sortie-dogs 0.9.10 → 0.9.12

package/README.md

@@ -20,9 +20,42 @@ bounded implementation, canonical validation, and evidence-backed completion.
 
 Requirements: Node.js 22.6 or newer, npm, and OpenCode.
 
-Guides: [日本語](docs/guide-ja.md) · [简体中文](docs/guide-zh-CN.md) · [CLI testing](docs/cli-testing.md)
+Guides: [日本語](docs/guide-ja.md) · [简体中文](docs/guide-zh-CN.md) · [テスト実行](docs/testing.md) · [CLI testing](docs/cli-testing.md)
 
-Release: [v0.9.10](https://github.com/zufall-upon/Sortie-dogs/releases/tag/v0.9.10)
+Release: [v0.9.12](https://github.com/zufall-upon/Sortie-dogs/releases/tag/v0.9.12)
+
+## Provisional quality–cost position
+
+**Reference values, not a successful benchmark claim.** Quality and end-to-end completion
+problems remain. The latest completed qualification attempt ended at `IN_PROGRESS`, so its official
+verifier was not run. Further benchmarks are frozen while completion defects are repaired.
+
+The last complete measured **Bare OpenCode vs Sortie** pair below used one frozen task,
+`datacurve/anko-typed-variable-bindings`, on 2026-09-10. It used Sortie **v0.9.5**, not the
+current release. Both candidates failed the official verifier.
+
+| Metric · one task, one trial per arm | Bare OpenCode | Sortie v0.9.5 |
+| --- | ---: | ---: |
+| Verified PASS | 0/1 | 0/1 |
+| Task-check completion · F2P | 55.6% · 5/9 | 88.9% · 8/9 |
+| Retained checks · P2P | 94/94 | 93/94 |
+| Estimated API-equivalent total cost | $5.42 | $1.46 |
+| Median agent wall · n=1 | 28.7 min | 10.0 min |
+| Premium-model token share · Sol | 100% | 20.1% |
+
+Later Sortie-only evidence is weaker: the v0.9.9 recovery candidate passed **5/9** task
+checks with **0/1 Verified PASS**; `0.9.11-bench.2` did not reach a gradeable completion.
+Those attempts are not pooled into the historical pair above.
+
+![Historical quality–cost reference: Bare at $5.42 and 55.6% task-check completion; Sortie v0.9.5 at $1.46 and 88.9%. Neither achieved Verified PASS.](docs/assets/quality-cost-reference.svg)
+
+The goal is **higher OpenCode task success with selective use of premium models**.
+These reference observations do not yet establish that success-rate claim: Sortie missed
+one task check and regressed one retained check. Codex, Pi, and Oh My OpenCode belong to
+separate methodologies and are not assigned comparable positions on this chart.
+
+[Definitions, frozen inputs, current failure status, and limitations](docs/benchmark-reference.md)
+· [Machine-readable reference values](docs/benchmarks/provisional-reference.json)
 
 ## Why Sortie-dogs?
 
@@ -41,6 +74,15 @@ models are reserved for implementation, escalation, and independent review.
 Writes stay scoped, and completion requires validation evidence. Every completed
 run can return a concise Speed / Cost / Proof debrief.
 
+### Use only as much harness as the task needs
+
+Small changes can skip Scout and independent review when one worker and targeted
+validation are sufficient. Larger work can be decomposed into multiple units;
+units that are safely independent can use a Luna fabric DAG for bounded parallel
+execution. Higher-risk candidates add independent review, while full-suite and
+package verification are reserved for release work. Not every task pays the
+cost of the heaviest workflow.
+
 ## Designed to coexist with OpenCode
 
 Sortie-dogs adds a workflow to your existing setup rather than replacing it.
@@ -266,6 +308,18 @@ dog-coordinator: completion evidence accepted
    progress; repeated batches remain bounded rather than becoming endless
    delegation.
 
+## Built to work on itself
+
+Self-improvement keeps the same scoped manifests, worker ownership, validation,
+and review gates as other work. A loaded plugin is not treated as hot-reloadable:
+source changes are validated first, then packaged into an isolated `_testenv`
+fixture and exercised through the real OpenCode CLI. Continuation and compaction
+changes must demonstrate same-session recovery and terminal completion there.
+
+`npm run test:full` is reserved for explicit release validation; ordinary
+changes run targeted tests and `npm test`. The control plane coordinating a run
+is not replaced while that run is in flight.
+
 ## A visual walkthrough
 
 ### Control complexity

package/dist/asset-version.d.ts

@@ -2,5 +2,5 @@
  * Version of the installable runtime assets. Kept in its own module so the plugin can compare an
  * installed project marker without importing every asset body.
  */
-export declare const RUNTIME_ASSET_VERSION = "0.3.86-codegen-proof-v1";
+export declare const RUNTIME_ASSET_VERSION = "0.3.89-completion-proof-v1";
 export type RuntimeAssetVersion = typeof RUNTIME_ASSET_VERSION;

package/dist/asset-version.js

@@ -2,4 +2,4 @@
  * Version of the installable runtime assets. Kept in its own module so the plugin can compare an
  * installed project marker without importing every asset body.
  */
-export const RUNTIME_ASSET_VERSION = "0.3.86-codegen-proof-v1";
+export const RUNTIME_ASSET_VERSION = "0.3.89-completion-proof-v1";

package/dist/core/worktree-commit-artifact.js

@@ -978,11 +978,11 @@ async function stagedPaths(context) {
     return new Set(fields);
 }
 async function assertNoSubmodules(context, entries) {
-    for (const entry of entries) {
-        const index = await context.git(["ls-files", "--stage", "-z", "--", entry.path]);
-        if (index.toString("utf8").startsWith("160000 ")) {
-            throw new WorktreeCommitArtifactError("invalid-state", "Submodule changes are forbidden.");
-        }
+    if (entries.length === 0)
+        return;
+    const index = await context.git(["ls-files", "--stage", "-z", "--", ...entries.map(({ path }) => path)]);
+    if (index.toString("utf8").split("\0").some((entry) => entry.startsWith("160000 "))) {
+        throw new WorktreeCommitArtifactError("invalid-state", "Submodule changes are forbidden.");
     }
 }
 function decodeGitOutput(source, message) {
@@ -1261,7 +1261,7 @@ export async function produceWorktreeCommitArtifact(request, control) {
         throw new WorktreeCommitArtifactError("validation-failed", "Validation was cancelled.");
     control?.enterProtectedPhase();
     await context.git(["add", "--", ...before.map(({ path }) => path)]);
-    await assertValidatedIndex(context, before, beforeFingerprint);
+    // Keep one fresh index check after staging, immediately before the protected commit.
     await assertValidatedIndex(context, before, beforeFingerprint);
     await context.git([
         "-c", "user.name=Sortie Fabric",

package/dist/core/worktree-parallel-dispatch.js

@@ -2138,57 +2138,79 @@ export class ParallelDispatchCoordinator {
         };
     }
     async buildFabricCandidate(runID, base, tasks) {
+        if (tasks.length === 0)
+            return base;
         let head = base;
-        for (const task of tasks) {
-            const artifact = task.artifact;
-            const nonce = randomUUID();
-            const indexPath = join(this.stateRoot, `.fabric-index-${nonce}`);
-            const patchPath = join(this.stateRoot, `.fabric-patch-${nonce}`);
-            const messagePath = join(this.stateRoot, `.fabric-message-${nonce}`);
-            const indexEnvironment = this.cleanGitEnvironment({ GIT_INDEX_FILE: indexPath });
-            try {
-                await this.gitBuffer(["read-tree", head], indexEnvironment);
-                const patch = await this.gitBuffer([
-                    "diff-tree", "-p", "--binary", "--full-index", "--no-ext-diff", "--no-renames",
-                    artifact.base_sha, artifact.commit_sha, "--",
-                ]);
-                if (patch.byteLength > 8 * 1024 * 1024) {
-                    throw new ParallelDispatchError("wave-integration-failed", "Fabric artifact patch exceeds the practical bound.");
-                }
-                await writeFile(patchPath, patch, { flag: "wx", mode: 0o600 });
-                await this.gitBuffer(["apply", "--cached", "--3way", "--binary", patchPath], indexEnvironment);
-                const tree = (await this.gitBuffer(["write-tree"], indexEnvironment)).toString("utf8").trim();
-                const timestamp = (await this.gitBuffer(["show", "-s", "--format=%ct", artifact.commit_sha])).toString("utf8").trim();
-                if (!SHA.test(tree) || !/^\d{1,12}$/u.test(timestamp))
+        const indexPath = join(this.stateRoot, `.fabric-index-${randomUUID()}`);
+        const indexEnvironment = this.cleanGitEnvironment({ GIT_INDEX_FILE: indexPath });
+        try {
+            // Commit objects are immutable; collect their timestamps once for this build only.
+            const commits = [...new Set(tasks.map((task) => task.artifact.commit_sha))];
+            const metadata = (await this.gitBuffer([
+                "show", "-s", "--no-walk=unsorted", "--format=%H%x00%ct", ...commits, "--",
+            ])).toString("utf8").trimEnd();
+            const timestamps = new Map();
+            for (const line of metadata.split(/\r?\n/u)) {
+                const fields = line.split("\0");
+                if (fields.length !== 2 || !SHA.test(fields[0]) || !/^\d{1,12}$/u.test(fields[1]) || timestamps.has(fields[0])) {
                     throw new Error("identity");
-                const message = `Sortie fabric integration: ${task.descriptor.task_id}\n\n` +
-                    `Sortie-Run: ${runID}\nSortie-Task: ${task.descriptor.task_id}\n` +
-                    `Sortie-Artifact: ${artifact.commit_sha}\nSortie-Base: ${artifact.base_sha}\n`;
-                await writeFile(messagePath, message, { flag: "wx", mode: 0o600 });
-                const commitEnvironment = this.cleanGitEnvironment({
-                    GIT_AUTHOR_NAME: "Sortie Fabric",
-                    GIT_AUTHOR_EMAIL: "sortie@example.invalid",
-                    GIT_COMMITTER_NAME: "Sortie Fabric",
-                    GIT_COMMITTER_EMAIL: "sortie@example.invalid",
-                    GIT_AUTHOR_DATE: `${timestamp} +0000`,
-                    GIT_COMMITTER_DATE: `${timestamp} +0000`,
-                });
-                const commit = (await this.gitBuffer(["commit-tree", tree, "-p", head, "-F", messagePath], commitEnvironment))
-                    .toString("utf8").trim();
-                if (!SHA.test(commit))
-                    throw new Error("commit");
-                head = commit;
-            }
-            catch (error) {
-                if (error instanceof ParallelDispatchError)
-                    throw error;
-                throw new ParallelDispatchError("wave-integration-failed", "Fabric artifact could not be applied to the hidden candidate.");
+                }
+                timestamps.set(fields[0], fields[1]);
             }
-            finally {
-                await Promise.all([indexPath, `${indexPath}.lock`, patchPath, messagePath].map((path) => rm(path, { force: true }).catch(() => undefined)));
+            if (timestamps.size !== commits.length || commits.some((commit) => !timestamps.has(commit)))
+                throw new Error("identity");
+            await this.gitBuffer(["read-tree", base], indexEnvironment);
+            for (const task of tasks) {
+                const artifact = task.artifact;
+                const nonce = randomUUID();
+                const patchPath = join(this.stateRoot, `.fabric-patch-${nonce}`);
+                const messagePath = join(this.stateRoot, `.fabric-message-${nonce}`);
+                try {
+                    const patch = await this.gitBuffer([
+                        "diff-tree", "-p", "--binary", "--full-index", "--no-ext-diff", "--no-renames",
+                        artifact.base_sha, artifact.commit_sha, "--",
+                    ]);
+                    if (patch.byteLength > 8 * 1024 * 1024) {
+                        throw new ParallelDispatchError("wave-integration-failed", "Fabric artifact patch exceeds the practical bound.");
+                    }
+                    await writeFile(patchPath, patch, { flag: "wx", mode: 0o600 });
+                    await this.gitBuffer(["apply", "--cached", "--3way", "--binary", patchPath], indexEnvironment);
+                    const tree = (await this.gitBuffer(["write-tree"], indexEnvironment)).toString("utf8").trim();
+                    const timestamp = timestamps.get(artifact.commit_sha);
+                    if (!SHA.test(tree) || timestamp === undefined)
+                        throw new Error("identity");
+                    const message = `Sortie fabric integration: ${task.descriptor.task_id}\n\n` +
+                        `Sortie-Run: ${runID}\nSortie-Task: ${task.descriptor.task_id}\n` +
+                        `Sortie-Artifact: ${artifact.commit_sha}\nSortie-Base: ${artifact.base_sha}\n`;
+                    await writeFile(messagePath, message, { flag: "wx", mode: 0o600 });
+                    const commitEnvironment = this.cleanGitEnvironment({
+                        GIT_AUTHOR_NAME: "Sortie Fabric",
+                        GIT_AUTHOR_EMAIL: "sortie@example.invalid",
+                        GIT_COMMITTER_NAME: "Sortie Fabric",
+                        GIT_COMMITTER_EMAIL: "sortie@example.invalid",
+                        GIT_AUTHOR_DATE: `${timestamp} +0000`,
+                        GIT_COMMITTER_DATE: `${timestamp} +0000`,
+                    });
+                    const commit = (await this.gitBuffer(["commit-tree", tree, "-p", head, "-F", messagePath], commitEnvironment))
+                        .toString("utf8").trim();
+                    if (!SHA.test(commit))
+                        throw new Error("commit");
+                    head = commit;
+                }
+                finally {
+                    await Promise.all([patchPath, messagePath].map((path) => rm(path, { force: true }).catch(() => undefined)));
+                }
             }
+            return head;
+        }
+        catch (error) {
+            if (error instanceof ParallelDispatchError)
+                throw error;
+            throw new ParallelDispatchError("wave-integration-failed", "Fabric artifact could not be applied to the hidden candidate.");
+        }
+        finally {
+            await Promise.all([indexPath, `${indexPath}.lock`].map((path) => rm(path, { force: true }).catch(() => undefined)));
         }
-        return head;
     }
     cleanGitEnvironment(extra = {}) {
         const environment = { ...process.env };

package/dist/plugin/continuation.d.ts

@@ -199,7 +199,7 @@ export interface ContinuationHooks {
     toolStarted(sessionID: string, tool: string): void;
     blocksTool(sessionID: string): boolean;
     sessionIdle(sessionID: string): Promise<void>;
-    stopAutomaticRecovery(sessionID: string, abortSession?: boolean): Promise<void>;
+    stopAutomaticRecovery(sessionID: string, abortSession?: boolean, resumeOnRealUserTurn?: boolean): Promise<void>;
     recoverStalledTask(sessionID: string, callIDs: readonly string[]): Promise<"recovered" | "identity-rejected" | "capability-unavailable" | "request-rejected">;
     forgetSession(sessionID: string): void;
 }

package/dist/plugin/continuation.js

@@ -225,6 +225,7 @@ export function createContinuationHooks(client, directory, policySource, timings
     const sessions = new Map();
     const warned = new Set();
     const stoppedSessions = new Set();
+    const realTurnResumableStops = new Set();
     function observeTransition(type, sessionID, state, reason) {
         try {
             transitionObserver?.({
@@ -811,8 +812,9 @@ export function createContinuationHooks(client, directory, policySource, timings
         }
         sessions.delete(sessionID);
         stoppedSessions.delete(sessionID);
+        realTurnResumableStops.delete(sessionID);
     }
-    async function stopAutomaticRecovery(sessionID, abortSession = true) {
+    async function stopAutomaticRecovery(sessionID, abortSession = true, resumeOnRealUserTurn = false) {
         const state = sessions.get(sessionID);
         if (state !== undefined) {
             clearTimer(state.cooldownTimer);
@@ -830,8 +832,14 @@ export function createContinuationHooks(client, directory, policySource, timings
         }
         stoppedSessions.delete(sessionID);
         stoppedSessions.add(sessionID);
+        if (resumeOnRealUserTurn)
+            realTurnResumableStops.add(sessionID);
+        else
+            realTurnResumableStops.delete(sessionID);
         while (stoppedSessions.size > MAX_TRACKED_SESSIONS) {
-            stoppedSessions.delete(stoppedSessions.values().next().value);
+            const oldest = stoppedSessions.values().next().value;
+            stoppedSessions.delete(oldest);
+            realTurnResumableStops.delete(oldest);
         }
         const abort = abortSession ? client?.session?.abort : undefined;
         if (abort !== undefined) {
@@ -842,6 +850,8 @@ export function createContinuationHooks(client, directory, policySource, timings
         }
     }
     async function recoverStalledTask(sessionID, callIDs) {
+        if (stoppedSessions.has(sessionID))
+            return "request-rejected";
         const active = policy();
         const resolution = resolveContinuation({
             identity: await readIdentity(sessionID),
@@ -863,6 +873,8 @@ export function createContinuationHooks(client, directory, policySource, timings
             // Reserve durable authority before mutating the host session. A configured authority that
             // cannot issue a ticket must not cause an abort/retry loop or an unproven synthetic send.
             const metadata = await ticketMetadata(sessionID, `watchdog:${callIDs.join(",")}`);
+            if (stoppedSessions.has(sessionID))
+                return "request-rejected";
             await abort.call(client.session, {
                 path: { id: sessionID },
                 query: { directory },
@@ -1164,8 +1176,11 @@ export function createContinuationHooks(client, directory, policySource, timings
                 output.enabled = false;
         },
         observeModel(sessionID, model, synthetic = false) {
-            if (stoppedSessions.has(sessionID))
-                return;
+            if (stoppedSessions.has(sessionID)) {
+                if (synthetic || !realTurnResumableStops.delete(sessionID))
+                    return;
+                stoppedSessions.delete(sessionID);
+            }
             if (!nonEmpty(model.providerID) || !nonEmpty(model.modelID))
                 return;
             const state = stateFor(sessionID);

package/dist/plugin/index.d.ts

@@ -91,6 +91,11 @@ export type HandoffDenialReason = "configuration-unavailable" | "path-invalid" |
 export type FreshSessionReason = "child-lineage" | "asset-contract-skew";
 export type FreshSessionAction = "open-fresh-root" | "install-assets-then-open-fresh-root" | "restart-host-after-install";
 export type FreshSessionResult = Readonly<{
+    status: "redispatch-queued";
+    reason: FreshSessionReason;
+    source_session_id: string;
+    retry_same_session: false;
+}> | Readonly<{
     status: "redispatched";
     reason: FreshSessionReason;
     source_session_id: string;

package/dist/plugin/index.js

@@ -483,15 +483,58 @@ function sameRelativePath(left, right) {
 function textPart(part) {
     return isRecord(part) && typeof part.text === "string" ? part.text : undefined;
 }
+const FILE_PART_KEYS = new Set(["id", "sessionID", "messageID", "type", "mime", "filename", "url", "source"]);
+function safeFilePart(part) {
+    if (Object.keys(part).some((key) => !FILE_PART_KEYS.has(key)) ||
+        typeof part.mime !== "string" || part.mime.length === 0 ||
+        typeof part.url !== "string" || part.url.length === 0 ||
+        (part.filename !== undefined && typeof part.filename !== "string") ||
+        (part.id !== undefined && typeof part.id !== "string") ||
+        (part.sessionID !== undefined && typeof part.sessionID !== "string") ||
+        (part.messageID !== undefined && typeof part.messageID !== "string") ||
+        (part.source !== undefined && !isRecord(part.source)))
+        return undefined;
+    return {
+        type: "file",
+        mime: part.mime,
+        ...(part.filename === undefined ? {} : { filename: part.filename }),
+        url: part.url,
+    };
+}
+/** OpenCode expands text attachments into synthetic explanatory text alongside the real file.
+ * Those derived text parts carry no user authority and are never copied into a fresh prompt.
+ * A ticket-bearing synthetic turn (or a turn without real user text and a safe file) stays synthetic.
+ */
+function realAttachmentParts(parts) {
+    const hasUserText = parts.some((part) => isRecord(part) && part.type === "text" &&
+        part.synthetic !== true && typeof part.text === "string" && part.text.trim().length > 0);
+    const hasFile = parts.some((part) => isRecord(part) && part.type === "file" && part.synthetic !== true && safeFilePart(part) !== undefined);
+    if (!hasUserText || !hasFile || parts.some((part) => isRecord(part) && part.synthetic === true &&
+        (part.type !== "text" || typeof part.text !== "string" || part.metadata !== undefined)))
+        return parts;
+    return parts.filter((part) => !isRecord(part) || part.synthetic !== true);
+}
+function syntheticPrompt(parts) {
+    return realAttachmentParts(parts).some((part) => isRecord(part) && part.synthetic === true);
+}
 function freshSessionPrompt(parts) {
     const prompt = [];
-    for (const part of parts) {
-        if (!isRecord(part) || part.type !== "text" || part.synthetic === true || typeof part.text !== "string") {
+    for (const part of realAttachmentParts(parts)) {
+        if (!isRecord(part) || part.synthetic === true)
             return undefined;
+        if (part.type === "text" && typeof part.text === "string")
+            prompt.push({ type: "text", text: part.text });
+        else if (part.type === "file") {
+            const attachment = safeFilePart(part);
+            if (attachment === undefined)
+                return undefined;
+            prompt.push(attachment);
         }
-        prompt.push({ type: "text", text: part.text });
+        else
+            return undefined;
     }
-    return prompt.length > 0 && prompt.some(({ text }) => text.trim().length > 0) ? prompt : undefined;
+    return prompt.length > 0 && prompt.some((part) => part.type === "text" && part.text.trim().length > 0)
+        ? prompt : undefined;
 }
 /**
  * One handoff entry per line. The coordinator asset emits inline digests as `key: value`, often
@@ -1282,10 +1325,6 @@ export const SortieDogsPlugin = async (input, options) => {
         const messages = input.client?.session?.messages;
         if (messages === undefined)
             return undefined;
-        const currentText = currentParts.filter(isRecord).filter((part) => part.type === "text" && part.synthetic !== true)
-            .map((part) => part.text).filter((text) => typeof text === "string");
-        if (currentText.length === 0)
-            return undefined;
         for (const delay of [0, 10, 50]) {
             if (delay > 0)
                 await new Promise((resolve) => setTimeout(resolve, delay));
@@ -1303,11 +1342,10 @@ export const SortieDogsPlugin = async (input, options) => {
                     continue;
                 const info = isRecord(message.info) ? message.info : undefined;
                 if ((info?.role ?? message.role) !== "user" || (info?.agent ?? message.agent) !== selectedAgent ||
-                    !Array.isArray(message.parts) || message.parts.some((part) => isRecord(part) && part.synthetic === true))
+                    !Array.isArray(message.parts) || syntheticPrompt(message.parts))
                     continue;
-                const persistedText = message.parts.filter(isRecord).filter((part) => part.type === "text")
-                    .map((part) => part.text).filter((text) => typeof text === "string");
-                if (JSON.stringify(persistedText) !== JSON.stringify(currentText))
+                const persistedParts = freshSessionPrompt(message.parts);
+                if (persistedParts === undefined || JSON.stringify(persistedParts) !== JSON.stringify(currentParts))
                     continue;
                 const messageID = info?.id ?? message.id;
                 if (typeof messageID === "string" && messageID.length > 0)
@@ -1503,16 +1541,24 @@ export const SortieDogsPlugin = async (input, options) => {
         }
     }
     async function acceptRealGoalTurn(sessionID, messageID, selectedAgent, parts) {
+        // Root recovery can prove the persisted user message identity before its parts are available.
+        // Preserve that existing empty projection while every observed non-empty shape stays strict.
+        const safeParts = parts.length === 0
+            ? []
+            : freshSessionPrompt(parts);
+        if (safeParts === undefined)
+            throw new Error("SORTIE_GOAL_CONTROL_DENIED: unsafe-message-parts");
         await recoverCompletedGoalReservations(sessionID);
         const ledger = await goalLedger(sessionID);
         const state = (await ledger.readGoal()).state;
         if (state.latest_user_message_id === messageID)
             return;
         const at = new Date().toISOString();
-        const explicitContinuation = parts.map(textPart).filter((value) => value !== undefined).join("\n")
+        const explicitContinuation = safeParts.filter((part) => part.type === "text")
+            .map((part) => part.text).join("\n")
             .split(/\r?\n/u).some((line) => /^\s*(?:goal_acceptance_fingerprint|goal_budget_(?:units|time_ms|cost_usd))\s*:/iu.test(line));
         if (state.goal_id === null || state.phase === "terminal" || (state.phase === "stopped" && !explicitContinuation)) {
-            const acceptance = goalFingerprint({ message_id: messageID, parts: parts.map(textPart).filter((value) => value !== undefined) });
+            const acceptance = goalFingerprint({ message_id: messageID, parts: safeParts.map((part) => part.type === "text" ? part.text : part) });
             await ledger.appendGoal({ kind: "goal.accepted", at,
                 goal_id: goalFingerprint({ root: goalRoot(sessionID), origin_user_message_id: messageID }),
                 revision: 1, scope_epoch: 1, acceptance_fingerprint: acceptance,
@@ -1544,7 +1590,7 @@ export const SortieDogsPlugin = async (input, options) => {
         const persistedInfo = isRecord(message.info) ? message.info : undefined;
         if ((persistedInfo?.role ?? message.role) !== "user" ||
             (persistedInfo?.agent ?? message.agent) !== COORDINATOR_AGENT || !Array.isArray(message.parts) ||
-            message.parts.some((part) => isRecord(part) && part.synthetic === true))
+            syntheticPrompt(message.parts))
             return;
         await acceptRealGoalTurn(sessionID, info.id, COORDINATOR_AGENT, message.parts);
         goalDeclarationAuthority.set(sessionID, info.id);
@@ -1685,9 +1731,6 @@ export const SortieDogsPlugin = async (input, options) => {
         else if (receipt === undefined && proved) {
             receipt = await terminalGoal(sessionID, "completed", "succeeded").catch(() => undefined);
         }
-        else if (receipt === undefined && outcome === "DONE") {
-            receipt = await terminalGoal(sessionID, "completed", "succeeded").catch(() => undefined);
-        }
         else if (receipt === undefined && outcome === "INTERRUPTED") {
             receipt = await terminalGoal(sessionID, "stopped", "stopped").catch(() => undefined);
         }
@@ -2054,6 +2097,9 @@ export const SortieDogsPlugin = async (input, options) => {
             entry.execution.units.includes(reservation.unitID));
         const newEvidence = acceptedEvidence.filter((entry) => entry.measurement.criterion_ids.some((criterionID) => !state.satisfied_criteria.includes(criterionID)));
         const progress = newEvidence.length > 0;
+        // Revalidation can succeed for an already-satisfied criterion after a candidate change.
+        // New criterion coverage controls progress accounting, not the validation disposition.
+        const validated = acceptedEvidence.length > 0;
         const metadata = isRecord(output.metadata) ? output.metadata : undefined;
         const interrupted = metadata?.status === "cancel" || metadata?.status === "cancelled" ||
             output.status === "cancel" || output.status === "cancelled";
@@ -2063,12 +2109,12 @@ export const SortieDogsPlugin = async (input, options) => {
             execution.endedAt !== undefined && Date.parse(execution.startedAt) >= reservation.started - 1000 &&
             execution.outcome === "fail");
         const processDefect = !failedAcceptanceExecution && (childSessionID === undefined || hostBindingDefect ||
-            goalValidationDefects.has(childSessionID));
-        const resultClass = progress ? "acceptance" : interrupted ? "interrupted" : processDefect ? "process-defect" : "acceptance";
+            goalValidationDefects.has(childSessionID) || !validated);
+        const resultClass = validated ? "acceptance" : interrupted ? "interrupted" : processDefect ? "process-defect" : "acceptance";
         await ledger.appendGoal({ kind: "unit.settled", at: new Date().toISOString(),
             reservation_id: reservation.reservationID, receipt_id: goalFingerprint({ call_id: callID, output: outputText.slice(0, 2048) }),
             goal_id: state.goal_id, unit_id: reservation.unitID,
-            disposition: progress ? "succeeded" : interrupted ? "cancelled" : "failed", result_class: resultClass,
+            disposition: validated ? "succeeded" : interrupted ? "cancelled" : "failed", result_class: resultClass,
             progress_fingerprint: progress ? goalFingerprint(acceptedEvidence) : null,
             evidence: acceptedEvidence, elapsed_ms: Math.max(0, Date.now() - reservation.started), cost_usd: null });
         if (childSessionID !== undefined) {
@@ -2181,57 +2227,67 @@ export const SortieDogsPlugin = async (input, options) => {
         }
         const key = `${sourceSessionID}\u0000${reason}`;
         const existing = freshSessionRedispatches.get(key);
+        const queued = { status: "redispatch-queued", reason,
+            source_session_id: sourceSessionID, retry_same_session: false };
         if (existing !== undefined)
-            return await existing.operation;
-        const operation = (async () => {
-            let targetSessionID;
-            try {
-                const created = await create.call(input.client.session, {
-                    query: { directory: input.worktree ?? input.directory },
-                    body: {},
-                });
-                const payload = isRecord(created) && "data" in created ? created.data : created;
-                if (!isRecord(payload) || typeof payload.id !== "string" || payload.id.length === 0) {
-                    return freshSessionFallback(reason, fallbackAction);
-                }
-                targetSessionID = payload.id;
-                const parentID = typeof payload.parentID === "string" ? payload.parentID
-                    : typeof payload.parentId === "string" ? payload.parentId
-                        : undefined;
-                if (parentID !== undefined) {
-                    await deleteFreshSession(targetSessionID);
-                    return freshSessionFallback(reason, fallbackAction);
-                }
-                goalRootSessions.set(targetSessionID, goalRoot(sourceSessionID));
-                const ticket = await issueGoalTicket(targetSessionID, `fresh-root:${reason}`);
-                if (!ticket.issued)
-                    throw new Error("fresh coordinator ticket already outstanding");
-                const sent = await send.call(input.client.session, {
-                    path: { id: targetSessionID },
-                    query: { directory: input.worktree ?? input.directory },
-                    body: { agent: COORDINATOR_AGENT, parts: prompt.map((part) => ({ ...part, synthetic: true,
-                            metadata: ticket.metadata })) },
-                });
-                if (!promptAccepted(sent))
-                    throw new Error("fresh coordinator prompt rejected");
-                appLogInfo("fresh-session.redispatched", sourceSessionID, {
-                    reason,
-                    targetSessionID: targetSessionID.slice(0, 128),
-                });
-                return {
-                    status: "redispatched",
-                    reason,
-                    source_session_id: sourceSessionID,
-                    target_session_id: targetSessionID,
-                    retry_same_session: false,
-                };
-            }
-            catch {
-                if (targetSessionID !== undefined)
-                    await deleteFreshSession(targetSessionID);
-                return freshSessionFallback(reason, fallbackAction);
-            }
-        })();
+            return existing.settled ? await existing.operation : queued;
+        // Both session.create and promptAsync enter the host request scheduler. Defer the entire
+        // redispatch until the child chat hook has returned its typed control error.
+        const operation = new Promise((accept) => {
+            setTimeout(() => {
+                void (async () => {
+                    let targetSessionID;
+                    try {
+                        const created = await create.call(input.client.session, {
+                            query: { directory: input.worktree ?? input.directory },
+                            body: {},
+                        });
+                        const payload = isRecord(created) && "data" in created ? created.data : created;
+                        if (!isRecord(payload) || typeof payload.id !== "string" || payload.id.length === 0) {
+                            return freshSessionFallback(reason, fallbackAction);
+                        }
+                        targetSessionID = payload.id;
+                        const parentID = typeof payload.parentID === "string" ? payload.parentID
+                            : typeof payload.parentId === "string" ? payload.parentId
+                                : undefined;
+                        if (parentID !== undefined) {
+                            await deleteFreshSession(targetSessionID);
+                            return freshSessionFallback(reason, fallbackAction);
+                        }
+                        goalRootSessions.set(targetSessionID, goalRoot(sourceSessionID));
+                        const ticket = await issueGoalTicket(targetSessionID, `fresh-root:${reason}`);
+                        if (!ticket.issued)
+                            throw new Error("fresh coordinator ticket already outstanding");
+                        const sendFresh = send;
+                        const sent = await sendFresh.call(input.client.session, {
+                            path: { id: targetSessionID },
+                            query: { directory: input.worktree ?? input.directory },
+                            body: { agent: COORDINATOR_AGENT, parts: prompt.map((part) => part.type === "text"
+                                    ? { ...part, synthetic: true, metadata: ticket.metadata }
+                                    : part) },
+                        });
+                        if (!promptAccepted(sent))
+                            throw new Error("fresh coordinator prompt rejected");
+                        appLogInfo("fresh-session.redispatched", sourceSessionID, {
+                            reason,
+                            targetSessionID: targetSessionID.slice(0, 128),
+                        });
+                        return {
+                            status: "redispatched",
+                            reason,
+                            source_session_id: sourceSessionID,
+                            target_session_id: targetSessionID,
+                            retry_same_session: false,
+                        };
+                    }
+                    catch {
+                        if (targetSessionID !== undefined)
+                            await deleteFreshSession(targetSessionID);
+                        return freshSessionFallback(reason, fallbackAction);
+                    }
+                })().then(accept);
+            }, 0);
+        });
         const entry = { operation, settled: false };
         freshSessionRedispatches.set(key, entry);
         void operation.then((result) => {
@@ -2254,7 +2310,7 @@ export const SortieDogsPlugin = async (input, options) => {
                 break;
             freshSessionRedispatches.delete(completed[0]);
         }
-        return await operation;
+        return queued;
     }
     async function ensureLoaded() {
         if (loaded?.gate !== undefined)
@@ -3425,9 +3481,6 @@ export const SortieDogsPlugin = async (input, options) => {
                     contract_fingerprint: structuralAdmission.contract_fingerprint, experience: experience.trace });
             }
             const coordinator = await getParallelCoordinator();
-            if (await coordinator.targetCheckedOut(`refs/heads/${structuralAdmission.contract.provenance.target_branch}`)) {
-                return JSON.stringify({ status: "sol-serial", reason: "target-checked-out", experience: experience.trace });
-            }
             const result = await coordinator.prepareFabric(contract, ownerRoot, executionPlanPath === undefined ? undefined : await readJson(resolve(executionPlanPath), INPUT_LIMITS.parallel));
             if (result.status === "sol-serial")
                 return JSON.stringify({ ...result, experience: experience.trace });
@@ -5017,15 +5070,19 @@ export const SortieDogsPlugin = async (input, options) => {
                 const agent = info?.agent ?? message.agent;
                 if (typeof agent !== "string")
                     return undefined;
-                if (message.parts !== undefined && !Array.isArray(message.parts))
+                const parts = message.parts === undefined ? [] : message.parts;
+                if (!Array.isArray(parts))
                     return undefined;
-                const synthetic = Array.isArray(message.parts) && message.parts.some((part) => isRecord(part) && part.synthetic === true);
+                const synthetic = syntheticPrompt(parts);
                 if (synthetic)
                     continue;
+                const persistedParts = parts.length === 0 ? [] : freshSessionPrompt(parts);
+                if (persistedParts === undefined)
+                    return undefined;
                 const messageID = info?.id ?? message.id;
                 if (typeof messageID !== "string" || messageID.length === 0)
                     return undefined;
-                persistedTurn = { agent, synthetic: false, messageID };
+                persistedTurn = { agent, synthetic: false, messageID, parts: persistedParts };
                 break;
             }
             return { hasForeignUserTurn: persistedTurn?.agent !== undefined && persistedTurn.agent !== COORDINATOR_AGENT,
@@ -5121,7 +5178,7 @@ export const SortieDogsPlugin = async (input, options) => {
             return false;
         await rememberCoordinatorRoot(sessionID);
         if (persistedTurn !== undefined) {
-            await acceptRealGoalTurn(sessionID, persistedTurn.messageID, persistedTurn.agent, []);
+            await acceptRealGoalTurn(sessionID, persistedTurn.messageID, persistedTurn.agent, persistedTurn.parts);
             goalDeclarationAuthority.set(sessionID, persistedTurn.messageID);
         }
         await pinAssetVersion(sessionID);
@@ -5560,16 +5617,23 @@ export const SortieDogsPlugin = async (input, options) => {
                     .replaceAll(CONTINUATION_MARKER, "")
                     .trimEnd();
             }
+            const hostRunOutcome = terminalRunOutcome(textOutput.text);
             textOutput.text = await preserveActiveGoalContinuation(textInput.sessionID, textOutput.text, textInput.messageID);
-            const runOutcome = terminalRunOutcome(textOutput.text);
+            // Preserve a host DONE claim for proof checks while allowing an ordinary local INTERRUPTED
+            // response to remain presentation-only IN_PROGRESS continuation.
+            const runOutcome = hostRunOutcome === "DONE" ? hostRunOutcome : terminalRunOutcome(textOutput.text);
             const terminal = runOutcome === undefined || !isCoordinatorSession(textInput.sessionID)
                 ? undefined
                 : await terminalGoalFromHostText(textInput.sessionID, textOutput.text);
-            if (runOutcome === "DONE" && terminal !== undefined && (terminal.delivery === "running" ||
-                (terminal.receipt === undefined && terminal.goal !== undefined && terminal.goal.goal_id !== null))) {
-                textOutput.text = replaceDoneTerminalStatus(textOutput.text, terminal.delivery === "running"
-                    ? "status: IN_PROGRESS — durable delivery active; same sessionでjoinまたはstale reconcileが必要"
-                    : "status: IN_PROGRESS\ngoal_control: accepted criteria remain unproved");
+            if (runOutcome === "DONE" && terminal?.delivery === "running") {
+                textOutput.text = replaceDoneTerminalStatus(textOutput.text, "status: IN_PROGRESS — durable delivery active; same sessionでjoinまたはstale reconcileが必要");
+                await continuation.stopAutomaticRecovery(textInput.sessionID, false, true);
+            }
+            else if (runOutcome === "DONE" && terminal?.receipt === undefined &&
+                terminal?.goal !== undefined && terminal.goal.goal_id !== null) {
+                textOutput.text = replaceDoneTerminalStatus(textOutput.text, "status: INTERRUPTED — accepted criteria remain unproved\n" +
+                    "TRUE_INTERRUPTION: internal: accepted criteria remain unproved");
+                await continuation.stopAutomaticRecovery(textInput.sessionID, false, true);
             }
             if (runOutcome !== "DONE" && terminal?.delivery === "ready" && terminal.receipt?.status === "succeeded") {
                 textOutput.text = replaceTerminalStatus(textOutput.text, "status: DONE");
@@ -5663,7 +5727,7 @@ export const SortieDogsPlugin = async (input, options) => {
             observedChildTerminals.delete(chatInput.sessionID);
             await serializeChatTransition(chatInput.sessionID, async () => {
                 const parentID = chatParentID(chatInput);
-                const synthetic = output.parts.some((part) => isRecord(part) && part.synthetic === true);
+                const synthetic = syntheticPrompt(output.parts);
                 const selectedAgent = chatInput.agent ?? output.message.agent;
                 if (parentID !== undefined)
                     rememberParent(chatInput.sessionID, parentID);
@@ -5673,8 +5737,15 @@ export const SortieDogsPlugin = async (input, options) => {
                     output.message.agent = chatInput.agent;
                 }
                 const requestedCoordinator = selectedAgent === COORDINATOR_AGENT;
+                const projectedCoordinatorParts = requestedCoordinator && !synthetic && output.parts.length > 0
+                    ? freshSessionPrompt(output.parts)
+                    : undefined;
+                if (requestedCoordinator && !synthetic && output.parts.length > 0 && projectedCoordinatorParts === undefined) {
+                    throw new Error("SORTIE_GOAL_CONTROL_DENIED: unsafe-message-parts");
+                }
                 const messageID = realMessageID(chatInput, output) ?? (synthetic ? undefined
-                    : await persistedCurrentRealMessageID(chatInput.sessionID, selectedAgent, output.parts));
+                    : projectedCoordinatorParts === undefined ? undefined
+                        : await persistedCurrentRealMessageID(chatInput.sessionID, selectedAgent, projectedCoordinatorParts));
                 if (requestedCoordinator && !coordinatorRoot && (parentID !== undefined || knownChildSessions.has(chatInput.sessionID)) &&
                     !synthetic && messageID !== undefined) {
                     await acceptRealGoalTurn(chatInput.sessionID, messageID, selectedAgent, output.parts);
@@ -5721,11 +5792,11 @@ export const SortieDogsPlugin = async (input, options) => {
                         // Some native hosts persist the user message only after this hook returns. Defer to the
                         // system-transform boundary, but retain no synthetic authority and accept only the exact
                         // final persisted real-user parts through persistedCurrentRealMessageID.
-                        pendingRealGoalTurns.set(chatInput.sessionID, { selectedAgent, parts: [...output.parts] });
+                        pendingRealGoalTurns.set(chatInput.sessionID, { selectedAgent, parts: projectedCoordinatorParts ?? [] });
                         pruneParallelChildMap(pendingRealGoalTurns);
                         schedulePendingRealGoalRecovery(chatInput.sessionID);
                     }
-                    const prompt = synthetic ? undefined : freshSessionPrompt(output.parts);
+                    const prompt = projectedCoordinatorParts;
                     if (prompt !== undefined)
                         coordinatorPrompts.set(chatInput.sessionID, prompt);
                     // Synthetic coordinator turns reach here only after consumeGoalTicket accepted the
@@ -6676,14 +6747,19 @@ export const SortieDogsPlugin = async (input, options) => {
                 return;
             // Deletion is terminal for watchdog recovery. Disarm synchronously before any generic event
             // processing can await, touch activity, or let a queued sweep recover the cancelled root.
-            if (event.type === "session.deleted")
+            if (event.type === "session.deleted") {
                 disarmDeletedCoordinatorTaskWatchdog(eventSessionID);
+                await continuation.stopAutomaticRecovery(eventSessionID, false);
+            }
             if (event.type === "message.updated" && info !== undefined) {
                 rememberCoordinatorInterruption(eventSessionID, info);
-                await acceptPersistedRealGoalEvent(eventSessionID, info);
+                if (pendingRealGoalTurns.has(eventSessionID))
+                    await recoverPendingRealGoalTurn(eventSessionID);
+                else
+                    await acceptPersistedRealGoalEvent(eventSessionID, info);
             }
             if (pendingRealGoalTurns.has(eventSessionID) &&
-                (event.type === "message.updated" || event.type === "message.part.updated")) {
+                event.type === "message.part.updated") {
                 await recoverPendingRealGoalTurn(eventSessionID);
             }
             const eventPartTime = isRecord(eventPart?.time) ? eventPart.time : undefined;

package/dist/plugin/run-metrics.d.ts

@@ -112,7 +112,15 @@ type SortieGoalSnapshot = Pick<GoalFlightState, "acceptance_contract" | "consume
 export declare function createSortieResult(receipt: GoalTerminalReceipt, goal: SortieGoalSnapshot, metrics: RunMetrics | undefined, asOf?: string, records?: readonly GoalFlightEventRecord[]): SortieResult;
 export type RunTerminalOutcome = "DONE" | "INTERRUPTED" | "BLOCKED" | "NEED_DECISION";
 export declare function collectRunMetrics(client: RunMetricsClient | undefined, rootSessionID: string, directory?: string, now?: number, window?: RunMetricsWindow): Promise<RunMetrics | undefined>;
-export declare function formatSortieResult(result: SortieResult): string;
+export interface SortieResultPresentation {
+    readonly implementation?: string;
+    readonly pending?: string;
+    readonly next?: string;
+    readonly commit?: string;
+    readonly statusSummary?: string;
+    readonly stopReason?: string;
+}
+export declare function formatSortieResult(result: SortieResult, presentation?: SortieResultPresentation): string;
 export declare function formatRunMetrics(metrics: RunMetrics): string;
 export declare function isDoneTerminalText(text: string): boolean;
 export declare function terminalRunOutcome(text: string): RunTerminalOutcome | undefined;

package/dist/plugin/run-metrics.js

@@ -1,4 +1,4 @@
-import { buildDebrief, renderDebrief, observeDebriefSession } from "./sortie-debrief.js";
+import { buildDebrief, renderDebrief, renderDebriefProof, observeDebriefSession } from "./sortie-debrief.js";
 import { goalFingerprint } from "../core/goal-bound.js";
 import { renderCareer } from "./sortie-career.js";
 const unavailable = (reason) => ({ availability: "unavailable", value: null, reason });
@@ -341,25 +341,61 @@ function duration(milliseconds) {
 function metricText(metric, render) {
     return metric.availability === "available" ? render(metric.value) : "計測不可";
 }
-export function formatSortieResult(result) {
+const displayText = (value) => value?.replace(/[\r\n\t]+/gu, " ").trim() || "未取得";
+const reportFence = (body) => {
+    const longest = Math.max(0, ...[...body.matchAll(/^~+/gmu)].map((match) => match[0].length));
+    const fence = "~".repeat(Math.max(3, longest + 1));
+    return `${fence}text\n${body}\n${fence}`;
+};
+export function formatSortieResult(result, presentation = {}) {
     const criteria = metricText(result.proof.criteria, (entries) => {
         const passing = entries.filter(({ status }) => status === "PASS").length;
         return `${passing}/${entries.length}`;
     });
-    const achievement = result.mission.status === "COMPLETED" ? "完了"
+    const achievement = result.mission.status === "COMPLETED" ? "COMPLETED"
+        : result.mission.status === "INTERRUPTED" ? "INTERRUPTED"
+            : result.mission.status === "EXTERNAL_BLOCKER" ? "EXTERNAL_BLOCKER" : "USER_DECISION";
+    const summaryLabel = result.mission.status === "COMPLETED" ? "完了"
         : result.mission.status === "INTERRUPTED" ? "中断（未完了）"
-            : result.mission.status === "EXTERNAL_BLOCKER" ? "外部要因で未完了"
-                : "ユーザー判断待ち（未完了）";
+            : result.mission.status === "EXTERNAL_BLOCKER" ? "外部要因で未完了" : "ユーザー判断待ち（未完了）";
     const color = result.mission.status === "COMPLETED" ? "🟢" : result.mission.status === "EXTERNAL_BLOCKER" ? "🔴" : "🟡";
+    const proof = renderDebriefProof(result.debrief);
     const body = [
-        `**⚡ 任務経過（待機含む）:** **${metricText(result.speed.goal_wall_ms, duration)}**`,
-        `**🪙 使用量:** **${metricText(result.cost.total_tokens, (value) => `${value.toLocaleString("ja-JP")} tokens`)}** · host推定額 ${metricText(result.cost.cost_usd, (value) => `$${value.toFixed(4)}`)}（実課金換算なし）`,
+        "🐾 SORTIE DOGS — 帰還報告",
+        result.result_id[0],
+        "",
+        `${color} ${achievement} — ${displayText(presentation.statusSummary)}`,
+        "",
+        "⚔️ MISSION",
+        `経過          ⏱ ${metricText(result.speed.goal_wall_ms, duration)} ※待機含む`,
+        `最終達成条件  ◔ ${criteria}`,
+        `対象検証      ${proof.validation}`,
+        `SourceReview  ${proof.review}`,
+        `Commit        ${displayText(presentation.commit)}`,
+        "",
+        "🔧 実装",
+        displayText(presentation.implementation),
+        "",
+        "⏳ 未実施",
+        displayText(presentation.pending),
+        "",
+        "➡️ NEXT",
+        displayText(presentation.next),
+        "",
+        "🪙 COST / PACK",
+        `使用量        ${metricText(result.cost.total_tokens, (value) => `${value.toLocaleString("ja-JP")} tokens`)}`,
+        `host推定額    ${metricText(result.cost.cost_usd, (value) => `$${value.toFixed(4)}`)} ※実課金換算なし`,
+        "",
         ...renderDebrief(result.debrief),
-        `**🛡 達成:** ${color} **${achievement}** · 達成条件 **${criteria}**`,
-        "*最終応答生成前の計測*",
+        "",
         ...renderCareer(result.career),
-    ].join("\n\n");
-    return `<details>\n<summary><strong>🐾 SORTIE DOGS — 帰還報告｜${color} ${achievement}</strong></summary>\n\n${body}\n\n</details>`;
+        "",
+        "🛑 STOP REASON",
+        displayText(presentation.stopReason ?? result.mission.stop_reason),
+        "",
+        "※使用量は最終応答生成前の計測",
+    ].join("\n");
+    return `<details>\n<summary><strong>🐾 SORTIE DOGS — 帰還報告｜${color} ${summaryLabel}</strong></summary>\n\n${reportFence(body)}\n\n</details>`;
 }
 export function formatRunMetrics(metrics) {
     const elapsed = metrics.durationMilliseconds === undefined ? "duration unavailable" : `${duration(metrics.durationMilliseconds)} wall-clock`;
@@ -476,6 +512,7 @@ export function insertRunMetrics(text, metrics) {
     return lines.join(newline);
 }
 export function insertSortieResult(text, result) {
+    const presentation = extractSortiePresentation(text);
     const visible = sanitizeTerminalReport(text);
     const checkpoint = terminalCheckpoint(visible);
     if (checkpoint === undefined)
@@ -506,7 +543,7 @@ export function insertSortieResult(text, result) {
         cleaned.splice(checkpoint.index + 1, 1);
     let card;
     try {
-        card = formatSortieResult(result);
+        card = formatSortieResult(result, presentation);
     }
     catch {
         card = "<details>\n<summary><strong>🐾 SORTIE DOGS — 帰還報告</strong></summary>\n\n**確認:** 表示集計を取得できません。任務結果は先頭の状態を参照。\n\n</details>";
@@ -514,6 +551,23 @@ export function insertSortieResult(text, result) {
     cleaned.splice(checkpoint.index + 1, 0, "", card, "");
     return cleaned.join(newline).trimEnd();
 }
+function extractSortiePresentation(text) {
+    const first = topLevelLines(text).find(({ line }) => line.trim().length > 0)?.line ?? "";
+    const statusSummary = first.split(/\s+[—-]\s+/u).slice(1).join(" — ").trim() || undefined;
+    const section = (names) => {
+        const expression = new RegExp(`^[ \\t]*(?:#{1,6}[ \\t]*)?(?:\\*\\*)?(?:${names})(?:\\*\\*)?[ \\t]*:?[ \\t]*(?:\\*\\*)?[ \\t]*(.*)$`, "imu");
+        const match = expression.exec(text);
+        if (match === null)
+            return undefined;
+        if (match[1]?.trim())
+            return match[1].trim();
+        const tail = text.slice(match.index + match[0].length).split(/\r?\n/u);
+        return tail.find((line) => line.trim().length > 0 && !/^[ \\t]*(?:#{1,6}[ \\t]*)?(?:\\*\\*)?(?:変更点|実装|未実施|次|NEXT|Commit|コミット)/iu.test(line))?.trim();
+    };
+    const explicitStop = /^(?:TRUE_INTERRUPTION|TRUE_BLOCKER)[ \\t]*:[ \\t]*(.+)$/imu.exec(text)?.[1]?.trim();
+    return { statusSummary, implementation: section("変更点|実装"), pending: section("未実施"), next: section("次|NEXT"),
+        commit: section("Commit|コミット"), stopReason: explicitStop };
+}
 export function createGoalReport(result, receipt) {
     const tokens = result.cost.total_tokens.availability === "available" && Number.isSafeInteger(result.cost.total_tokens.value)
         ? result.cost.total_tokens.value : null;

package/dist/plugin/sortie-career.js

@@ -124,21 +124,25 @@ export async function collectCareer(directories, currentPath, current, read, max
 }
 export function renderCareer(career) {
     if (career === undefined)
-        return ["**📜 PACK RECORD:** 保存履歴を取得できません"];
+        return ["📜 PACK RECORD", "保存履歴を取得できません"];
     const terminal = career.goals - career.active;
-    const firstPass = career.firstPass.eligible === 0 ? "計測不可（対象0件）" : `${career.firstPass.count}/${career.firstPass.eligible}件`;
-    const minutes = (metric) => metric.covered === 0 ? "計測不可" : `${(metric.sum / 60000).toFixed(1)}分（${metric.covered}/${terminal}任務）`;
-    const models = [...career.models].sort((a, b) => b.tokens - a.tokens || a.model.localeCompare(b.model));
-    const modelText = models.slice(0, 4).map((entry) => `${entry.model.replace(/[\\`*_{}\[\]()<>!|\r\n]/gu, "").slice(0, 120)} ${entry.tokens.toLocaleString("ja-JP")}`).join(" · ");
+    const minutes = (metric) => metric.covered === 0 ? "計測不可" : `${(metric.sum / 60000).toFixed(1)}分 ※${metric.covered}/${terminal}任務`;
     return [
-        "**📜 PACK RECORD — 記録済み戦績**",
-        `**戦績:** 完了 ${career.completed} · 中断 ${career.interrupted} · 外部待機 ${career.external} · 指示待ち ${career.decision} · 進行中 ${career.active}`,
-        `**初回完遂:** ${firstPass} · 復帰 ${career.recoveries}件（計測記録 ${career.telemetryCovered}/${terminal}任務）`,
-        `**累積使用量:** ${career.tokens.covered === 0 ? "計測不可" : `${career.tokens.sum.toLocaleString("ja-JP")} tokens`}（計測 ${career.tokens.covered}/${terminal}任務）`,
-        `**累積モデル:** ${career.modelCovered === 0 ? "計測不可" : modelText || "出力なし"}${models.length > 4 ? " · ほか" : ""}（token計測 ${career.modelCovered}/${terminal}任務）`,
-        `**累積時間:** worker ${minutes(career.workerTime)} · goal期間合計 ${minutes(career.goalWall)}（待機含む・同時刻重複あり）`,
-        `**累積実行重複率:** ${career.overlap.ratio === null ? "計測不可" : `${career.overlap.ratio.toFixed(2)}×`}（総和の比・${career.overlap.covered}/${terminal}任務・速度倍率ではありません）`,
-        `**保存範囲:** ${career.since?.slice(0, 10) ?? "開始日不明"}以降の現存履歴 · ${career.coverage.included}/${career.coverage.files}ファイル${career.coverage.unavailable || career.coverage.truncated ? " · 部分集計" : ""} · 生涯戦績ではありません`,
-        ...(career.titles.length ? [`**🎖 隊の称号:** ${career.titles.join(" · ")}`] : []),
+        "📜 PACK RECORD",
+        `🏁 完了 ${career.completed}`,
+        `🟡 中断 ${career.interrupted}`,
+        `⏳ 外部待機 ${career.external}`,
+        `❓ 指示待ち ${career.decision}`,
+        `🔄 進行中 ${career.active}`,
+        `↩️ 復帰 ${career.recoveries}`,
+        "",
+        `🪙 累積使用量 ${career.tokens.covered === 0 ? "計測不可" : `${career.tokens.sum.toLocaleString("ja-JP")} tokens`} ※${career.tokens.covered}/${terminal}任務`,
+        `⏱ 累積worker ${minutes(career.workerTime)}`,
+        `🕰 累積goal ${minutes(career.goalWall)} ※待機・重複含む`,
+        `⚡ 累積重複率 ${career.overlap.ratio === null ? "計測不可" : `${career.overlap.ratio.toFixed(2)}×`} ※${career.overlap.covered}/${terminal}任務・速度倍率ではありません`,
+        "",
+        "📦 保存範囲",
+        `${career.since?.slice(0, 10) ?? "開始日不明"}以降 / ${career.coverage.included} of ${career.coverage.files} files${career.coverage.unavailable || career.coverage.truncated ? " ※部分集計" : ""}`,
+        "※生涯戦績ではありません",
     ];
 }

package/dist/plugin/sortie-debrief.d.ts

@@ -55,4 +55,8 @@ export interface Debrief {
 export declare function observeDebriefSession(id: string, root: boolean, messages: readonly Record<string, unknown>[], window?: Span): DebriefSession;
 export declare function buildDebrief(receipt: GoalTerminalReceipt, contract: GoalAcceptanceContract | null, observation: DebriefObservation | undefined, records?: readonly GoalFlightEventRecord[]): Debrief;
 export declare function renderDebrief(debrief: Debrief | undefined): string[];
+export declare function renderDebriefProof(debrief: Debrief | undefined): {
+    validation: string;
+    review: string;
+};
 export {};

package/dist/plugin/sortie-debrief.js

@@ -254,6 +254,12 @@ export function buildDebrief(receipt, contract, observation, records) {
                 wallMilliseconds: unionDuration(spans) } } : {}) };
 }
 const label = (text) => text.replace(/[\r\n\t]/gu, " ").replace(/[\\`*_{}\[\]()<>!|]/gu, "").slice(0, 120);
+const gauge = (percent) => {
+    const eighths = Math.max(0, Math.min(80, Math.round(percent * 0.8)));
+    const whole = Math.floor(eighths / 8), remainder = eighths % 8;
+    const partial = ["", "▏", "▎", "▍", "▌", "▋", "▊", "▉"][remainder];
+    return `${"█".repeat(whole)}${partial}${" ".repeat(10 - whole - (remainder === 0 ? 0 : 1))}`;
+};
 export function renderDebrief(debrief) {
     const pack = debrief?.pack == null ? null : [...debrief.pack].sort((a, b) => b.count - a.count || a.model.localeCompare(b.model));
     const packVisible = pack?.slice(0, 4) ?? [];
@@ -264,21 +270,24 @@ export function renderDebrief(debrief) {
     if (mix !== null && mix.length > 4)
         visible.push({ model: "その他", tokens: mix.slice(4).reduce((sum, entry) => sum + entry.tokens, 0),
             percent: mix.slice(4).reduce((sum, entry) => sum + entry.percent, 0) });
-    const bars = (percent) => {
-        const filled = Math.max(0, Math.min(10, Math.round(percent / 10)));
-        return "█".repeat(filled) + "░".repeat(10 - filled);
-    };
-    const status = (value) => value === "PASS" ? "🟢 **PASS**" : value === "FAIL" ? "🔴 **FAIL**"
+    const status = (value) => value === "PASS" ? "🟢 PASS" : value === "FAIL" ? "🔴 FAIL"
         : value === "WAIVED" ? "免除" : "未記録";
+    const counts = new Map(packVisible.map((entry) => [entry.model, entry.count]));
     return [
-        `**🐕 出撃隊:** ${pack === null ? "履歴未取得" : pack.length === 0 ? "出撃なし" : packVisible.map((entry) => `${label(entry.model)} **×${entry.count}**`).join(" · ")}`,
-        `**モデル別token内訳:** ${mix === null ? "usage未取得" : ""}`,
-        ...visible.map((entry) => `**↳** ${label(entry.model)} \`${bars(entry.percent)}\` ${entry.percent.toFixed(1)}%`),
-        `**実行重複率:** ${debrief?.overlap !== undefined && debrief.overlap.wallMilliseconds > 0
-            ? `**${(debrief.overlap.workerMilliseconds / debrief.overlap.wallMilliseconds).toFixed(2)}×**（worker区間・速度倍率ではありません）`
+        ...(mix === null ? ["モデル内訳    usage未取得"] : visible.map((entry) => {
+            const count = counts.get(entry.model);
+            return `🐕 ${label(entry.model)} ${gauge(entry.percent)} ${entry.percent.toFixed(1)}% ${entry.tokens.toLocaleString("ja-JP")} tokens${count === undefined ? "" : ` ×${count}`}`;
+        })),
+        `⚡ 実行重複率 ${debrief?.overlap !== undefined && debrief.overlap.wallMilliseconds > 0
+            ? `${(debrief.overlap.workerMilliseconds / debrief.overlap.wallMilliseconds).toFixed(2)}×`
             : pack?.length === 0 ? "対象なし（出撃なし）" : "稼働区間の記録不足"}`,
-        `**確認:** 対象検証 ${status(debrief?.validation ?? "未確認")} · 直近Review ${status(debrief?.review ?? "未確認")}${debrief?.reviewSource === "reviewer" ? "（reviewer報告）" : ""}`,
-        ...(debrief?.notes?.length ? [`**計測範囲:** ${debrief.notes.join(" · ")}`] : []),
-        ...(debrief?.traits.length ? [`**🏅 今回の戦績:** ${debrief.traits.join(" · ")}`] : []),
+        "   ※worker区間・速度倍率ではありません",
+        ...(debrief?.notes?.length ? [`計測範囲      ${debrief.notes.join(" · ")}`] : []),
     ];
 }
+export function renderDebriefProof(debrief) {
+    const status = (value) => value === "PASS" ? "🟢 PASS" : value === "FAIL" ? "🔴 FAIL"
+        : value === "WAIVED" ? "免除" : "未記録";
+    return { validation: status(debrief?.validation ?? "未確認"),
+        review: `${status(debrief?.review ?? "未確認")}${debrief?.reviewSource === "reviewer" ? "（reviewer報告）" : ""}` };
+}

package/dist/runtime-assets.d.ts

@@ -7,7 +7,7 @@ export interface RuntimeAsset {
 }
 export declare const runtimeAssets: readonly [{
     readonly name: "dog-coordinator";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-coordinator.md";
     readonly content: `---
 description: Canonical MkII coordinator packaged by Sortie-dogs
@@ -1391,6 +1391,9 @@ criterion -> changed or inspected implementation path -> concrete exercising tes
 An aggregate validation command without that mapping is insufficient. Multi-form criteria must cover their
 materially distinct syntax, value-shape, scope, and error paths; any missing path remains UNPROVEN and requires
 continued implementation or validation. Include the complete trace in every high-risk SourceReview artifact.
+For accepted failure behavior, require result/error/state evidence and a valid case; for accepted composite
+or wrapped-value behavior, require public-entry-point evidence. Accept justified N/A dimensions and existing
+sufficient evidence without extra validation or review. Do not accept expectations copied from the candidate's output.
 Any generated input/output pair in the changed manifest is high risk and requires SourceReview. DONE requires
 generator command evidence, post-generation candidate identity, generated-output stability, and canonical
 validation after generation. Reject evidence produced only before regeneration.
@@ -1468,10 +1471,10 @@ TERMINAL_STATUS_SEMANTICS_FIXTURE
 END_TERMINAL_STATUS_SEMANTICS_FIXTURE
 
 RUNTIME_ASSET_VERSION_SYNC_FIXTURE
-    runtime_version: 0.3.86-codegen-proof-v1
+    runtime_version: 0.3.89-completion-proof-v1
     shared_marker: src/asset-version.ts
-    packaged_expectation: test/plugin-loader.test.ts uses 0.3.86-codegen-proof-v1
-    initialize_expectation: test/initialize.test.ts uses 0.3.86-codegen-proof-v1
+    packaged_expectation: test/plugin-loader.test.ts uses 0.3.89-completion-proof-v1
+    initialize_expectation: test/initialize.test.ts uses 0.3.89-completion-proof-v1
     rule: runtime asset versions, shared marker, packaged expectation, and initialize expectation change together
 END_RUNTIME_ASSET_VERSION_SYNC_FIXTURE
 
@@ -1494,32 +1497,32 @@ END_INTERNAL_TERMINAL_PROOF_FIXTURE
 `;
 }, {
     readonly name: "dog-worker";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-worker.md";
     readonly content: string;
 }, {
     readonly name: "dog-luna-worker";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-luna-worker.md";
     readonly content: string;
 }, {
     readonly name: "dog-scout";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-scout.md";
     readonly content: "---\ndescription: Bounded evidence scout for dog-coordinator\nmode: subagent\nsteps: 8\npermission:\n  bash: deny\n  webfetch: deny\n  task: deny\n  question: deny\n  glob: deny\n  grep: deny\n  edit: deny\n  list: deny\n  write: deny\n  patch: deny\ntools:\n  bash: false\n  webfetch: false\n  task: false\n  question: false\n  glob: false\n  grep: false\n  edit: false\n  list: false\n  write: false\n  patch: false\n---\n# dog-scout\n\nAccept one concrete missing_evidence_code: manifest, validation, or owner-risk. Accept only an\nexplicit absolute project_root and a known_paths list of at most four paths from dog-coordinator.\nResolve only that evidence key from those paths under project_root; never resolve a path against the\nsession directory. Use Read only, with at most 120 lines and no more than one read per supplied path.\nDo not resolve a second key, explore, invoke another tool, retry, edit, stage, commit, or become user-facing.\n\nWhen project_root is missing, or a supplied path does not resolve under it, or a resolved path is\nunreadable, report that dispatch defect as the facts for the requested key and name the exact paths.\nDo not retry, guess another root, or answer from an unread path.\n\nReturn exactly one concise JSON object of at most 800 characters with exactly these keys:\nmissing_evidence_code, facts, evidence_paths, risks. Use no Markdown, code fence, commentary, or raw log. Return it only\nto dog-coordinator. Write the facts and risks prose in the language the dispatch uses for its own\nprose; keep the keys, paths, commands, and identifiers verbatim.\n";
 }, {
     readonly name: "dog-reviewer";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-reviewer.md";
-    readonly content: "---\ndescription: Independent source reviewer for dog-coordinator\nmode: subagent\npermission:\n  bash: deny\n  webfetch: deny\n  task: deny\n  question: deny\n  glob: deny\n  grep: deny\n  edit: deny\n  list: deny\n  write: deny\n  patch: deny\n  read: deny\ntools:\n  bash: false\n  webfetch: false\n  task: false\n  question: false\n  glob: false\n  grep: false\n  edit: false\n  list: false\n  write: false\n  patch: false\n  read: false\n---\n# dog-reviewer\n\nAccept only one bounded SourceReview request from dog-coordinator after canonical\nvalidation for one high-risk candidate. Review only the supplied acceptance criteria, exact\nmanifest, changedLogicSummary, supplied changed-code excerpts, and validation evidence. Confirm every acceptance item explicitly\nmaps to at least one changedLogicSummary entry and assess that changed logic against the mapped\nacceptance item. Missing or incomplete coverage is a concrete finding, never PASS.\nRequire one indexed acceptance[i] -> changedLogicSummary[j] mapping line per acceptance item and\nreject a missing index or unequal mapping count before assessing the changed logic.\nAlso require each acceptance item to map to a concrete exercising test/input/branch and result. A broad\nsuite PASS without criterion-level exercise evidence is insufficient. When one item contains materially\ndifferent syntax forms, value shapes, scopes, or error paths, reject PASS unless representative traces cover\neach path or the artifact proves they share one implementation path.\nWhen changed files include generator inputs or checked-in generated outputs, require the canonical generator\ncommand, a stable post-generation diff, and validation executed after generation. Reject PASS if helper logic\nexists only in a generated output, regeneration removes behavior, or validation predates the generated candidate.\nDo not request raw logs or full source files, review low-risk candidates, expand scope, or dispatch\nanother agent.\nTreat those supplied fields as the complete bounded SourceReview artifact; use only that artifact and invoke no tools.\nDo not infer that a branch or exemption is absent from source because a prose summary omits it.\nIf the supplied excerpts do not establish a claim, report an evidence gap and request the exact\nbranch/helper excerpt in the next artifact; do not prescribe a source fix for an unproven defect.\n\nReturn one concise PASS or concrete-finding response only to dog-coordinator before the\ncoordinator commit. Write every finding, evidence, and required-fix sentence in the language the\nsupplied artifact uses for its own prose, one statement per line, and keep verdict values,\nidentifiers, paths, and commands verbatim. Do not implement, remediate, resolve blockers, edit,\nstage, commit, or become user-facing. Remain host-routed: do not require or identify a provider, vendor, model, variant,\nor transport.\n";
+    readonly content: "---\ndescription: Independent source reviewer for dog-coordinator\nmode: subagent\npermission:\n  bash: deny\n  webfetch: deny\n  task: deny\n  question: deny\n  glob: deny\n  grep: deny\n  edit: deny\n  list: deny\n  write: deny\n  patch: deny\n  read: deny\ntools:\n  bash: false\n  webfetch: false\n  task: false\n  question: false\n  glob: false\n  grep: false\n  edit: false\n  list: false\n  write: false\n  patch: false\n  read: false\n---\n# dog-reviewer\n\nAccept only one bounded SourceReview request from dog-coordinator after canonical\nvalidation for one high-risk candidate. Review only the supplied acceptance criteria, exact\nmanifest, changedLogicSummary, supplied changed-code excerpts, and validation evidence. Confirm every acceptance item explicitly\nmaps to at least one changedLogicSummary entry and assess that changed logic against the mapped\nacceptance item. Missing or incomplete coverage is a concrete finding, never PASS.\nRequire one indexed acceptance[i] -> changedLogicSummary[j] mapping line per acceptance item and\nreject a missing index or unequal mapping count before assessing the changed logic.\nAlso require each acceptance item to map to a concrete exercising test/input/branch and result. A broad\nsuite PASS without criterion-level exercise evidence is insufficient. When one item contains materially\ndifferent syntax forms, value shapes, scopes, or error paths, reject PASS unless representative traces cover\neach path or the artifact proves they share one implementation path.\nCheck contract-derived expectations only for accepted behavior. For an applicable failure or composite-value\ncriterion, error-only or helper-only assertions can leave public result/state behavior unproved. Accept justified\nN/A dimensions; do not demand new behavior, new review rounds, or redundant checks outside accepted scope.\nWhen changed files include generator inputs or checked-in generated outputs, require the canonical generator\ncommand, a stable post-generation diff, and validation executed after generation. Reject PASS if helper logic\nexists only in a generated output, regeneration removes behavior, or validation predates the generated candidate.\nDo not request raw logs or full source files, review low-risk candidates, expand scope, or dispatch\nanother agent.\nTreat those supplied fields as the complete bounded SourceReview artifact; use only that artifact and invoke no tools.\nDo not infer that a branch or exemption is absent from source because a prose summary omits it.\nIf the supplied excerpts do not establish a claim, report an evidence gap and request the exact\nbranch/helper excerpt in the next artifact; do not prescribe a source fix for an unproven defect.\n\nReturn one concise PASS or concrete-finding response only to dog-coordinator before the\ncoordinator commit. Write every finding, evidence, and required-fix sentence in the language the\nsupplied artifact uses for its own prose, one statement per line, and keep verdict values,\nidentifiers, paths, and commands verbatim. Do not implement, remediate, resolve blockers, edit,\nstage, commit, or become user-facing. Remain host-routed: do not require or identify a provider, vendor, model, variant,\nor transport.\n";
 }, {
     readonly name: "dog-advisor";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "agent/dog-advisor.md";
     readonly content: "---\ndescription: Focused technical advisor for dog-coordinator\nmode: subagent\npermission:\n  bash: deny\n  webfetch: deny\n  task: deny\n  question: deny\n  glob: deny\n  grep: deny\n  edit: deny\n  list: deny\n  write: deny\n  patch: deny\n  read: deny\ntools:\n  bash: false\n  webfetch: false\n  task: false\n  question: false\n  glob: false\n  grep: false\n  edit: false\n  list: false\n  write: false\n  patch: false\n  read: false\n---\n# dog-advisor\n\nAccept only one bounded Strategy request from dog-coordinator for one candidate and one focused\nquestion. Use only the supplied acceptance criteria, exact manifest, constraints, and concise\nevidence. Do not request raw logs or full source files, expand scope, or dispatch another agent.\nTreat those supplied fields as the complete bounded Strategy artifact; use only that artifact and invoke no tools.\nReject every SourceReview request and return the rejection only to dog-coordinator; SourceReview is\ndog-reviewer-only work.\n\nReturn concise options and one recommendation only to dog-coordinator. Write every option,\nrecommendation, and consideration in the language the supplied request uses for its own prose, one\nstatement per line, and keep identifiers, paths, and commands verbatim. Do not perform\nSourceReview, implement, remediate, resolve blockers, edit, stage, commit, or become user-facing.\nImplementation remains dog-worker work. Remain host-routed: do not require or identify a\nprovider, vendor, model, variant, or transport.\n";
 }, {
     readonly name: "sortie";
-    readonly version: "0.3.86-codegen-proof-v1";
+    readonly version: "0.3.89-completion-proof-v1";
     readonly installPath: "command/sortie.md";
     readonly content: "---\ndescription: Start the canonical Sortie-dogs MkII workflow\nagent: dog-coordinator\n---\nRequest: $ARGUMENTS\n\n1. If $ARGUMENTS is empty, request task context and stop; give project init guidance first.\n2. Do not preflight installed runtime assets. The plugin reports version skew without adding model\n   turns; proceed from task evidence and project instructions.\n3. On restart or re-entry, reconstruct context from project-local durable artifacts and the\n   latest bounded handoff or checkpoint. Preserve both manifests and ordered validation history;\n   resume the same task through dog-coordinator with only the required delta.\n4. Otherwise transfer request and project context to dog-coordinator. Frontmatter is the single coordinator\n   transfer; never route a worker to the user.\n";
 }];

package/dist/runtime-assets.js

@@ -1,5 +1,5 @@
 import { GOAL_DECLARATION_FORMAT } from "./core/goal-declaration-format.js";
-const ASSET_VERSION = "0.3.86-codegen-proof-v1";
+const ASSET_VERSION = "0.3.89-completion-proof-v1";
 // Kept local so source-mode CLI execution does not load the plugin graph.
 const BACKLOG_DRAIN_CAPABILITY = "sortie_enable_backlog_drain";
 const PARALLEL_PREPARE_CAPABILITY = "sortie_prepare_parallel_dispatch";
@@ -101,6 +101,8 @@ Validation budget exhaustion, host counters, local routing, and unavailable host
 defects, never TRUE_BLOCKER: external and never a reason to ask the user for an internal route. Return the
 typed defect to dog-coordinator for autonomous repair. A changed candidate may run the next declared
 validation; an unchanged duplicate remains forbidden.
+Run the exact declared canonical validation string in its own tool call. Do not prepend or append
+formatting, generation, or cleanup commands: host evidence must match the declared command boundary.
 
 Before returning canonical PASS, build a criterion-level trace for every accepted criterion. Each trace
 must name the criterion, the changed implementation path or inspected existing path, the concrete test
@@ -108,6 +110,11 @@ case/input form or static branch that exercises it, and PASS or UNPROVEN. A broa
 not prove every criterion. Split criteria that cover multiple syntax forms, value shapes, scopes, or error
 paths into representative paths. If any accepted edge remains UNPROVEN, add a manifest-authorized check or
 return the evidence gap; never report completion from aggregate validation alone.
+Derive expected behavior from the accepted contract and existing public semantics, not the candidate.
+Only when an accepted criterion covers failure behavior, check the public return/result, error, and
+observable state together, paired with a valid case. For an accepted composite or wrapped-value feature,
+exercise its public entry point as well as its helper. Mark inapplicable dimensions N/A with a short reason;
+do not invent failure behavior, widen acceptance, or repeat a proved check to satisfy this guidance.
 Treat generated-source boundaries as high risk. If a manifest changes a generator input, grammar, schema,
 template, or a checked-in generated output, identify the repository's canonical generator and run it before
 the final validation. Prove the regenerated output is stable and that canonical validation ran against that
@@ -1613,6 +1620,9 @@ criterion -> changed or inspected implementation path -> concrete exercising tes
 An aggregate validation command without that mapping is insufficient. Multi-form criteria must cover their
 materially distinct syntax, value-shape, scope, and error paths; any missing path remains UNPROVEN and requires
 continued implementation or validation. Include the complete trace in every high-risk SourceReview artifact.
+For accepted failure behavior, require result/error/state evidence and a valid case; for accepted composite
+or wrapped-value behavior, require public-entry-point evidence. Accept justified N/A dimensions and existing
+sufficient evidence without extra validation or review. Do not accept expectations copied from the candidate's output.
 Any generated input/output pair in the changed manifest is high risk and requires SourceReview. DONE requires
 generator command evidence, post-generation candidate identity, generated-output stability, and canonical
 validation after generation. Reject evidence produced only before regeneration.
@@ -1821,6 +1831,9 @@ Also require each acceptance item to map to a concrete exercising test/input/bra
 suite PASS without criterion-level exercise evidence is insufficient. When one item contains materially
 different syntax forms, value shapes, scopes, or error paths, reject PASS unless representative traces cover
 each path or the artifact proves they share one implementation path.
+Check contract-derived expectations only for accepted behavior. For an applicable failure or composite-value
+criterion, error-only or helper-only assertions can leave public result/state behavior unproved. Accept justified
+N/A dimensions; do not demand new behavior, new review rounds, or redundant checks outside accepted scope.
 When changed files include generator inputs or checked-in generated outputs, require the canonical generator
 command, a stable post-generation diff, and validation executed after generation. Reject PASS if helper logic
 exists only in a generated output, regeneration removes behavior, or validation predates the generated candidate.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sortie-dogs",
-  "version": "0.9.10",
+  "version": "0.9.12",
   "description": "Bounded agent harness and validated orchestration loop plugin for OpenCode",
   "keywords": [
     "opencode",