@shrkcrft/cli 0.1.0-alpha.29 → 0.1.0-alpha.31

package/dist/commands/baseline.command.js

@@ -17,35 +17,47 @@
 import { spawnSync } from 'node:child_process';
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
 import * as nodePath from 'node:path';
-import { failsWhenEmpty } from '@shrkcrft/core';
-import { baselineCount, baselineFails, computeBaselineFromExtractor, diffBaseline, matchesAny, } from '@shrkcrft/boundaries';
+import { coverageShortfall, failsWhenEmpty, formatEmptyRuleAdvice, normalizeRuleList, normalizeWiringSource, resolveSourceGlobs, ruleAssertsEmptyOutput, settleRuleEmptiness, unitStateLists, UnitLivenessState, } from '@shrkcrft/core';
+import { settleVerdict } from "../gates/settle-verdict.js";
+import { baselineCount, baselineFails, ceilingValue, clearFileReadCache, computeBaselineFromExtractor, diffBaseline, evaluateCeiling, globListSelects, planeScanExcludeDirs, readGlobListUnits, readScopeCoverage, readScopeHasUnread, settleGlobLists, sourceLivenessRequest, } from '@shrkcrft/boundaries';
 import { resolveChangedFiles, resolveProjectConfig } from '@shrkcrft/inspector';
 import { flagBool, flagString, resolveCwd, } from "../command-registry.js";
+import { PositionalMode } from "../dispatch/positional-mode.js";
 import { ExitCode } from "../exit-codes.js";
 import { asJson, header, kv } from "../output/format-output.js";
 import { buildGateEnvelope } from "../gates/gate-envelope.js";
+import { verdictLine } from "../gates/verdict-line.js";
+import { acceptedEmptyNote } from "../gates/accepted-empty-note.js";
+import { emptyRuleAdviceLines } from "../gates/empty-rule-advice-lines.js";
+import { qualifyCleanForUnits } from "../gates/qualify-clean-for-units.js";
+import { unitStateNotes } from "../gates/unit-state-notes.js";
+import { planeVerdictForExit } from "../gates/plane-verdict.js";
+import { seamRejectedRules } from "../gates/seam-rejected-rules.js";
 const SCHEMA = 'sharkcraft.baseline/v1';
 const DEFAULT_TIMEOUT_MS = 60_000;
 async function loadBaselines(cwd) {
     const loaded = await resolveProjectConfig(cwd);
     if (!loaded.ok)
         return { ok: false, message: loaded.error.message };
-    const rel = nodePath.relative(cwd, loaded.value.sharkcraftDir).split(nodePath.sep).join('/');
     return {
         ok: true,
         value: {
             rules: loaded.value.config.baselines ?? [],
             planeDiagnostics: loaded.value.planeDiagnostics,
-            sharkcraftDirRel: rel && !rel.startsWith('..') ? rel : '',
+            rejected: seamRejectedRules(loaded.value, ['baseline']),
+            // THE plane scan scope (one entry, or none).
+            sharkcraftDirRel: planeScanExcludeDirs(cwd, loaded.value.sharkcraftDir)[0] ?? '',
         },
     };
 }
 /** Narrow to `--id`, refusing an unknown id rather than silently selecting nothing. */
-function selectRules(rules, id) {
+function selectRules(rules, id, 
+/** Ids that are declared but did not load (merge-seam rejections) — selectable by a verdict verb. */
+extraKnown = []) {
     if (!id)
         return { ok: true, rules };
     const wanted = id.split(',').map((s) => s.trim()).filter(Boolean);
-    const known = new Set(rules.map((r) => r.id));
+    const known = new Set([...rules.map((r) => r.id), ...extraKnown]);
     const unknown = wanted.filter((w) => !known.has(w));
     if (unknown.length > 0) {
         return {
@@ -59,15 +71,16 @@ function computeCurrent(cwd, rule, excludeDirs) {
     if (rule.compute.kind === 'extractor') {
         const source = rule.compute.source;
         if (!source)
-            return { text: '', error: 'compute.kind "extractor" but no `source` declared' };
+            return { text: '', error: 'compute.kind "extractor" but no `source` declared', unitCount: 0 };
         const res = computeBaselineFromExtractor(cwd, source, excludeDirs);
+        const readScope = res.unread.length > 0 ? { readScope: { read: res.filesScanned, unread: res.unread } } : {};
         return res.error
-            ? { text: '', error: res.error, filesScanned: res.filesScanned }
-            : { text: res.text, filesScanned: res.filesScanned };
+            ? { text: '', error: res.error, filesScanned: res.filesScanned, unitCount: 0 }
+            : { text: res.text, filesScanned: res.filesScanned, unitCount: res.ids.length, ...readScope };
     }
     const run = rule.compute.run;
     if (!run || run.trim() === '') {
-        return { text: '', error: 'compute.kind "command" but no `run` declared' };
+        return { text: '', error: 'compute.kind "command" but no `run` declared', unitCount: 0 };
     }
     const child = spawnSync(run, {
         cwd,
@@ -76,18 +89,135 @@ function computeCurrent(cwd, rule, excludeDirs) {
         timeout: rule.compute.timeoutMs ?? DEFAULT_TIMEOUT_MS,
         maxBuffer: 16 * 1024 * 1024,
     });
-    if (child.error)
-        return { text: '', error: `compute command failed to start: ${child.error.message}` };
+    // The command may rewrite files anywhere: drop every memoised tree read, so
+    // a read-cache window (`withFileReadCache`, e.g. `shrk quality`'s plane
+    // section) never serves a pre-spawn snapshot after it.
+    clearFileReadCache();
+    if (child.error) {
+        return { text: '', error: `compute command failed to start: ${child.error.message}`, unitCount: 0 };
+    }
     if (child.status !== 0) {
         const tail = String(child.stderr ?? '').trim().split('\n').slice(-3).join(' | ');
         return {
             text: '',
             error: `compute command exited ${child.status ?? 'null'}${tail ? ` — ${tail}` : ''}`,
+            unitCount: 0,
         };
     }
-    return { text: String(child.stdout ?? '') };
+    const text = String(child.stdout ?? '');
+    return { text, unitCount: text.trim() === '' ? 0 : 1 };
 }
-function evaluateRule(cwd, rule, excludeDirs, changedFiles) {
+/** A ledger whose recompute and committed side are both empty, in its words. */
+const LEDGER_EMPTY = 'the recompute produced no entries (and the committed baseline is empty too)';
+/** A ceiling over a measurement that measured nothing, in its words. */
+const CEILING_EMPTY = 'the compute produced nothing — a ceiling over an empty measurement proves nothing';
+/** A bless (`baseline update`) whose recompute measured nothing, in its words. */
+const UPDATE_EMPTY = 'the recompute produced no entries';
+/**
+ * THE glob lists of a baseline's INPUTS (round 13): an extractor compute's
+ * `source` (`compute.source.files`, and `to.files` for an import-edges fence)
+ * and the rule's `watchFiles` — with their `expectEmpty` markers. `baseline
+ * check` settles it (`settleGlobLists`) and `gates coverage` builds the same
+ * request, so the two read one baseline's units alike. `primaryLists` names
+ * the lists whose liveness decides the rule's emptiness — a command compute
+ * has none (its `watchFiles` is a footprint, not an input it reads).
+ */
+export function baselineLivenessRequest(cwd, rule, excludeDirs) {
+    const source = rule.compute.kind === 'extractor' ? rule.compute.source : undefined;
+    const sources = sourceLivenessRequest(cwd, source !== undefined ? [{ label: 'compute.source', source }] : [], excludeDirs, rule.id);
+    const watch = rule.watchFiles ?? [];
+    return {
+        request: {
+            subject: rule.id,
+            lists: [
+                ...sources.lists,
+                ...(watch.length > 0
+                    ? [
+                        {
+                            list: 'watchFiles',
+                            ...(sources.lists.length > 0 ? { label: 'watchFiles' } : {}),
+                            globs: watch,
+                            units: readGlobListUnits(cwd, watch, new Set(excludeDirs)),
+                        },
+                    ]
+                    : []),
+            ],
+            marks: [...sources.marks, ...(rule.expectEmptyUnits ?? [])],
+        },
+        primaryLists: source !== undefined ? ['compute.source.files'] : [],
+    };
+}
+/** The unit fields every outcome carries when the rule's inputs were settled. */
+function unitFieldsOf(liveness) {
+    if (liveness === undefined)
+        return {};
+    const nonLive = liveness.units.filter((u) => u.state !== UnitLivenessState.Live);
+    return {
+        ...(liveness.acceptance !== undefined ? { unitAcceptance: liveness.acceptance } : {}),
+        ...(liveness.dead.length + liveness.intendedEmpty.length + liveness.wentLive.length > 0
+            ? { units: unitStateLists(liveness) }
+            : {}),
+        ...(nonLive.length > 0 ? { unitLiveness: nonLive } : {}),
+    };
+}
+/**
+ * THE rule-emptiness settle of a baseline whose compute measured nothing
+ * (round 13): the ledger's empty recompute, the ceiling's empty measurement
+ * (P1: decided on the extractor's UNIT count, never its text — an extractor
+ * serialises an empty set as `[]`, so the old text test could never fire for
+ * one) and `baseline update`'s refusal all ask it. A fence (`expectEmpty:
+ * true`) is accepted only over live or intended-empty inputs; over a dead
+ * input it is Stale (it proves nothing).
+ */
+function settleBaselineEmptiness(rule, computed, liveness, primaryLists, reason) {
+    return settleRuleEmptiness({
+        subject: rule.id,
+        unitLabel: 'entries',
+        filesMatched: computed.filesScanned ?? 0,
+        unitsMatched: 0,
+        unread: false,
+        liveness,
+        primaryLists,
+        assertsEmptyOutput: ruleAssertsEmptyOutput(rule),
+        failOnEmpty: failsWhenEmpty(rule),
+        noFilesReason: reason,
+        noUnitsReason: reason,
+    });
+}
+/**
+ * The engine entry's idempotent normalisation (round 13): the extractor
+ * source's markable lists and `watchFiles` into plain lists plus their
+ * `expectEmptyUnits`. A malformed marker is an errored rule, never a crash.
+ */
+function normalizeBaselineRule(rule) {
+    const source = rule.compute.source !== undefined ? normalizeWiringSource(rule.compute.source) : undefined;
+    if (source !== undefined && !source.ok)
+        return { ok: false, error: `compute.source ${source.error.message}` };
+    const withSource = source !== undefined ? { ...rule, compute: { ...rule.compute, source: source.value } } : rule;
+    const watch = normalizeRuleList(withSource, 'watchFiles');
+    if (!watch.ok)
+        return { ok: false, error: `watchFiles ${watch.error.message}` };
+    return { ok: true, rule: watch.value };
+}
+export function evaluateBaselineRule(cwd, authored, excludeDirs, changedFiles) {
+    const normalized = normalizeBaselineRule(authored);
+    if (!normalized.ok) {
+        return { rule: authored, status: 'error', committedCount: 0, currentCount: 0, error: normalized.error };
+    }
+    const rule = normalized.rule;
+    // The rule's input globs, settled with their markers (round 13) — only when
+    // a unit is marked (its acceptance / went-live line rides on the outcome) or
+    // the compute measured nothing (what the empty rule is decided from).
+    const marked = (rule.expectEmptyUnits?.length ?? 0) > 0 || (rule.compute.source?.expectEmptyUnits?.length ?? 0) > 0;
+    let input;
+    const inputOf = () => {
+        if (input === undefined) {
+            const built = baselineLivenessRequest(cwd, rule, excludeDirs);
+            input = { liveness: settleGlobLists(built.request), primaryLists: built.primaryLists };
+        }
+        return input;
+    };
+    const markedFields = () => marked ? unitFieldsOf(inputOf().liveness) : {};
     // --changed-only is honest about what it CANNOT scope: a command compute with
     // no `watchFiles` has no file footprint, so it is reported as skipped rather
     // than quietly passing.
@@ -95,7 +225,7 @@ function evaluateRule(cwd, rule, excludeDirs, changedFiles) {
         const globs = rule.watchFiles && rule.watchFiles.length > 0
             ? rule.watchFiles
             : rule.compute.kind === 'extractor'
-                ? (rule.compute.source?.files ?? [])
+                ? (resolveSourceGlobs(rule.compute.source ?? { files: [] }))
                 : undefined;
         if (globs === undefined) {
             return {
@@ -106,7 +236,9 @@ function evaluateRule(cwd, rule, excludeDirs, changedFiles) {
                 skipReason: 'command compute with no `watchFiles` cannot be scoped to a changeset',
             };
         }
-        if (!changedFiles.some((f) => matchesAny(f, globs))) {
+        // One list (watchFiles, else the extractor's files): a changed file the
+        // list EXCLUDES cannot move the value, so it does not select the rule.
+        if (!changedFiles.some((f) => globListSelects(f, globs))) {
             return {
                 rule,
                 status: 'skipped',
@@ -116,6 +248,64 @@ function evaluateRule(cwd, rule, excludeDirs, changedFiles) {
             };
         }
     }
+    // A ceiling rule pins a NUMBER declared in the config, so it has no committed
+    // artifact to read and no entry diff to compute — the comparison is the whole
+    // check. Branching here (rather than in a parallel engine) keeps `baseline
+    // check`, `gates check` and `quality` running the identical evaluation.
+    if (rule.mode === 'ceiling') {
+        const computed = computeCurrent(cwd, rule, excludeDirs);
+        if (computed.error) {
+            return { rule, status: 'error', committedCount: 0, currentCount: 0, error: computed.error, ...markedFields() };
+        }
+        const value = ceilingValue(rule, computed.text);
+        // A compute that MEASURED nothing measured nothing. Reporting `0 ≤ 200` as
+        // a pass is how a broken extractor greens a ratchet forever. Decided on the
+        // UNIT count (round 13, P1 — an extractor serialises its empty set as `[]`,
+        // so the old `text.trim() === ''` test could never fire for one) through
+        // THE rule-emptiness settle, the one `gates coverage` reads too.
+        if (computed.unitCount === 0 && !readScopeHasUnread(computed.readScope)) {
+            const { liveness, primaryLists } = inputOf();
+            const emptiness = settleBaselineEmptiness(rule, computed, liveness, primaryLists, CEILING_EMPTY);
+            if (emptiness.skipped) {
+                return {
+                    rule,
+                    status: emptiness.fails ? 'failed' : 'skipped',
+                    current: computed.text,
+                    committedCount: rule.ceiling ?? 0,
+                    currentCount: 0,
+                    skipReason: emptiness.skipReason ?? CEILING_EMPTY,
+                    ...unitFieldsOf(liveness),
+                };
+            }
+            // Accepted: the rule asserts an empty output (`expectEmpty: true` is
+            // honoured on a ceiling) or every input glob is intended-empty. The
+            // ceiling is still judged — on the measured 0.
+            const judged = evaluateCeiling(rule, value);
+            return {
+                rule,
+                status: judged.failed ? 'failed' : 'passed',
+                current: computed.text,
+                committedCount: judged.ceiling,
+                currentCount: judged.value,
+                ceiling: judged,
+                ...(!judged.failed && emptiness.coverage !== undefined ? { emptyCoverage: emptiness.coverage } : {}),
+                ...unitFieldsOf(liveness),
+            };
+        }
+        const verdict = evaluateCeiling(rule, value);
+        return {
+            rule,
+            status: verdict.failed ? 'failed' : 'passed',
+            current: computed.text,
+            committedCount: verdict.ceiling,
+            currentCount: verdict.value,
+            ceiling: verdict,
+            // An unread file makes the measurement a lower bound: over the ceiling
+            // is still a real failure, under it is not verified.
+            ...(computed.readScope ? { readScope: computed.readScope } : {}),
+            ...markedFields(),
+        };
+    }
     const abs = nodePath.resolve(cwd, rule.baseline);
     if (!existsSync(abs)) {
         // No artifact yet. `check` must still fail (nothing to compare against),
@@ -157,15 +347,53 @@ function evaluateRule(cwd, rule, excludeDirs, changedFiles) {
     // to prevent. But this only holds when the COMMITTED side is empty too: if
     // the baseline has entries and the recompute has none, that is real drift
     // (everything vanished) and must be reported as such, not swallowed as a skip.
+    const readScope = computed.readScope ? { readScope: computed.readScope } : {};
     if (currentCount === 0 && committedCount === 0) {
+        // An empty recompute over a file the reader could not read is not "matched
+        // nothing": PARTIAL (the coverage names the file), never failOnEmpty's
+        // failure and never a pass.
+        if (readScopeHasUnread(computed.readScope)) {
+            return {
+                rule,
+                status: 'passed',
+                committed,
+                current: computed.text,
+                committedCount,
+                currentCount,
+                ...readScope,
+                ...markedFields(),
+            };
+        }
+        // THE rule-emptiness settle (round 13). A FENCE (`expectEmpty: true`)
+        // asserts the empty OUTPUT, so over live (or intended-empty) inputs its
+        // empty set is the verified pass — otherwise the rule could never be green.
+        // Over a DEAD input it proves nothing: Stale, and failOnEmpty's 1 (V2 f1d:
+        // it used to be accepted). An extractor whose every source glob is marked
+        // `expectEmpty` and matched no file is intended-empty, accepted. Anything
+        // else is the loud skip it always was.
+        const { liveness, primaryLists } = inputOf();
+        const emptiness = settleBaselineEmptiness(rule, computed, liveness, primaryLists, LEDGER_EMPTY);
+        if (!emptiness.skipped) {
+            return {
+                rule,
+                status: 'passed',
+                committed,
+                current: computed.text,
+                committedCount,
+                currentCount,
+                ...(emptiness.coverage !== undefined ? { emptyCoverage: emptiness.coverage } : {}),
+                ...unitFieldsOf(liveness),
+            };
+        }
         return {
             rule,
-            status: failsWhenEmpty(rule) ? 'failed' : 'skipped',
+            status: emptiness.fails ? 'failed' : 'skipped',
             committed,
             current: computed.text,
             committedCount,
             currentCount,
-            skipReason: 'the recompute produced no entries (and the committed baseline is empty too)',
+            skipReason: emptiness.skipReason ?? LEDGER_EMPTY,
+            ...unitFieldsOf(liveness),
         };
     }
     const diff = diffBaseline(rule, committed, computed.text);
@@ -180,8 +408,52 @@ function evaluateRule(cwd, rule, excludeDirs, changedFiles) {
         // A total wipe is far more often a broken compute than a real emptying —
         // say so next to the diff so it is not blessed by reflex.
         ...(currentCount === 0 ? { emptyCompute: true } : {}),
+        ...readScope,
+        ...markedFields(),
     };
 }
+/**
+ * What one baseline rule examined — shared by `baseline check` and the
+ * aggregate (`gates check`, `quality`), so both report the same scope.
+ *
+ * A ledger examines the union of its committed and recomputed entries; a
+ * ceiling examines one measurement. A fence (`expectEmpty`) over an empty set
+ * is the asserted state, so its empty scope is ACCEPTED by that field — and
+ * printed as accepted, never silent. A skipped or errored rule examined nothing.
+ *
+ * An extractor compute that matched a file the reader did not read folds it
+ * in through `readScopeCoverage`, the one rule every plane uses: the record
+ * becomes `examined N of M files`, naming the unread file.
+ */
+export function baselineCoverage(o) {
+    return readScopeCoverage(planeBaselineCoverage(o), o.readScope);
+}
+function planeBaselineCoverage(o) {
+    // The accepted empty (round 13): the fence's asserted empty set, or an
+    // intended-empty input — the rule-emptiness settle's own record, one wording
+    // on every surface (`the rule asserts an empty set` / `asserted empty — …`).
+    if (o.emptyCoverage !== undefined)
+        return o.emptyCoverage;
+    if (o.ceiling !== undefined)
+        return { unit: 'measurements', expected: 1, examined: 1 };
+    if (o.skipReason !== undefined) {
+        return { unit: 'entries', expected: 0, examined: 0, reason: o.skipReason };
+    }
+    if (o.status === 'error') {
+        return { unit: 'entries', expected: 0, examined: 0, reason: 'the baseline could not be evaluated' };
+    }
+    const size = Math.max(o.committedCount, o.currentCount);
+    return { unit: 'entries', expected: size, examined: size };
+}
+/**
+ * What this rule pins, for display. A ledger names its committed file; a
+ * ceiling names the number in the config, because that IS its committed value.
+ */
+function pinLabel(rule) {
+    return rule.mode === 'ceiling'
+        ? `ceiling ${rule.ceiling ?? 0} (sharkcraft.config.ts)`
+        : (rule.baseline ?? '(no artifact)');
+}
 function hintFor(rule) {
     return rule.hint ?? `review the diff, then bless it with \`shrk baseline update --id ${rule.id}\``;
 }
@@ -189,9 +461,11 @@ function outcomeJson(o) {
     return {
         id: o.rule.id,
         ...(o.rule.description ? { description: o.rule.description } : {}),
-        baseline: o.rule.baseline,
+        ...(o.rule.baseline ? { baseline: o.rule.baseline } : {}),
+        mode: o.rule.mode ?? 'ledger',
         severity: o.rule.severity ?? 'error',
-        direction: o.rule.direction ?? 'two-way',
+        direction: o.rule.direction ?? (o.rule.mode === 'ceiling' ? 'at-most' : 'two-way'),
+        ...(o.ceiling ? { ceiling: o.ceiling } : {}),
         status: o.status,
         committedCount: o.committedCount,
         currentCount: o.currentCount,
@@ -228,7 +502,13 @@ async function prepare(args, opts) {
             process.stderr.write(`Could not load config: ${loaded.message}\n  Run \`shrk doctor\` for details.\n`);
         return { ok: false, code: ExitCode.UsageError };
     }
-    const selected = selectRules(loaded.value.rules, flagString(args, 'id') ?? undefined);
+    const idFlag = flagString(args, 'id') ?? undefined;
+    // A verdict verb may select a rejected baseline by id (its errored row); a
+    // verb that acts on a rule (update / diff / explain) cannot — there is none.
+    const rejectedKnown = opts.rejectedKnown === true ? loaded.value.rejected.map((r) => r.id) : [];
+    const selected = selectRules(loaded.value.rules, idFlag, rejectedKnown);
+    const wantedIds = idFlag ? idFlag.split(',').map((s) => s.trim()).filter(Boolean) : undefined;
+    const rejected = wantedIds ? loaded.value.rejected.filter((r) => wantedIds.includes(r.id)) : loaded.value.rejected;
     if (!selected.ok) {
         process.stderr.write(selected.message + '\n');
         return { ok: false, code: ExitCode.UsageError };
@@ -252,19 +532,53 @@ async function prepare(args, opts) {
         excludeDirs: loaded.value.sharkcraftDirRel ? [loaded.value.sharkcraftDirRel] : [],
         ...(changedFiles !== undefined ? { changedFiles } : {}),
         planeDiagnostics: loaded.value.planeDiagnostics,
+        rejected,
     };
 }
-/** The "no baselines declared" landing, shared by every subverb. */
-function writeNoRules(json) {
+/** `  ✗ <id>  REJECTED — <why>` — a pack rule the merge seam refused (declared, never run). */
+function writeRejected(rejected) {
+    for (const r of rejected)
+        process.stdout.write(`  ✗ ${r.id}  REJECTED — ${r.error ?? 'failed validation'}\n`);
+}
+/**
+ * The "no baselines declared" landing, shared by every subverb. A VERDICT verb
+ * passes its name so its JSON still carries the settled `gate` envelope —
+ * nothing declared is `2`, never a pass.
+ */
+function writeNoRules(json, verb) {
+    // A VERDICT verb settles first and renders second, in text AND JSON: nothing
+    // declared proposes 0 and the run coverage (expected 0) settles it to 2. The
+    // exit comes from the envelope — never a hard-coded code — so text, JSON and
+    // gate.exit cannot disagree. List / explain subverbs stay informational.
+    const gate = verb !== undefined
+        ? buildGateEnvelope(verb, ExitCode.VerifiedPass, [], {
+            unit: 'baselines',
+            expected: 0,
+            examined: 0,
+            reason: 'no baselines[] declared',
+        })
+        : undefined;
+    const exit = gate?.exit ?? ExitCode.NotVerified;
     if (json) {
-        process.stdout.write(asJson({ schema: SCHEMA, results: [], evaluated: 0, verdict: 'not-verified' }) + '\n');
-        return ExitCode.NotVerified;
+        process.stdout.write(asJson({
+            schema: SCHEMA,
+            results: [],
+            evaluated: 0,
+            verdict: gate ? planeVerdictForExit(gate.exit) : 'not-verified',
+            ...(gate ? { exitCode: gate.exit, gate } : {}),
+        }) + '\n');
+        return exit;
     }
     process.stdout.write(header('Baselines'));
     process.stdout.write('  No baselines declared. Add `baselines[]` to sharkcraft.config.ts to replace a\n' +
         '  hand-rolled "committed file + recompute script + drift test" trio with one\n' +
         '  two-way engine (see docs/baseline-drift.md).\n');
-    return ExitCode.NotVerified;
+    if (gate) {
+        const line = verdictLine(gate, 'Nothing declared — accepted.');
+        if (line)
+            process.stdout.write(`\n${line}\n`);
+    }
+    return exit;
 }
 export const baselineListCommand = {
     name: 'list',
@@ -276,11 +590,13 @@ export const baselineListCommand = {
         if (!prep.ok)
             return prep.code;
         const json = flagBool(args, 'json');
-        if (prep.all.length === 0)
+        if (prep.all.length === 0 && prep.rejected.length === 0)
             return writeNoRules(json);
         if (json) {
             process.stdout.write(asJson({
                 schema: SCHEMA,
+                // Declared by a pack, refused by the merge seam: never computed.
+                rejected: prep.rejected.map((r) => ({ id: r.id, error: r.error ?? null })),
                 baselines: prep.all.map((r) => ({
                     id: r.id,
                     description: r.description ?? null,
@@ -288,7 +604,9 @@ export const baselineListCommand = {
                     compute: r.compute.kind,
                     direction: r.direction ?? 'two-way',
                     keyBy: r.keyBy ?? null,
-                    failOnEmpty: r.failOnEmpty === true,
+                    // The EFFECTIVE failOnEmpty (round 13) — THE authority, never the
+                    // raw field: an error rule fails on empty by default.
+                    failOnEmpty: failsWhenEmpty(r),
                 })),
                 diagnostics: prep.planeDiagnostics,
             }) + '\n');
@@ -299,10 +617,14 @@ export const baselineListCommand = {
             process.stdout.write(`  • ${r.id}  →  ${r.baseline}\n`);
             process.stdout.write(`      compute ${r.compute.kind}${r.compute.kind === 'command' ? ` (${r.compute.run})` : ''}` +
                 `  ·  direction ${r.direction ?? 'two-way'}${r.keyBy ? `  ·  keyBy ${r.keyBy}` : ''}` +
-                `${r.failOnEmpty ? '  ·  failOnEmpty' : ''}\n`);
+                `${failsWhenEmpty(r) ? '  ·  failOnEmpty' : ''}\n`);
             if (r.description)
                 process.stdout.write(`      ${r.description}\n`);
         }
+        if (prep.rejected.length > 0) {
+            process.stdout.write(`\n  rejected at the pack-plane merge seam — never computed (${prep.rejected.length}):\n`);
+            writeRejected(prep.rejected);
+        }
         for (const d of prep.planeDiagnostics)
             process.stdout.write(`  ! ${d}\n`);
         return ExitCode.VerifiedPass;
@@ -314,52 +636,87 @@ export const baselineCheckCommand = {
     usage: 'shrk baseline check [--id <ids>] [--changed-only] [--since <ref>] [--json]',
     booleanFlags: new Set(['json', 'changed-only']),
     async run(args) {
-        const prep = await prepare(args, { changedAware: true });
+        const prep = await prepare(args, { changedAware: true, rejectedKnown: true });
         if (!prep.ok)
             return prep.code;
         const json = flagBool(args, 'json');
-        if (prep.rules.length === 0)
-            return writeNoRules(json);
-        const outcomes = prep.rules.map((r) => evaluateRule(prep.cwd, r, prep.excludeDirs, prep.changedFiles));
+        // A pack baseline the merge seam rejected is a declared baseline that was
+        // never computed — an errored row, exit 1 (round 12 review, R12-X1).
+        const rejected = prep.rejected;
+        if (prep.rules.length === 0 && rejected.length === 0)
+            return writeNoRules(json, 'baseline check');
+        const outcomes = prep.rules.map((r) => evaluateBaselineRule(prep.cwd, r, prep.excludeDirs, prep.changedFiles));
         const failed = outcomes.filter((o) => o.status === 'failed' || (o.status === 'error' && (o.rule.severity ?? 'error') === 'error'));
         const evaluated = outcomes.filter((o) => o.status !== 'skipped').length;
         // 0 only when a NON-EMPTY scope was actually compared; 2 when nothing was.
         // A skipped baseline is "partially verified", never a green 0.
         const skippedCount = outcomes.filter((o) => o.status === 'skipped').length;
-        const exit = failed.length > 0
+        const proposed = failed.length > 0 || rejected.length > 0
             ? ExitCode.Failure
             : evaluated === 0 || skippedCount > 0
                 ? ExitCode.NotVerified
                 : ExitCode.VerifiedPass;
+        // Settle first, render second: one envelope for text AND JSON, and the ✓
+        // line is printed only from the settled verdict.
+        const unexamined = [
+            ...outcomes.filter((o) => o.status === 'skipped' || o.status === 'error').map((o) => o.rule.id),
+            ...rejected.map((r) => r.id),
+        ];
+        const env = buildGateEnvelope('baseline check', proposed, [...outcomes.map((o) => ({
+                id: o.rule.id,
+                type: 'baseline',
+                status: o.status,
+                severity: o.rule.severity ?? 'error',
+                counts: { committed: o.committedCount, current: o.currentCount },
+                violations: [
+                    ...(o.diff?.added ?? []).map((id) => ({ id, message: 'added', hint: hintFor(o.rule) })),
+                    ...(o.diff?.removed ?? []).map((id) => ({ id, message: 'removed', hint: hintFor(o.rule) })),
+                ],
+                ...(o.skipReason ? { skipReason: o.skipReason } : {}),
+                ...(o.error ? { error: o.error } : {}),
+                coverage: baselineCoverage(o),
+                // The rule's `expectEmpty` acceptance and unit lines (round 13), folded
+                // into the envelope's one settle.
+                ...(o.unitAcceptance !== undefined ? { unitAcceptance: o.unitAcceptance } : {}),
+                ...(o.units !== undefined ? { units: o.units } : {}),
+            })), ...rejected], {
+            unit: 'baselines',
+            expected: outcomes.length + rejected.length,
+            examined: outcomes.length + rejected.length - unexamined.length,
+            ...(unexamined.length > 0 ? { unexamined, reason: 'compared nothing or could not run' } : {}),
+        });
+        const exit = env.exit;
         if (json) {
             process.stdout.write(asJson({
                 schema: SCHEMA,
                 results: outcomes.map(outcomeJson),
+                // Pack baselines the merge seam refused — errored rows in `gate.rules`.
+                rejected: rejected.map((r) => ({ id: r.id, error: r.error ?? null })),
                 evaluated,
-                skipped: outcomes.filter((o) => o.status === 'skipped').length,
-                verdict: failed.length > 0 ? 'errors' : evaluated === 0 ? 'not-verified' : 'pass',
+                skipped: skippedCount,
+                verdict: exit === ExitCode.Failure ? 'errors' : exit === ExitCode.VerifiedPass ? 'pass' : 'not-verified',
                 diagnostics: prep.planeDiagnostics,
-                gate: buildGateEnvelope('baseline check', exit, outcomes.map((o) => ({
-                    id: o.rule.id,
-                    type: 'baseline',
-                    status: o.status,
-                    severity: o.rule.severity ?? 'error',
-                    counts: { committed: o.committedCount, current: o.currentCount },
-                    violations: [
-                        ...(o.diff?.added ?? []).map((id) => ({ id, message: 'added', hint: hintFor(o.rule) })),
-                        ...(o.diff?.removed ?? []).map((id) => ({ id, message: 'removed', hint: hintFor(o.rule) })),
-                    ],
-                    ...(o.skipReason ? { skipReason: o.skipReason } : {}),
-                    ...(o.error ? { error: o.error } : {}),
-                }))),
+                exitCode: exit,
+                gate: env,
             }) + '\n');
             return exit;
         }
         process.stdout.write(header('Baseline drift'));
-        process.stdout.write(kv('evaluated', `${evaluated} of ${prep.rules.length}`) + '\n');
+        // Round 13 (K6): the printed count is the envelope's (`gate.evaluated`), which
+        // never counts a rule accepted as intended-empty — it is named apart.
+        process.stdout.write(kv('evaluated', `${env.evaluated} of ${prep.rules.length + rejected.length}${acceptedEmptyNote(env.acceptedEmpty)}` +
+            (rejected.length > 0 ? ` (${rejected.length} rejected at the pack-plane merge seam — NOT evaluated)` : '')) + '\n');
+        writeRejected(rejected);
         for (const o of outcomes) {
             if (o.status === 'passed') {
-                process.stdout.write(`  ✓ ${o.rule.id}  (${o.currentCount} entries, no drift)\n`);
+                // Rendered from the SETTLED rule: a pass over part of its scope is
+                // `partial`, never a ✓ (the keystone emitter pattern, step 5).
+                const settledRule = env.rules.find((x) => x.id === o.rule.id);
+                process.stdout.write(settledRule?.status === 'partial'
+                    ? `  ~ ${o.rule.id}  PARTIAL — ${settledRule.shortfall ?? 'part of its scope was not examined'}\n`
+                    : o.ceiling
+                        ? `  ✓ ${o.rule.id}  (${o.ceiling.value} ${o.ceiling.direction} ${o.ceiling.ceiling} — ${o.ceiling.slack} to spare)\n`
+                        : `  ✓ ${o.rule.id}  (${o.currentCount} entries, no drift)\n`);
                 continue;
             }
             if (o.status === 'skipped') {
@@ -374,6 +731,24 @@ export const baselineCheckCommand = {
                 }
                 continue;
             }
+            // A rule that measured nothing and FAILS on it (failOnEmpty, or a fence
+            // over a dead input) has no diff — its reason is the only true line. It is
+            // never rendered as DRIFT with a bless hint (round 13, P1): blessing it
+            // would commit the empty measurement the loud skip refused.
+            if (o.status === 'failed' && o.skipReason !== undefined) {
+                process.stdout.write(`  ✗ ${o.rule.id}  FAILED — ${o.skipReason}\n`);
+                if (!ruleAssertsEmptyOutput(o.rule)) {
+                    process.stdout.write(`      → ${formatEmptyRuleAdvice({ fails: true })}\n`);
+                }
+                continue;
+            }
+            if (o.ceiling) {
+                const over = -o.ceiling.slack;
+                process.stdout.write(`  ✗ ${o.rule.id}  OVER CEILING — ${o.ceiling.value} is ${over} ` +
+                    `${o.ceiling.direction === 'at-most' ? 'above' : 'below'} the limit of ${o.ceiling.ceiling}\n` +
+                    `      lower the measurement, or raise the ceiling deliberately: \`shrk baseline update --id ${o.rule.id}\`\n`);
+                continue;
+            }
             const added = o.diff?.added.length ?? 0;
             const removed = o.diff?.removed.length ?? 0;
             process.stdout.write(`  ✗ ${o.rule.id}  DRIFT — ${added} added, ${removed} removed ` +
@@ -386,12 +761,29 @@ export const baselineCheckCommand = {
         }
         for (const d of prep.planeDiagnostics)
             process.stdout.write(`  ! ${d}\n`);
-        if (exit === ExitCode.NotVerified) {
-            process.stdout.write('\nNothing was compared — this is NOT a pass. Every selected baseline was skipped.\n');
-        }
-        else if (exit === ExitCode.VerifiedPass) {
-            process.stdout.write('\nEvery baseline matches its committed artifact. ✓\n');
+        // THE empty-rule advice (round 13) for a SOFT skip: a failing one is
+        // advised inline (above); a rule that matched nothing and does not fail on
+        // it was told nothing.
+        for (const a of emptyRuleAdviceLines(outcomes.filter((o) => o.status === 'skipped' && o.skipReason !== undefined).map(() => ({ fails: false })))) {
+            process.stdout.write(`  ${a}.\n`);
         }
+        // THE shared unit-state block (round 13, K2): a dead input glob — a dead
+        // import-edges `to.files` under an accepted fence included — and a LOCAL
+        // expectEmpty marker whose target appeared withhold the ✓ (exit
+        // unchanged); a pack marker is INFO.
+        const unitNotes = unitStateNotes(outcomes.map((o) => ({
+            id: o.rule.id,
+            ...(o.unitLiveness !== undefined ? { unitLiveness: o.unitLiveness } : {}),
+            reportedEmpty: o.status === 'skipped' || o.skipReason !== undefined,
+        })));
+        process.stdout.write(unitNotes.text);
+        const line = verdictLine(env, qualifyCleanForUnits(outcomes.some((o) => o.ceiling)
+            ? 'Every baseline is within its pinned value. ✓'
+            : 'Every baseline matches its committed artifact. ✓', unitNotes), proposed === ExitCode.NotVerified && evaluated === 0
+            ? 'Nothing was compared — this is NOT a pass. Every selected baseline was skipped.'
+            : undefined);
+        if (line)
+            process.stdout.write(`\n${line}\n`);
         return exit;
     },
 };
@@ -407,7 +799,7 @@ export const baselineDiffCommand = {
         const json = flagBool(args, 'json');
         if (prep.rules.length === 0)
             return writeNoRules(json);
-        const outcomes = prep.rules.map((r) => evaluateRule(prep.cwd, r, prep.excludeDirs, undefined));
+        const outcomes = prep.rules.map((r) => evaluateBaselineRule(prep.cwd, r, prep.excludeDirs, undefined));
         if (json) {
             process.stdout.write(asJson({ schema: SCHEMA, results: outcomes.map(outcomeJson), inspection: true }) + '\n');
             return ExitCode.VerifiedPass;
@@ -416,7 +808,13 @@ export const baselineDiffCommand = {
         for (const o of outcomes) {
             const added = o.diff?.added.length ?? 0;
             const removed = o.diff?.removed.length ?? 0;
-            process.stdout.write(`\n${o.rule.id}  (${o.rule.baseline})  ${o.status === 'error' ? `! ${o.error}` : `+${added} / -${removed}`}\n`);
+            process.stdout.write(`\n${o.rule.id}  (${pinLabel(o.rule)})  ` +
+                (o.status === 'error'
+                    ? `! ${o.error}`
+                    : o.ceiling
+                        ? `${o.ceiling.value} vs ${o.ceiling.direction} ${o.ceiling.ceiling} (slack ${o.ceiling.slack})`
+                        : `+${added} / -${removed}`) +
+                '\n');
             writeDiff(o, 100);
         }
         return ExitCode.VerifiedPass;
@@ -437,19 +835,101 @@ export const baselineUpdateCommand = {
             return writeNoRules(json);
         const written = [];
         const errors = [];
+        // A ceiling's pinned value lives in `sharkcraft.config.ts`, so blessing it
+        // is a config edit, not a file write. `update` prints the exact one-line
+        // change instead of touching the config — the same explicit, reviewable
+        // bless, without this command growing the ability to rewrite the rules it
+        // is enforcing.
+        const reblessed = [];
+        // A value recomputed from an incomplete read (a file over the read cap)
+        // misses whatever the unread file holds. Blessing it would commit a number
+        // the loud-skip rule forbids reporting, and the next `baseline check`
+        // would be NOT VERIFIED anyway. So the rule is refused and named, and the
+        // run settles NOT VERIFIED (2), never "wrote … (exit 0)".
+        const unverified = [];
+        const unverifiedCoverage = [];
+        // Round 13 (K9): ceilings whose measurement is an EMPTY one the settle does
+        // not accept — the loud skip `baseline check` reports, never a proposal.
+        const emptyCeilings = [];
         for (const rule of prep.rules) {
             const computed = computeCurrent(prep.cwd, rule, prep.excludeDirs);
             if (computed.error) {
                 errors.push({ id: rule.id, error: computed.error });
                 continue;
             }
-            if (computed.text.trim() === '' && failsWhenEmpty(rule)) {
-                errors.push({
+            if (readScopeHasUnread(computed.readScope)) {
+                const record = {
+                    ...readScopeCoverage({ unit: 'baseline entries', expected: 1, examined: 1 }, computed.readScope),
+                    subject: rule.id,
+                };
+                unverifiedCoverage.push(record);
+                unverified.push({
+                    id: rule.id,
+                    shortfall: coverageShortfall(record) ?? '',
+                    unread: (computed.readScope?.unread ?? []).map((u) => u.path),
+                });
+                continue;
+            }
+            if (rule.mode === 'ceiling') {
+                // K9: never propose `ceiling: <measured>` over an EMPTY measurement.
+                // THE rule-emptiness settle `baseline check` reads decides it (on the
+                // extractor's UNIT count): an accepted empty (a fence, an intended-
+                // empty input) is judged like any value; a stale one is the loud skip
+                // — 2 NOT VERIFIED, or 1 under failOnEmpty — never `ceiling: 0`.
+                if (computed.unitCount === 0) {
+                    const input = baselineLivenessRequest(prep.cwd, rule, prep.excludeDirs);
+                    const emptiness = settleBaselineEmptiness(rule, computed, settleGlobLists(input.request), input.primaryLists, CEILING_EMPTY);
+                    if (emptiness.skipped) {
+                        const reason = emptiness.skipReason ?? CEILING_EMPTY;
+                        if (emptiness.fails) {
+                            errors.push({
+                                id: rule.id,
+                                error: `refusing to propose a ceiling from an EMPTY measurement for a \`failOnEmpty\` rule — ${reason} — fix the compute first`,
+                            });
+                        }
+                        else {
+                            unverifiedCoverage.push({ unit: 'measurements', expected: 1, examined: 0, subject: rule.id, reason });
+                            emptyCeilings.push({ id: rule.id, reason });
+                        }
+                        continue;
+                    }
+                }
+                const value = ceilingValue(rule, computed.text);
+                reblessed.push({
                     id: rule.id,
-                    error: 'refusing to write an EMPTY baseline for a `failOnEmpty` rule — fix the compute first',
+                    from: rule.ceiling ?? 0,
+                    to: value,
+                    edit: `ceiling: ${value},   // was ${rule.ceiling ?? 0}`,
                 });
                 continue;
             }
+            // An EMPTY bless is decided by the compute's UNIT count and THE
+            // rule-emptiness settle `baseline check` reads (round 13, P1) — never by
+            // its text: an extractor serialises its empty set as `[]`, so the old
+            // `text.trim() === ''` test never refused one, and a failOnEmpty ledger
+            // over a dead selector was blessed empty. A fence's asserted-empty output
+            // over live inputs, or an intended-empty input, is blessable; a stale
+            // empty (a dead or unmarked selector, files that yielded nothing) of a
+            // failOnEmpty rule is refused, naming why.
+            if (computed.unitCount === 0) {
+                const input = baselineLivenessRequest(prep.cwd, rule, prep.excludeDirs);
+                const liveness = settleGlobLists(input.request);
+                // Name the dead input the settle found (its own `.dead` lines), so the
+                // refusal says WHICH selector to fix.
+                const deadInputs = liveness.dead.map((u) => `${u.label}: ${u.deadReason ?? 'matched 0 files'}`);
+                const reason = deadInputs.length > 0
+                    ? `${UPDATE_EMPTY} — its input selector matched nothing (${deadInputs.join('; ')})`
+                    : UPDATE_EMPTY;
+                const emptiness = settleBaselineEmptiness(rule, computed, liveness, input.primaryLists, reason);
+                if (emptiness.skipped && emptiness.fails) {
+                    errors.push({
+                        id: rule.id,
+                        error: 'refusing to write an EMPTY baseline for a `failOnEmpty` rule — ' +
+                            `${emptiness.skipReason ?? reason} — fix the compute first`,
+                    });
+                    continue;
+                }
+            }
             const abs = nodePath.resolve(prep.cwd, rule.baseline);
             const previous = existsSync(abs) ? readFileSync(abs, 'utf8') : undefined;
             const changed = previous !== computed.text;
@@ -459,20 +939,56 @@ export const baselineUpdateCommand = {
             }
             written.push({ id: rule.id, path: rule.baseline, bytes: computed.text.length, changed });
         }
+        const settled = settleVerdict(errors.length > 0 ? ExitCode.Failure : ExitCode.VerifiedPass, unverifiedCoverage);
+        const exit = settled.exit;
         if (json) {
-            process.stdout.write(asJson({ schema: SCHEMA, dryRun, written, errors }) + '\n');
-            return errors.length > 0 ? ExitCode.Failure : ExitCode.VerifiedPass;
+            process.stdout.write(asJson({
+                schema: SCHEMA,
+                dryRun,
+                written,
+                reblessed,
+                errors,
+                ...(unverified.length > 0 ? { unverified } : {}),
+                // K9: a ceiling over an empty measurement is the loud skip, never a
+                // proposal — named here with the reason `baseline check` prints.
+                ...(emptyCeilings.length > 0 ? { skipped: emptyCeilings } : {}),
+                ...(unverified.length > 0 || emptyCeilings.length > 0 ? { shortfalls: settled.shortfalls } : {}),
+                exitCode: exit,
+            }) + '\n');
+            return exit;
         }
         process.stdout.write(header(dryRun ? 'Baseline update (dry run)' : 'Baseline update'));
         for (const w of written) {
             process.stdout.write(`  ${w.changed ? (dryRun ? 'would write' : 'wrote') : 'unchanged '} ${w.path}  (${w.bytes} bytes)\n`);
         }
+        for (const r of reblessed) {
+            const verb = r.from === r.to ? 'unchanged ' : 'edit config';
+            process.stdout.write(`  ${verb} ${r.id}  (ceiling ${r.from} → ${r.to})\n`);
+            if (r.from !== r.to)
+                process.stdout.write(`             ${r.edit}\n`);
+        }
         for (const e of errors)
             process.stdout.write(`  ! ${e.id}: ${e.error}\n`);
+        for (const u of unverified) {
+            process.stdout.write(`  ! ${u.id}: not written — computed from an incomplete read (${u.unread.join(', ')})\n`);
+        }
+        for (const s of emptyCeilings) {
+            process.stdout.write(`  – ${s.id}: no ceiling proposed — ${s.reason}\n`);
+        }
+        if (reblessed.some((r) => r.from !== r.to)) {
+            process.stdout.write('\nA ceiling lives in sharkcraft.config.ts — apply the line above by hand so raising it stays a reviewed diff.\n');
+        }
         if (written.some((w) => w.changed) && !dryRun) {
             process.stdout.write('\nReview the diff before committing — this is the bless step.\n');
         }
-        return errors.length > 0 ? ExitCode.Failure : ExitCode.VerifiedPass;
+        const line = verdictLine(settled, '', unverified.length > 0
+            ? 'Nothing was blessed for the baseline(s) above: their value came from an incomplete read.'
+            : emptyCeilings.length > 0
+                ? 'No ceiling was proposed for the baseline(s) above: their measurement was empty — fix the compute, or mark a planned input { pattern, expectEmpty: true }.'
+                : undefined);
+        if (line)
+            process.stdout.write(`\n${line}\n`);
+        return exit;
     },
 };
 export const baselineExplainCommand = {
@@ -494,7 +1010,7 @@ export const baselineExplainCommand = {
             process.stderr.write(`No baseline "${id}". Declared: ${prep.all.map((r) => r.id).join(', ') || '(none)'}\n`);
             return ExitCode.UsageError;
         }
-        const outcome = evaluateRule(prep.cwd, rule, prep.excludeDirs, undefined);
+        const outcome = evaluateBaselineRule(prep.cwd, rule, prep.excludeDirs, undefined);
         if (flagBool(args, 'json')) {
             process.stdout.write(asJson({
                 schema: 'sharkcraft.baseline-explain/v1',
@@ -507,15 +1023,21 @@ export const baselineExplainCommand = {
         process.stdout.write(header(`Baseline: ${rule.id}`));
         if (rule.description)
             process.stdout.write(`  ${rule.description}\n`);
-        process.stdout.write(kv('committed', rule.baseline) + '\n');
-        process.stdout.write(kv('compute', rule.compute.kind === 'command' ? `command · ${rule.compute.run}` : `extractor · ${rule.compute.source?.extract ?? 'sugar'}`) + '\n');
-        process.stdout.write(kv('direction', rule.direction ?? 'two-way') + '\n');
+        process.stdout.write(kv('committed', pinLabel(rule)) + '\n');
+        process.stdout.write(kv('compute', rule.compute.kind === 'command'
+            ? `command · ${rule.compute.run}`
+            : `extractor · ${rule.compute.source?.extract ?? 'sugar'}` +
+                (rule.compute.source?.$use ? `  (via $use:${rule.compute.source.$use})` : '')) + '\n');
+        process.stdout.write(kv('direction', rule.direction ?? (rule.mode === 'ceiling' ? 'at-most' : 'two-way')) + '\n');
         process.stdout.write(kv('canonical', outcome.diff?.canonical ?? rule.compute.canonical ?? 'auto') + '\n');
         if (rule.keyBy)
             process.stdout.write(kv('keyBy', rule.keyBy) + '\n');
-        process.stdout.write(kv('entries', outcome.missingBaseline
-            ? `committed (none yet) → ${outcome.currentCount} now`
-            : `${outcome.committedCount} committed → ${outcome.currentCount} now`) + '\n');
+        process.stdout.write(kv(outcome.ceiling ? 'measured' : 'entries', outcome.ceiling
+            ? `${outcome.ceiling.value} ${outcome.ceiling.direction} ${outcome.ceiling.ceiling}` +
+                `  (${outcome.ceiling.slack >= 0 ? `${outcome.ceiling.slack} to spare` : `${-outcome.ceiling.slack} over`})`
+            : outcome.missingBaseline
+                ? `committed (none yet) → ${outcome.currentCount} now`
+                : `${outcome.committedCount} committed → ${outcome.currentCount} now`) + '\n');
         process.stdout.write(kv('status', outcome.status) + '\n');
         if (outcome.error)
             process.stdout.write(`  ! ${outcome.error}\n`);
@@ -532,11 +1054,13 @@ export const baselineCommand = {
     name: 'baseline',
     description: 'Committed-baseline drift engine: recompute a ledger/digest/allow-list and fail on drift in BOTH directions. Read-only except `update`.',
     usage: 'shrk baseline list | check | diff | update | explain --id <id>',
+    // Every subverb is a registered trie child; any other bare token is refused
+    // by the dispatcher guard (closest match named) before this body runs — the
+    // one unknown-subcommand authority. Only a bare `shrk baseline` lands here.
+    positionals: PositionalMode.None,
     booleanFlags: new Set(['json', 'changed-only', 'dry-run']),
-    async run(args) {
-        const sub = args.positional[0];
-        process.stderr.write((sub ? `Unknown subcommand "${sub}". ` : '') +
-            'Usage: shrk baseline list | check [--id X] | diff | update | explain --id <id>\n');
+    async run() {
+        process.stderr.write('Usage: shrk baseline list | check [--id X] | diff | update | explain --id <id>\n');
         return ExitCode.UsageError;
     },
 };