@axiom-lattice/local-stores 2.0.4 → 2.0.6

package/jest.config.js

@@ -0,0 +1,19 @@
+module.exports = {
+  preset: "ts-jest",
+  testEnvironment: "node",
+  roots: ["<rootDir>/src"],
+  testMatch: ["**/__tests__/**/*.test.ts"],
+  transform: {
+    "^.+\\.ts$": [
+      "ts-jest",
+      {
+        isolatedModules: true,
+      },
+    ],
+  },
+  moduleNameMapper: {
+    "^@axiom-lattice/protocols$": "<rootDir>/../protocols/src/index.ts",
+  },
+  transformIgnorePatterns: [],
+  testTimeout: 10000,
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@axiom-lattice/local-stores",
-  "version": "2.0.4",
+  "version": "2.0.6",
   "description": "Local SQLite-based stores for Axiom Lattice framework",
   "main": "dist/index.js",
   "module": "dist/index.mjs",
@@ -25,13 +25,19 @@
     "@types/sql.js": "^1.4.11",
     "sql.js": "^1.14.1",
     "uuid": "^14.0.1",
-    "@axiom-lattice/core": "3.0.4",
+    "@axiom-lattice/core": "3.0.6",
     "@axiom-lattice/protocols": "3.0.2"
   },
   "devDependencies": {
+    "@types/jest": "^29.5.14",
     "@types/node": "^20.11.24",
     "@types/uuid": "^9.0.8",
+    "@typescript-eslint/eslint-plugin": "^7.2.0",
+    "@typescript-eslint/parser": "^7.2.0",
+    "eslint": "^8.57.0",
+    "jest": "^29.7.0",
     "rimraf": "^5.0.5",
+    "ts-jest": "^29.4.0",
     "tsup": "^8.0.1",
     "typescript": "^5.4.2"
   },
@@ -39,6 +45,7 @@
     "build": "tsup src/index.ts --format cjs,esm --dts --sourcemap",
     "dev": "tsup src/index.ts --format cjs,esm --dts --watch",
     "lint": "eslint src",
+    "test": "jest",
     "clean": "rimraf dist"
   }
 }

package/src/__tests__/LocalEvalStore.test.ts

@@ -0,0 +1,96 @@
+/**
+ * LocalEvalStore project name uniqueness tests
+ *
+ * Verifies the per-tenant unique name constraint enforced by the
+ * `idx_lt_eval_projects_tenant_name` index (eval-{agent-id} convention).
+ */
+
+import { describe, expect, it, beforeEach, afterEach } from "@jest/globals";
+import { tmpdir } from "os";
+import path from "path";
+import { initDatabase, closeDatabase, ensureTable } from "../database";
+import { LocalEvalStore, DDL } from "../stores/LocalEvalStore";
+
+// sql.js persists to disk via DatabaseWrapper.save(), so each test needs a
+// unique file path (":memory:" is treated as a plain relative path).
+let dbPath = "";
+const nextDbPath = () => {
+  dbPath = path.join(tmpdir(), `lt-eval-test-${Date.now()}-${Math.random().toString(36).slice(2)}.db`);
+  return dbPath;
+};
+
+describe("LocalEvalStore project name uniqueness", () => {
+  let store: LocalEvalStore;
+
+  beforeEach(async () => {
+    const db = await initDatabase({ dbPath: nextDbPath() });
+    store = new LocalEvalStore(db);
+  });
+
+  afterEach(() => {
+    closeDatabase();
+  });
+
+  it("rejects a duplicate project name in the same tenant", async () => {
+    await store.createProject("tenant-1", "uuid-1", {
+      name: "eval-a1",
+      judgeModelConfig: {},
+      targetServerConfig: {},
+    });
+
+    await expect(
+      store.createProject("tenant-1", "uuid-2", {
+        name: "eval-a1",
+        judgeModelConfig: {},
+        targetServerConfig: {},
+      })
+    ).rejects.toThrow();
+  });
+
+  it("allows the same name in a different tenant", async () => {
+    await store.createProject("tenant-1", "uuid-1", {
+      name: "eval-a1",
+      judgeModelConfig: {},
+      targetServerConfig: {},
+    });
+
+    await expect(
+      store.createProject("tenant-2", "uuid-2", {
+        name: "eval-a1",
+        judgeModelConfig: {},
+        targetServerConfig: {},
+      })
+    ).resolves.toBeDefined();
+  });
+
+  it("deduplicates existing duplicate rows on construction and keeps child history", async () => {
+    closeDatabase();
+    const db = await initDatabase({ dbPath: nextDbPath() });
+    // Recreate the schema WITHOUT the unique index, simulating an old install
+    // that already holds duplicate rows.
+    ensureTable(db, DDL);
+    db.exec(`
+      INSERT INTO lt_eval_projects (id, tenant_id, name, judge_model_config, target_server_config, concurrency, created_at, updated_at)
+      VALUES
+        ('dup-old', 'tenant-1', 'eval-a1', '{}', '{}', 3, '2026-01-01T00:00:00.000Z', '2026-01-01T00:00:00.000Z'),
+        ('dup-new', 'tenant-1', 'eval-a1', '{}', '{}', 3, '2026-01-02T00:00:00.000Z', '2026-01-02T00:00:00.000Z');
+      INSERT INTO lt_eval_suites (id, tenant_id, project_id, name, created_at, updated_at)
+      VALUES ('suite-1', 'tenant-1', 'dup-old', 'refund-path', '2026-01-01T00:00:00.000Z', '2026-01-01T00:00:00.000Z');
+      INSERT INTO lt_eval_runs (id, project_id, tenant_id, status, concurrency, total_cases, created_at)
+      VALUES ('run-1', 'dup-old', 'tenant-1', 'completed', 1, 1, '2026-01-01T00:00:00.000Z');
+    `);
+
+    const cleaned = new LocalEvalStore(db);
+    const projects = await cleaned.getProjectsByTenant("tenant-1");
+
+    // Kept the newest duplicate (dup-new, created 2026-01-02).
+    expect(projects).toHaveLength(1);
+    expect(projects[0].id).toBe("dup-new");
+
+    // Child history was repointed to the kept project, not deleted.
+    const suites = db.prepare(`SELECT project_id FROM lt_eval_suites WHERE id = 'suite-1'`).get() as { project_id: string };
+    expect(suites.project_id).toBe("dup-new");
+    const runs = db.prepare(`SELECT project_id FROM lt_eval_runs WHERE id = 'run-1'`).get() as { project_id: string };
+    expect(runs.project_id).toBe("dup-new");
+  });
+});

package/src/stores/LocalEvalStore.ts

@@ -15,11 +15,12 @@ import {
   CreateEvalRunRequest,
   EvalRunResult,
   EvalProjectReport,
+  InterruptPolicy,
 } from "@axiom-lattice/protocols";
 import { ensureTable, nowISO, parseISO } from "../database";
 import { randomUUID } from "crypto";
 
-const DDL = `
+export const DDL = `
 CREATE TABLE IF NOT EXISTS lt_eval_projects (
   id TEXT NOT NULL,
   tenant_id TEXT NOT NULL,
@@ -56,6 +57,7 @@ CREATE TABLE IF NOT EXISTS lt_eval_cases (
   output_type TEXT NOT NULL DEFAULT 'message_content',
   content_assertion TEXT NOT NULL DEFAULT '',
   rubrics TEXT,
+  interrupt_policy TEXT,
   created_at TEXT NOT NULL,
   updated_at TEXT NOT NULL,
   PRIMARY KEY (tenant_id, id)
@@ -71,6 +73,7 @@ CREATE TABLE IF NOT EXISTS lt_eval_runs (
   total_cases INTEGER NOT NULL DEFAULT 0,
   passed_cases INTEGER NOT NULL DEFAULT 0,
   failed_cases INTEGER NOT NULL DEFAULT 0,
+  interrupted_cases INTEGER NOT NULL DEFAULT 0,
   avg_score REAL NOT NULL DEFAULT 0,
   error TEXT,
   holdout INTEGER NOT NULL DEFAULT 0,
@@ -96,6 +99,7 @@ CREATE TABLE IF NOT EXISTS lt_eval_run_results (
   messages TEXT DEFAULT '[]',
   logs TEXT DEFAULT '[]',
   error TEXT,
+  interrupted INTEGER NOT NULL DEFAULT 0,
   created_at TEXT NOT NULL,
   PRIMARY KEY (run_id, id)
 );
@@ -123,6 +127,85 @@ export class LocalEvalStore implements EvalStore {
   constructor(db: DatabaseWrapper) {
     this.db = db;
     ensureTable(db, DDL);
+    // Existing installs: add HITL interrupt tracking columns.
+    // sql.js bundles an older SQLite that does not support
+    // `ADD COLUMN IF NOT EXISTS`, so check via PRAGMA instead.
+    this.ensureColumn("lt_eval_runs", "interrupted_cases", "interrupted_cases INTEGER NOT NULL DEFAULT 0");
+    this.ensureColumn("lt_eval_run_results", "interrupted", "interrupted INTEGER NOT NULL DEFAULT 0");
+    this.ensureColumn("lt_eval_cases", "interrupt_policy", "interrupt_policy TEXT");
+    // Project name must be unique per tenant (eval-{agent-id} convention).
+    // Repoint child rows (suites, runs) of duplicate projects onto the kept
+    // project (newest by created_at, tie-break rowid — mirrors the pg
+    // migration's (created_at, id) DESC policy) BEFORE deleting, so eval
+    // history is preserved. SQLite has no FK cascade here, so this is manual.
+    this.db.exec(`
+      UPDATE lt_eval_suites
+      SET project_id = (
+        SELECT p2.id FROM lt_eval_projects p2
+        WHERE p2.tenant_id = (
+          SELECT p.tenant_id FROM lt_eval_projects p WHERE p.id = lt_eval_suites.project_id
+        )
+        AND p2.name = (
+          SELECT p.name FROM lt_eval_projects p WHERE p.id = lt_eval_suites.project_id
+        )
+        ORDER BY p2.created_at DESC, p2.rowid DESC
+        LIMIT 1
+      )
+      WHERE project_id IN (
+        SELECT p3.id FROM lt_eval_projects p3
+        WHERE p3.id <> (
+          SELECT p2.id FROM lt_eval_projects p2
+          WHERE p2.tenant_id = p3.tenant_id AND p2.name = p3.name
+          ORDER BY p2.created_at DESC, p2.rowid DESC
+          LIMIT 1
+        )
+      )
+    `);
+    this.db.exec(`
+      UPDATE lt_eval_runs
+      SET project_id = (
+        SELECT p2.id FROM lt_eval_projects p2
+        WHERE p2.tenant_id = (
+          SELECT p.tenant_id FROM lt_eval_projects p WHERE p.id = lt_eval_runs.project_id
+        )
+        AND p2.name = (
+          SELECT p.name FROM lt_eval_projects p WHERE p.id = lt_eval_runs.project_id
+        )
+        ORDER BY p2.created_at DESC, p2.rowid DESC
+        LIMIT 1
+      )
+      WHERE project_id IN (
+        SELECT p3.id FROM lt_eval_projects p3
+        WHERE p3.id <> (
+          SELECT p2.id FROM lt_eval_projects p2
+          WHERE p2.tenant_id = p3.tenant_id AND p2.name = p3.name
+          ORDER BY p2.created_at DESC, p2.rowid DESC
+          LIMIT 1
+        )
+      )
+    `);
+    this.db.exec(`
+      DELETE FROM lt_eval_projects
+      WHERE id <> (
+        SELECT p2.id FROM lt_eval_projects p2
+        WHERE p2.tenant_id = lt_eval_projects.tenant_id
+          AND p2.name = lt_eval_projects.name
+        ORDER BY p2.created_at DESC, p2.rowid DESC
+        LIMIT 1
+      )
+    `);
+    this.db.exec(`
+      CREATE UNIQUE INDEX IF NOT EXISTS idx_lt_eval_projects_tenant_name
+      ON lt_eval_projects(tenant_id, name)
+    `);
+  }
+
+  /** Add a column if it does not exist (SQLite version compatible). */
+  private ensureColumn(table: string, column: string, ddl: string): void {
+    const cols = this.db.prepare(`PRAGMA table_info(${table})`).all() as Array<{ name: string }>;
+    if (!cols.some((c) => c.name === column)) {
+      this.db.exec(`ALTER TABLE ${table} ADD COLUMN ${ddl}`);
+    }
   }
 
   // -------------------------------------------------------------------------
@@ -252,13 +335,14 @@ export class LocalEvalStore implements EvalStore {
   async createCase(tenantId: string, suiteId: string, id: string, data: CreateEvalCaseRequest): Promise<EvalCase> {
     const actualId = id || randomUUID();
     this.db.prepare(
-      `INSERT INTO lt_eval_cases (id, tenant_id, suite_id, input_message, input_files, steps, output_type, content_assertion, rubrics, created_at, updated_at)
-       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
+      `INSERT INTO lt_eval_cases (id, tenant_id, suite_id, input_message, input_files, steps, output_type, content_assertion, rubrics, interrupt_policy, created_at, updated_at)
+       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
     ).run(actualId, tenantId, suiteId, data.inputMessage,
       data.inputFiles ? JSON.stringify(data.inputFiles) : null,
       JSON.stringify(data.steps), data.outputType || "message_content",
       data.contentAssertion || "",
       data.rubrics ? JSON.stringify(data.rubrics) : null,
+      data.interruptPolicy ? JSON.stringify(data.interruptPolicy) : null,
       nowISO(), nowISO());
     return (await this.getCaseById(tenantId, actualId))!;
   }
@@ -275,6 +359,7 @@ export class LocalEvalStore implements EvalStore {
     if (updates.outputType !== undefined) { set.push("output_type = ?"); vals.push(updates.outputType); }
     if (updates.contentAssertion !== undefined) { set.push("content_assertion = ?"); vals.push(updates.contentAssertion); }
     if (updates.rubrics !== undefined) { set.push("rubrics = ?"); vals.push(JSON.stringify(updates.rubrics)); }
+    if (updates.interruptPolicy !== undefined) { set.push("interrupt_policy = ?"); vals.push(updates.interruptPolicy ? JSON.stringify(updates.interruptPolicy) : null); }
     set.push("updated_at = ?"); vals.push(nowISO());
     vals.push(tenantId, id);
 
@@ -320,7 +405,7 @@ export class LocalEvalStore implements EvalStore {
   }
 
   async updateRunStatus(tenantId: string, id: string, updates: {
-    status?: EvalRun["status"]; passedCases?: number; failedCases?: number; avgScore?: number; error?: string; completedAt?: Date;
+    status?: EvalRun["status"]; passedCases?: number; failedCases?: number; interruptedCases?: number; avgScore?: number; error?: string; completedAt?: Date;
   }): Promise<EvalRun | null> {
     const existing = await this.getRunById(tenantId, id);
     if (!existing) return null;
@@ -330,6 +415,7 @@ export class LocalEvalStore implements EvalStore {
     if (updates.status !== undefined) { set.push("status = ?"); vals.push(updates.status); }
     if (updates.passedCases !== undefined) { set.push("passed_cases = ?"); vals.push(updates.passedCases); }
     if (updates.failedCases !== undefined) { set.push("failed_cases = ?"); vals.push(updates.failedCases); }
+    if (updates.interruptedCases !== undefined) { set.push("interrupted_cases = ?"); vals.push(updates.interruptedCases); }
     if (updates.avgScore !== undefined) { set.push("avg_score = ?"); vals.push(updates.avgScore); }
     if (updates.error !== undefined) { set.push("error = ?"); vals.push(updates.error || null); }
     if (updates.completedAt !== undefined) { set.push("completed_at = ?"); vals.push(updates.completedAt ? updates.completedAt.toISOString() : null); }
@@ -371,13 +457,13 @@ export class LocalEvalStore implements EvalStore {
   async createRunResult(tenantId: string, runId: string, id: string, data: Omit<EvalRunResult, "id" | "runId" | "createdAt">): Promise<EvalRunResult> {
     const actualId = id || randomUUID();
     this.db.prepare(
-      `INSERT INTO lt_eval_run_results (id, run_id, suite_name, case_id, pass, score, summary, dimension_results, duration_ms, messages, logs, error, created_at)
-       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
+      `INSERT INTO lt_eval_run_results (id, run_id, suite_name, case_id, pass, score, summary, dimension_results, duration_ms, messages, logs, error, interrupted, created_at)
+       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
     ).run(actualId, runId, data.suiteName, data.caseId || null,
       data.pass ? 1 : 0, data.score, data.summary || null,
       JSON.stringify(data.dimensionResults || []), data.durationMs ?? null,
       JSON.stringify(data.messages || []), JSON.stringify(data.logs || []),
-      data.error || null, nowISO());
+      data.error || null, data.interrupted ? 1 : 0, nowISO());
     return (await this.getRunResultById(tenantId, actualId))!;
   }
 
@@ -396,6 +482,7 @@ export class LocalEvalStore implements EvalStore {
     if (updates.messages !== undefined) { set.push("messages = ?"); vals.push(JSON.stringify(updates.messages)); }
     if (updates.logs !== undefined) { set.push("logs = ?"); vals.push(JSON.stringify(updates.logs)); }
     if (updates.error !== undefined) { set.push("error = ?"); vals.push(updates.error || null); }
+    if (updates.interrupted !== undefined) { set.push("interrupted = ?"); vals.push(updates.interrupted ? 1 : 0); }
     if (set.length === 0) return this.getRunResultById(tenantId, id);
 
     vals.push(id, tenantId);
@@ -478,6 +565,7 @@ export class LocalEvalStore implements EvalStore {
       outputType: (row.output_type as EvalCase["outputType"]) || "message_content",
       contentAssertion: (row.content_assertion as string) || "",
       rubrics: parseOptionalJson<Array<{ name: string; weight: number; description: string }>>(row.rubrics),
+      interruptPolicy: parseOptionalJson<InterruptPolicy>(row.interrupt_policy),
       createdAt: parseISO(row.created_at as string),
       updatedAt: parseISO(row.updated_at as string),
     };
@@ -493,6 +581,7 @@ export class LocalEvalStore implements EvalStore {
       totalCases: (row.total_cases as number) ?? 0,
       passedCases: (row.passed_cases as number) ?? 0,
       failedCases: (row.failed_cases as number) ?? 0,
+      interruptedCases: (row.interrupted_cases as number) ?? 0,
       avgScore: (row.avg_score as number) ?? 0,
       error: row.error as string | undefined,
       holdout: row.holdout ? Boolean(row.holdout) : undefined,
@@ -518,6 +607,7 @@ export class LocalEvalStore implements EvalStore {
       messages: parseOptionalJson<Array<{ role: string; content: string; id?: string }>>(row.messages),
       logs: parseOptionalJson<Array<{ timestamp: string; level: string; message: string; data?: unknown }>>(row.logs),
       error: row.error as string | undefined,
+      interrupted: (row.interrupted as number) === 1,
       createdAt: parseISO(row.created_at as string),
     };
   }