@gaunt-sloth/batch 2.0.0-beta.1 → 2.0.0-beta.11

package/dist/toolCoverage.d.ts

@@ -0,0 +1,154 @@
+/**
+ * One advertised tool, as the batch layer carries it — a structural mirror of core's
+ * `GthAdvertisedTool`, kept local for the same reason `ToolResultRecord` (`#src/types.js`) is: this
+ * package's outcome shapes stay independent of the runner/LLM types.
+ */
+export interface AdvertisedToolRecord {
+    /** The registered tool name, exactly as the model would call it. */
+    name: string;
+    /**
+     * The `mcpServers` key that explains the name, resolved against the configured keys at capture.
+     * Absent for a tool outside the MCP namespace (a built-in or a user-configured one); the empty
+     * string for an MCP-namespaced name no configured key explains.
+     */
+    server?: string;
+}
+/**
+ * One run's advertised-tool inventory, as captured by the agent at `init` — a structural mirror of
+ * core's `GthAdvertisedTools`.
+ */
+export interface AdvertisedToolInventory {
+    /** Every named tool the agent loaded, BEFORE any `allowedTools` narrowing. */
+    tools: AdvertisedToolRecord[];
+    /** The subset an `allowedTools` allow-list removed. Empty when no allow-list is configured. */
+    filteredOut: AdvertisedToolRecord[];
+    /** How many loaded tools carry no name — counted in neither half of the fraction. */
+    unnamed: number;
+}
+/**
+ * A suite's `tool_coverage:` declaration: which tools are deliberately not covered, and what would
+ * make the run fail.
+ */
+export interface ToolCoverageSpec {
+    /**
+     * Globs (the `must_call` matcher) whose matching advertised tools LEAVE the denominator.
+     *
+     * Load-bearing, not a nicety: a read-only suite must not be shamed for never calling the mutating
+     * tools, and without waivers any real server's coverage starts red and is then ignored — which is
+     * the same failure as a metric nobody reads. A deliberate decision not to cover something belongs
+     * in the suite, where it is reviewable, rather than in a comment.
+     */
+    waive: string[];
+    /**
+     * Optional floor: the minimum percentage (0-100) of the post-waiver denominator that must have
+     * been exercised. Breaching it is a product signal, graded like a declared metric gate.
+     */
+    min?: number;
+    /**
+     * Globs that must EACH match at least one exercised tool, whatever the percentage says. A floor
+     * answers "is enough of the surface covered"; this answers "was this specific tool reached", which
+     * a percentage can always satisfy by covering something else.
+     */
+    require: string[];
+}
+/** One bucket of the per-server breakdown. */
+export interface ToolCoverageServerReport {
+    /**
+     * Which kind of bucket this is — explicit rather than encoded in {@link server}, so a JSON reader
+     * never has to tell "no server" from "a server named empty", and so a configured key cannot
+     * collide with a label this file made up.
+     */
+    kind: 'mcp' | 'builtin' | 'unresolved';
+    /** The `mcpServers` key, for a `mcp` bucket only. */
+    server?: string;
+    /** Exercised tool names in this bucket (post-waiver). */
+    covered: string[];
+    /** Advertised-but-never-called names in this bucket (post-waiver). */
+    uncovered: string[];
+}
+/**
+ * The coverage block written to `results.json` and rendered on the console.
+ *
+ * Name LISTS rather than counts, deliberately: a directory run's aggregate is a set union across
+ * suites (the same 41-tool server advertised to every suite must be counted once, not once per
+ * suite), and counts cannot be unioned. `n/N` is `covered.length` over `covered.length +
+ * uncovered.length`.
+ */
+export interface ToolCoverageReport {
+    /** Advertised AND exercised, after waivers — the numerator. */
+    covered: string[];
+    /** Advertised, not waived, never called — the names the report exists to print. */
+    uncovered: string[];
+    /**
+     * Advertised names a `waive:` glob removed from the denominator, reported BESIDE the fraction.
+     *
+     * A waived tool leaves the denominator, so a suite waiving 38 of 41 reports 100% while covering
+     * three. Printing the waived count next to the number is what stops that reading as full coverage
+     * — the blind-denominator shape a metric facility cannot afford, because its whole value is being
+     * trusted.
+     */
+    waived: string[];
+    /** Advertised names an `allowedTools` allow-list removed before the agent bound them. */
+    filteredOut: string[];
+    /** Nameless provider-native tools, counted in neither half (there is no name to count). */
+    unnamed: number;
+    /** Per-server breakdown — a single percentage hides which server is uncovered. */
+    byServer: ToolCoverageServerReport[];
+    /** Everything that makes the figure mean less than it appears to. Never silently dropped. */
+    warnings: string[];
+    /**
+     * Breached `min` / unmet `require:` entries. Non-empty forces the run's exit code to 1 through the
+     * same contract a breached metric gate uses — a product signal, never a harness one.
+     */
+    gateFailures: string[];
+}
+/**
+ * The waived share at which the fraction stops describing the surface and starts describing the
+ * exemptions. Half is the point where the tools NOT being measured outnumber the tools that are, so
+ * a reader who takes the percentage at face value is wrong more often than right.
+ *
+ * A threshold rather than a hard failure because a high waiver share is legitimate — a read-only
+ * suite against a mostly-mutating server is exactly that — and failing it would push authors to
+ * delete the waivers, which is how the denominator goes blind in the first place.
+ */
+export declare const WAIVER_SHARE_WARN_THRESHOLD = 0.5;
+/** Inputs to {@link computeToolCoverage}. */
+export interface ToolCoverageInput {
+    /**
+     * One entry per cell that reported an inventory. **Empty means no cell reported one**, which is
+     * not the same as a cell reporting an empty inventory: the first cannot supply a denominator at
+     * all (an external target, or a run whose SUT never initialised) and yields no report; the second
+     * is a real denominator of zero and does.
+     */
+    inventories: readonly AdvertisedToolInventory[];
+    /** Every tool name any cell invoked, in any order, deduplicated or not. */
+    exercised: readonly string[];
+    /** The suite's `tool_coverage:` declaration, when it made one. */
+    spec?: ToolCoverageSpec;
+}
+/**
+ * Compute the coverage report, or `undefined` when no inventory was observed at all.
+ *
+ * `undefined` is the honest answer for a target that cannot supply a denominator, and is why this
+ * returns a value rather than a zeroed report: **`0/0` printed as a coverage figure is the "metric
+ * nobody can trust" failure in its purest form** — it looks like a measurement, it is green, and it
+ * measured nothing. A caller that declared gates on a target that cannot be measured is rejected at
+ * parse time instead (`evalSuite.ts`), so a silent pass is unreachable from both ends.
+ */
+export declare function computeToolCoverage(input: ToolCoverageInput): ToolCoverageReport | undefined;
+/**
+ * Union several suites' reports into the run-level figure, or `undefined` when none was produced.
+ *
+ * **A set union, never a sum.** A directory run points every suite at the same agent, so the same 41
+ * advertised tools appear in each report; summing would report 123 tools and make one suite's
+ * coverage of a tool count three times. Coverage is a directory-level figure precisely because one
+ * suite covering 3 tools is fine if its sibling covers the other 38 — which only reads correctly
+ * when a tool covered anywhere is covered once.
+ *
+ * **Carries no warnings and no gate failures, deliberately.** A gate is declared in a suite and
+ * graded against that suite (the `metrics:` precedent), and the run's exit is the OR of those. If a
+ * declared floor were re-applied to the aggregate, the same suite would pass when run alone and fail
+ * when run as part of a directory, with nothing in either output saying the threshold had moved.
+ * This is reporting only.
+ */
+export declare function aggregateToolCoverage(reports: readonly ToolCoverageReport[]): ToolCoverageReport | undefined;

package/dist/toolCoverage.js

@@ -0,0 +1,304 @@
+/**
+ * @packageDocumentation
+ * BATCH-32 — **tool coverage**: which of the tools the agent actually advertised to the model a
+ * suite exercised, and which it never touched.
+ *
+ * The numerator has existed since GS2-16 (`runStats.tools`, per cell, already in every
+ * `<case>.json`). What was missing is the DENOMINATOR: a run that calls 3 of 41 advertised tools
+ * prints the same all-green summary as one that calls all 41, so "every cell passed" reads as
+ * reassurance it has not earned, and coverage decays silently as the server grows a 42nd tool
+ * nobody wrote a case for.
+ *
+ * Pure and dependency-free on purpose (the `classificationReport.ts` precedent): every input is
+ * handed in, so the whole facility is unit-testable without an agent, an MCP server or a model.
+ *
+ * ## The denominator is the FULL advertised list, and that is the design decision here
+ *
+ * The list the agent binds is narrowed by `allowedTools` before it is bound. Taking the denominator
+ * from the narrowed list would let any suite reach 100% by narrowing the allow-list to the tools it
+ * already calls — the tools nobody exercises would leave the bottom of the fraction instead of being
+ * named. So the denominator is the pre-filter inventory, and the removed tools are reported as their
+ * own {@link ToolCoverageReport.filteredOut} category.
+ *
+ * ## What "covered" means, stated in the output rather than assumed
+ *
+ * A tool counts as covered when it was **called**. A call whose result came back an error still
+ * counts: the tool was reached, which is what a coverage figure measures. Grading what a tool
+ * RETURNED is what `must_error` / `tool_result_json_path` assertions are for, and they are a
+ * different question asked per case.
+ */
+import { toolNameMatchesPattern } from '@gaunt-sloth/core/utils/toolMatching.js';
+/**
+ * The waived share at which the fraction stops describing the surface and starts describing the
+ * exemptions. Half is the point where the tools NOT being measured outnumber the tools that are, so
+ * a reader who takes the percentage at face value is wrong more often than right.
+ *
+ * A threshold rather than a hard failure because a high waiver share is legitimate — a read-only
+ * suite against a mostly-mutating server is exactly that — and failing it would push authors to
+ * delete the waivers, which is how the denominator goes blind in the first place.
+ */
+export const WAIVER_SHARE_WARN_THRESHOLD = 0.5;
+/** Percentage of `total` that `part` represents, to one decimal place. `total` of 0 yields 0. */
+function percent(part, total) {
+    if (total === 0)
+        return 0;
+    return Math.round((part / total) * 1000) / 10;
+}
+/** Bucket key for the per-server breakdown — one stable string per bucket identity. */
+function bucketKeyFor(record) {
+    if (record.server === undefined)
+        return 'builtin';
+    if (record.server === '')
+        return 'unresolved';
+    return `mcp:${record.server}`;
+}
+/**
+ * Fold the per-cell inventories into one, and say so when they disagreed.
+ *
+ * **The union is the denominator, not the first cell's list.** Each cell resolves its own tools (a
+ * fresh MCP client per cell), so a cell whose server failed to connect advertises a shorter list. A
+ * denominator taken from that cell would quietly shrink, and coverage would IMPROVE because a server
+ * broke — the most misleading direction a coverage number can move. The union keeps the tool in the
+ * denominator, where it shows up as uncovered.
+ *
+ * Disagreement is still a warning, because a varying inventory means the cells did not all see the
+ * same surface and no single fraction describes the run exactly.
+ */
+function unionInventories(inventories) {
+    const tools = new Map();
+    const filteredOut = new Map();
+    let unnamed = 0;
+    let disagreed = false;
+    let firstSignature;
+    for (const inventory of inventories) {
+        for (const record of inventory.tools) {
+            if (!tools.has(record.name))
+                tools.set(record.name, record);
+        }
+        for (const record of inventory.filteredOut) {
+            if (!filteredOut.has(record.name))
+                filteredOut.set(record.name, record);
+        }
+        // The same session's inventory repeated per cell must not multiply, so take the largest count
+        // rather than the sum — `unnamed` describes one agent's surface, like every other field here.
+        unnamed = Math.max(unnamed, inventory.unnamed);
+        // Compared as JSON rather than as a joined string: any separator character that could itself
+        // occur in a tool name would make two different inventories compare equal.
+        const signature = JSON.stringify(inventory.tools.map((record) => record.name).sort());
+        if (firstSignature === undefined)
+            firstSignature = signature;
+        else if (signature !== firstSignature)
+            disagreed = true;
+    }
+    return {
+        tools: [...tools.values()],
+        filteredOut: [...filteredOut.values()],
+        unnamed,
+        disagreed,
+    };
+}
+/**
+ * Compute the coverage report, or `undefined` when no inventory was observed at all.
+ *
+ * `undefined` is the honest answer for a target that cannot supply a denominator, and is why this
+ * returns a value rather than a zeroed report: **`0/0` printed as a coverage figure is the "metric
+ * nobody can trust" failure in its purest form** — it looks like a measurement, it is green, and it
+ * measured nothing. A caller that declared gates on a target that cannot be measured is rejected at
+ * parse time instead (`evalSuite.ts`), so a silent pass is unreachable from both ends.
+ */
+export function computeToolCoverage(input) {
+    if (input.inventories.length === 0)
+        return undefined;
+    const { tools, filteredOut, unnamed, disagreed } = unionInventories(input.inventories);
+    const spec = input.spec;
+    const waivePatterns = spec?.waive ?? [];
+    const exercised = new Set(input.exercised);
+    const warnings = [];
+    const gateFailures = [];
+    const waived = [];
+    const counted = [];
+    for (const record of tools) {
+        if (waivePatterns.some((pattern) => toolNameMatchesPattern(record.name, pattern))) {
+            waived.push(record);
+        }
+        else {
+            counted.push(record);
+        }
+    }
+    const covered = counted.filter((record) => exercised.has(record.name));
+    const uncovered = counted.filter((record) => !exercised.has(record.name));
+    const total = counted.length;
+    // Per-server buckets over the counted (post-waiver) tools — the same population the headline
+    // fraction describes, so the buckets sum to it.
+    const buckets = new Map();
+    for (const record of counted) {
+        const key = bucketKeyFor(record);
+        let bucket = buckets.get(key);
+        if (!bucket) {
+            bucket =
+                record.server === undefined
+                    ? { kind: 'builtin', covered: [], uncovered: [] }
+                    : record.server === ''
+                        ? { kind: 'unresolved', covered: [], uncovered: [] }
+                        : { kind: 'mcp', server: record.server, covered: [], uncovered: [] };
+            buckets.set(key, bucket);
+        }
+        if (exercised.has(record.name))
+            bucket.covered.push(record.name);
+        else
+            bucket.uncovered.push(record.name);
+    }
+    if (disagreed) {
+        warnings.push('the cells did not all advertise the same tools — the denominator is their union, so a ' +
+            'tool missing from some cells still counts (check for an MCP server that failed to connect)');
+    }
+    // A waiver that matches nothing is dead config, and the likely cause is the dangerous one: the
+    // tool was renamed, so it is back in the denominator under its new name while the author believes
+    // it is still waived. Cheap to say, and it is the decay this whole feature exists to catch.
+    for (const pattern of waivePatterns) {
+        if (!tools.some((record) => toolNameMatchesPattern(record.name, pattern))) {
+            warnings.push(`waive "${pattern}" matched no advertised tool — stale waiver, or a typo`);
+        }
+    }
+    if (waived.length > 0 && waived.length / tools.length >= WAIVER_SHARE_WARN_THRESHOLD) {
+        warnings.push(`${waived.length} of ${tools.length} advertised tool(s) are waived ` +
+            `(${percent(waived.length, tools.length)}%) — the figure describes the exemptions more ` +
+            'than the surface');
+    }
+    if (filteredOut.length > 0) {
+        warnings.push(`${filteredOut.length} advertised tool(s) were removed by allowedTools and never reached ` +
+            'the model — they are in the denominator, so they can only ever read as uncovered');
+    }
+    // Exercised names the inventory never listed. Not folded into the numerator (that would let a
+    // fraction exceed 1 and would credit coverage of a tool nobody advertised), and not dropped
+    // either — a name that ran but was never advertised means the two halves are measuring different
+    // populations, which is exactly the kind of quiet mismatch this report must surface.
+    const advertisedNames = new Set(tools.map((record) => record.name));
+    const unexpected = [...exercised].filter((name) => !advertisedNames.has(name)).sort();
+    if (unexpected.length > 0) {
+        warnings.push(`${unexpected.length} exercised tool(s) were not in the advertised inventory ` +
+            `(${unexpected.join(', ')}) — they are not counted in the numerator`);
+    }
+    if (spec) {
+        // Deliberately asymmetric with `waive`, which warns when a pattern matches nothing. A `require`
+        // pattern that matches nothing — the mistyped or stale entry — lands in the first branch and
+        // fails the gate, which already stops the run and names the pattern, so a stale-config warning
+        // beside it would only be quieter duplication. The warning below is the narrower case the gate
+        // cannot state: the pattern WAS satisfied by an exercised tool, but no advertised tool matches
+        // it, so the run passes while the two halves disagree about what exists.
+        for (const pattern of spec.require) {
+            if (![...exercised].some((name) => toolNameMatchesPattern(name, pattern))) {
+                gateFailures.push(`require "${pattern}": no case exercised a tool matching it`);
+            }
+            else if (!tools.some((record) => toolNameMatchesPattern(record.name, pattern))) {
+                warnings.push(`require "${pattern}" matched no ADVERTISED tool — it was satisfied by a tool the ` +
+                    'inventory does not list');
+            }
+        }
+        if (spec.min !== undefined) {
+            if (total === 0) {
+                // A floor over an empty denominator cannot be met, and must not be treated as met. This is
+                // the case where every tool went missing (an MCP server that never connected, or a waiver
+                // list that swallowed the whole surface) — the one run where a vacuous pass would be most
+                // misleading, because it is indistinguishable from perfect coverage.
+                gateFailures.push(`min ${spec.min}%: no tools remain in the denominator — ` +
+                    (tools.length === 0
+                        ? 'the agent advertised none'
+                        : `all ${tools.length} advertised tool(s) are waived`));
+            }
+            else if (percent(covered.length, total) < spec.min) {
+                gateFailures.push(`min ${spec.min}%: covered ${covered.length}/${total} ` +
+                    `(${percent(covered.length, total)}%)`);
+            }
+        }
+    }
+    return {
+        covered: covered.map((record) => record.name),
+        uncovered: uncovered.map((record) => record.name),
+        waived: waived.map((record) => record.name),
+        filteredOut: filteredOut.map((record) => record.name),
+        unnamed,
+        byServer: [...buckets.values()],
+        warnings,
+        gateFailures,
+    };
+}
+/**
+ * Union several suites' reports into the run-level figure, or `undefined` when none was produced.
+ *
+ * **A set union, never a sum.** A directory run points every suite at the same agent, so the same 41
+ * advertised tools appear in each report; summing would report 123 tools and make one suite's
+ * coverage of a tool count three times. Coverage is a directory-level figure precisely because one
+ * suite covering 3 tools is fine if its sibling covers the other 38 — which only reads correctly
+ * when a tool covered anywhere is covered once.
+ *
+ * **Carries no warnings and no gate failures, deliberately.** A gate is declared in a suite and
+ * graded against that suite (the `metrics:` precedent), and the run's exit is the OR of those. If a
+ * declared floor were re-applied to the aggregate, the same suite would pass when run alone and fail
+ * when run as part of a directory, with nothing in either output saying the threshold had moved.
+ * This is reporting only.
+ */
+export function aggregateToolCoverage(reports) {
+    if (reports.length === 0)
+        return undefined;
+    const covered = new Set();
+    const uncovered = new Set();
+    const waived = new Set();
+    const filteredOut = new Set();
+    const buckets = new Map();
+    let unnamed = 0;
+    for (const report of reports) {
+        for (const name of report.covered)
+            covered.add(name);
+        for (const name of report.uncovered)
+            uncovered.add(name);
+        for (const name of report.waived)
+            waived.add(name);
+        for (const name of report.filteredOut)
+            filteredOut.add(name);
+        unnamed = Math.max(unnamed, report.unnamed);
+        for (const bucket of report.byServer) {
+            const key = bucket.kind === 'mcp' ? `mcp:${bucket.server}` : bucket.kind;
+            let merged = buckets.get(key);
+            if (!merged) {
+                merged = {
+                    kind: bucket.kind,
+                    ...(bucket.server !== undefined ? { server: bucket.server } : {}),
+                    covered: [],
+                    uncovered: [],
+                };
+                buckets.set(key, merged);
+            }
+            merged.covered.push(...bucket.covered);
+            merged.uncovered.push(...bucket.uncovered);
+        }
+    }
+    // A tool covered by ANY suite is covered: that is what makes the aggregate the interesting
+    // number. Without this, a tool one suite exercised would still be listed as uncovered because a
+    // sibling suite never touched it.
+    for (const name of covered)
+        uncovered.delete(name);
+    // Likewise a tool that is waived in one suite but counted in another stays counted — a waiver is
+    // one suite's statement about its own scope, and the run as a whole did measure that tool.
+    for (const name of [...covered, ...uncovered])
+        waived.delete(name);
+    return {
+        covered: [...covered],
+        uncovered: [...uncovered],
+        waived: [...waived],
+        filteredOut: [...filteredOut],
+        unnamed,
+        byServer: [...buckets.values()].map((bucket) => {
+            const bucketCovered = [...new Set(bucket.covered)];
+            const coveredSet = new Set(bucketCovered);
+            return {
+                ...bucket,
+                covered: bucketCovered,
+                uncovered: [...new Set(bucket.uncovered)].filter((name) => !coveredSet.has(name)),
+            };
+        }),
+        warnings: [],
+        gateFailures: [],
+    };
+}
+//# sourceMappingURL=toolCoverage.js.map

package/dist/toolCoverage.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"toolCoverage.js","sourceRoot":"","sources":["../src/toolCoverage.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,OAAO,EAAE,sBAAsB,EAAE,MAAM,yCAAyC,CAAC;AA+GjF;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC;AAE/C,iGAAiG;AACjG,SAAS,OAAO,CAAC,IAAY,EAAE,KAAa;IAC1C,IAAI,KAAK,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC1B,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;AAChD,CAAC;AAED,uFAAuF;AACvF,SAAS,YAAY,CAAC,MAA4B;IAChD,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAClD,IAAI,MAAM,CAAC,MAAM,KAAK,EAAE;QAAE,OAAO,YAAY,CAAC;IAC9C,OAAO,OAAO,MAAM,CAAC,MAAM,EAAE,CAAC;AAChC,CAAC;AAED;;;;;;;;;;;GAWG;AACH,SAAS,gBAAgB,CAAC,WAA+C;IAMvE,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgC,CAAC;IACtD,MAAM,WAAW,GAAG,IAAI,GAAG,EAAgC,CAAC;IAC5D,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,SAAS,GAAG,KAAK,CAAC;IACtB,IAAI,cAAkC,CAAC;IAEvC,KAAK,MAAM,SAAS,IAAI,WAAW,EAAE,CAAC;QACpC,KAAK,MAAM,MAAM,IAAI,SAAS,CAAC,KAAK,EAAE,CAAC;YACrC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;gBAAE,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QAC9D,CAAC;QACD,KAAK,MAAM,MAAM,IAAI,SAAS,CAAC,WAAW,EAAE,CAAC;YAC3C,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;gBAAE,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QAC1E,CAAC;QACD,8FAA8F;QAC9F,8FAA8F;QAC9F,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC;QAE/C,6FAA6F;QAC7F,2EAA2E;QAC3E,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QACtF,IAAI,cAAc,KAAK,SAAS;YAAE,cAAc,GAAG,SAAS,CAAC;aACxD,IAAI,SAAS,KAAK,cAAc;YAAE,SAAS,GAAG,IAAI,CAAC;IAC1D,CAAC;IAED,OAAO;QACL,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;QAC1B,WAAW,EAAE,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC;QACtC,OAAO;QACP,SAAS;KACV,CAAC;AACJ,CAAC;AAiBD;;;;;;;;GAQG;AACH,MAAM,UAAU,mBAAmB,CAAC,KAAwB;IAC1D,IAAI,KAAK,CAAC,WAAW,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAErD,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,gBAAgB,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;IACvF,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC;IACxB,MAAM,aAAa,GAAG,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;IACxC,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAC3C,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,MAAM,YAAY,GAAa,EAAE,CAAC;IAElC,MAAM,MAAM,GAA2B,EAAE,CAAC;IAC1C,MAAM,OAAO,GAA2B,EAAE,CAAC;IAC3C,KAAK,MAAM,MAAM,IAAI,KAAK,EAAE,CAAC;QAC3B,IAAI,aAAa,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,sBAAsB,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC;YAClF,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACtB,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACvB,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IACvE,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IAC1E,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;IAE7B,6FAA6F;IAC7F,gDAAgD;IAChD,MAAM,OAAO,GAAG,IAAI,GAAG,EAAoC,CAAC;IAC5D,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,GAAG,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC;QACjC,IAAI,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC9B,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM;gBACJ,MAAM,CAAC,MAAM,KAAK,SAAS;oBACzB,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;oBACjD,CAAC,CAAC,MAAM,CAAC,MAAM,KAAK,EAAE;wBACpB,CAAC,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,OAAO,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;wBACpD,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;YAC3E,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC3B,CAAC;QACD,IAAI,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;YAAE,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;;YAC5D,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,SAAS,EAAE,CAAC;QACd,QAAQ,CAAC,IAAI,CACX,wFAAwF;YACtF,4FAA4F,CAC/F,CAAC;IACJ,CAAC;IAED,+FAA+F;IAC/F,kGAAkG;IAClG,4FAA4F;IAC5F,KAAK,MAAM,OAAO,IAAI,aAAa,EAAE,CAAC;QACpC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,sBAAsB,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC;YAC1E,QAAQ,CAAC,IAAI,CAAC,UAAU,OAAO,wDAAwD,CAAC,CAAC;QAC3F,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,IAAI,2BAA2B,EAAE,CAAC;QACrF,QAAQ,CAAC,IAAI,CACX,GAAG,MAAM,CAAC,MAAM,OAAO,KAAK,CAAC,MAAM,iCAAiC;YAClE,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,gDAAgD;YACxF,kBAAkB,CACrB,CAAC;IACJ,CAAC;IAED,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,QAAQ,CAAC,IAAI,CACX,GAAG,WAAW,CAAC,MAAM,qEAAqE;YACxF,kFAAkF,CACrF,CAAC;IACJ,CAAC;IAED,8FAA8F;IAC9F,4FAA4F;IAC5F,iGAAiG;IACjG,qFAAqF;IACrF,MAAM,eAAe,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,eAAe,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACtF,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC1B,QAAQ,CAAC,IAAI,CACX,GAAG,UAAU,CAAC,MAAM,0DAA0D;YAC5E,IAAI,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,2CAA2C,CACvE,CAAC;IACJ,CAAC;IAED,IAAI,IAAI,EAAE,CAAC;QACT,gGAAgG;QAChG,6FAA6F;QAC7F,+FAA+F;QAC/F,+FAA+F;QAC/F,+FAA+F;QAC/F,yEAAyE;QACzE,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACnC,IAAI,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,sBAAsB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC;gBAC1E,YAAY,CAAC,IAAI,CAAC,YAAY,OAAO,yCAAyC,CAAC,CAAC;YAClF,CAAC;iBAAM,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,sBAAsB,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC;gBACjF,QAAQ,CAAC,IAAI,CACX,YAAY,OAAO,gEAAgE;oBACjF,yBAAyB,CAC5B,CAAC;YACJ,CAAC;QACH,CAAC;QAED,IAAI,IAAI,CAAC,GAAG,KAAK,SAAS,EAAE,CAAC;YAC3B,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;gBAChB,2FAA2F;gBAC3F,0FAA0F;gBAC1F,0FAA0F;gBAC1F,qEAAqE;gBACrE,YAAY,CAAC,IAAI,CACf,OAAO,IAAI,CAAC,GAAG,0CAA0C;oBACvD,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC;wBACjB,CAAC,CAAC,2BAA2B;wBAC7B,CAAC,CAAC,OAAO,KAAK,CAAC,MAAM,gCAAgC,CAAC,CAC3D,CAAC;YACJ,CAAC;iBAAM,IAAI,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,KAAK,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;gBACrD,YAAY,CAAC,IAAI,CACf,OAAO,IAAI,CAAC,GAAG,cAAc,OAAO,CAAC,MAAM,IAAI,KAAK,GAAG;oBACrD,IAAI,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,CACzC,CAAC;YACJ,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO;QACL,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC;QAC7C,SAAS,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC;QACjD,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC;QAC3C,WAAW,EAAE,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC;QACrD,OAAO;QACP,QAAQ,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC;QAC/B,QAAQ;QACR,YAAY;KACb,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,qBAAqB,CACnC,OAAsC;IAEtC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAE3C,MAAM,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;IAClC,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IACpC,MAAM,MAAM,GAAG,IAAI,GAAG,EAAU,CAAC;IACjC,MAAM,WAAW,GAAG,IAAI,GAAG,EAAU,CAAC;IACtC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAoC,CAAC;IAC5D,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,OAAO;YAAE,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACrD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,SAAS;YAAE,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACzD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,MAAM;YAAE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACnD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,WAAW;YAAE,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC7D,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,CAAC;QAC5C,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrC,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,OAAO,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC;YACzE,IAAI,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC9B,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,MAAM,GAAG;oBACP,IAAI,EAAE,MAAM,CAAC,IAAI;oBACjB,GAAG,CAAC,MAAM,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;oBACjE,OAAO,EAAE,EAAE;oBACX,SAAS,EAAE,EAAE;iBACd,CAAC;gBACF,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YAC3B,CAAC;YACD,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC;YACvC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC;IAED,2FAA2F;IAC3F,gGAAgG;IAChG,kCAAkC;IAClC,KAAK,MAAM,IAAI,IAAI,OAAO;QAAE,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IACnD,iGAAiG;IACjG,2FAA2F;IAC3F,KAAK,MAAM,IAAI,IAAI,CAAC,GAAG,OAAO,EAAE,GAAG,SAAS,CAAC;QAAE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAEnE,OAAO;QACL,OAAO,EAAE,CAAC,GAAG,OAAO,CAAC;QACrB,SAAS,EAAE,CAAC,GAAG,SAAS,CAAC;QACzB,MAAM,EAAE,CAAC,GAAG,MAAM,CAAC;QACnB,WAAW,EAAE,CAAC,GAAG,WAAW,CAAC;QAC7B,OAAO;QACP,QAAQ,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE;YAC7C,MAAM,aAAa,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC;YACnD,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,aAAa,CAAC,CAAC;YAC1C,OAAO;gBACL,GAAG,MAAM;gBACT,OAAO,EAAE,aAAa;gBACtB,SAAS,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;aAClF,CAAC;QACJ,CAAC,CAAC;QACF,QAAQ,EAAE,EAAE;QACZ,YAAY,EAAE,EAAE;KACjB,CAAC;AACJ,CAAC"}

package/dist/toolCoverageRender.d.ts

@@ -0,0 +1,31 @@
+/**
+ * @packageDocumentation
+ * BATCH-32 — render a {@link ToolCoverageReport} as plain lines.
+ *
+ * A pure string builder, separate from the reporter that prints it (the `classificationRender.ts`
+ * precedent): the console rendering is unit-testable without mocking `consoleUtils`, and another
+ * reporter can emit the same block without reimplementing the layout.
+ *
+ * Total over the report — no "…and 12 more" anywhere. The uncovered list IS the output; a renderer
+ * that truncates it turns the one actionable part of the block into a teaser, and a truncated list
+ * reads as a complete one.
+ */
+import type { ToolCoverageReport } from '#src/toolCoverage.js';
+/** Options for {@link renderToolCoverage}. */
+export interface RenderToolCoverageOptions {
+    /**
+     * Prefix the headline with this word — `'TOTAL'` for the run-level aggregate that follows
+     * `EVAL TOTAL:`, so the two numbers on screen are never mistaken for each other.
+     */
+    scope?: string;
+}
+/**
+ * Render the whole block: the fraction, what qualifies it, the uncovered names, the per-server
+ * breakdown, any warnings, and any gate failures.
+ *
+ * The waived count sits on the headline rather than below it. A suite waiving 38 of 41 reports
+ * 100%, and a reader who sees only the fraction has been told the opposite of the truth — so the
+ * qualifier has to arrive in the same glance as the number it qualifies, not in a detail line that
+ * scrolls or gets skimmed.
+ */
+export declare function renderToolCoverage(report: ToolCoverageReport, options?: RenderToolCoverageOptions): string[];

package/dist/toolCoverageRender.js

@@ -0,0 +1,71 @@
+/** The headline label, shared by the per-suite line and the run-level aggregate. */
+const HEADLINE = 'TOOL COVERAGE';
+/** How many tools the report's fraction is over (covered + uncovered). */
+function denominatorOf(report) {
+    return report.covered.length + report.uncovered.length;
+}
+/** A bucket's human label: the server key, or what the bucket holds instead of one. */
+function serverLabel(bucket) {
+    if (bucket.kind === 'mcp')
+        return bucket.server ?? '';
+    if (bucket.kind === 'builtin')
+        return 'built-in / config tools';
+    return 'unattributed MCP tools';
+}
+/**
+ * Render the whole block: the fraction, what qualifies it, the uncovered names, the per-server
+ * breakdown, any warnings, and any gate failures.
+ *
+ * The waived count sits on the headline rather than below it. A suite waiving 38 of 41 reports
+ * 100%, and a reader who sees only the fraction has been told the opposite of the truth — so the
+ * qualifier has to arrive in the same glance as the number it qualifies, not in a detail line that
+ * scrolls or gets skimmed.
+ */
+export function renderToolCoverage(report, options = {}) {
+    const lines = [];
+    const total = denominatorOf(report);
+    const headline = options.scope ? `${HEADLINE} ${options.scope}` : HEADLINE;
+    const qualifiers = [];
+    if (report.waived.length > 0)
+        qualifiers.push(`${report.waived.length} waived`);
+    if (report.filteredOut.length > 0) {
+        qualifiers.push(`${report.filteredOut.length} filtered out by allowedTools`);
+    }
+    if (report.unnamed > 0)
+        qualifiers.push(`${report.unnamed} unnamed (not counted)`);
+    lines.push(`${headline}: ${report.covered.length}/${total} tools exercised` +
+        (qualifiers.length > 0 ? ` — ${qualifiers.join(', ')}` : ''));
+    // Coverage counts a CALL. Said in the output rather than left for the reader to assume, because
+    // the alternative reading (the tool was called AND returned something usable) is the one a person
+    // reaching for a coverage number is likely to want.
+    lines.push('  a tool counts as covered once a case CALLED it, error result or not');
+    if (report.uncovered.length > 0) {
+        lines.push(`  uncovered: ${[...report.uncovered].sort().join(', ')}`);
+    }
+    if (report.waived.length > 0) {
+        lines.push(`  waived: ${[...report.waived].sort().join(', ')}`);
+    }
+    if (report.filteredOut.length > 0) {
+        lines.push(`  filtered out by allowedTools: ${[...report.filteredOut].sort().join(', ')}`);
+    }
+    // Per-server only once more than one MCP server is configured — with a single server the
+    // breakdown restates the headline, and printing it twice is noise rather than information. The
+    // report still carries every bucket in `results.json`, where a reader is not paying for lines.
+    const mcpBuckets = report.byServer.filter((bucket) => bucket.kind === 'mcp');
+    if (mcpBuckets.length > 1) {
+        lines.push('  by server:');
+        for (const bucket of report.byServer) {
+            const bucketTotal = bucket.covered.length + bucket.uncovered.length;
+            lines.push(`    ${serverLabel(bucket)}: ${bucket.covered.length}/${bucketTotal}` +
+                (bucket.uncovered.length > 0
+                    ? ` — uncovered: ${[...bucket.uncovered].sort().join(', ')}`
+                    : ''));
+        }
+    }
+    for (const warning of report.warnings)
+        lines.push(`  ! ${warning}`);
+    for (const failure of report.gateFailures)
+        lines.push(`TOOL COVERAGE GATE FAILED — ${failure}`);
+    return lines;
+}
+//# sourceMappingURL=toolCoverageRender.js.map

package/dist/toolCoverageRender.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"toolCoverageRender.js","sourceRoot":"","sources":["../src/toolCoverageRender.ts"],"names":[],"mappings":"AAcA,oFAAoF;AACpF,MAAM,QAAQ,GAAG,eAAe,CAAC;AAEjC,0EAA0E;AAC1E,SAAS,aAAa,CAAC,MAA0B;IAC/C,OAAO,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC;AACzD,CAAC;AAED,uFAAuF;AACvF,SAAS,WAAW,CAAC,MAAgC;IACnD,IAAI,MAAM,CAAC,IAAI,KAAK,KAAK;QAAE,OAAO,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC;IACtD,IAAI,MAAM,CAAC,IAAI,KAAK,SAAS;QAAE,OAAO,yBAAyB,CAAC;IAChE,OAAO,wBAAwB,CAAC;AAClC,CAAC;AAWD;;;;;;;;GAQG;AACH,MAAM,UAAU,kBAAkB,CAChC,MAA0B,EAC1B,OAAO,GAA8B,EAAE;IAEvC,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,KAAK,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC;IACpC,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,QAAQ,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC;IAE3E,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,SAAS,CAAC,CAAC;IAChF,IAAI,MAAM,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAClC,UAAU,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,WAAW,CAAC,MAAM,+BAA+B,CAAC,CAAC;IAC/E,CAAC;IACD,IAAI,MAAM,CAAC,OAAO,GAAG,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,OAAO,wBAAwB,CAAC,CAAC;IAEnF,KAAK,CAAC,IAAI,CACR,GAAG,QAAQ,KAAK,MAAM,CAAC,OAAO,CAAC,MAAM,IAAI,KAAK,kBAAkB;QAC9D,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAC/D,CAAC;IAEF,gGAAgG;IAChG,kGAAkG;IAClG,oDAAoD;IACpD,KAAK,CAAC,IAAI,CAAC,uEAAuE,CAAC,CAAC;IAEpF,IAAI,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7B,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAClE,CAAC;IACD,IAAI,MAAM,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAClC,KAAK,CAAC,IAAI,CAAC,mCAAmC,CAAC,GAAG,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC7F,CAAC;IAED,yFAAyF;IACzF,+FAA+F;IAC/F,+FAA+F;IAC/F,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,KAAK,CAAC,CAAC;IAC7E,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC1B,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QAC3B,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrC,MAAM,WAAW,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC;YACpE,KAAK,CAAC,IAAI,CACR,OAAO,WAAW,CAAC,MAAM,CAAC,KAAK,MAAM,CAAC,OAAO,CAAC,MAAM,IAAI,WAAW,EAAE;gBACnE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC;oBAC1B,CAAC,CAAC,iBAAiB,CAAC,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;oBAC5D,CAAC,CAAC,EAAE,CAAC,CACV,CAAC;QACJ,CAAC;IACH,CAAC;IAED,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,QAAQ;QAAE,KAAK,CAAC,IAAI,CAAC,OAAO,OAAO,EAAE,CAAC,CAAC;IACpE,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,YAAY;QAAE,KAAK,CAAC,IAAI,CAAC,+BAA+B,OAAO,EAAE,CAAC,CAAC;IAEhG,OAAO,KAAK,CAAC;AACf,CAAC"}

package/dist/types.d.ts

@@ -7,6 +7,7 @@
  * `packages/app/src/commands/batchCommand.ts`, which is what keeps {@link RunCellFn} injectable —
  * unit tests here fake it, the CLI wires the real one.
  */
+import type { AdvertisedToolInventory } from '#src/toolCoverage.js';
 /** One row/record parsed from `--over <path.csv|path.jsonl>`. Values are always strings for csv. */
 export type MatrixRow = Record<string, string>;
 /**
@@ -68,6 +69,13 @@ export interface CellRunOutcome {
      * {@link tools} but un-deduped and carrying each result's error status + payload. Only the
      * in-process `gth-agent` runner can populate this (external adk-agent/ag-ui targets never do). */
     toolResults?: ToolResultRecord[];
+    /**
+     * BATCH-32 — what the agent ADVERTISED to the model for this cell: the tool-coverage denominator,
+     * to {@link tools}'s numerator. Only the in-process `gth-agent` runner can populate it (an
+     * external adk-agent/ag-ui target never advertises its inventory over the wire), and a test fake
+     * leaves it unset — which is why absence means "no denominator available", never "no tools".
+     */
+    advertisedTools?: AdvertisedToolInventory;
     /** A human-readable failure reason, set when `ok` is `false`. */
     error?: string;
 }

package/dist/types.js

@@ -1,12 +1,3 @@
-/**
- * @packageDocumentation
- * BATCH-1 — the shapes shared across matrix construction, the concurrency/retry runner, and the
- * structured output writer. Deliberately independent of any LLM/runner types (`GthConfig`,
- * `runSingleShot`, …): this package only knows about *cells* and *outcomes*. The production
- * adapter that turns a cell into an actual `runSingleShot` call lives in
- * `packages/app/src/commands/batchCommand.ts`, which is what keeps {@link RunCellFn} injectable —
- * unit tests here fake it, the CLI wires the real one.
- */
 /**
  * BATCH-24 — the default **cell** concurrency cap for `gth batch` / `gth eval` when
  * `-j/--concurrency` is not supplied. Serial by default, for every provider (deliberately NOT

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAkHH;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG,CAAC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAkIA;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG,CAAC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gaunt-sloth/batch",
-  "version": "2.0.0-beta.1",
+  "version": "2.0.0-beta.11",
   "description": "Batch matrix runtime for Gaunt Sloth (BATCH-1): run a prompt-executable over a matrix of models and/or content-bound inputs",
   "license": "MIT",
   "author": "Andrew Kondratev",
@@ -34,11 +34,11 @@
     "#src/*.js": "./dist/*.js"
   },
   "dependencies": {
-    "@langchain/core": "^1.2.5",
-    "yaml": "^2.9.0",
-    "zod": "^4.4.3",
-    "@gaunt-sloth/agent": "2.0.0-beta.1",
-    "@gaunt-sloth/core": "2.0.0-beta.1"
+    "@langchain/core": "^1.2.11",
+    "yaml": "^2.9.1",
+    "zod": "^4.6.3",
+    "@gaunt-sloth/agent": "2.0.0-beta.11",
+    "@gaunt-sloth/core": "2.0.0-beta.11"
   },
   "files": [
     "./dist/*"