswered question stands — answer or close it first (it is on the task page)",
             "incident-open": "an unresolved incident stands — resolve it first",
@@ -11400,7 +9606,7 @@ async function scopeTask(positional, flags, context) {
     const id = positional[0];
     const goal = text(flags, "goal");
     if (id === undefined || goal === undefined) {
-        return fail(write, json, "task scope", "usage", "`toolroll task scope <id> --goal <what success is> --acceptance <rubric> [--not <text>] [--touches a,b] [--candidate <commit>] [--budget-usd <n>] [--race provider:model[,provider:model…]] [--race-count 2..4] [--race-per-usd <n>] [--race-total-usd <n>]`", EXIT.usage);
+        return fail(write, json, "task scope", "usage", "`toolroll task scope <id> --goal <what success is> --acceptance <rubric> [--not <text>] [--touches a,b] [--candidate <commit>] [--budget-usd <n>]`", EXIT.usage);
     }
     if (store.getTask(id) === null) {
         return fail(write, json, "task scope", "unknown-task", `no task \`${id}\``, EXIT.refused);
@@ -11416,10 +9622,6 @@ async function scopeTask(positional, flags, context) {
     if (acceptanceParse.problems.length > 0) {
         return fail(write, json, "task scope", "bad-acceptance", `${acceptanceParse.problems.map(p => p.message).join("; ")} — each criterion is \`<statement>|<evidence,kinds>\`, criteria are \`;\`-separated, and \`--acceptance plan\` asks the planner to write the rubric`, EXIT.usage);
     }
-    const riskGiven = text(flags, "risk");
-    if (riskGiven !== undefined && !isRiskLevel(riskGiven)) {
-        return fail(write, json, "task scope", "usage", `--risk is one of ${RISK_LEVELS.join(", ")}`, EXIT.usage);
-    }
     // v69: a prepared commit. The machine checks it out as the attempt and
     // runs the gate and the review; no agent is dispatched for it.
     const candidateGiven = text(flags, "candidate");
@@ -11433,9 +9635,6 @@ async function scopeTask(positional, flags, context) {
     const badText = validateScopeText({ goal, outOfScope: text(flags, "not") ?? null, touches });
     if (badText !== null)
         return fail(write, json, "task scope", badText.reason, badText.message, EXIT.usage);
-    // A tournament rides the same filing (stage 3): --race names the agents,
-    // the dollar terms are REQUIRED, and everything lands unapproved — the
-    // one yes later covers scope AND race terms as a single fingerprint.
     const budgetGiven = text(flags, "budget-usd");
     const defaults = store.getSpendDefaults();
     const budgetUsd = budgetGiven !== undefined
@@ -11446,84 +9645,6 @@ async function scopeTask(positional, flags, context) {
     if (budgetGiven !== undefined && (!Number.isFinite(Number(budgetGiven)) || Number(budgetGiven) <= 0)) {
         return fail(write, json, "task scope", "bad-budget", "--budget-usd is a positive dollar amount", EXIT.usage);
     }
-    const raceGiven = text(flags, "race");
-    const raceCountGiven = text(flags, "race-count");
-    // The comparison road (Phase 3 slice B): labeled lanes, no dollar terms,
-    // any registered provider — refused outright when every lane could hold
-    // a real budget (the discipline gate: race those instead).
-    const compareGiven = text(flags, "compare");
-    const permissionMode = store.refFor(BUILT_IN, id).permissionMode ?? store.permissionDefault().mode;
-    let plannedComparison = null;
-    if (compareGiven !== undefined) {
-        if (raceGiven !== undefined || raceCountGiven !== undefined || text(flags, "race-per-usd") !== undefined || text(flags, "race-total-usd") !== undefined) {
-            return fail(write, json, "task scope", "usage", "--compare and the --race flags are different ceremonies — file one or the other", EXIT.usage);
-        }
-        if (store.mirrorByTask(id) !== null) {
-            return fail(write, json, "task scope", "external-race", "external work compares in a follow-up release — file the comparison on a local task", EXIT.refused);
-        }
-        const lanes = compareGiven.split(",").map(one => {
-            const [provider = "", model = ""] = one.trim().split(":");
-            return { provider, model, permissionMode };
-        });
-        plannedComparison = planComparison({ agents: lanes });
-        if (!plannedComparison.ok) {
-            return fail(write, json, "task scope", plannedComparison.reason, plannedComparison.message, EXIT.refused);
-        }
-    }
-    let plannedRace = null;
-    if (raceCountGiven !== undefined && raceGiven === undefined) {
-        return fail(write, json, "task scope", "usage", "--race-count needs --race to name the competing agent, e.g. `--race claude:claude-sonnet-5 --race-count 3`", EXIT.usage);
-    }
-    if (raceGiven !== undefined) {
-        // Explicit flags win; absent ones fall back to the configured defaults
-        // (operator request) — the digest binds the ACTUAL numbers either way,
-        // and the approval restates them. A budget that is simply MISSING is
-        // named as the missing flag here, before planTournament's generic
-        // positive-amount backstop turns it into a riddle (round-4 finding 11).
-        if (text(flags, "race-per-usd") === undefined && defaults?.racePerAgentMicrousd == null) {
-            return fail(write, json, "task scope", "bad-budget", "--race-per-usd is missing and no default is set — pass it, or set one with `toolroll config set budgets --race-per-usd <n>`", EXIT.usage);
-        }
-        if (text(flags, "race-total-usd") === undefined && defaults?.raceTotalMicrousd == null) {
-            return fail(write, json, "task scope", "bad-budget", "--race-total-usd is missing and no default is set — pass it, or set one with `toolroll config set budgets --race-total-usd <n>`", EXIT.usage);
-        }
-        const perUsd = Number(text(flags, "race-per-usd") ?? (defaults?.racePerAgentMicrousd == null ? Number.NaN : defaults.racePerAgentMicrousd / 1_000_000));
-        const totalUsd = Number(text(flags, "race-total-usd") ?? (defaults?.raceTotalMicrousd == null ? Number.NaN : defaults.raceTotalMicrousd / 1_000_000));
-        if (store.mirrorByTask(id) !== null) {
-            return fail(write, json, "task scope", "external-race", "external work races in a follow-up release — file the tournament on a local task", EXIT.refused);
-        }
-        let agents = raceGiven.split(",").map(one => {
-            const [provider = "", model = ""] = one.trim().split(":");
-            return { provider, model, permissionMode };
-        });
-        // The competing-agent COUNT (operator request): an explicit --race-count
-        // replicates a single named agent; with several named agents it may only
-        // agree with the list — a count that contradicts an explicit lineup is a
-        // question, not an instruction. Absent both, the configured default
-        // count replicates a single agent; an explicit list is always itself.
-        if (raceCountGiven !== undefined) {
-            const count = Number(raceCountGiven);
-            if (!Number.isInteger(count) || count < 2 || count > 4) {
-                return fail(write, json, "task scope", "usage", "--race-count is how many agents compete: a whole number from 2 to 4", EXIT.usage);
-            }
-            if (agents.length === 1) {
-                agents = Array.from({ length: count }, () => ({ ...agents[0] }));
-            }
-            else if (agents.length !== count) {
-                return fail(write, json, "task scope", "usage", `--race names ${agents.length} agents but --race-count says ${count} — make them agree, or name one agent and let the count replicate it`, EXIT.usage);
-            }
-        }
-        else if (agents.length === 1 && defaults?.raceAgents != null) {
-            agents = Array.from({ length: defaults.raceAgents }, () => ({ ...agents[0] }));
-        }
-        plannedRace = planTournament({
-            agents,
-            perAgentBudgetUsd: perUsd,
-            totalBudgetUsd: totalUsd,
-        });
-        if (!plannedRace.ok) {
-            return fail(write, json, "task scope", plannedRace.reason, plannedRace.message, EXIT.refused);
-        }
-    }
     // v24 routing flags (foundations finding 5 — these used to be silently
     // swallowed by the global parser): explicit flags resolve HERE, and an
     // explicit ask that cannot resolve refuses rather than filing unresolved.
@@ -11537,11 +9658,6 @@ async function scopeTask(positional, flags, context) {
         }
         explicitProfile = resolvedRouting.profile;
     }
-    // The race branch shares ONE transaction with the scope save (round-6
-    // finding 5): the attended exclusion refuses BEFORE anything writes, and
-    // a filing failure rolls the proposal back — a refused race never leaves
-    // a rewritten scope behind it.
-    //
     // C1/M3, the credentialed-CLI road: when --as/--token authenticate the
     // MODE'S SIGNER and their live mode auto-approves filings, the scope
     // seals in the same transaction it files — plain scopes only, and the
@@ -11559,8 +9675,7 @@ async function scopeTask(positional, flags, context) {
         const previous = store.getScope(id);
         if (ref === null || ref.deliverable === "report" || candidateGiven !== undefined ||
             ref.sizing?.source === "person" || approvalOf(previous).approved ||
-            previous?.proposedVia === "coordinator" || store.filedViaOf(id)?.startsWith("mcp:") ||
-            ((plannedRace !== null || plannedComparison !== null) && store.openAuthorizationFor(ref.id) !== null))
+            previous?.proposedVia === "coordinator" || store.filedViaOf(id)?.startsWith("mcp:"))
             return null;
         const input = { title: store.getTask(id).title, goal, outOfScope: text(flags, "not") ?? null, touches };
         const sized = store.applySizing(id, heuristicSizing(input), { followPlanning: true }, now);
@@ -11571,22 +9686,12 @@ async function scopeTask(positional, flags, context) {
     // Wait for bounded classification before filing so any approval in the
     // transaction below binds the final size and route.
     await refinement;
-    // ONE replayed composite (surfaces round 1, finding 3): the filing, any
-    // race/comparison terms, AND the mode seal record as a single operation —
+    // ONE replayed composite (surfaces round 1, finding 3): the filing AND
+    // the mode seal record as a single operation —
     // a replayed key returns the FIRST answer whole instead of re-sealing
     // whatever scope happens to be current.
     const filed = store.replay(mutationFrom(flags, now), "task-scope-filed", () => store.transact(() => {
-        if (plannedRace !== null && plannedRace.ok) {
-            const raceRef = store.refFor(BUILT_IN, id);
-            if (store.openAuthorizationFor(raceRef.id) !== null) {
-                return {
-                    ok: false,
-                    reason: "attended-open",
-                    message: "an attended authorization is open on this task — revoke it before filing a tournament",
-                };
-            }
-        }
-        const coverage = plannedRace === null && plannedComparison === null && actor !== null
+        const coverage = actor !== null
             ? modeFilingCoverage(store, store.refFor(BUILT_IN, id).repo, actor, now)
             : null;
         const proposed = propose(store, {
@@ -11596,8 +9701,6 @@ async function scopeTask(positional, flags, context) {
             outOfScope: text(flags, "not") ?? null,
             touches,
             acceptance: acceptanceParse.criteria,
-            // v47: the declared risk — a durable task choice the route reads.
-            ...(riskGiven === undefined ? {} : { riskLevel: riskGiven }),
             ...(budgetUsd !== null
                 ? { budgetMicrousd: Math.round(budgetUsd * 1_000_000) }
                 : coverage?.defaultBudgetMicrousd != null
@@ -11625,42 +9728,6 @@ async function scopeTask(positional, flags, context) {
             sealedUnderMode = store.sealScopeApproval(id, actor, now, {}, { kind: "mode", modeDigest: coverage.digest });
             modeRefusedCoordinator = !sealedUnderMode;
         }
-        if (plannedRace !== null && plannedRace.ok) {
-            const plan = plannedRace.plan;
-            const raceRef = store.refFor(BUILT_IN, id);
-            store.fileTournamentTerms({
-                taskRef: raceRef.id,
-                raceDigest: plan.raceDigest,
-                agents: plan.agents,
-                perAgentBudgetMicrousd: plan.perAgentBudgetMicrousd,
-                overrunReserveMicrousd: plan.overrunReserveMicrousd,
-                totalBudgetMicrousd: plan.totalBudgetMicrousd,
-                priceVersion: plan.priceVersion,
-                publicationPolicy: plan.publicationPolicy,
-            }, now);
-        }
-        if (plannedComparison !== null && plannedComparison.ok) {
-            const plan = plannedComparison.plan;
-            const compareRef = store.refFor(BUILT_IN, id);
-            if (store.openAuthorizationFor(compareRef.id) !== null) {
-                return {
-                    ok: false,
-                    reason: "attended-open",
-                    message: "an attended authorization is open on this task — revoke it before filing a comparison",
-                };
-            }
-            store.fileTournamentTerms({
-                taskRef: compareRef.id,
-                kind: "comparison",
-                raceDigest: plan.comparisonDigest,
-                agents: plan.agents,
-                perAgentBudgetMicrousd: 0,
-                overrunReserveMicrousd: 0,
-                totalBudgetMicrousd: 0,
-                priceVersion: 0,
-                publicationPolicy: plan.publicationPolicy,
-            }, now);
-        }
         return { ok: true, scope: proposed, sealedUnderMode, modeRefusedCoordinator };
     }));
     if (!filed.ok) {
@@ -11679,33 +9746,6 @@ async function scopeTask(positional, flags, context) {
             ...describeScope(scope),
         ]);
     }
-    if (plannedComparison !== null && plannedComparison.ok) {
-        const plan = plannedComparison.plan;
-        return succeed(write, json, "task scope", { scope, comparison: plan }, () => [
-            `Scope and comparison written for ${id}. Nothing builds until somebody approves BOTH, with one yes:`,
-            ...describeScope(scope),
-            "",
-            ...plan.laneWords.map(lane => `  ${lane}`),
-            "  no dollar caps exist on a comparison — each agent runs until it finishes or stops making progress;",
-            "  spend lands measured only where the harness reports dollars",
-            "",
-            `  toolroll task approve ${id} --yes`,
-        ]);
-    }
-    if (plannedRace !== null && plannedRace.ok) {
-        const plan = plannedRace.plan;
-        const worst = plan.perAgentReserveMicrousd.reduce((sum, reserve) => sum + plan.perAgentBudgetMicrousd + reserve, 0);
-        return succeed(write, json, "task scope", { scope, race: plan }, () => [
-            `Scope and tournament written for ${id}. Nothing builds until somebody approves BOTH, with one yes:`,
-            ...describeScope(scope),
-            "",
-            `  tournament: ${plan.agents.map(agent => `${agent.provider} · ${agent.model}`).join("  vs  ")}`,
-            `  each agent may spend $${(plan.perAgentBudgetMicrousd / 1_000_000).toFixed(2)}, plus its stated overrun reserve;` +
-                ` worst case $${(worst / 1_000_000).toFixed(2)} total`,
-            "",
-            `  toolroll task approve ${id} --yes`,
-        ]);
-    }
     return succeed(write, json, "task scope", { scope }, () => [
         `Scope written for ${id}. Nothing will build it until somebody approves it.`,
         ...describeScope(scope),
@@ -11724,8 +9764,8 @@ async function scopeTask(positional, flags, context) {
 /**
  * `toolroll task route <id>` (v47): the explainable phase route —
  * shown from the ONE projection every surface renders, and edited only by
- * an approver: `--risk <routine|elevated|high>` declares the task's risk;
- * `--phase <p> --provider <p> --model <m>` records a per-phase override
+ * an approver: `--size <small|medium|large>` and `--risky yes|no` set the
+ * task's size; `--phase <p> --provider <p> --model <m>` records a per-phase override
  * (an exact pair — approvals bind exact routing) with attribution;
  * `--clear-phase <p>` removes one; `--digest <d>` names the scope digest
  * you read, so the edit lands only on what you saw. The edit is ONE
@@ -11739,13 +9779,12 @@ async function routeTaskCommand(positional, flags, context) {
     const { store, write, json, clock } = context;
     const id = positional[0];
     if (id === undefined) {
-        return fail(write, json, "task route", "usage", "`toolroll task route <id> [--risk <level>] [--phase <p> --provider <p> --model <m> | --clear-phase <p>] [--digest <d>] --as <you> --token <t>`", EXIT.usage);
+        return fail(write, json, "task route", "usage", "`toolroll task route <id> [--size <s>] [--risky yes|no] [--phase <p> --provider <p> --model <m> | --clear-phase <p>] [--digest <d>] --as <you> --token <t>`", EXIT.usage);
     }
     if (store.getTask(id) === null) {
         return fail(write, json, "task route", "unknown-task", `no task \`${id}\``, EXIT.refused);
     }
     const ref = store.refFor(BUILT_IN, id);
-    const riskGiven = text(flags, "risk");
     const phaseGiven = text(flags, "phase");
     const clearGiven = text(flags, "clear-phase");
     const providerGiven = text(flags, "provider");
@@ -11753,7 +9792,7 @@ async function routeTaskCommand(positional, flags, context) {
     const digestGiven = text(flags, "digest");
     const sizeGiven = text(flags, "size");
     const riskyGiven = text(flags, "risky");
-    const editing = riskGiven !== undefined || phaseGiven !== undefined || clearGiven !== undefined || sizeGiven !== undefined || riskyGiven !== undefined;
+    const editing = phaseGiven !== undefined || clearGiven !== undefined || sizeGiven !== undefined || riskyGiven !== undefined;
     const show = () => {
         const scope = store.getScope(id);
         const current = store.refFor(BUILT_IN, id);
@@ -11761,7 +9800,7 @@ async function routeTaskCommand(positional, flags, context) {
         const readiness = store.readinessLookupFor(ref.repo, ref.assignedRunner, clock());
         const view = taskRouteView(routed, readiness);
         const projection = routed !== null && routed.kind === "route" ? projectRoute(routed.route, readiness) : null;
-        return succeed(write, json, "task route", { id, risk: current.riskLevel ?? scope?.riskLevel ?? "routine", riskConsequence: riskConsequence(current.riskLevel ?? scope?.riskLevel ?? "routine"), size: current.sizing ?? null, source: routed !== null && routed.kind === "route" ? routed.source : routed?.kind ?? null, route: view, overrides: current.routeOverrides ?? [], digest: scope?.digest ?? null, approval: approvalOf(scope) }, () => routed === null
+        return succeed(write, json, "task route", { id, size: current.sizing ?? null, source: routed !== null && routed.kind === "route" ? routed.source : routed?.kind ?? null, route: view, overrides: current.routeOverrides ?? [], digest: scope?.digest ?? null, approval: approvalOf(scope) }, () => routed === null
             ? [`${id}: no route can be recommended yet — ${scope === null ? "place the task in a repository and file a scope" : scope.unresolvedReason ?? "the phase configuration cannot resolve"}`]
             : [
                 ...(projection === null
@@ -11769,10 +9808,8 @@ async function routeTaskCommand(positional, flags, context) {
                     : [
                         `${id}: route ${routed.kind === "route" && routed.source === "approved" ? "SEALED by the approval" : routed.kind === "route" && routed.source === "proposed" ? "proposed — the next approval seals it" : "recommended live — no scope filed yet"}`,
                         ...routeWords(projection),
-                        // The declared risk, in the words the console and chat
-                        // use — and the one CLI hint that belongs here, not in
-                        // the route's own reasons: how a stronger tier is named.
-                        `  risk         ${current.riskLevel ?? scope?.riskLevel ?? "routine"} — ${riskConsequence(current.riskLevel ?? scope?.riskLevel ?? "routine")}`,
+                        // The one CLI hint that belongs here, not in the
+                        // route's own reasons: how a stronger tier is named.
                         ...(projection.legs.some(leg => leg.reasons.some(reason => reason.startsWith("no stronger")))
                             ? [`               name a stronger agent once with \`config set <phase> --tier strong --provider … --model …\``]
                             : []),
@@ -11795,9 +9832,6 @@ async function routeTaskCommand(positional, flags, context) {
     if (acting === null) {
         return fail(write, json, "task route", "usage", "changing a route takes `--as <you> --token <t>` — it reroutes spend", EXIT.usage);
     }
-    if (riskGiven !== undefined && !isRiskLevel(riskGiven)) {
-        return fail(write, json, "task route", "usage", `--risk is one of ${RISK_LEVELS.join(", ")}`, EXIT.usage);
-    }
     if (sizeGiven !== undefined && !isTaskSize(sizeGiven)) {
         return fail(write, json, "task route", "usage", `--size is one of ${TASK_SIZES.join(", ")}`, EXIT.usage);
     }
@@ -11835,7 +9869,6 @@ async function routeTaskCommand(positional, flags, context) {
             const authenticated = authenticateApprover(store, acting.name, acting.token);
             return authenticated.ok ? { ok: true } : { ok: false, reason: authenticated.reason };
         },
-        ...(riskGiven === undefined ? {} : { risk: riskGiven }),
         ...(sizeEdit === undefined ? {} : { size: sizeEdit }),
         ...(phase === undefined
             ? {}
@@ -11846,7 +9879,7 @@ async function routeTaskCommand(positional, flags, context) {
         if (edited.reason === "unauthenticated") {
             return fail(write, json, "task route", edited.detail, describeApproveFailure(edited.detail, id), EXIT.refused);
         }
-        const code = { "no-task": "unknown-task", "live-claim": "claimed", "contest-open": "contest-open", changed: "changed", nothing: "usage", "not-configured": "not-configured" };
+        const code = { "no-task": "unknown-task", "live-claim": "claimed", changed: "changed", nothing: "usage", "not-configured": "not-configured" };
         return fail(write, json, "task route", code[edited.reason], `${id}: ${edited.detail}`, edited.reason === "nothing" ? EXIT.usage : EXIT.refused);
     }
     const code = show();
@@ -11888,14 +9921,6 @@ async function approveTask(positional, flags, context) {
             write(line);
         for (const line of planContractLines(store, context.evidenceRoot, id))
             write(line);
-        const interactiveRace = store.activeTournamentTerms(store.refFor(BUILT_IN, id).id);
-        if (interactiveRace !== null) {
-            write("");
-            write(`  AND it starts a tournament: ${interactiveRace.n} agents build this independently —`);
-            write(`  ${interactiveRace.agents.map(agent => `${agent.provider} · ${agent.model}`).join("  vs  ")}`);
-            write(`  each may spend $${(interactiveRace.perAgentBudgetMicrousd / 1_000_000).toFixed(2)} plus its overrun reserve;`);
-            write(`  the whole tournament is capped at $${(interactiveRace.totalBudgetMicrousd / 1_000_000).toFixed(2)}. You pick the winner; only the winner publishes.`);
-        }
         write("");
         const agreed = await confirm("Approve exactly this?");
         if (!agreed) {
@@ -11924,14 +9949,8 @@ async function approveTask(positional, flags, context) {
         for (const line of planContractLines(store, context.evidenceRoot, id))
             write(line);
         write("");
-        const previewRace = store.activeTournamentTerms(store.refFor(BUILT_IN, id).id);
-        if (previewRace !== null) {
-            write(`  AND the tournament: ${previewRace.agents.map(agent => `${agent.provider} · ${agent.model}`).join("  vs  ")}`);
-            write(`  each capped at $${(previewRace.perAgentBudgetMicrousd / 1_000_000).toFixed(2)} + reserve, total $${(previewRace.totalBudgetMicrousd / 1_000_000).toFixed(2)}`);
-            write("");
-        }
         write("Nothing has been approved. Agree to this exact scope with:");
-        write(`  toolroll task approve ${id} --yes --digest ${previewRace === null ? scope.digest : jointApprovalDigest(scope.digest, previewRace.raceDigest)} --as <you> --token <your password>`);
+        write(`  toolroll task approve ${id} --yes --digest ${scope.digest} --as <you> --token <your password>`);
         // Unconfirmed is "no, not yet" — exit 3 in both modes (round-4 finding 10).
         return EXIT.refused;
     }
@@ -11943,38 +9962,6 @@ async function approveTask(positional, flags, context) {
     // The credential is required for a different reason: an agent that can run
     // these commands can read the digest out of `task show`, and an approval
     // nobody has to authenticate would let it agree to its own brief.
-    //
-    // A tournament task's yes covers BOTH documents (finding 31): the named
-    // digest is tournament-approval/v1 = H(scope, race), and the scope and
-    // the race terms approve together, in one transaction, or not at all.
-    const raceTerms = store.activeTournamentTerms(store.refFor(BUILT_IN, id).id);
-    if (raceTerms !== null) {
-        const joint = jointApprovalDigest(scope.digest, raceTerms.raceDigest);
-        if (saw !== joint && saw !== scope.digest) {
-            return fail(write, json, "task approve", "changed", `this task races a tournament — approve the JOINT fingerprint: ${joint}`, EXIT.refused);
-        }
-        if (saw === scope.digest && !confirmedAloud) {
-            return fail(write, json, "task approve", "changed", `this task races a tournament — the yes must name the joint fingerprint ${joint}, which covers the race terms too`, EXIT.refused);
-        }
-        const both = store.transact(() => {
-            const scopeApproved = approve(store, id, asWho, now, scope.digest, token, mutationFrom(flags, now));
-            if (!scopeApproved.ok)
-                return scopeApproved;
-            if (!store.approveTournamentTerms(raceTerms.id, asWho, raceTerms.raceDigest, now)) {
-                throw new Error("the race terms changed while you were reading — nothing was approved");
-            }
-            return scopeApproved;
-        });
-        if (!both.ok) {
-            return fail(write, json, "task approve", both.reason, both.reason === "policy" ? `${id}: ${both.message}` : describeApproveFailure(both.reason, id), EXIT.refused);
-        }
-        return succeed(write, json, "task approve", { scope: both.scope, race: raceTerms }, () => [
-            `Approved — scope AND tournament, with one yes. ${raceTerms.n} agents will build ${id} independently:`,
-            ...describeScope(both.scope),
-            `  ${raceTerms.agents.map(agent => `${agent.provider} · ${agent.model}`).join("  vs  ")}`,
-            `  each may spend $${(raceTerms.perAgentBudgetMicrousd / 1_000_000).toFixed(2)} plus its overrun reserve; total cap $${(raceTerms.totalBudgetMicrousd / 1_000_000).toFixed(2)}`,
-        ]);
-    }
     const approved = approve(store, id, asWho, now, saw, token, mutationFrom(flags, now));
     if (!approved.ok) {
         if (approved.reason === "policy")
@@ -11993,7 +9980,7 @@ function describeApproveFailure(reason, id) {
     if (reason === "requester")
         return `you filed ${id}, and this project needs someone else to approve it`;
     if (reason === "person-required")
-        return `${id} is protected work: a person has to approve it, not an operating mode, a routine or an AI teammate`;
+        return `${id} is protected work: a person has to approve it, not an operating mode, a schedule or an AI teammate`;
     if (reason === "second-approver")
         return `your approval of ${id} is recorded; it is protected work, so a second person needs to approve it (toolroll task approve ${id} as them)`;
     if (reason === "no-approvers") {
@@ -12004,7 +9991,7 @@ function describeApproveFailure(reason, id) {
     if (reason === "unrouted")
         return `${id} was filed before agent routing and its old approval no longer stands — an approval now names exactly which agent plans, builds, repairs, and reviews: file the scope again (\`task scope ${id} …\`) so it is routed under today's agents, then approve it`;
     if (reason === "profile-unresolved")
-        return `${id} cannot name an exact agent for every role — configure the project's agents (\`config set <phase> --provider … --model …\`), then \`routine refresh ${id}\` and approve it again`;
+        return `${id} cannot name an exact agent for every role — configure the project's agents (\`config set <phase> --provider … --model …\`), then file the scope again (\`task scope ${id} …\`) and approve it`;
     return `${id} has no scope to approve`;
 }
 /**
@@ -12354,9 +10341,6 @@ function unholdTask(positional, flags, context) {
     if (store.getTask(id) === null) {
         return fail(write, json, "task unhold", "unknown-task", `no task \`${id}\``, EXIT.refused);
     }
-    const racingGuard = refuseWhileRacing(context, "task unhold", id);
-    if (racingGuard !== null)
-        return racingGuard;
     const lifted = store.unhold(store.refFor(BUILT_IN, id).id, mutationFrom(flags, context.now));
     if (!lifted)
         return fail(write, json, "task unhold", "not-held", `${id} was not on hold`, EXIT.refused);
@@ -13304,14 +11288,6 @@ function text(flags, name) {
 function flag(flags, name) {
     return flags.get(name) === true || flags.get(name) === "true";
 }
-/** A running tournament owns its task (round-1 finding 5): the generic
- * doors refuse until it finishes, is picked, or is abandoned. */
-function refuseWhileRacing(context, command, taskId) {
-    const ref = context.store.lookupRef(taskId);
-    if (ref === null || context.store.openContestFor(ref.id) === null)
-        return null;
-    return fail(context.write, context.json, command, "contest-open", "a tournament is running on this task — let it finish, then pick or abandon it from the tournament screen in the console (the task's page links to it)", EXIT.refused);
-}
 /**
  * The demo fence (Codex adoption review, finding 8): a database stamped as
  * a demo sandbox NEVER spends money or touches the world outside — no
@@ -13321,9 +11297,8 @@ function refuseWhileRacing(context, command, taskId) {
  * is the enforcement.
  */
 /**
- * Deep links (attended A5): when a console URL is configured (`webhook set
- * console-url`), CLI answers print it beside ids so the attended eye can
- * jump. Absent configuration prints nothing — a link nobody configured is
+ * Deep links: when a console URL is configured (`webhook set
+ * console-url`), CLI answers print it beside ids so a reader can jump. Absent configuration prints nothing — a link nobody configured is
  * a guess, and stale guesses are worse than none.
  */
 function consoleLinkFor(context, path) {