@pome-sh/cli 0.20.0 → 0.21.1

package/dist/src/hosted/errors.d.ts

@@ -1,60 +0,0 @@
-export declare class HostedAuthError extends Error {
-    readonly requestId?: string | undefined;
-    constructor(message: string, requestId?: string | undefined);
-}
-export declare class HostedQuotaError extends Error {
-    readonly requestId?: string | undefined;
-    readonly details?: Record<string, unknown> | undefined;
-    /** `details` mirrors the cloud error envelope's machine-readable details
-     *  (e.g. `{ kind: "daily_judge_cap" }`) so `pome demo` (FDRS-643/662) can
-     *  render honest labeled at-capacity states. Optional: older responses and
-     *  the twin-pod 401 shape carry none. */
-    constructor(message: string, requestId?: string | undefined, details?: Record<string, unknown> | undefined);
-}
-export declare class HostedOrchError extends Error {
-    readonly requestId?: string | undefined;
-    readonly status?: number | undefined;
-    /** Machine-readable `error.type` from the cloud envelope, when present
-     *  (e.g. `multi_twin_unsupported` so the CLI can map an old-cloud
-     *  multi-twin rejection to a friendly hint). Undefined for
-     *  network/parse failures and the twin-pod 401 shape. */
-    readonly type?: string | undefined;
-    /** `status` is the HTTP status that produced this error, when one exists
-     *  (network/parse failures leave it undefined). `pome eval` uses it to
-     *  scope its reaped-session retry to 404/410 only (FDRS-656 review). */
-    constructor(message: string, requestId?: string | undefined, status?: number | undefined, 
-    /** Machine-readable `error.type` from the cloud envelope, when present
-     *  (e.g. `multi_twin_unsupported` so the CLI can map an old-cloud
-     *  multi-twin rejection to a friendly hint). Undefined for
-     *  network/parse failures and the twin-pod 401 shape. */
-    type?: string | undefined);
-}
-/** Bad invocation — missing scenario file, unknown flag value, etc. Maps
- *  to documented exit code 5 ("usage error"). F0-5a. */
-export declare class HostedUsageError extends Error {
-    constructor(message: string);
-}
-/** FDRS-636 — a group trial errored before a cloud verdict existed
- *  (preflight failure, agent timeout, agent crash). The session was
- *  abandoned (POST /v1/sessions/:id/abandon) with `errorCode`; the group
- *  runner renders the trial as an errored row EXCLUDED from the verdict
- *  fraction. `message` is the short human reason shown on that row. */
-export declare class HostedTrialError extends Error {
-    readonly errorCode: string;
-    constructor(message: string, errorCode: string);
-}
-/** F-983 — `DELETE /v1/sessions/:id` refused to destroy a session whose run
- *  has not been graded. Pome creates the `runs` row at finalize, so an open
- *  session holds an ungraded run and deleting it discards the evidence. The
- *  control plane issues `discardToken`; replaying it confirms the discard.
- *  Deliberately NOT swallowed by `bestEffort` — that flag covers transport
- *  noise and already-closed races, not a deliberate server refusal. */
-export declare class HostedDiscardRefusedError extends Error {
-    readonly sessionId: string;
-    readonly state: string;
-    readonly taskName: string | null;
-    readonly openSeconds: number;
-    readonly discardToken: string;
-    constructor(message: string, sessionId: string, state: string, taskName: string | null, openSeconds: number, discardToken: string);
-}
-export declare function exitCodeFor(err: unknown): number;

package/dist/src/hosted/errors.js

@@ -1,120 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// Cloud → CLI exit-code mapping per docs/05-api-spec.md §1 "pome run".
-//   0 — pass; 1 — below pass-threshold; 2 — twin/orch error;
-//   3 — auth (401/403); 4 — quota (402/429); 5 — usage error.
-//
-// Session A (Linear FDRS-423) audit confirmed cloud-side responses map
-// cleanly: 401 only from `lib/auth.ts` (CLI exit 3), 402/429 = quota (CLI
-// exit 4), 502 = downstream/gateway (CLI exit 2), 410/426 are not auth
-// errors (CLI exit 2). F0-5 covers the CLI-side mapping gaps the test
-// plan walkthrough surfaced:
-//   - F0-5a: `pome run /does/not/exist.md` was returning exit 2 instead
-//     of 5 because file-not-found threw a plain Error.
-//   - F0-5c: `pome logout && pome run` was returning exit 2 instead of 3
-//     because credential-resolution failure threw a plain Error.
-//   - F18: hosted sub-threshold runs sometimes returned exit 3 because
-//     `runTaskHosted.ts` mapped any non-zero agent exit to exit 3,
-//     stealing the auth slot.
-//
-// FDRS-636 — trial groups (`pome run -n k`, k>1) map the WHOLE GROUP to one
-// exit code ([DECISION 2026-07-05]):
-//   0 — at least one trial completed AND every completed trial passed;
-//   1 — at least one completed trial failed its pass threshold;
-//   2 — no trial completed (every trial errored/was abandoned).
-// Errored trials are excluded from the verdict fraction and never lower a
-// passing group below 0 on their own. Single-run (k=1) semantics above are
-// untouched.
-export class HostedAuthError extends Error {
-    requestId;
-    constructor(message, requestId) {
-        super(message);
-        this.requestId = requestId;
-        this.name = "HostedAuthError";
-    }
-}
-export class HostedQuotaError extends Error {
-    requestId;
-    details;
-    /** `details` mirrors the cloud error envelope's machine-readable details
-     *  (e.g. `{ kind: "daily_judge_cap" }`) so `pome demo` (FDRS-643/662) can
-     *  render honest labeled at-capacity states. Optional: older responses and
-     *  the twin-pod 401 shape carry none. */
-    constructor(message, requestId, details) {
-        super(message);
-        this.requestId = requestId;
-        this.details = details;
-        this.name = "HostedQuotaError";
-    }
-}
-export class HostedOrchError extends Error {
-    requestId;
-    status;
-    type;
-    /** `status` is the HTTP status that produced this error, when one exists
-     *  (network/parse failures leave it undefined). `pome eval` uses it to
-     *  scope its reaped-session retry to 404/410 only (FDRS-656 review). */
-    constructor(message, requestId, status, 
-    /** Machine-readable `error.type` from the cloud envelope, when present
-     *  (e.g. `multi_twin_unsupported` so the CLI can map an old-cloud
-     *  multi-twin rejection to a friendly hint). Undefined for
-     *  network/parse failures and the twin-pod 401 shape. */
-    type) {
-        super(message);
-        this.requestId = requestId;
-        this.status = status;
-        this.type = type;
-        this.name = "HostedOrchError";
-    }
-}
-/** Bad invocation — missing scenario file, unknown flag value, etc. Maps
- *  to documented exit code 5 ("usage error"). F0-5a. */
-export class HostedUsageError extends Error {
-    constructor(message) {
-        super(message);
-        this.name = "HostedUsageError";
-    }
-}
-/** FDRS-636 — a group trial errored before a cloud verdict existed
- *  (preflight failure, agent timeout, agent crash). The session was
- *  abandoned (POST /v1/sessions/:id/abandon) with `errorCode`; the group
- *  runner renders the trial as an errored row EXCLUDED from the verdict
- *  fraction. `message` is the short human reason shown on that row. */
-export class HostedTrialError extends Error {
-    errorCode;
-    constructor(message, errorCode) {
-        super(message);
-        this.errorCode = errorCode;
-        this.name = "HostedTrialError";
-    }
-}
-/** F-983 — `DELETE /v1/sessions/:id` refused to destroy a session whose run
- *  has not been graded. Pome creates the `runs` row at finalize, so an open
- *  session holds an ungraded run and deleting it discards the evidence. The
- *  control plane issues `discardToken`; replaying it confirms the discard.
- *  Deliberately NOT swallowed by `bestEffort` — that flag covers transport
- *  noise and already-closed races, not a deliberate server refusal. */
-export class HostedDiscardRefusedError extends Error {
-    sessionId;
-    state;
-    taskName;
-    openSeconds;
-    discardToken;
-    constructor(message, sessionId, state, taskName, openSeconds, discardToken) {
-        super(message);
-        this.sessionId = sessionId;
-        this.state = state;
-        this.taskName = taskName;
-        this.openSeconds = openSeconds;
-        this.discardToken = discardToken;
-        this.name = "HostedDiscardRefusedError";
-    }
-}
-export function exitCodeFor(err) {
-    if (err instanceof HostedAuthError)
-        return 3;
-    if (err instanceof HostedQuotaError)
-        return 4;
-    if (err instanceof HostedUsageError)
-        return 5;
-    return 2;
-}

package/dist/src/hosted/evalResultCache.d.ts

@@ -1,71 +0,0 @@
-import type { CriterionResult } from "../types/shared.js";
-export declare const VERDICT_ARTIFACT_VERSION = 1;
-export declare const VERDICT_FILENAME = "verdict.json";
-export interface VerdictArtifact {
-    version: number;
-    /** Provenance: the only writer is the /finalize response path. */
-    source: "cloud-finalize";
-    task_name: string;
-    /** The task path as the run invocation saw it (may move later — readers
-     *  must tolerate a dangling path). */
-    task_path: string;
-    /** Shared trial-group id (`grp_` + nanoid21); null on single runs. */
-    group_id: string | null;
-    session_id: string;
-    cloud_run_id: string;
-    cloud_dashboard_url: string;
-    judge_model: string | null;
-    /** Cloud-authoritative satisfaction score, 0-100. */
-    score: number;
-    pass_threshold: number;
-    passed: boolean;
-    criteria_results: CriterionResult[];
-    duration_ms: number;
-    finalized_at: string;
-}
-export interface TrialVerdict {
-    runDir: string;
-    verdict: VerdictArtifact;
-}
-export declare function writeVerdictArtifact(runDir: string, verdict: VerdictArtifact): Promise<void>;
-export declare function readVerdictArtifact(runDir: string): Promise<TrialVerdict | null>;
-/** Scan an artifacts root for finalized trials — exactly the two-level
- *  layout `<root>/<task-slug>/<session-id>/verdict.json`. Anything
- *  unreadable is skipped. */
-export declare function scanVerdictArtifacts(artifactsRoot: string): Promise<TrialVerdict[]>;
-export interface RunSet {
-    /** null = a single run that never had a group. */
-    groupId: string | null;
-    taskName: string;
-    /** The task path recorded at run time (first trial's). */
-    taskPath: string;
-    /** Trials sorted by finalized_at ascending. */
-    trials: TrialVerdict[];
-    latestFinalizedAt: string;
-    anyFailed: boolean;
-}
-/** Group trials into run sets: trials sharing a group_id form one set; a
- *  null group_id is its own single-run set. */
-export declare function groupRunSets(trials: TrialVerdict[]): RunSet[];
-/** The newest run set with at least one failed (completed) trial. */
-export declare function latestFailedRunSet(sets: RunSet[]): RunSet | null;
-export interface RunSetDiscovery {
-    kind: "trial-dir" | "root";
-    /** The set to build a fix prompt from (per `kind` semantics: a trial dir
-     *  targets ITS set regardless of outcome; a root targets the latest
-     *  FAILED set). Null when nothing matches. */
-    set: RunSet | null;
-    /** Total finalized run sets seen — lets the caller distinguish "no runs
-     *  at all" from "runs exist but none failed". */
-    totalSets: number;
-}
-/**
- * Resolve what `pome fix-prompt [target]` should read.
- * - `target` = a trial run dir (has verdict.json): that trial's whole set —
- *   the user pointed at it, outcome doesn't matter.
- * - `target` = an artifacts root: the latest failed set under it.
- */
-export declare function discoverRunSet(target: string): Promise<RunSetDiscovery>;
-/** Load a trial's captured events.jsonl (raw trace) for prompt assembly.
- *  Missing/corrupt lines are skipped — the prompt degrades, never throws. */
-export declare function loadTrialEvents(runDir: string): Promise<unknown[]>;

package/dist/src/hosted/evalResultCache.js

@@ -1,219 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// FDRS-644 — the per-trial CLOUD verdict artifact.
-//
-// Hosted `pome run` persists each finalized trial's cloud verdict payload
-// (what /finalize returned and the terminal rendered) as `verdict.json`
-// next to the trial's raw trace. This is a provenance-labeled CACHE of the
-// cloud judge's output — NOT a local score: the OSS CLI still has no
-// scoring engine (FDRS-657 / no-eval-in-oss), score.json is still never
-// written, and deleting verdict.json loses nothing the dashboard doesn't
-// hold. `pome fix-prompt` reads these to hand grouped failure signatures
-// to the user's coding agent without a credentialed cloud round-trip —
-// the F0-3 / L5 intent recorded in uploadAndFinalize.ts ("so the CLI can
-// render per-criterion verdicts … without a follow-up cloud round-trip").
-//
-// Artifact-layout knowledge lives HERE and only here:
-//   <artifacts-root>/<task-slug>/<session-id>/verdict.json
-// (runDir shape from recorder/artifacts.ts `writeRunArtifactsCore`).
-//
-// F-689/D16 — moved from `src/recorder/verdictArtifact.ts` to here (and
-// renamed: the repo-wide no-eval gate denies any module NAMED verdict*, so
-// this file itself can't be called that anymore) and grouped under
-// `hosted/` since it caches a CLOUD response, not a recorder concern.
-import { existsSync } from "node:fs";
-import { readdir, readFile, writeFile } from "node:fs/promises";
-import { join } from "node:path";
-export const VERDICT_ARTIFACT_VERSION = 1;
-export const VERDICT_FILENAME = "verdict.json";
-export async function writeVerdictArtifact(runDir, verdict) {
-    await writeFile(join(runDir, VERDICT_FILENAME), `${JSON.stringify(verdict, null, 2)}\n`, "utf8");
-}
-/** Read one trial's verdict.json. Returns null when absent or when the file
- *  isn't a recognizable verdict artifact (foreign/corrupt files are skipped,
- *  never thrown on — fix-prompt discovery must survive a messy runs/). */
-/** Every field the fix-prompt pipeline dereferences must hold its declared
- *  shape, or the FILE is rejected — discovery treats a half-recognizable
- *  verdict.json as foreign rather than crashing downstream on it. */
-function isVerdictArtifact(parsed) {
-    if (typeof parsed !== "object" || parsed === null)
-        return false;
-    const v = parsed;
-    if (v.source !== "cloud-finalize")
-        return false;
-    if (typeof v.session_id !== "string")
-        return false;
-    if (typeof v.task_name !== "string")
-        return false;
-    if (typeof v.task_path !== "string" && typeof v.scenario_path !== "string")
-        return false;
-    if (v.group_id !== null && typeof v.group_id !== "string")
-        return false;
-    if (typeof v.finalized_at !== "string")
-        return false;
-    if (typeof v.passed !== "boolean")
-        return false;
-    if (typeof v.score !== "number")
-        return false;
-    if (!Array.isArray(v.criteria_results))
-        return false;
-    return v.criteria_results.every((r) => {
-        if (typeof r !== "object" || r === null)
-            return false;
-        const result = r;
-        const criterion = result.criterion;
-        return (typeof criterion === "object" &&
-            criterion !== null &&
-            typeof criterion.text === "string" &&
-            typeof result.reason === "string" &&
-            typeof result.passed === "boolean" &&
-            typeof result.skipped === "boolean");
-    });
-}
-/** F-933 — collapse the legacy `scenario_path` onto `task_path` so callers
- *  only ever deal with one spelling. Validation ran first, so at least one of
- *  the two is a string. */
-function normalizeVerdictArtifact(parsed) {
-    const { scenario_path: legacyPath, task_path: taskPath, ...rest } = parsed;
-    return { ...rest, task_path: taskPath ?? legacyPath };
-}
-export async function readVerdictArtifact(runDir) {
-    const path = join(runDir, VERDICT_FILENAME);
-    let raw;
-    try {
-        raw = await readFile(path, "utf8");
-    }
-    catch {
-        return null;
-    }
-    try {
-        const parsed = JSON.parse(raw);
-        if (!isVerdictArtifact(parsed))
-            return null;
-        return { runDir, verdict: normalizeVerdictArtifact(parsed) };
-    }
-    catch {
-        return null;
-    }
-}
-/** Scan an artifacts root for finalized trials — exactly the two-level
- *  layout `<root>/<task-slug>/<session-id>/verdict.json`. Anything
- *  unreadable is skipped. */
-export async function scanVerdictArtifacts(artifactsRoot) {
-    const found = [];
-    let slugs;
-    try {
-        slugs = await readdir(artifactsRoot);
-    }
-    catch {
-        return found;
-    }
-    for (const slug of slugs) {
-        const slugDir = join(artifactsRoot, slug);
-        let runIds;
-        try {
-            runIds = await readdir(slugDir);
-        }
-        catch {
-            continue;
-        }
-        for (const runId of runIds) {
-            const trial = await readVerdictArtifact(join(slugDir, runId));
-            if (trial)
-                found.push(trial);
-        }
-    }
-    return found;
-}
-/** Group trials into run sets: trials sharing a group_id form one set; a
- *  null group_id is its own single-run set. */
-export function groupRunSets(trials) {
-    const byKey = new Map();
-    for (const trial of trials) {
-        const key = trial.verdict.group_id ?? `solo:${trial.verdict.session_id}`;
-        const bucket = byKey.get(key);
-        if (bucket)
-            bucket.push(trial);
-        else
-            byKey.set(key, [trial]);
-    }
-    const sets = [];
-    for (const bucket of byKey.values()) {
-        bucket.sort((a, b) => a.verdict.finalized_at.localeCompare(b.verdict.finalized_at));
-        const last = bucket[bucket.length - 1];
-        sets.push({
-            groupId: bucket[0].verdict.group_id,
-            taskName: bucket[0].verdict.task_name,
-            taskPath: bucket[0].verdict.task_path,
-            trials: bucket,
-            latestFinalizedAt: last.verdict.finalized_at,
-            anyFailed: bucket.some((t) => !t.verdict.passed),
-        });
-    }
-    sets.sort((a, b) => a.latestFinalizedAt.localeCompare(b.latestFinalizedAt));
-    return sets;
-}
-/** The newest run set with at least one failed (completed) trial. */
-export function latestFailedRunSet(sets) {
-    for (let i = sets.length - 1; i >= 0; i -= 1) {
-        if (sets[i].anyFailed)
-            return sets[i];
-    }
-    return null;
-}
-/**
- * Resolve what `pome fix-prompt [target]` should read.
- * - `target` = a trial run dir (has verdict.json): that trial's whole set —
- *   the user pointed at it, outcome doesn't matter.
- * - `target` = an artifacts root: the latest failed set under it.
- */
-export async function discoverRunSet(target) {
-    const anchor = await readVerdictArtifact(target);
-    if (anchor) {
-        // Trial dir → its set. Layout is <root>/<slug>/<runId>, so the root is
-        // two levels up; a moved/isolated dir degrades to a set of one.
-        const root = join(target, "..", "..");
-        const sets = groupRunSets(await scanVerdictArtifacts(root));
-        const own = sets.find((s) => (anchor.verdict.group_id !== null &&
-            s.groupId === anchor.verdict.group_id) ||
-            (anchor.verdict.group_id === null &&
-                s.trials.length === 1 &&
-                s.trials[0].verdict.session_id === anchor.verdict.session_id)) ?? groupRunSets([anchor])[0];
-        return { kind: "trial-dir", set: own, totalSets: Math.max(sets.length, 1) };
-    }
-    if (!existsSync(target))
-        return { kind: "root", set: null, totalSets: 0 };
-    const sets = groupRunSets(await scanVerdictArtifacts(target));
-    return {
-        kind: "root",
-        set: latestFailedRunSet(sets),
-        totalSets: sets.length,
-    };
-}
-/** Load a trial's captured events.jsonl (raw trace) for prompt assembly.
- *  Missing/corrupt lines are skipped — the prompt degrades, never throws. */
-export async function loadTrialEvents(runDir) {
-    let raw;
-    try {
-        raw = await readFile(join(runDir, "events.jsonl"), "utf8");
-    }
-    catch {
-        return [];
-    }
-    const events = [];
-    for (const line of raw.split("\n")) {
-        const trimmed = line.trim();
-        if (!trimmed)
-            continue;
-        try {
-            const parsed = JSON.parse(trimmed);
-            // Valid JSON but not an event object (`null`, `3`, `"x"`) is corrupt
-            // for our purposes — renderEvent dereferences fields on it.
-            if (typeof parsed === "object" && parsed !== null)
-                events.push(parsed);
-        }
-        catch {
-            // skip corrupt row
-        }
-    }
-    return events;
-}

package/dist/src/hosted/evalResultView.d.ts

@@ -1,40 +0,0 @@
-import type { z } from "zod";
-import type { criterionSchema } from "../types/shared.js";
-type WireCriterion = z.infer<typeof criterionSchema>;
-export type CriterionOutcome = "passed" | "failed" | "skipped" | "errored";
-export type CriterionResult = {
-    criterion: WireCriterion;
-    outcome?: CriterionOutcome;
-    passed: boolean;
-    skipped: boolean;
-    reason: string;
-    confidence?: number;
-    judge_model?: string;
-    judge_tokens_in?: number;
-    judge_tokens_out?: number;
-    judge_has_usage?: boolean;
-};
-export type Score = {
-    satisfaction: number;
-    passed: number;
-    failed: number;
-    skipped: number;
-    errored: number;
-    total_required: number;
-    evaluated: boolean;
-    can_pass: boolean;
-    results: CriterionResult[];
-    judge_model: string | null;
-    judge_tokens_in: number | null;
-    judge_tokens_out: number | null;
-};
-export declare function outcomeOf(result: CriterionResult): CriterionOutcome;
-export type ScoreStatus = "pass" | "fail" | "incomplete";
-export declare function scoreStatus(score: Score, passThreshold: number): ScoreStatus;
-export declare function taskPassed(score: Score, passThreshold: number): boolean;
-export declare function markerFor(outcome: CriterionOutcome): string;
-export declare function criterionMarkerLabel(criterion: WireCriterion): string;
-export declare function twinSkipSuffix(result: CriterionResult): string;
-export declare function scoreCountsSummary(score: Score): string;
-export declare function runScoreLine(score: Score, passThreshold: number, unevaluatedNumericLabel: string): string;
-export {};

package/dist/src/hosted/evalResultView.js

@@ -1,103 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-//
-// Cloud-verdict DISPLAY model + pure label/render helpers (FDRS-656/657).
-//
-// The OSS CLI is CAPTURE-ONLY: it never computes a score, never runs a judge,
-// and never correlates locally. Every helper here operates on a score object
-// that ORIGINATES FROM THE CLOUD (a hosted `pome run` /finalize response, or a
-// `pome eval` upload verdict — see `scoreFromFinalizeResponse`). Nothing in
-// this module derives a verdict from twin state, events, or a judge call.
-//
-// This is the relocation of the former `src/evaluator/score.ts` + the
-// verdict-rendering half of `src/cli/render.ts`. The local scoring engine
-// (`scoreResults`, the deterministic matchers, the BYOK LLM judge) was deleted
-// under FDRS-657; only the pure display model survives, moved out of the
-// `evaluator/` tree so the `no-eval-in-oss` gate can assert that tree is gone.
-//
-// F-689/D16 — moved AGAIN from `src/score/view.ts` to here. `score/` (a
-// module-name stem the repo-wide gate now denies outright) has to cease to
-// exist, so this pure display model lives under `hosted/` with the rest of
-// the cloud-facing surface it renders.
-// Derive the four-state outcome from a cloud result. Prefers the explicit
-// `outcome` field; falls back to the legacy passed/skipped booleans. PURE.
-export function outcomeOf(result) {
-    if (result.outcome)
-        return result.outcome;
-    if (result.skipped)
-        return "skipped";
-    return result.passed ? "passed" : "failed";
-}
-// Single source of truth for "did this run pass?", applied to a CLOUD score.
-// Encodes the A5 guard: a run is only a PASS when it was evaluated, every
-// required criterion was evaluated (can_pass), AND satisfaction cleared the
-// threshold. PURE — no computation of the score itself.
-//
-// F-932 renamed the third state from `unevaluated` to `incomplete` and CHANGED
-// NOTHING ELSE HERE. The guard is the one place the CLI refuses to inflate a
-// partial run into a pass — the same refusal pome-cloud added server-side in
-// F-925 — so the rename must not become a loosening.
-//
-// One rule, two repos: `can_pass` is false for ANY abstention
-// (`uploadAndFinalize.ts`), and pome-cloud's `isRunIncomplete` says
-// `notEvaluated > 0` over the same `criteria_results`. Deliberately NOT read
-// from the wire's `all_skipped`, which is the narrower every-abstained
-// predicate and would loosen this guard.
-export function scoreStatus(score, passThreshold) {
-    if (!score.evaluated || !score.can_pass)
-        return "incomplete";
-    return score.satisfaction >= passThreshold ? "pass" : "fail";
-}
-export function taskPassed(score, passThreshold) {
-    return scoreStatus(score, passThreshold) === "pass";
-}
-// FDRS-591/611 per-criterion marker: ✓ passed, ✗ failed, - skipped, ! errored.
-export function markerFor(outcome) {
-    switch (outcome) {
-        case "passed":
-            return "✓";
-        case "failed":
-            return "✗";
-        case "errored":
-            return "!";
-        default:
-            return "-";
-    }
-}
-// Multi-twin (M3): the per-criterion bracket for terminal display —
-// `[code]` / `[model]`, plus the `:<twin>` suffix when the criterion attributes
-// to a specific twin (so a `[code:slack]`/`[model:github]` marker survives into the
-// INCOMPLETE / criteria list). A bare (primary-twin) criterion renders `[code]`
-// unchanged.
-export function criterionMarkerLabel(criterion) {
-    return criterion.twin ? `[${criterion.type}:${criterion.twin}]` : `[${criterion.type}]`;
-}
-// Multi-twin (M3): when the cloud could not evaluate a criterion for a
-// twin-related reason (a twin-tagged criterion, or a `no_matching_predicate`
-// skip), name the twin inline so the INCOMPLETE line explains WHICH twin's timeline
-// came up empty. Returns "" when there's nothing twin-specific to add.
-export function twinSkipSuffix(result) {
-    const twin = result.criterion.twin;
-    if (!twin)
-        return "";
-    const outcome = outcomeOf(result);
-    const twinRelated = outcome === "skipped" ||
-        outcome === "errored" ||
-        /no_matching_predicate|no matching predicate/i.test(result.reason);
-    return twinRelated ? ` (twin: ${twin})` : "";
-}
-export function scoreCountsSummary(score) {
-    return `${score.passed ?? 0} passed, ${score.failed ?? 0} failed, ${score.skipped ?? 0} skipped, ${score.errored ?? 0} errored`;
-}
-export function runScoreLine(score, passThreshold, unevaluatedNumericLabel) {
-    const status = scoreStatus(score, passThreshold);
-    if (status === "incomplete") {
-        // Leads with the COUNT, which is the fact the reader needs and the same
-        // fact the cloud's own header now states. The old copy said "cannot pass",
-        // which is a verdict about the AGENT for a gap in the GRADER — the exact
-        // inversion F-925 exists to stop, one surface over.
-        const notEvaluated = score.skipped + score.errored;
-        const total = score.total_required + notEvaluated;
-        return `score: incomplete — ${notEvaluated} of ${total} criteria not evaluated; ${scoreCountsSummary(score)}; ${unevaluatedNumericLabel}: ${score.satisfaction}/100`;
-    }
-    return `score: ${score.satisfaction}/100`;
-}