@cat-factory/sandbox 0.7.2 → 0.7.3

package/dist/baselines.d.ts

@@ -0,0 +1,35 @@
+import type { SandboxPromptVersion } from '@cat-factory/kernel';
+import type { SandboxTaskType } from './rubrics.js';
+export type SandboxAgentBucket = 'inline' | 'container';
+export interface SandboxAgentKindMeta {
+    /** The agent kind (matches `AgentKind` strings used across the product). */
+    agentKind: string;
+    /** A short human label for the Sandbox prompt browser. */
+    label: string;
+    /** Inline kinds run a single LLM call; container kinds need a real checkout. */
+    bucket: SandboxAgentBucket;
+    /** Which rubric the judge grades this kind's output against. */
+    rubric: SandboxTaskType;
+    /**
+     * The version-controlled baseline prompt id (a `PROMPT_VERSIONS` key) this kind's
+     * system prompt comes from. When null, the baseline text is read from
+     * `systemPromptFor(agentKind)` and labelled `<kind>@v1`.
+     */
+    basePromptId: string | null;
+}
+/** The testable-kind catalog. Ordered for stable display (inline-first, then container). */
+export declare const SANDBOX_AGENT_KINDS: readonly SandboxAgentKindMeta[];
+/** Metadata for a testable agent kind, or undefined if the kind is not in the catalog. */
+export declare function sandboxKindMeta(agentKind: string): SandboxAgentKindMeta | undefined;
+/** The current shipped system-prompt text + `id@vN` label for a catalog kind. */
+export declare function baselinePromptText(meta: SandboxAgentKindMeta): {
+    text: string;
+    label: string;
+};
+/**
+ * Enumerate every shipped baseline as a synthetic (un-persisted) {@link SandboxPromptVersion}.
+ * These are version 0, origin `baseline`, with no parent/lineage of their own — the prompt
+ * browser groups them by agent kind and offers "clone" to start an editable candidate lineage.
+ */
+export declare function listBaselines(now: number): SandboxPromptVersion[];
+//# sourceMappingURL=baselines.d.ts.map

package/dist/baselines.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"baselines.d.ts","sourceRoot":"","sources":["../src/baselines.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAA;AAC/D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,cAAc,CAAA;AAWnD,MAAM,MAAM,kBAAkB,GAAG,QAAQ,GAAG,WAAW,CAAA;AAEvD,MAAM,WAAW,oBAAoB;IACnC,4EAA4E;IAC5E,SAAS,EAAE,MAAM,CAAA;IACjB,0DAA0D;IAC1D,KAAK,EAAE,MAAM,CAAA;IACb,gFAAgF;IAChF,MAAM,EAAE,kBAAkB,CAAA;IAC1B,gEAAgE;IAChE,MAAM,EAAE,eAAe,CAAA;IACvB;;;;OAIG;IACH,YAAY,EAAE,MAAM,GAAG,IAAI,CAAA;CAC5B;AAED,4FAA4F;AAC5F,eAAO,MAAM,mBAAmB,EAAE,SAAS,oBAAoB,EAwC9D,CAAA;AAMD,0FAA0F;AAC1F,wBAAgB,eAAe,CAAC,SAAS,EAAE,MAAM,GAAG,oBAAoB,GAAG,SAAS,CAEnF;AAED,iFAAiF;AACjF,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,oBAAoB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAM9F;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,GAAG,EAAE,MAAM,GAAG,oBAAoB,EAAE,CAmBjE"}

package/dist/baselines.js

@@ -0,0 +1,82 @@
+import { PROMPT_VERSIONS, promptVersionLabel, systemPromptFor } from '@cat-factory/agents';
+/** The testable-kind catalog. Ordered for stable display (inline-first, then container). */
+export const SANDBOX_AGENT_KINDS = [
+    {
+        agentKind: 'requirements-review',
+        label: 'Requirements review',
+        bucket: 'inline',
+        rubric: 'requirement-review',
+        basePromptId: 'requirement-review',
+    },
+    {
+        agentKind: 'clarity-review',
+        label: 'Clarity (bug-report) review',
+        bucket: 'inline',
+        rubric: 'requirement-review',
+        basePromptId: 'clarity-review',
+    },
+    {
+        agentKind: 'reviewer',
+        label: 'Code reviewer',
+        bucket: 'inline',
+        rubric: 'code-review',
+        basePromptId: 'review',
+    },
+    {
+        // Reviews an `architect`'s design proposal (the architect-companion grades it). A
+        // proposal critique is graded on the same axes as a requirements review — gap
+        // coverage, no-hallucination, specificity. No numbered baseline prompt: the text is
+        // read live from `systemPromptFor('architect-companion')`.
+        agentKind: 'architect-companion',
+        label: 'Architecture-proposal review',
+        bucket: 'inline',
+        rubric: 'requirement-review',
+        basePromptId: null,
+    },
+    {
+        agentKind: 'coder',
+        label: 'Coder (implementation)',
+        bucket: 'container',
+        rubric: 'implementation',
+        basePromptId: 'build',
+    },
+];
+const BY_KIND = new Map(SANDBOX_AGENT_KINDS.map((m) => [m.agentKind, m]));
+/** Metadata for a testable agent kind, or undefined if the kind is not in the catalog. */
+export function sandboxKindMeta(agentKind) {
+    return BY_KIND.get(agentKind);
+}
+/** The current shipped system-prompt text + `id@vN` label for a catalog kind. */
+export function baselinePromptText(meta) {
+    if (meta.basePromptId && meta.basePromptId in PROMPT_VERSIONS) {
+        const versioned = PROMPT_VERSIONS[meta.basePromptId];
+        return { text: versioned.text, label: promptVersionLabel(versioned.id, versioned.version) };
+    }
+    return { text: systemPromptFor(meta.agentKind), label: promptVersionLabel(meta.agentKind, 1) };
+}
+/**
+ * Enumerate every shipped baseline as a synthetic (un-persisted) {@link SandboxPromptVersion}.
+ * These are version 0, origin `baseline`, with no parent/lineage of their own — the prompt
+ * browser groups them by agent kind and offers "clone" to start an editable candidate lineage.
+ */
+export function listBaselines(now) {
+    return SANDBOX_AGENT_KINDS.map((meta) => {
+        const { text, label } = baselinePromptText(meta);
+        return {
+            id: `baseline:${meta.basePromptId ?? meta.agentKind}`,
+            lineageId: `baseline:${meta.basePromptId ?? meta.agentKind}`,
+            agentKind: meta.agentKind,
+            name: label,
+            origin: 'baseline',
+            systemText: text,
+            basePromptId: meta.basePromptId,
+            version: 0,
+            parentId: null,
+            labels: [],
+            createdAt: now,
+            createdBy: null,
+            archivedAt: null,
+        };
+    });
+}
+//# sourceMappingURL=baselines.js.map

package/dist/baselines.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"baselines.js","sourceRoot":"","sources":["../src/baselines.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAA;AAgC1F,4FAA4F;AAC5F,MAAM,CAAC,MAAM,mBAAmB,GAAoC;IAClE;QACE,SAAS,EAAE,qBAAqB;QAChC,KAAK,EAAE,qBAAqB;QAC5B,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,oBAAoB;KACnC;IACD;QACE,SAAS,EAAE,gBAAgB;QAC3B,KAAK,EAAE,6BAA6B;QACpC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,gBAAgB;KAC/B;IACD;QACE,SAAS,EAAE,UAAU;QACrB,KAAK,EAAE,eAAe;QACtB,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,aAAa;QACrB,YAAY,EAAE,QAAQ;KACvB;IACD;QACE,kFAAkF;QAClF,8EAA8E;QAC9E,oFAAoF;QACpF,2DAA2D;QAC3D,SAAS,EAAE,qBAAqB;QAChC,KAAK,EAAE,8BAA8B;QACrC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,oBAAoB;QAC5B,YAAY,EAAE,IAAI;KACnB;IACD;QACE,SAAS,EAAE,OAAO;QAClB,KAAK,EAAE,wBAAwB;QAC/B,MAAM,EAAE,WAAW;QACnB,MAAM,EAAE,gBAAgB;QACxB,YAAY,EAAE,OAAO;KACtB;CACF,CAAA;AAED,MAAM,OAAO,GAAG,IAAI,GAAG,CACrB,mBAAmB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,CACjD,CAAA;AAED,0FAA0F;AAC1F,MAAM,UAAU,eAAe,CAAC,SAAiB;IAC/C,OAAO,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAA;AAC/B,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,kBAAkB,CAAC,IAA0B;IAC3D,IAAI,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,YAAY,IAAI,eAAe,EAAE,CAAC;QAC9D,MAAM,SAAS,GAAG,eAAe,CAAC,IAAI,CAAC,YAA4C,CAAC,CAAA;QACpF,OAAO,EAAE,IAAI,EAAE,SAAS,CAAC,IAAI,EAAE,KAAK,EAAE,kBAAkB,CAAC,SAAS,CAAC,EAAE,EAAE,SAAS,CAAC,OAAO,CAAC,EAAE,CAAA;IAC7F,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,eAAe,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,kBAAkB,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC,EAAE,CAAA;AAChG,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAAC,GAAW;IACvC,OAAO,mBAAmB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;QACtC,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,kBAAkB,CAAC,IAAI,CAAC,CAAA;QAChD,OAAO;YACL,EAAE,EAAE,YAAY,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,SAAS,EAAE;YACrD,SAAS,EAAE,YAAY,IAAI,CAAC,YAAY,IAAI,IAAI,CAAC,SAAS,EAAE;YAC5D,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,IAAI,EAAE,KAAK;YACX,MAAM,EAAE,UAAU;YAClB,UAAU,EAAE,IAAI;YAChB,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,OAAO,EAAE,CAAC;YACV,QAAQ,EAAE,IAAI;YACd,MAAM,EAAE,EAAE;YACV,SAAS,EAAE,GAAG;YACd,SAAS,EAAE,IAAI;YACf,UAAU,EAAE,IAAI;SACjB,CAAA;IACH,CAAC,CAAC,CAAA;AACJ,CAAC"}

package/dist/fixtures.d.ts

@@ -0,0 +1,41 @@
+import type { CreateSandboxExperimentInput, SandboxFixture } from '@cat-factory/contracts';
+export { BUILTIN_SANDBOX_FIXTURES, builtinFixturesFor, builtinFixture, toSandboxFixture, type SandboxFixtureDefinition, type SandboxFixtureDifficulty, } from '@cat-factory/sandbox-fixtures';
+/**
+ * The default-loaded builtin fixtures as wire `SandboxFixture`s (the runtime seeds these
+ * when a workspace has no custom fixtures yet). `now` stamps `createdAt`.
+ */
+export declare function listBuiltinFixtures(now: number): SandboxFixture[];
+/** The synthetic baseline prompt-version id for a catalog agent kind (matches `listBaselines`). */
+export declare function baselineVersionId(agentKind: string): string;
+export interface SuggestExperimentInput {
+    /** The agent kind every cell exercises (a Sandbox catalog kind). */
+    agentKind: string;
+    /** Model catalog ids to test (the user's selection — e.g. `anthropic:claude-opus-4-8`). */
+    models: string[];
+    /** Fixture ids to run against (one or more). */
+    fixtureIds: string[];
+    /**
+     * Prompt-version ids to test. Defaults to just the shipped baseline for the agent, so the
+     * suggestion answers "which model is best?" out of the box; pass candidate lineage ids to
+     * also answer "does a better prompt help?".
+     */
+    promptVersionIds?: string[];
+    /** Judge model catalog id; omit to let the API default it (latest Claude). */
+    judgeModel?: string;
+    /** Repeats per cell (variance); defaults to 1. */
+    repeats?: number;
+    /** Experiment name; defaults to a label derived from the agent. */
+    name?: string;
+    /** Optional hard token budget for the whole experiment. */
+    budgetTokens?: number | null;
+}
+/**
+ * Build a ready-to-create experiment for "run these selected models and prompts against
+ * these selected fixtures, mapped to this selected agent". Pure: it assembles a
+ * {@link CreateSandboxExperimentInput} (the matrix is the cartesian product of prompt
+ * versions × models × fixtures) without dispatching anything — the caller POSTs it to the
+ * experiments API. Throws on an empty model/fixture selection so a non-runnable suggestion
+ * can't be created.
+ */
+export declare function suggestExperiment(input: SuggestExperimentInput): CreateSandboxExperimentInput;
+//# sourceMappingURL=fixtures.d.ts.map

package/dist/fixtures.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixtures.d.ts","sourceRoot":"","sources":["../src/fixtures.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,4BAA4B,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAA;AAS1F,OAAO,EACL,wBAAwB,EACxB,kBAAkB,EAClB,cAAc,EACd,gBAAgB,EAChB,KAAK,wBAAwB,EAC7B,KAAK,wBAAwB,GAC9B,MAAM,+BAA+B,CAAA;AAEtC;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,MAAM,GAAG,cAAc,EAAE,CAEjE;AAED,mGAAmG;AACnG,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAG3D;AAED,MAAM,WAAW,sBAAsB;IACrC,oEAAoE;IACpE,SAAS,EAAE,MAAM,CAAA;IACjB,2FAA2F;IAC3F,MAAM,EAAE,MAAM,EAAE,CAAA;IAChB,gDAAgD;IAChD,UAAU,EAAE,MAAM,EAAE,CAAA;IACpB;;;;OAIG;IACH,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC3B,8EAA8E;IAC9E,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,kDAAkD;IAClD,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,mEAAmE;IACnE,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,2DAA2D;IAC3D,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;CAC7B;AAED;;;;;;;GAOG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,sBAAsB,GAAG,4BAA4B,CAwB7F"}

package/dist/fixtures.js

@@ -0,0 +1,50 @@
+import { BUILTIN_SANDBOX_FIXTURES, toSandboxFixture } from '@cat-factory/sandbox-fixtures';
+import { sandboxKindMeta } from './baselines.js';
+// The Sandbox loads its builtin fixtures from the workspace `@cat-factory/sandbox-fixtures`
+// package by default — that package is the single source of truth for the hand-authored,
+// graded no-repo fixtures. Re-exported here so every consumer imports them (and the
+// experiment-suggestion helper) from `@cat-factory/sandbox`.
+export { BUILTIN_SANDBOX_FIXTURES, builtinFixturesFor, builtinFixture, toSandboxFixture, } from '@cat-factory/sandbox-fixtures';
+/**
+ * The default-loaded builtin fixtures as wire `SandboxFixture`s (the runtime seeds these
+ * when a workspace has no custom fixtures yet). `now` stamps `createdAt`.
+ */
+export function listBuiltinFixtures(now) {
+    return BUILTIN_SANDBOX_FIXTURES.map((def) => toSandboxFixture(def, now));
+}
+/** The synthetic baseline prompt-version id for a catalog agent kind (matches `listBaselines`). */
+export function baselineVersionId(agentKind) {
+    const meta = sandboxKindMeta(agentKind);
+    return `baseline:${meta?.basePromptId ?? agentKind}`;
+}
+/**
+ * Build a ready-to-create experiment for "run these selected models and prompts against
+ * these selected fixtures, mapped to this selected agent". Pure: it assembles a
+ * {@link CreateSandboxExperimentInput} (the matrix is the cartesian product of prompt
+ * versions × models × fixtures) without dispatching anything — the caller POSTs it to the
+ * experiments API. Throws on an empty model/fixture selection so a non-runnable suggestion
+ * can't be created.
+ */
+export function suggestExperiment(input) {
+    if (input.models.length === 0)
+        throw new Error('suggestExperiment: at least one model is required');
+    if (input.fixtureIds.length === 0)
+        throw new Error('suggestExperiment: at least one fixture is required');
+    const meta = sandboxKindMeta(input.agentKind);
+    const promptVersionIds = input.promptVersionIds && input.promptVersionIds.length > 0
+        ? input.promptVersionIds
+        : [baselineVersionId(input.agentKind)];
+    return {
+        name: input.name ?? `${meta?.label ?? input.agentKind} — sandbox run`,
+        agentKind: input.agentKind,
+        matrix: {
+            promptVersionIds,
+            models: input.models,
+            fixtureIds: input.fixtureIds,
+        },
+        ...(input.judgeModel ? { judgeModel: input.judgeModel } : {}),
+        repeats: input.repeats ?? 1,
+        budgetTokens: input.budgetTokens ?? null,
+    };
+}
+//# sourceMappingURL=fixtures.js.map

package/dist/fixtures.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixtures.js","sourceRoot":"","sources":["../src/fixtures.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,wBAAwB,EAAE,gBAAgB,EAAE,MAAM,+BAA+B,CAAA;AAC1F,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAA;AAEhD,4FAA4F;AAC5F,yFAAyF;AACzF,oFAAoF;AACpF,6DAA6D;AAE7D,OAAO,EACL,wBAAwB,EACxB,kBAAkB,EAClB,cAAc,EACd,gBAAgB,GAGjB,MAAM,+BAA+B,CAAA;AAEtC;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAW;IAC7C,OAAO,wBAAwB,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,gBAAgB,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;AAC1E,CAAC;AAED,mGAAmG;AACnG,MAAM,UAAU,iBAAiB,CAAC,SAAiB;IACjD,MAAM,IAAI,GAAG,eAAe,CAAC,SAAS,CAAC,CAAA;IACvC,OAAO,YAAY,IAAI,EAAE,YAAY,IAAI,SAAS,EAAE,CAAA;AACtD,CAAC;AAyBD;;;;;;;GAOG;AACH,MAAM,UAAU,iBAAiB,CAAC,KAA6B;IAC7D,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC;QAC3B,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAA;IACtE,IAAI,KAAK,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC;QAC/B,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAA;IAExE,MAAM,IAAI,GAAG,eAAe,CAAC,KAAK,CAAC,SAAS,CAAC,CAAA;IAC7C,MAAM,gBAAgB,GACpB,KAAK,CAAC,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC;QACzD,CAAC,CAAC,KAAK,CAAC,gBAAgB;QACxB,CAAC,CAAC,CAAC,iBAAiB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAA;IAE1C,OAAO;QACL,IAAI,EAAE,KAAK,CAAC,IAAI,IAAI,GAAG,IAAI,EAAE,KAAK,IAAI,KAAK,CAAC,SAAS,gBAAgB;QACrE,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,MAAM,EAAE;YACN,gBAAgB;YAChB,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,UAAU,EAAE,KAAK,CAAC,UAAU;SAC7B;QACD,GAAG,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7D,OAAO,EAAE,KAAK,CAAC,OAAO,IAAI,CAAC;QAC3B,YAAY,EAAE,KAAK,CAAC,YAAY,IAAI,IAAI;KACzC,CAAA;AACH,CAAC"}

package/dist/index.d.ts

@@ -0,0 +1,6 @@
+export { type SandboxTaskType, type Rubric, type RubricDimension, type ExpectationScore, HIGH_IMPACT_THRESHOLD, TRICKY_THRESHOLD, rubricFor, weightedTotal, scoreExpectations, renderExpectationBrief, } from './rubrics.js';
+export { type SandboxAgentBucket, type SandboxAgentKindMeta, SANDBOX_AGENT_KINDS, sandboxKindMeta, baselinePromptText, listBaselines, } from './baselines.js';
+export { type NewVersionFields, firstVersionFromBaseline, nextVersion, versionLabel, filterByLabels, } from './promptVersions.logic.js';
+export { type ExpandDeps, cellCount, expandMatrix, isRunnableMatrix } from './matrix.logic.js';
+export { type SuggestExperimentInput, type SandboxFixtureDefinition, type SandboxFixtureDifficulty, BUILTIN_SANDBOX_FIXTURES, builtinFixturesFor, builtinFixture, toSandboxFixture, listBuiltinFixtures, baselineVersionId, suggestExperiment, } from './fixtures.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAMA,OAAO,EACL,KAAK,eAAe,EACpB,KAAK,MAAM,EACX,KAAK,eAAe,EACpB,KAAK,gBAAgB,EACrB,qBAAqB,EACrB,gBAAgB,EAChB,SAAS,EACT,aAAa,EACb,iBAAiB,EACjB,sBAAsB,GACvB,MAAM,cAAc,CAAA;AAErB,OAAO,EACL,KAAK,kBAAkB,EACvB,KAAK,oBAAoB,EACzB,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,aAAa,GACd,MAAM,gBAAgB,CAAA;AAEvB,OAAO,EACL,KAAK,gBAAgB,EACrB,wBAAwB,EACxB,WAAW,EACX,YAAY,EACZ,cAAc,GACf,MAAM,2BAA2B,CAAA;AAElC,OAAO,EAAE,KAAK,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAA;AAE9F,OAAO,EACL,KAAK,sBAAsB,EAC3B,KAAK,wBAAwB,EAC7B,KAAK,wBAAwB,EAC7B,wBAAwB,EACxB,kBAAkB,EAClB,cAAc,EACd,gBAAgB,EAChB,mBAAmB,EACnB,iBAAiB,EACjB,iBAAiB,GAClB,MAAM,eAAe,CAAA"}

package/dist/index.js

@@ -0,0 +1,11 @@
+// @cat-factory/sandbox — the parallel prompt/model testing surface. This package is
+// deliberately isolated: it depends on kernel (ports), contracts (wire types) and
+// agents (baseline prompts), and nothing in the core product depends on it, so the
+// whole feature can be lifted out later. This entry re-exports the pure domain logic;
+// the run driver + judge service (which consume the executor seams) build on top.
+export { HIGH_IMPACT_THRESHOLD, TRICKY_THRESHOLD, rubricFor, weightedTotal, scoreExpectations, renderExpectationBrief, } from './rubrics.js';
+export { SANDBOX_AGENT_KINDS, sandboxKindMeta, baselinePromptText, listBaselines, } from './baselines.js';
+export { firstVersionFromBaseline, nextVersion, versionLabel, filterByLabels, } from './promptVersions.logic.js';
+export { cellCount, expandMatrix, isRunnableMatrix } from './matrix.logic.js';
+export { BUILTIN_SANDBOX_FIXTURES, builtinFixturesFor, builtinFixture, toSandboxFixture, listBuiltinFixtures, baselineVersionId, suggestExperiment, } from './fixtures.js';
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,oFAAoF;AACpF,kFAAkF;AAClF,mFAAmF;AACnF,sFAAsF;AACtF,kFAAkF;AAElF,OAAO,EAKL,qBAAqB,EACrB,gBAAgB,EAChB,SAAS,EACT,aAAa,EACb,iBAAiB,EACjB,sBAAsB,GACvB,MAAM,cAAc,CAAA;AAErB,OAAO,EAGL,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,aAAa,GACd,MAAM,gBAAgB,CAAA;AAEvB,OAAO,EAEL,wBAAwB,EACxB,WAAW,EACX,YAAY,EACZ,cAAc,GACf,MAAM,2BAA2B,CAAA;AAElC,OAAO,EAAmB,SAAS,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAA;AAE9F,OAAO,EAIL,wBAAwB,EACxB,kBAAkB,EAClB,cAAc,EACd,gBAAgB,EAChB,mBAAmB,EACnB,iBAAiB,EACjB,iBAAiB,GAClB,MAAM,eAAe,CAAA"}

package/dist/matrix.logic.d.ts

@@ -0,0 +1,19 @@
+import type { SandboxExperiment, SandboxMatrix, SandboxRun } from '@cat-factory/kernel';
+/** The number of cells a matrix expands to (for the pre-launch cost estimate). */
+export declare function cellCount(matrix: SandboxMatrix, repeats: number): number;
+export interface ExpandDeps {
+    /** Mint a unique run id; called once per cell (pass index so callers can vary it). */
+    makeId: (index: number) => string;
+    /** The frozen `name@vN` label for a prompt version id (resolved by the service). */
+    labelFor: (promptVersionId: string) => string;
+    now: number;
+}
+/**
+ * Expand an experiment into queued {@link SandboxRun} cells. The product (prompt ×
+ * model × fixture × repeat) is emitted in a stable order (prompt-major) so a results
+ * grid renders consistently. Each cell starts `queued` with all outcome fields null.
+ */
+export declare function expandMatrix(experiment: Pick<SandboxExperiment, 'id' | 'matrix' | 'repeats'>, deps: ExpandDeps): SandboxRun[];
+/** A non-empty matrix references at least one of each axis. */
+export declare function isRunnableMatrix(matrix: SandboxMatrix): boolean;
+//# sourceMappingURL=matrix.logic.d.ts.map

package/dist/matrix.logic.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"matrix.logic.d.ts","sourceRoot":"","sources":["../src/matrix.logic.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,aAAa,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAA;AAOvF,kFAAkF;AAClF,wBAAgB,SAAS,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAExE;AAED,MAAM,WAAW,UAAU;IACzB,sFAAsF;IACtF,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,MAAM,CAAA;IACjC,oFAAoF;IACpF,QAAQ,EAAE,CAAC,eAAe,EAAE,MAAM,KAAK,MAAM,CAAA;IAC7C,GAAG,EAAE,MAAM,CAAA;CACZ;AAED;;;;GAIG;AACH,wBAAgB,YAAY,CAC1B,UAAU,EAAE,IAAI,CAAC,iBAAiB,EAAE,IAAI,GAAG,QAAQ,GAAG,SAAS,CAAC,EAChE,IAAI,EAAE,UAAU,GACf,UAAU,EAAE,CAkCd;AAED,+DAA+D;AAC/D,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,aAAa,GAAG,OAAO,CAI/D"}

package/dist/matrix.logic.js

@@ -0,0 +1,53 @@
+// Pure expansion of an experiment's matrix into individual run cells. One cell per
+// (prompt version × model × fixture × repeat). The durable fan-out driver consumes
+// these queued skeletons; everything time/identity-dependent is injected so the
+// expansion is deterministic and unit-testable.
+/** The number of cells a matrix expands to (for the pre-launch cost estimate). */
+export function cellCount(matrix, repeats) {
+    return matrix.promptVersionIds.length * matrix.models.length * matrix.fixtureIds.length * repeats;
+}
+/**
+ * Expand an experiment into queued {@link SandboxRun} cells. The product (prompt ×
+ * model × fixture × repeat) is emitted in a stable order (prompt-major) so a results
+ * grid renders consistently. Each cell starts `queued` with all outcome fields null.
+ */
+export function expandMatrix(experiment, deps) {
+    const { promptVersionIds, models, fixtureIds } = experiment.matrix;
+    const runs = [];
+    let index = 0;
+    for (const promptVersionId of promptVersionIds) {
+        for (const model of models) {
+            for (const fixtureId of fixtureIds) {
+                for (let repeatIndex = 0; repeatIndex < experiment.repeats; repeatIndex++) {
+                    runs.push({
+                        id: deps.makeId(index),
+                        experimentId: experiment.id,
+                        promptVersionId,
+                        model,
+                        fixtureId,
+                        repeatIndex,
+                        status: 'queued',
+                        outputText: null,
+                        usage: null,
+                        latencyMs: null,
+                        branch: null,
+                        prUrl: null,
+                        diff: null,
+                        error: null,
+                        seedSha: null,
+                        promptLabel: deps.labelFor(promptVersionId),
+                        startedAt: null,
+                        finishedAt: null,
+                    });
+                    index++;
+                }
+            }
+        }
+    }
+    return runs;
+}
+/** A non-empty matrix references at least one of each axis. */
+export function isRunnableMatrix(matrix) {
+    return (matrix.promptVersionIds.length > 0 && matrix.models.length > 0 && matrix.fixtureIds.length > 0);
+}
+//# sourceMappingURL=matrix.logic.js.map

package/dist/matrix.logic.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"matrix.logic.js","sourceRoot":"","sources":["../src/matrix.logic.ts"],"names":[],"mappings":"AAEA,mFAAmF;AACnF,mFAAmF;AACnF,gFAAgF;AAChF,gDAAgD;AAEhD,kFAAkF;AAClF,MAAM,UAAU,SAAS,CAAC,MAAqB,EAAE,OAAe;IAC9D,OAAO,MAAM,CAAC,gBAAgB,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,OAAO,CAAA;AACnG,CAAC;AAUD;;;;GAIG;AACH,MAAM,UAAU,YAAY,CAC1B,UAAgE,EAChE,IAAgB;IAEhB,MAAM,EAAE,gBAAgB,EAAE,MAAM,EAAE,UAAU,EAAE,GAAG,UAAU,CAAC,MAAM,CAAA;IAClE,MAAM,IAAI,GAAiB,EAAE,CAAA;IAC7B,IAAI,KAAK,GAAG,CAAC,CAAA;IACb,KAAK,MAAM,eAAe,IAAI,gBAAgB,EAAE,CAAC;QAC/C,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;gBACnC,KAAK,IAAI,WAAW,GAAG,CAAC,EAAE,WAAW,GAAG,UAAU,CAAC,OAAO,EAAE,WAAW,EAAE,EAAE,CAAC;oBAC1E,IAAI,CAAC,IAAI,CAAC;wBACR,EAAE,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC;wBACtB,YAAY,EAAE,UAAU,CAAC,EAAE;wBAC3B,eAAe;wBACf,KAAK;wBACL,SAAS;wBACT,WAAW;wBACX,MAAM,EAAE,QAAQ;wBAChB,UAAU,EAAE,IAAI;wBAChB,KAAK,EAAE,IAAI;wBACX,SAAS,EAAE,IAAI;wBACf,MAAM,EAAE,IAAI;wBACZ,KAAK,EAAE,IAAI;wBACX,IAAI,EAAE,IAAI;wBACV,KAAK,EAAE,IAAI;wBACX,OAAO,EAAE,IAAI;wBACb,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,eAAe,CAAC;wBAC3C,SAAS,EAAE,IAAI;wBACf,UAAU,EAAE,IAAI;qBACjB,CAAC,CAAA;oBACF,KAAK,EAAE,CAAA;gBACT,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAA;AACb,CAAC;AAED,+DAA+D;AAC/D,MAAM,UAAU,gBAAgB,CAAC,MAAqB;IACpD,OAAO,CACL,MAAM,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAC/F,CAAA;AACH,CAAC"}

package/dist/promptVersions.logic.d.ts

@@ -0,0 +1,17 @@
+import type { SandboxPromptVersion } from '@cat-factory/kernel';
+export interface NewVersionFields {
+    /** The id to assign the new version row. */
+    id: string;
+    createdAt: number;
+    createdBy: string | null;
+    labels?: string[];
+}
+/** Clone a baseline (or any version's text) into a fresh candidate lineage at version 1. */
+export declare function firstVersionFromBaseline(source: Pick<SandboxPromptVersion, 'agentKind' | 'systemText' | 'basePromptId'>, name: string, fields: NewVersionFields): SandboxPromptVersion;
+/** Append a new version onto an existing lineage from an edited system prompt. */
+export declare function nextVersion(parent: SandboxPromptVersion, systemText: string, fields: NewVersionFields): SandboxPromptVersion;
+/** The canonical `name@vN` label for a stored version (frozen onto each run). */
+export declare function versionLabel(version: SandboxPromptVersion): string;
+/** Filter versions to those carrying every one of the given labels (AND semantics). */
+export declare function filterByLabels(versions: SandboxPromptVersion[], labels: string[]): SandboxPromptVersion[];
+//# sourceMappingURL=promptVersions.logic.d.ts.map

package/dist/promptVersions.logic.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"promptVersions.logic.d.ts","sourceRoot":"","sources":["../src/promptVersions.logic.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAA;AAQ/D,MAAM,WAAW,gBAAgB;IAC/B,4CAA4C;IAC5C,EAAE,EAAE,MAAM,CAAA;IACV,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAA;IACxB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,4FAA4F;AAC5F,wBAAgB,wBAAwB,CACtC,MAAM,EAAE,IAAI,CAAC,oBAAoB,EAAE,WAAW,GAAG,YAAY,GAAG,cAAc,CAAC,EAC/E,IAAI,EAAE,MAAM,EACZ,MAAM,EAAE,gBAAgB,GACvB,oBAAoB,CAgBtB;AAED,kFAAkF;AAClF,wBAAgB,WAAW,CACzB,MAAM,EAAE,oBAAoB,EAC5B,UAAU,EAAE,MAAM,EAClB,MAAM,EAAE,gBAAgB,GACvB,oBAAoB,CAgBtB;AAED,iFAAiF;AACjF,wBAAgB,YAAY,CAAC,OAAO,EAAE,oBAAoB,GAAG,MAAM,CAElE;AAED,uFAAuF;AACvF,wBAAgB,cAAc,CAC5B,QAAQ,EAAE,oBAAoB,EAAE,EAChC,MAAM,EAAE,MAAM,EAAE,GACf,oBAAoB,EAAE,CAOxB"}

package/dist/promptVersions.logic.js

@@ -0,0 +1,51 @@
+/** Clone a baseline (or any version's text) into a fresh candidate lineage at version 1. */
+export function firstVersionFromBaseline(source, name, fields) {
+    return {
+        id: fields.id,
+        lineageId: fields.id, // v1 roots its own lineage
+        agentKind: source.agentKind,
+        name,
+        origin: 'candidate',
+        systemText: source.systemText,
+        basePromptId: source.basePromptId,
+        version: 1,
+        parentId: null,
+        labels: fields.labels ?? [],
+        createdAt: fields.createdAt,
+        createdBy: fields.createdBy,
+        archivedAt: null,
+    };
+}
+/** Append a new version onto an existing lineage from an edited system prompt. */
+export function nextVersion(parent, systemText, fields) {
+    return {
+        id: fields.id,
+        lineageId: parent.lineageId,
+        agentKind: parent.agentKind,
+        name: parent.name,
+        origin: 'candidate',
+        systemText,
+        basePromptId: parent.basePromptId,
+        version: parent.version + 1,
+        parentId: parent.id,
+        labels: fields.labels ?? [],
+        createdAt: fields.createdAt,
+        createdBy: fields.createdBy,
+        archivedAt: null,
+    };
+}
+/** The canonical `name@vN` label for a stored version (frozen onto each run). */
+export function versionLabel(version) {
+    return `${version.name}@v${version.version}`;
+}
+/** Filter versions to those carrying every one of the given labels (AND semantics). */
+export function filterByLabels(versions, labels) {
+    if (labels.length === 0)
+        return versions;
+    const wanted = labels.map((l) => l.trim().toLowerCase()).filter(Boolean);
+    return versions.filter((vsn) => {
+        const have = new Set(vsn.labels.map((l) => l.toLowerCase()));
+        return wanted.every((w) => have.has(w));
+    });
+}
+//# sourceMappingURL=promptVersions.logic.js.map

package/dist/promptVersions.logic.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"promptVersions.logic.js","sourceRoot":"","sources":["../src/promptVersions.logic.ts"],"names":[],"mappings":"AAgBA,4FAA4F;AAC5F,MAAM,UAAU,wBAAwB,CACtC,MAA+E,EAC/E,IAAY,EACZ,MAAwB;IAExB,OAAO;QACL,EAAE,EAAE,MAAM,CAAC,EAAE;QACb,SAAS,EAAE,MAAM,CAAC,EAAE,EAAE,2BAA2B;QACjD,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,IAAI;QACJ,MAAM,EAAE,WAAW;QACnB,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,YAAY,EAAE,MAAM,CAAC,YAAY;QACjC,OAAO,EAAE,CAAC;QACV,QAAQ,EAAE,IAAI;QACd,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,EAAE;QAC3B,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,UAAU,EAAE,IAAI;KACjB,CAAA;AACH,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,WAAW,CACzB,MAA4B,EAC5B,UAAkB,EAClB,MAAwB;IAExB,OAAO;QACL,EAAE,EAAE,MAAM,CAAC,EAAE;QACb,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,MAAM,EAAE,WAAW;QACnB,UAAU;QACV,YAAY,EAAE,MAAM,CAAC,YAAY;QACjC,OAAO,EAAE,MAAM,CAAC,OAAO,GAAG,CAAC;QAC3B,QAAQ,EAAE,MAAM,CAAC,EAAE;QACnB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,EAAE;QAC3B,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,UAAU,EAAE,IAAI;KACjB,CAAA;AACH,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,YAAY,CAAC,OAA6B;IACxD,OAAO,GAAG,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,OAAO,EAAE,CAAA;AAC9C,CAAC;AAED,uFAAuF;AACvF,MAAM,UAAU,cAAc,CAC5B,QAAgC,EAChC,MAAgB;IAEhB,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,QAAQ,CAAA;IACxC,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;IACxE,OAAO,QAAQ,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE;QAC7B,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAA;QAC5D,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAA;IACzC,CAAC,CAAC,CAAA;AACJ,CAAC"}

package/dist/rubrics.d.ts

@@ -0,0 +1,63 @@
+import type { SandboxExpectation } from '@cat-factory/contracts';
+/** The grading task a Sandbox agent kind maps to (drives which rubric is used). */
+export type SandboxTaskType = 'requirement-review' | 'code-review' | 'implementation';
+export interface RubricDimension {
+    key: string;
+    label: string;
+    description: string;
+    weight: number;
+}
+export interface Rubric {
+    task: SandboxTaskType;
+    dimensions: RubricDimension[];
+}
+export declare function rubricFor(task: SandboxTaskType): Rubric;
+/** Weighted mean of dimension scores (1–5), using the rubric weights. */
+export declare function weightedTotal(task: SandboxTaskType, scores: {
+    key: string;
+    score: number;
+}[]): number;
+/** An expectation is "high-impact" (a serious miss) at or above this impact rating. */
+export declare const HIGH_IMPACT_THRESHOLD = 4;
+/** An expectation is "tricky" (its catch earns the wow bonus) at or above this rating. */
+export declare const TRICKY_THRESHOLD = 4;
+export interface ExpectationScore {
+    /** Expectations the candidate output surfaced. */
+    caught: SandboxExpectation[];
+    /** Expectations the candidate output missed. */
+    missed: SandboxExpectation[];
+    /**
+     * Impact-weighted recall in [0,1]: `1 − Σ(impact of missed) / Σ(impact of all)`. Missing
+     * a high-impact item moves this far more than missing a low-impact one — the asymmetry the
+     * fixtures are graded on. 1 when there are no expectations.
+     */
+    impactRecall: number;
+    /**
+     * Trickiness-weighted "wow" bonus in [0,1]: `Σ(trickiness of caught tricky items) /
+     * Σ(trickiness of all tricky items)`. Only the genuinely tricky items (trickiness ≥
+     * {@link TRICKY_THRESHOLD}) contribute, so catching a hard-to-spot finding is rewarded
+     * while missing one is not penalized here (impact handles penalties). 1 when nothing is
+     * tricky (no wow on offer).
+     */
+    wowBonus: number;
+    /** Ids of missed expectations with impact ≥ {@link HIGH_IMPACT_THRESHOLD}. */
+    missedHighImpact: string[];
+}
+/**
+ * Deterministic, asymmetric objective score for `findings` fixtures. An expectation is
+ * "caught" when any of its `matchHints` (defaulting to its `summary`) appears in the
+ * candidate output as a contiguous run of word tokens — case/whitespace/punctuation
+ * insensitive, so `reset logic` does not match inside `preset logic`. Recorded ALONGSIDE
+ * the judge grade (never blended in); it intentionally does not penalize extra findings
+ * (that is the judge's `false_positives` dimension). The two signals are deliberately
+ * different: `impactRecall` punishes missing what matters, `wowBonus` rewards catching what
+ * is hard to spot. See {@link SandboxExpectation}.
+ */
+export declare function scoreExpectations(expectations: readonly SandboxExpectation[], output: string): ExpectationScore;
+/**
+ * Render the graded expectations into a Markdown section to append to the judge prompt —
+ * "what the judge should expect to see", with the scoring guidance the asymmetry implies.
+ * Returns an empty string when there are no expectations (an un-graded fixture).
+ */
+export declare function renderExpectationBrief(expectations: readonly SandboxExpectation[]): string;
+//# sourceMappingURL=rubrics.d.ts.map

package/dist/rubrics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"rubrics.d.ts","sourceRoot":"","sources":["../src/rubrics.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,wBAAwB,CAAA;AAQhE,mFAAmF;AACnF,MAAM,MAAM,eAAe,GAAG,oBAAoB,GAAG,aAAa,GAAG,gBAAgB,CAAA;AAErF,MAAM,WAAW,eAAe;IAC9B,GAAG,EAAE,MAAM,CAAA;IACX,KAAK,EAAE,MAAM,CAAA;IACb,WAAW,EAAE,MAAM,CAAA;IACnB,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,MAAM;IACrB,IAAI,EAAE,eAAe,CAAA;IACrB,UAAU,EAAE,eAAe,EAAE,CAAA;CAC9B;AA6GD,wBAAgB,SAAS,CAAC,IAAI,EAAE,eAAe,GAAG,MAAM,CAEvD;AAED,yEAAyE;AACzE,wBAAgB,aAAa,CAC3B,IAAI,EAAE,eAAe,EACrB,MAAM,EAAE;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,EAAE,GACvC,MAAM,CAYR;AAED,uFAAuF;AACvF,eAAO,MAAM,qBAAqB,IAAI,CAAA;AACtC,0FAA0F;AAC1F,eAAO,MAAM,gBAAgB,IAAI,CAAA;AAEjC,MAAM,WAAW,gBAAgB;IAC/B,kDAAkD;IAClD,MAAM,EAAE,kBAAkB,EAAE,CAAA;IAC5B,gDAAgD;IAChD,MAAM,EAAE,kBAAkB,EAAE,CAAA;IAC5B;;;;OAIG;IACH,YAAY,EAAE,MAAM,CAAA;IACpB;;;;;;OAMG;IACH,QAAQ,EAAE,MAAM,CAAA;IAChB,8EAA8E;IAC9E,gBAAgB,EAAE,MAAM,EAAE,CAAA;CAC3B;AAED;;;;;;;;;GASG;AACH,wBAAgB,iBAAiB,CAC/B,YAAY,EAAE,SAAS,kBAAkB,EAAE,EAC3C,MAAM,EAAE,MAAM,GACb,gBAAgB,CA2BlB;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,YAAY,EAAE,SAAS,kBAAkB,EAAE,GAAG,MAAM,CAgB1F"}

package/dist/rubrics.js

@@ -0,0 +1,207 @@
+const REQUIREMENT_REVIEW = [
+    {
+        key: 'gap_coverage',
+        label: 'Gap coverage',
+        description: 'Surfaces the genuine gaps, ambiguities and risks that would block confident implementation.',
+        weight: 3,
+    },
+    {
+        key: 'specificity',
+        label: 'Specificity & actionability',
+        description: 'Each item is concrete and phrased so a product owner can answer it directly.',
+        weight: 2,
+    },
+    {
+        key: 'no_hallucination',
+        label: 'No invented requirements',
+        description: 'Does not fabricate requirements or answers; raises questions instead of guessing.',
+        weight: 3,
+    },
+    {
+        key: 'severity_calibration',
+        label: 'Severity calibration',
+        description: 'Severity/category labels are sensible and ordered high-impact first.',
+        weight: 1,
+    },
+    {
+        key: 'signal_noise',
+        label: 'Signal vs noise',
+        description: 'Avoids trivial or duplicate items; volume matches the actual ambiguity.',
+        weight: 1,
+    },
+];
+const CODE_REVIEW = [
+    {
+        key: 'issue_detection',
+        label: 'Real-issue detection',
+        description: 'Finds the genuine correctness, security and edge-case problems in the work.',
+        weight: 3,
+    },
+    {
+        key: 'correctness',
+        label: 'Correctness of findings',
+        description: 'Findings are technically accurate and the proposed fixes are sound.',
+        weight: 3,
+    },
+    {
+        key: 'severity_order',
+        label: 'Severity ordering',
+        description: 'Orders findings blocker → nit and separates must-fix from optional.',
+        weight: 1,
+    },
+    {
+        key: 'actionability',
+        label: 'Actionability',
+        description: 'References the specific code each finding concerns; fixes are concrete.',
+        weight: 2,
+    },
+    {
+        key: 'false_positives',
+        label: 'Few false positives',
+        description: 'Does not invent problems; acknowledges sound code rather than nit-picking.',
+        weight: 2,
+    },
+];
+const IMPLEMENTATION = [
+    {
+        key: 'faithfulness',
+        label: 'Design faithfulness',
+        description: 'Implements the agreed design and resolved decisions without silent redesign.',
+        weight: 3,
+    },
+    {
+        key: 'correctness',
+        label: 'Correctness',
+        description: 'The diff is correct, handles errors/edge cases, and would plausibly pass CI.',
+        weight: 3,
+    },
+    {
+        key: 'completeness',
+        label: 'Completeness',
+        description: 'Covers the requested scope; no obvious missing pieces or stubs left behind.',
+        weight: 2,
+    },
+    {
+        key: 'scope_discipline',
+        label: 'Scope discipline',
+        description: 'Stays within scope; no speculative abstraction or unrelated churn.',
+        weight: 1,
+    },
+    {
+        key: 'code_quality',
+        label: 'Code quality',
+        description: 'Cohesive, readable, idiomatic to the surrounding codebase.',
+        weight: 1,
+    },
+];
+const RUBRICS = {
+    'requirement-review': REQUIREMENT_REVIEW,
+    'code-review': CODE_REVIEW,
+    implementation: IMPLEMENTATION,
+};
+export function rubricFor(task) {
+    return { task, dimensions: RUBRICS[task] };
+}
+/** Weighted mean of dimension scores (1–5), using the rubric weights. */
+export function weightedTotal(task, scores) {
+    const dims = RUBRICS[task];
+    let sum = 0;
+    let weight = 0;
+    for (const dim of dims) {
+        const score = scores.find((s) => s.key === dim.key)?.score;
+        if (typeof score === 'number') {
+            sum += score * dim.weight;
+            weight += dim.weight;
+        }
+    }
+    return weight === 0 ? 0 : Math.round((sum / weight) * 100) / 100;
+}
+/** An expectation is "high-impact" (a serious miss) at or above this impact rating. */
+export const HIGH_IMPACT_THRESHOLD = 4;
+/** An expectation is "tricky" (its catch earns the wow bonus) at or above this rating. */
+export const TRICKY_THRESHOLD = 4;
+/**
+ * Deterministic, asymmetric objective score for `findings` fixtures. An expectation is
+ * "caught" when any of its `matchHints` (defaulting to its `summary`) appears in the
+ * candidate output as a contiguous run of word tokens — case/whitespace/punctuation
+ * insensitive, so `reset logic` does not match inside `preset logic`. Recorded ALONGSIDE
+ * the judge grade (never blended in); it intentionally does not penalize extra findings
+ * (that is the judge's `false_positives` dimension). The two signals are deliberately
+ * different: `impactRecall` punishes missing what matters, `wowBonus` rewards catching what
+ * is hard to spot. See {@link SandboxExpectation}.
+ */
+export function scoreExpectations(expectations, output) {
+    const haystack = tokenize(output);
+    const caught = [];
+    const missed = [];
+    for (const expectation of expectations) {
+        const hints = expectation.matchHints.length > 0 ? expectation.matchHints : [expectation.summary];
+        const hit = hints.some((hint) => {
+            const needle = tokenize(hint);
+            return needle.length > 0 && containsSequence(haystack, needle);
+        });
+        (hit ? caught : missed).push(expectation);
+    }
+    const totalImpact = expectations.reduce((sum, e) => sum + e.impact, 0);
+    const missedImpact = missed.reduce((sum, e) => sum + e.impact, 0);
+    const impactRecall = totalImpact === 0 ? 1 : round2(1 - missedImpact / totalImpact);
+    const trickyTotal = expectations
+        .filter((e) => e.trickiness >= TRICKY_THRESHOLD)
+        .reduce((sum, e) => sum + e.trickiness, 0);
+    const trickyCaught = caught
+        .filter((e) => e.trickiness >= TRICKY_THRESHOLD)
+        .reduce((sum, e) => sum + e.trickiness, 0);
+    const wowBonus = trickyTotal === 0 ? 1 : round2(trickyCaught / trickyTotal);
+    const missedHighImpact = missed.filter((e) => e.impact >= HIGH_IMPACT_THRESHOLD).map((e) => e.id);
+    return { caught, missed, impactRecall, wowBonus, missedHighImpact };
+}
+/**
+ * Render the graded expectations into a Markdown section to append to the judge prompt —
+ * "what the judge should expect to see", with the scoring guidance the asymmetry implies.
+ * Returns an empty string when there are no expectations (an un-graded fixture).
+ */
+export function renderExpectationBrief(expectations) {
+    if (expectations.length === 0)
+        return '';
+    const lines = [
+        '## Expected findings (grading reference)',
+        '',
+        'A strong response should surface the following. Each is rated by **impact** (how bad it',
+        'is to miss, 1–5) and **trickiness** (how hard it is to spot, 1–5). Reward catching',
+        'high-trickiness items — those are the impressive catches. Penalize missing high-impact',
+        'items most heavily; missing a merely tricky item is a smaller concern.',
+        '',
+    ];
+    for (const e of expectations) {
+        lines.push(`- **${e.summary}** _(impact ${e.impact}, trickiness ${e.trickiness})_`);
+        if (e.detail.trim())
+            lines.push(`  - ${e.detail.trim()}`);
+    }
+    return lines.join('\n');
+}
+/** Round to 2 decimal places. */
+function round2(n) {
+    return Math.round(n * 100) / 100;
+}
+/** Lowercase alphanumeric word tokens (drops punctuation/whitespace). */
+function tokenize(text) {
+    return text.toLowerCase().match(/[a-z0-9]+/g) ?? [];
+}
+/** Whether `needle`'s tokens appear as a contiguous run within `haystack`'s tokens. */
+function containsSequence(haystack, needle) {
+    if (needle.length === 0)
+        return false;
+    for (let i = 0; i + needle.length <= haystack.length; i++) {
+        let hit = true;
+        for (let j = 0; j < needle.length; j++) {
+            if (haystack[i + j] !== needle[j]) {
+                hit = false;
+                break;
+            }
+        }
+        if (hit)
+            return true;
+    }
+    return false;
+}
+//# sourceMappingURL=rubrics.js.map

package/dist/rubrics.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"rubrics.js","sourceRoot":"","sources":["../src/rubrics.ts"],"names":[],"mappings":"AAuBA,MAAM,kBAAkB,GAAsB;IAC5C;QACE,GAAG,EAAE,cAAc;QACnB,KAAK,EAAE,cAAc;QACrB,WAAW,EACT,6FAA6F;QAC/F,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,aAAa;QAClB,KAAK,EAAE,6BAA6B;QACpC,WAAW,EAAE,8EAA8E;QAC3F,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,kBAAkB;QACvB,KAAK,EAAE,0BAA0B;QACjC,WAAW,EACT,mFAAmF;QACrF,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,sBAAsB;QAC3B,KAAK,EAAE,sBAAsB;QAC7B,WAAW,EAAE,sEAAsE;QACnF,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,cAAc;QACnB,KAAK,EAAE,iBAAiB;QACxB,WAAW,EAAE,yEAAyE;QACtF,MAAM,EAAE,CAAC;KACV;CACF,CAAA;AAED,MAAM,WAAW,GAAsB;IACrC;QACE,GAAG,EAAE,iBAAiB;QACtB,KAAK,EAAE,sBAAsB;QAC7B,WAAW,EAAE,6EAA6E;QAC1F,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,aAAa;QAClB,KAAK,EAAE,yBAAyB;QAChC,WAAW,EAAE,qEAAqE;QAClF,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,gBAAgB;QACrB,KAAK,EAAE,mBAAmB;QAC1B,WAAW,EAAE,qEAAqE;QAClF,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,eAAe;QACpB,KAAK,EAAE,eAAe;QACtB,WAAW,EAAE,yEAAyE;QACtF,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,iBAAiB;QACtB,KAAK,EAAE,qBAAqB;QAC5B,WAAW,EAAE,4EAA4E;QACzF,MAAM,EAAE,CAAC;KACV;CACF,CAAA;AAED,MAAM,cAAc,GAAsB;IACxC;QACE,GAAG,EAAE,cAAc;QACnB,KAAK,EAAE,qBAAqB;QAC5B,WAAW,EAAE,8EAA8E;QAC3F,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,aAAa;QAClB,KAAK,EAAE,aAAa;QACpB,WAAW,EAAE,8EAA8E;QAC3F,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,cAAc;QACnB,KAAK,EAAE,cAAc;QACrB,WAAW,EAAE,6EAA6E;QAC1F,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,kBAAkB;QACvB,KAAK,EAAE,kBAAkB;QACzB,WAAW,EAAE,oEAAoE;QACjF,MAAM,EAAE,CAAC;KACV;IACD;QACE,GAAG,EAAE,cAAc;QACnB,KAAK,EAAE,cAAc;QACrB,WAAW,EAAE,4DAA4D;QACzE,MAAM,EAAE,CAAC;KACV;CACF,CAAA;AAED,MAAM,OAAO,GAA+C;IAC1D,oBAAoB,EAAE,kBAAkB;IACxC,aAAa,EAAE,WAAW;IAC1B,cAAc,EAAE,cAAc;CAC/B,CAAA;AAED,MAAM,UAAU,SAAS,CAAC,IAAqB;IAC7C,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,CAAC,IAAI,CAAC,EAAE,CAAA;AAC5C,CAAC;AAED,yEAAyE;AACzE,MAAM,UAAU,aAAa,CAC3B,IAAqB,EACrB,MAAwC;IAExC,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IAC1B,IAAI,GAAG,GAAG,CAAC,CAAA;IACX,IAAI,MAAM,GAAG,CAAC,CAAA;IACd,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC,GAAG,CAAC,EAAE,KAAK,CAAA;QAC1D,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC9B,GAAG,IAAI,KAAK,GAAG,GAAG,CAAC,MAAM,CAAA;YACzB,MAAM,IAAI,GAAG,CAAC,MAAM,CAAA;QACtB,CAAC;IACH,CAAC;IACD,OAAO,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,GAAG,MAAM,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,CAAA;AAClE,CAAC;AAED,uFAAuF;AACvF,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAA;AACtC,0FAA0F;AAC1F,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAA;AAyBjC;;;;;;;;;GASG;AACH,MAAM,UAAU,iBAAiB,CAC/B,YAA2C,EAC3C,MAAc;IAEd,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAA;IACjC,MAAM,MAAM,GAAyB,EAAE,CAAA;IACvC,MAAM,MAAM,GAAyB,EAAE,CAAA;IACvC,KAAK,MAAM,WAAW,IAAI,YAAY,EAAE,CAAC;QACvC,MAAM,KAAK,GAAG,WAAW,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,OAAO,CAAC,CAAA;QAChG,MAAM,GAAG,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE;YAC9B,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAA;YAC7B,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,gBAAgB,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAA;QAChE,CAAC,CAAC,CACD;QAAA,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAA;IAC5C,CAAC;IAED,MAAM,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAA;IACtE,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAA;IACjE,MAAM,YAAY,GAAG,WAAW,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,YAAY,GAAG,WAAW,CAAC,CAAA;IAEnF,MAAM,WAAW,GAAG,YAAY;SAC7B,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,IAAI,gBAAgB,CAAC;SAC/C,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAA;IAC5C,MAAM,YAAY,GAAG,MAAM;SACxB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,IAAI,gBAAgB,CAAC;SAC/C,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAA;IAC5C,MAAM,QAAQ,GAAG,WAAW,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,YAAY,GAAG,WAAW,CAAC,CAAA;IAE3E,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,qBAAqB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAA;IACjG,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,YAAY,EAAE,QAAQ,EAAE,gBAAgB,EAAE,CAAA;AACrE,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,sBAAsB,CAAC,YAA2C;IAChF,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAA;IACxC,MAAM,KAAK,GAAG;QACZ,0CAA0C;QAC1C,EAAE;QACF,yFAAyF;QACzF,oFAAoF;QACpF,wFAAwF;QACxF,wEAAwE;QACxE,EAAE;KACH,CAAA;IACD,KAAK,MAAM,CAAC,IAAI,YAAY,EAAE,CAAC;QAC7B,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,OAAO,eAAe,CAAC,CAAC,MAAM,gBAAgB,CAAC,CAAC,UAAU,IAAI,CAAC,CAAA;QACnF,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE;YAAE,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC,CAAA;IAC3D,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AACzB,CAAC;AAED,iCAAiC;AACjC,SAAS,MAAM,CAAC,CAAS;IACvB,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,CAAA;AAClC,CAAC;AAED,yEAAyE;AACzE,SAAS,QAAQ,CAAC,IAAY;IAC5B,OAAO,IAAI,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,CAAA;AACrD,CAAC;AAED,uFAAuF;AACvF,SAAS,gBAAgB,CAAC,QAAkB,EAAE,MAAgB;IAC5D,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAA;IACrC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC1D,IAAI,GAAG,GAAG,IAAI,CAAA;QACd,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACvC,IAAI,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;gBAClC,GAAG,GAAG,KAAK,CAAA;gBACX,MAAK;YACP,CAAC;QACH,CAAC;QACD,IAAI,GAAG;YAAE,OAAO,IAAI,CAAA;IACtB,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cat-factory/sandbox",
-  "version": "0.7.2",
+  "version": "0.7.3",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/kibertoad/cat-factory.git",
@@ -24,10 +24,10 @@
     "access": "public"
   },
   "dependencies": {
-    "@cat-factory/agents": "0.7.2",
+    "@cat-factory/contracts": "0.7.2",
     "@cat-factory/kernel": "0.7.2",
     "@cat-factory/sandbox-fixtures": "0.7.2",
-    "@cat-factory/contracts": "0.7.2"
+    "@cat-factory/agents": "0.7.2"
   },
   "devDependencies": {
     "typescript": "7.0.1-rc",