@deepagents/evals 7.1.2 → 7.1.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/store/index.ts"],"names":[],"mappings":"AAEA,OAAO,EACL,YAAY,EAGb,MAAM,aAAa,CAAC;AAIrB,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,MAAM;IACrB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,CAAC;IACvC,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,MAAM,EAAE,SAAS,CAAC;IAClB,OAAO,EAAE,UAAU,GAAG,IAAI,CAAC;CAC5B;AAED,MAAM,WAAW,OAAO;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,QAAQ,EAAE,OAAO,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAED,MAAM,WAAW,cAAe,SAAQ,OAAO;IAC7C,MAAM,EAAE,KAAK,CAAC;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CAC9E;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB;AAED,MAAM,WAAW,UAAU;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,cAAc,EAAE,MAAM,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,SAAS;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,SAAS;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,QAAA,MAAM,YAAY,6CAA8C,CAAC;AAEjE,KAAK,SAAS,GAAG,CAAC,OAAO,YAAY,CAAC,CAAC,MAAM,CAAC,CAAC;AAyL/C,qBAAa,QAAQ;;gBA6DP,QAAQ,CAAC,EAAE,MAAM,GAAG,YAAY;IA8G5C,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,QAAQ;IASnC,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS;IAS1C,WAAW,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI;IAI3C,SAAS,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI;IAIzC,SAAS,CAAC,GAAG,EAAE;QACb,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC;QACb,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KAClC,GAAG,MAAM;IAgBV,SAAS,CACP,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,WAAW,GAAG,QAAQ,EAC9B,OAAO,CAAC,EAAE,UAAU,GACnB,IAAI;IAMP,SAAS,CAAC,KAAK,EAAE,QAAQ,EAAE,GAAG,IAAI;IAsBlC,UAAU,CAAC,MAAM,EAAE,SAAS,EAAE,GAAG,IAAI;IAWrC,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS;IAKzC,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE;IAWpC,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,EAAE;IAQlC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,SAAM,GAAG,cAAc,EAAE;IAiCjE,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,SAAM,GAAG,UAAU;IA6DzD,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS;IASnD,qBAAqB,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS;IAkB1E,UAAU,IAAI,QAAQ,EAAE;IAOxB,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,SAAS;IAiBtD,WAAW,IAAI,SAAS,EAAE;IAO1B,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS;IAS5C,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;IAI9B,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;IAS1B,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;IAI3B,WAAW,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;CAG9B"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/store/index.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,YAAY,EAGb,MAAM,aAAa,CAAC;AAMrB,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,MAAM;IACrB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,CAAC;IACvC,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,MAAM,EAAE,SAAS,CAAC;IAClB,OAAO,EAAE,UAAU,GAAG,IAAI,CAAC;CAC5B;AAED,MAAM,WAAW,OAAO;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,QAAQ,EAAE,OAAO,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAED,MAAM,WAAW,cAAe,SAAQ,OAAO;IAC7C,MAAM,EAAE,KAAK,CAAC;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CAC9E;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB;AAED,MAAM,WAAW,UAAU;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACnC,cAAc,EAAE,MAAM,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,SAAS;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,SAAS;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,QAAA,MAAM,YAAY,6CAA8C,CAAC;AAEjE,KAAK,SAAS,GAAG,CAAC,OAAO,YAAY,CAAC,CAAC,MAAM,CAAC,CAAC;AAyL/C,qBAAa,QAAQ;;gBA6CP,QAAQ,CAAC,EAAE,MAAM,GAAG,YAAY;IA4B5C,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,QAAQ;IASnC,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS;IAS1C,WAAW,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI;IAI3C,SAAS,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI;IAIzC,SAAS,CAAC,GAAG,EAAE;QACb,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC;QACb,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KAClC,GAAG,MAAM;IAgBV,SAAS,CACP,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,WAAW,GAAG,QAAQ,EAC9B,OAAO,CAAC,EAAE,UAAU,GACnB,IAAI;IAMP,SAAS,CAAC,KAAK,EAAE,QAAQ,EAAE,GAAG,IAAI;IA0BlC,UAAU,CAAC,MAAM,EAAE,SAAS,EAAE,GAAG,IAAI;IAerC,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS;IAKzC,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE;IAWpC,QAAQ,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,EAAE;IAQlC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,SAAM,GAAG,cAAc,EAAE;IAiCjE,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,SAAM,GAAG,UAAU;IA6DzD,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS;IASnD,qBAAqB,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS;IAkB1E,UAAU,IAAI,QAAQ,EAAE;IAOxB,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,SAAS;IAiBtD,WAAW,IAAI,SAAS,EAAE;IAO1B,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS;IAS5C,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;IAI9B,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;IAa1B,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;IAI3B,WAAW,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI;CAG9B"}
@@ -1,4 +1,6 @@
1
1
  // packages/evals/src/store/index.ts
2
+ import { Declaration } from "@zukhruf/sqlite-drift";
3
+ import { transaction } from "@zukhruf/sqlite-transaction";
2
4
  import { mkdirSync } from "node:fs";
3
5
  import { dirname } from "node:path";
4
6
  import {
@@ -6,9 +8,10 @@ import {
6
8
  } from "node:sqlite";
7
9
 
8
10
  // packages/evals/src/store/ddl.sqlite.sql
9
- var ddl_sqlite_default = "PRAGMA journal_mode = WAL;\nPRAGMA synchronous = NORMAL;\nPRAGMA foreign_keys = ON;\n\nCREATE TABLE IF NOT EXISTS suites (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000)\n);\n\nCREATE TABLE IF NOT EXISTS runs (\n id TEXT PRIMARY KEY,\n suite_id TEXT NOT NULL,\n name TEXT NOT NULL,\n model TEXT NOT NULL,\n config TEXT,\n started_at INTEGER NOT NULL,\n finished_at INTEGER,\n status TEXT NOT NULL DEFAULT 'running' CHECK(status IN ('running', 'completed', 'failed')),\n summary TEXT,\n FOREIGN KEY (suite_id) REFERENCES suites(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_runs_suite_id ON runs(suite_id);\nCREATE INDEX IF NOT EXISTS idx_runs_started_at ON runs(started_at);\n\nCREATE TABLE IF NOT EXISTS cases (\n id TEXT PRIMARY KEY,\n run_id TEXT NOT NULL,\n idx INTEGER NOT NULL,\n input TEXT NOT NULL,\n output TEXT,\n expected TEXT,\n latency_ms INTEGER,\n tokens_in INTEGER,\n tokens_out INTEGER,\n error TEXT,\n FOREIGN KEY (run_id) REFERENCES runs(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_cases_run_id ON cases(run_id);\n\nCREATE TABLE IF NOT EXISTS scores (\n id TEXT PRIMARY KEY,\n case_id TEXT NOT NULL,\n scorer_name TEXT NOT NULL,\n score REAL NOT NULL,\n reason TEXT,\n FOREIGN KEY (case_id) REFERENCES cases(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_scores_case_id ON scores(case_id);\n\nCREATE TABLE IF NOT EXISTS prompts (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL UNIQUE,\n content TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000)\n);\n\nCREATE INDEX IF NOT EXISTS idx_prompts_created_at ON prompts(created_at);\n";
11
+ var ddl_sqlite_default = "CREATE TABLE IF NOT EXISTS suites (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000)\n);\n\nCREATE TABLE IF NOT EXISTS runs (\n id TEXT PRIMARY KEY,\n suite_id TEXT NOT NULL,\n name TEXT NOT NULL,\n model TEXT NOT NULL,\n config TEXT,\n started_at INTEGER NOT NULL,\n finished_at INTEGER,\n status TEXT NOT NULL DEFAULT 'running' CHECK(status IN ('running', 'completed', 'failed')),\n summary TEXT,\n FOREIGN KEY (suite_id) REFERENCES suites(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_runs_suite_id ON runs(suite_id);\nCREATE INDEX IF NOT EXISTS idx_runs_started_at ON runs(started_at);\n\nCREATE TABLE IF NOT EXISTS cases (\n id TEXT PRIMARY KEY,\n run_id TEXT NOT NULL,\n idx INTEGER NOT NULL,\n input TEXT NOT NULL,\n output TEXT,\n expected TEXT,\n latency_ms INTEGER,\n tokens_in INTEGER,\n tokens_out INTEGER,\n error TEXT,\n FOREIGN KEY (run_id) REFERENCES runs(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_cases_run_id ON cases(run_id);\n\nCREATE TABLE IF NOT EXISTS scores (\n id TEXT PRIMARY KEY,\n case_id TEXT NOT NULL,\n scorer_name TEXT NOT NULL,\n score REAL NOT NULL,\n reason TEXT,\n FOREIGN KEY (case_id) REFERENCES cases(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_scores_case_id ON scores(case_id);\n\nCREATE TABLE IF NOT EXISTS prompts (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n version INTEGER NOT NULL DEFAULT 1,\n content TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000),\n UNIQUE(name, version)\n);\n\nCREATE INDEX IF NOT EXISTS idx_prompts_created_at ON prompts(created_at);\nCREATE INDEX IF NOT EXISTS idx_prompts_name_version ON prompts(name, version DESC);\n";
10
12
 
11
13
  // packages/evals/src/store/index.ts
14
+ var schema = new Declaration([ddl_sqlite_default], { rows: "keep" });
12
15
  var RUN_STATUSES = ["running", "completed", "failed"];
13
16
  function isRunStatus(value) {
14
17
  return RUN_STATUSES.some((status) => status === value);
@@ -156,20 +159,6 @@ var RunStore = class {
156
159
  #all(columns, sql, ...params) {
157
160
  return this.#stmt(sql).all(...params).map((row) => readRow(row, columns, sql));
158
161
  }
159
- #transaction(fn) {
160
- this.#db.exec("BEGIN TRANSACTION");
161
- try {
162
- const result = fn();
163
- this.#db.exec("COMMIT");
164
- return result;
165
- } catch (error) {
166
- try {
167
- this.#db.exec("ROLLBACK");
168
- } catch {
169
- }
170
- throw error;
171
- }
172
- }
173
162
  constructor(pathOrDb) {
174
163
  if (pathOrDb instanceof DatabaseSync) {
175
164
  this.#db = pathOrDb;
@@ -178,93 +167,22 @@ var RunStore = class {
178
167
  mkdirSync(dirname(dbPath), { recursive: true });
179
168
  this.#db = new DatabaseSync(dbPath);
180
169
  }
181
- this.#db.exec(ddl_sqlite_default);
182
- this.#migrateRunsTableToSuiteRequired();
183
- this.#migratePromptsTableIfNeeded();
184
- this.#db.exec(
185
- "CREATE INDEX IF NOT EXISTS idx_prompts_name_version ON prompts(name, version DESC)"
170
+ this.#db.exec(`
171
+ PRAGMA journal_mode = WAL;
172
+ PRAGMA synchronous = NORMAL;
173
+ PRAGMA foreign_keys = ON;
174
+ PRAGMA busy_timeout = 5000;
175
+ `);
176
+ const runsTable = this.#get(
177
+ { name: "text" },
178
+ "SELECT name FROM sqlite_schema WHERE type = 'table' AND name = 'runs'"
186
179
  );
187
- }
188
- #migratePromptsTableIfNeeded() {
189
- const columns = this.#all({ name: "text" }, "PRAGMA table_info(prompts)");
190
- if (columns.length === 0) return;
191
- if (columns.some((column) => column.name === "version")) return;
192
- this.#transaction(() => {
193
- this.#db.exec("ALTER TABLE prompts RENAME TO prompts_legacy");
194
- this.#db.exec(`
195
- CREATE TABLE prompts (
196
- id TEXT PRIMARY KEY,
197
- name TEXT NOT NULL,
198
- version INTEGER NOT NULL,
199
- content TEXT NOT NULL,
200
- created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000),
201
- UNIQUE(name, version)
202
- )
203
- `);
204
- this.#db.exec(`
205
- INSERT INTO prompts (id, name, version, content, created_at)
206
- SELECT id, name, 1, content, created_at
207
- FROM prompts_legacy
208
- `);
209
- this.#db.exec("DROP TABLE prompts_legacy");
210
- this.#db.exec(
211
- "CREATE INDEX IF NOT EXISTS idx_prompts_created_at ON prompts(created_at)"
212
- );
213
- this.#db.exec(
214
- "CREATE INDEX IF NOT EXISTS idx_prompts_name_version ON prompts(name, version DESC)"
215
- );
216
- });
217
- }
218
- #migrateRunsTableToSuiteRequired() {
219
- const columns = this.#all(
220
- { name: "text", notnull: "number" },
221
- "PRAGMA table_info(runs)"
222
- );
223
- if (columns.length === 0) return;
224
- const suiteColumn = columns.find((column) => column.name === "suite_id");
225
- const hasNonNullSuite = suiteColumn?.notnull === 1;
226
- const runForeignKeys = this.#all(
227
- { from: "text", on_delete: "text", table: "text" },
228
- "PRAGMA foreign_key_list(runs)"
229
- );
230
- const suiteForeignKey = runForeignKeys.find(
231
- (fk) => fk.from === "suite_id" && fk.table === "suites"
232
- );
233
- const hasCascadeDelete = suiteForeignKey?.on_delete === "CASCADE";
234
- if (hasNonNullSuite && hasCascadeDelete) return;
235
- this.#statements.clear();
236
- this.#transaction(() => {
237
- this.#db.exec(`
238
- CREATE TABLE runs_next (
239
- id TEXT PRIMARY KEY,
240
- suite_id TEXT NOT NULL,
241
- name TEXT NOT NULL,
242
- model TEXT NOT NULL,
243
- config TEXT,
244
- started_at INTEGER NOT NULL,
245
- finished_at INTEGER,
246
- status TEXT NOT NULL DEFAULT 'running' CHECK(status IN ('running', 'completed', 'failed')),
247
- summary TEXT,
248
- FOREIGN KEY (suite_id) REFERENCES suites(id) ON DELETE CASCADE
249
- )
250
- `);
251
- this.#db.exec("DELETE FROM runs WHERE suite_id IS NULL");
252
- this.#db.exec(`
253
- INSERT INTO runs_next (id, suite_id, name, model, config, started_at, finished_at, status, summary)
254
- SELECT r.id, r.suite_id, r.name, r.model, r.config, r.started_at, r.finished_at, r.status, r.summary
255
- FROM runs r
256
- JOIN suites s ON s.id = r.suite_id
257
- `);
258
- this.#db.exec("DROP TABLE runs");
259
- this.#db.exec("ALTER TABLE runs_next RENAME TO runs");
260
- this.#db.exec(
261
- "CREATE INDEX IF NOT EXISTS idx_runs_suite_id ON runs(suite_id)"
262
- );
180
+ if (runsTable !== void 0) {
263
181
  this.#db.exec(
264
- "CREATE INDEX IF NOT EXISTS idx_runs_started_at ON runs(started_at)"
182
+ "DELETE FROM runs WHERE suite_id IS NULL OR NOT EXISTS (SELECT 1 FROM suites WHERE suites.id = runs.suite_id)"
265
183
  );
266
- });
267
- this.#statements.clear();
184
+ }
185
+ schema.apply(this.#db);
268
186
  }
269
187
  createSuite(name) {
270
188
  const id = crypto.randomUUID();
@@ -309,35 +227,43 @@ var RunStore = class {
309
227
  ).run(Date.now(), status, summary ? JSON.stringify(summary) : null, runId);
310
228
  }
311
229
  saveCases(cases) {
312
- this.#transaction(() => {
313
- const stmt = this.#stmt(
314
- "INSERT INTO cases (id, run_id, idx, input, output, expected, latency_ms, tokens_in, tokens_out, error) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)"
315
- );
316
- for (const c of cases) {
317
- stmt.run(
318
- c.id,
319
- c.run_id,
320
- c.idx,
321
- JSON.stringify(c.input),
322
- c.output,
323
- c.expected != null ? JSON.stringify(c.expected) : null,
324
- c.latency_ms,
325
- c.tokens_in,
326
- c.tokens_out,
327
- c.error ?? null
230
+ transaction(
231
+ this.#db,
232
+ () => {
233
+ const stmt = this.#stmt(
234
+ "INSERT INTO cases (id, run_id, idx, input, output, expected, latency_ms, tokens_in, tokens_out, error) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)"
328
235
  );
329
- }
330
- });
236
+ for (const c of cases) {
237
+ stmt.run(
238
+ c.id,
239
+ c.run_id,
240
+ c.idx,
241
+ JSON.stringify(c.input),
242
+ c.output,
243
+ c.expected != null ? JSON.stringify(c.expected) : null,
244
+ c.latency_ms,
245
+ c.tokens_in,
246
+ c.tokens_out,
247
+ c.error ?? null
248
+ );
249
+ }
250
+ },
251
+ { mode: "deferred" }
252
+ );
331
253
  }
332
254
  saveScores(scores) {
333
- this.#transaction(() => {
334
- const stmt = this.#stmt(
335
- "INSERT INTO scores (id, case_id, scorer_name, score, reason) VALUES (?, ?, ?, ?, ?)"
336
- );
337
- for (const s of scores) {
338
- stmt.run(s.id, s.case_id, s.scorer_name, s.score, s.reason ?? null);
339
- }
340
- });
255
+ transaction(
256
+ this.#db,
257
+ () => {
258
+ const stmt = this.#stmt(
259
+ "INSERT INTO scores (id, case_id, scorer_name, score, reason) VALUES (?, ?, ?, ?, ?)"
260
+ );
261
+ for (const s of scores) {
262
+ stmt.run(s.id, s.case_id, s.scorer_name, s.score, s.reason ?? null);
263
+ }
264
+ },
265
+ { mode: "deferred" }
266
+ );
341
267
  }
342
268
  getRun(runId) {
343
269
  const row = this.#get(runColumns, "SELECT * FROM runs WHERE id = ?", runId);
@@ -505,12 +431,16 @@ var RunStore = class {
505
431
  this.#stmt("DELETE FROM prompts WHERE id = ?").run(id);
506
432
  }
507
433
  resetRun(id) {
508
- this.#transaction(() => {
509
- this.#stmt("DELETE FROM cases WHERE run_id = ?").run(id);
510
- this.#stmt(
511
- "UPDATE runs SET status = ?, started_at = ?, finished_at = NULL, summary = NULL WHERE id = ?"
512
- ).run("running", Date.now(), id);
513
- });
434
+ transaction(
435
+ this.#db,
436
+ () => {
437
+ this.#stmt("DELETE FROM cases WHERE run_id = ?").run(id);
438
+ this.#stmt(
439
+ "UPDATE runs SET status = ?, started_at = ?, finished_at = NULL, summary = NULL WHERE id = ?"
440
+ ).run("running", Date.now(), id);
441
+ },
442
+ { mode: "deferred" }
443
+ );
514
444
  }
515
445
  deleteRun(id) {
516
446
  this.#stmt("DELETE FROM runs WHERE id = ?").run(id);
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "version": 3,
3
3
  "sources": ["../../src/store/index.ts", "../../src/store/ddl.sqlite.sql"],
4
- "sourcesContent": ["import { mkdirSync } from 'node:fs';\nimport { dirname } from 'node:path';\nimport {\n DatabaseSync,\n type SQLInputValue,\n type SQLOutputValue,\n} from 'node:sqlite';\n\nimport DDL from './ddl.sqlite.sql';\n\nexport interface SuiteRow {\n id: string;\n name: string;\n created_at: number;\n}\n\nexport interface RunRow {\n id: string;\n suite_id: string;\n name: string;\n model: string;\n config: Record<string, unknown> | null;\n started_at: number;\n finished_at: number | null;\n status: RunStatus;\n summary: RunSummary | null;\n}\n\nexport interface CaseRow {\n id: string;\n run_id: string;\n idx: number;\n input: unknown;\n output: string | null;\n expected: unknown | null;\n latency_ms: number | null;\n tokens_in: number | null;\n tokens_out: number | null;\n error: string | null;\n}\n\nexport interface CaseWithScores extends CaseRow {\n scores: Array<{ scorer_name: string; score: number; reason: string | null }>;\n}\n\nexport interface ScoreRow {\n id: string;\n case_id: string;\n scorer_name: string;\n score: number;\n reason: string | null;\n}\n\nexport interface RunSummary {\n totalCases: number;\n passCount: number;\n failCount: number;\n meanScores: Record<string, number>;\n totalLatencyMs: number;\n totalTokensIn: number;\n totalTokensOut: number;\n}\n\nexport interface PromptRow {\n id: string;\n name: string;\n version: number;\n content: string;\n created_at: number;\n}\n\nexport interface CaseData {\n id: string;\n run_id: string;\n idx: number;\n input: unknown;\n output: string | null;\n expected?: unknown;\n latency_ms: number;\n tokens_in: number;\n tokens_out: number;\n error?: string;\n}\n\nexport interface ScoreData {\n id: string;\n case_id: string;\n scorer_name: string;\n score: number;\n reason?: string;\n}\n\nconst RUN_STATUSES = ['running', 'completed', 'failed'] as const;\n\ntype RunStatus = (typeof RUN_STATUSES)[number];\n\nfunction isRunStatus(value: string): value is RunStatus {\n return RUN_STATUSES.some((status) => status === value);\n}\n\ntype SqlRow = Record<string, SQLOutputValue>;\n\n/** What node:sqlite returns for each kind of column the store reads. */\ninterface ColumnTypes {\n text: string;\n 'text?': string | null;\n number: number;\n 'number?': number | null;\n}\n\ntype ColumnType = keyof ColumnTypes;\n\ntype Columns = Readonly<Record<string, ColumnType>>;\n\ntype RowOf<C extends Columns> = { -readonly [K in keyof C]: ColumnTypes[C[K]] };\n\nfunction matchesColumn(\n value: SQLOutputValue | undefined,\n type: ColumnType,\n): boolean {\n switch (type) {\n case 'text':\n return typeof value === 'string';\n case 'text?':\n return value === null || typeof value === 'string';\n case 'number':\n return typeof value === 'number';\n case 'number?':\n return value === null || typeof value === 'number';\n }\n}\n\nfunction mismatchedColumns(row: SqlRow, columns: Columns): string[] {\n return Object.entries(columns)\n .filter(([name, type]) => !matchesColumn(row[name], type))\n .map(([name]) => name);\n}\n\nfunction hasColumns<C extends Columns>(\n row: SqlRow,\n columns: C,\n): row is SqlRow & RowOf<C> {\n return mismatchedColumns(row, columns).length === 0;\n}\n\nfunction readRow<C extends Columns>(\n row: SqlRow,\n columns: C,\n sql: string,\n): RowOf<C> {\n if (hasColumns(row, columns)) return row;\n throw new TypeError(\n `Unexpected values in columns ${mismatchedColumns(row, columns).join(', ')} of a row from: ${oneLine(sql)}`,\n );\n}\n\nfunction oneLine(sql: string): string {\n return sql.replace(/\\s+/g, ' ').trim();\n}\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n return typeof value === 'object' && value !== null && !Array.isArray(value);\n}\n\nfunction isRunSummary(value: unknown): value is RunSummary {\n return (\n isRecord(value) &&\n typeof value.totalCases === 'number' &&\n typeof value.passCount === 'number' &&\n typeof value.failCount === 'number' &&\n typeof value.totalLatencyMs === 'number' &&\n typeof value.totalTokensIn === 'number' &&\n typeof value.totalTokensOut === 'number' &&\n isRecord(value.meanScores) &&\n Object.values(value.meanScores).every((score) => typeof score === 'number')\n );\n}\n\nfunction parseJson<T>(\n text: string,\n guard: (value: unknown) => value is T,\n source: string,\n): T {\n const value: unknown = JSON.parse(text);\n if (guard(value)) return value;\n throw new TypeError(`Unexpected JSON in ${source}`);\n}\n\nconst suiteColumns = {\n id: 'text',\n name: 'text',\n created_at: 'number',\n} as const;\n\nconst runColumns = {\n id: 'text',\n suite_id: 'text',\n name: 'text',\n model: 'text',\n config: 'text?',\n started_at: 'number',\n finished_at: 'number?',\n status: 'text',\n summary: 'text?',\n} as const;\n\nconst caseColumns = {\n id: 'text',\n run_id: 'text',\n idx: 'number',\n input: 'text',\n output: 'text?',\n expected: 'text?',\n latency_ms: 'number?',\n tokens_in: 'number?',\n tokens_out: 'number?',\n error: 'text?',\n} as const;\n\nconst promptColumns = {\n id: 'text',\n name: 'text',\n version: 'number',\n content: 'text',\n created_at: 'number',\n} as const;\n\nfunction toSuite(row: RowOf<typeof suiteColumns>): SuiteRow {\n return { id: row.id, name: row.name, created_at: row.created_at };\n}\n\nfunction toRun(row: RowOf<typeof runColumns>): RunRow {\n if (!isRunStatus(row.status)) {\n throw new TypeError(`Unexpected status \"${row.status}\" of run ${row.id}`);\n }\n return {\n id: row.id,\n suite_id: row.suite_id,\n name: row.name,\n model: row.model,\n config: row.config\n ? parseJson(row.config, isRecord, `config of run ${row.id}`)\n : null,\n started_at: row.started_at,\n finished_at: row.finished_at,\n status: row.status,\n summary: row.summary\n ? parseJson(row.summary, isRunSummary, `summary of run ${row.id}`)\n : null,\n };\n}\n\nfunction toCase(row: RowOf<typeof caseColumns>): CaseRow {\n const input: unknown = JSON.parse(row.input);\n const expected: unknown = row.expected ? JSON.parse(row.expected) : null;\n return {\n id: row.id,\n run_id: row.run_id,\n idx: row.idx,\n input,\n output: row.output,\n expected,\n latency_ms: row.latency_ms,\n tokens_in: row.tokens_in,\n tokens_out: row.tokens_out,\n error: row.error,\n };\n}\n\nfunction toPrompt(row: RowOf<typeof promptColumns>): PromptRow {\n return {\n id: row.id,\n name: row.name,\n version: row.version,\n content: row.content,\n created_at: row.created_at,\n };\n}\n\nexport class RunStore {\n #db: DatabaseSync;\n #statements = new Map<string, ReturnType<DatabaseSync['prepare']>>();\n\n #stmt(sql: string): ReturnType<DatabaseSync['prepare']> {\n let stmt = this.#statements.get(sql);\n if (!stmt) {\n stmt = this.#db.prepare(sql);\n this.#statements.set(sql, stmt);\n }\n return stmt;\n }\n\n #get<const C extends Columns>(\n columns: C,\n sql: string,\n ...params: SQLInputValue[]\n ): RowOf<C> | undefined {\n const row = this.#stmt(sql).get(...params);\n return row === undefined ? undefined : readRow(row, columns, sql);\n }\n\n /** Reads an aggregate query, which returns exactly one row. */\n #one<const C extends Columns>(\n columns: C,\n sql: string,\n ...params: SQLInputValue[]\n ): RowOf<C> {\n const row = this.#get(columns, sql, ...params);\n if (row === undefined) {\n throw new Error(`Expected one row from: ${oneLine(sql)}`);\n }\n return row;\n }\n\n #all<const C extends Columns>(\n columns: C,\n sql: string,\n ...params: SQLInputValue[]\n ): RowOf<C>[] {\n return this.#stmt(sql)\n .all(...params)\n .map((row) => readRow(row, columns, sql));\n }\n\n #transaction<T>(fn: () => T): T {\n this.#db.exec('BEGIN TRANSACTION');\n try {\n const result = fn();\n this.#db.exec('COMMIT');\n return result;\n } catch (error) {\n try {\n this.#db.exec('ROLLBACK');\n } catch {\n // Preserve the operation error when SQLite already ended the transaction.\n }\n throw error;\n }\n }\n\n constructor(pathOrDb?: string | DatabaseSync) {\n if (pathOrDb instanceof DatabaseSync) {\n this.#db = pathOrDb;\n } else {\n const dbPath = pathOrDb ?? '.evals/store.db';\n mkdirSync(dirname(dbPath), { recursive: true });\n this.#db = new DatabaseSync(dbPath);\n }\n this.#db.exec(DDL);\n this.#migrateRunsTableToSuiteRequired();\n this.#migratePromptsTableIfNeeded();\n this.#db.exec(\n 'CREATE INDEX IF NOT EXISTS idx_prompts_name_version ON prompts(name, version DESC)',\n );\n }\n\n #migratePromptsTableIfNeeded(): void {\n const columns = this.#all({ name: 'text' }, 'PRAGMA table_info(prompts)');\n\n if (columns.length === 0) return;\n if (columns.some((column) => column.name === 'version')) return;\n\n this.#transaction(() => {\n this.#db.exec('ALTER TABLE prompts RENAME TO prompts_legacy');\n this.#db.exec(`\n CREATE TABLE prompts (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n version INTEGER NOT NULL,\n content TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000),\n UNIQUE(name, version)\n )\n `);\n this.#db.exec(`\n INSERT INTO prompts (id, name, version, content, created_at)\n SELECT id, name, 1, content, created_at\n FROM prompts_legacy\n `);\n this.#db.exec('DROP TABLE prompts_legacy');\n this.#db.exec(\n 'CREATE INDEX IF NOT EXISTS idx_prompts_created_at ON prompts(created_at)',\n );\n this.#db.exec(\n 'CREATE INDEX IF NOT EXISTS idx_prompts_name_version ON prompts(name, version DESC)',\n );\n });\n }\n\n #migrateRunsTableToSuiteRequired(): void {\n const columns = this.#all(\n { name: 'text', notnull: 'number' },\n 'PRAGMA table_info(runs)',\n );\n\n if (columns.length === 0) return;\n\n const suiteColumn = columns.find((column) => column.name === 'suite_id');\n const hasNonNullSuite = suiteColumn?.notnull === 1;\n\n const runForeignKeys = this.#all(\n { from: 'text', on_delete: 'text', table: 'text' },\n 'PRAGMA foreign_key_list(runs)',\n );\n const suiteForeignKey = runForeignKeys.find(\n (fk) => fk.from === 'suite_id' && fk.table === 'suites',\n );\n const hasCascadeDelete = suiteForeignKey?.on_delete === 'CASCADE';\n\n if (hasNonNullSuite && hasCascadeDelete) return;\n\n this.#statements.clear();\n this.#transaction(() => {\n this.#db.exec(`\n CREATE TABLE runs_next (\n id TEXT PRIMARY KEY,\n suite_id TEXT NOT NULL,\n name TEXT NOT NULL,\n model TEXT NOT NULL,\n config TEXT,\n started_at INTEGER NOT NULL,\n finished_at INTEGER,\n status TEXT NOT NULL DEFAULT 'running' CHECK(status IN ('running', 'completed', 'failed')),\n summary TEXT,\n FOREIGN KEY (suite_id) REFERENCES suites(id) ON DELETE CASCADE\n )\n `);\n\n // Drop legacy orphaned runs that do not belong to a suite.\n this.#db.exec('DELETE FROM runs WHERE suite_id IS NULL');\n\n this.#db.exec(`\n INSERT INTO runs_next (id, suite_id, name, model, config, started_at, finished_at, status, summary)\n SELECT r.id, r.suite_id, r.name, r.model, r.config, r.started_at, r.finished_at, r.status, r.summary\n FROM runs r\n JOIN suites s ON s.id = r.suite_id\n `);\n\n this.#db.exec('DROP TABLE runs');\n this.#db.exec('ALTER TABLE runs_next RENAME TO runs');\n this.#db.exec(\n 'CREATE INDEX IF NOT EXISTS idx_runs_suite_id ON runs(suite_id)',\n );\n this.#db.exec(\n 'CREATE INDEX IF NOT EXISTS idx_runs_started_at ON runs(started_at)',\n );\n });\n this.#statements.clear();\n }\n\n createSuite(name: string): SuiteRow {\n const id = crypto.randomUUID();\n const now = Date.now();\n this.#stmt(\n 'INSERT INTO suites (id, name, created_at) VALUES (?, ?, ?)',\n ).run(id, name, now);\n return { id, name, created_at: now };\n }\n\n getSuite(id: string): SuiteRow | undefined {\n const row = this.#get(\n suiteColumns,\n 'SELECT * FROM suites WHERE id = ?',\n id,\n );\n return row ? toSuite(row) : undefined;\n }\n\n renameSuite(id: string, name: string): void {\n this.#stmt('UPDATE suites SET name = ? WHERE id = ?').run(name, id);\n }\n\n renameRun(id: string, name: string): void {\n this.#stmt('UPDATE runs SET name = ? WHERE id = ?').run(name, id);\n }\n\n createRun(run: {\n suite_id: string;\n name: string;\n model: string;\n config?: Record<string, unknown>;\n }): string {\n const id = crypto.randomUUID();\n const now = Date.now();\n this.#stmt(\n 'INSERT INTO runs (id, suite_id, name, model, config, started_at) VALUES (?, ?, ?, ?, ?, ?)',\n ).run(\n id,\n run.suite_id,\n run.name,\n run.model,\n run.config ? JSON.stringify(run.config) : null,\n now,\n );\n return id;\n }\n\n finishRun(\n runId: string,\n status: 'completed' | 'failed',\n summary?: RunSummary,\n ): void {\n this.#stmt(\n 'UPDATE runs SET finished_at = ?, status = ?, summary = ? WHERE id = ?',\n ).run(Date.now(), status, summary ? JSON.stringify(summary) : null, runId);\n }\n\n saveCases(cases: CaseData[]): void {\n this.#transaction(() => {\n const stmt = this.#stmt(\n 'INSERT INTO cases (id, run_id, idx, input, output, expected, latency_ms, tokens_in, tokens_out, error) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)',\n );\n for (const c of cases) {\n stmt.run(\n c.id,\n c.run_id,\n c.idx,\n JSON.stringify(c.input),\n c.output,\n c.expected != null ? JSON.stringify(c.expected) : null,\n c.latency_ms,\n c.tokens_in,\n c.tokens_out,\n c.error ?? null,\n );\n }\n });\n }\n\n saveScores(scores: ScoreData[]): void {\n this.#transaction(() => {\n const stmt = this.#stmt(\n 'INSERT INTO scores (id, case_id, scorer_name, score, reason) VALUES (?, ?, ?, ?, ?)',\n );\n for (const s of scores) {\n stmt.run(s.id, s.case_id, s.scorer_name, s.score, s.reason ?? null);\n }\n });\n }\n\n getRun(runId: string): RunRow | undefined {\n const row = this.#get(runColumns, 'SELECT * FROM runs WHERE id = ?', runId);\n return row ? toRun(row) : undefined;\n }\n\n listRuns(suiteId?: string): RunRow[] {\n const rows = suiteId\n ? this.#all(\n runColumns,\n 'SELECT * FROM runs WHERE suite_id = ? ORDER BY started_at',\n suiteId,\n )\n : this.#all(runColumns, 'SELECT * FROM runs ORDER BY started_at');\n return rows.map(toRun);\n }\n\n getCases(runId: string): CaseRow[] {\n return this.#all(\n caseColumns,\n 'SELECT * FROM cases WHERE run_id = ? ORDER BY idx',\n runId,\n ).map(toCase);\n }\n\n getFailingCases(runId: string, threshold = 0.5): CaseWithScores[] {\n const rows = this.#all(\n {\n ...caseColumns,\n scorer_name: 'text',\n score: 'number',\n score_reason: 'text?',\n },\n `SELECT c.*, s.scorer_name, s.score, s.reason as score_reason\n FROM cases c\n JOIN scores s ON s.case_id = c.id\n WHERE c.run_id = ? AND s.score < ?\n ORDER BY c.idx`,\n runId,\n threshold,\n );\n\n const caseMap = new Map<string, CaseWithScores>();\n for (const row of rows) {\n let c = caseMap.get(row.id);\n if (!c) {\n c = { ...toCase(row), scores: [] };\n caseMap.set(row.id, c);\n }\n c.scores.push({\n scorer_name: row.scorer_name,\n score: row.score,\n reason: row.score_reason,\n });\n }\n return Array.from(caseMap.values());\n }\n\n getRunSummary(runId: string, threshold = 0.5): RunSummary {\n const totals = this.#one(\n {\n totalCases: 'number',\n totalLatencyMs: 'number',\n totalTokensIn: 'number',\n totalTokensOut: 'number',\n },\n `SELECT\n COUNT(DISTINCT c.id) as totalCases,\n COALESCE(SUM(c.latency_ms), 0) as totalLatencyMs,\n COALESCE(SUM(c.tokens_in), 0) as totalTokensIn,\n COALESCE(SUM(c.tokens_out), 0) as totalTokensOut\n FROM cases c WHERE c.run_id = ?`,\n runId,\n );\n\n const scorerMeans = this.#all(\n { scorer_name: 'text', meanScore: 'number' },\n `SELECT s.scorer_name, AVG(s.score) as meanScore\n FROM scores s\n JOIN cases c ON c.id = s.case_id\n WHERE c.run_id = ?\n GROUP BY s.scorer_name`,\n runId,\n );\n\n const meanScores: Record<string, number> = {};\n for (const row of scorerMeans) {\n meanScores[row.scorer_name] = row.meanScore;\n }\n\n const passFail = this.#all(\n { id: 'text', minScore: 'number' },\n `SELECT c.id,\n MIN(s.score) as minScore\n FROM cases c\n JOIN scores s ON s.case_id = c.id\n WHERE c.run_id = ?\n GROUP BY c.id`,\n runId,\n );\n\n let passCount = 0;\n let failCount = 0;\n for (const row of passFail) {\n if (row.minScore >= threshold) passCount++;\n else failCount++;\n }\n\n return {\n totalCases: totals.totalCases,\n passCount,\n failCount,\n meanScores,\n totalLatencyMs: totals.totalLatencyMs,\n totalTokensIn: totals.totalTokensIn,\n totalTokensOut: totals.totalTokensOut,\n };\n }\n\n findSuiteByName(name: string): SuiteRow | undefined {\n const row = this.#get(\n suiteColumns,\n 'SELECT * FROM suites WHERE name = ? ORDER BY created_at DESC LIMIT 1',\n name,\n );\n return row ? toSuite(row) : undefined;\n }\n\n getLatestCompletedRun(suiteId: string, model?: string): RunRow | undefined {\n const row = model\n ? this.#get(\n runColumns,\n 'SELECT * FROM runs WHERE suite_id = ? AND status = ? AND model = ? ORDER BY started_at DESC LIMIT 1',\n suiteId,\n 'completed',\n model,\n )\n : this.#get(\n runColumns,\n 'SELECT * FROM runs WHERE suite_id = ? AND status = ? ORDER BY started_at DESC LIMIT 1',\n suiteId,\n 'completed',\n );\n return row ? toRun(row) : undefined;\n }\n\n listSuites(): SuiteRow[] {\n return this.#all(\n suiteColumns,\n 'SELECT * FROM suites ORDER BY created_at DESC',\n ).map(toSuite);\n }\n\n createPrompt(name: string, content: string): PromptRow {\n const id = crypto.randomUUID();\n const now = Date.now();\n\n const { latestVersion } = this.#one(\n { latestVersion: 'number?' },\n 'SELECT MAX(version) as latestVersion FROM prompts WHERE name = ?',\n name,\n );\n const version = (latestVersion ?? 0) + 1;\n\n this.#stmt(\n 'INSERT INTO prompts (id, name, version, content, created_at) VALUES (?, ?, ?, ?, ?)',\n ).run(id, name, version, content, now);\n return { id, name, version, content, created_at: now };\n }\n\n listPrompts(): PromptRow[] {\n return this.#all(\n promptColumns,\n 'SELECT * FROM prompts ORDER BY name COLLATE NOCASE ASC, version DESC',\n ).map(toPrompt);\n }\n\n getPrompt(id: string): PromptRow | undefined {\n const row = this.#get(\n promptColumns,\n 'SELECT * FROM prompts WHERE id = ?',\n id,\n );\n return row ? toPrompt(row) : undefined;\n }\n\n deletePrompt(id: string): void {\n this.#stmt('DELETE FROM prompts WHERE id = ?').run(id);\n }\n\n resetRun(id: string): void {\n this.#transaction(() => {\n this.#stmt('DELETE FROM cases WHERE run_id = ?').run(id);\n this.#stmt(\n 'UPDATE runs SET status = ?, started_at = ?, finished_at = NULL, summary = NULL WHERE id = ?',\n ).run('running', Date.now(), id);\n });\n }\n\n deleteRun(id: string): void {\n this.#stmt('DELETE FROM runs WHERE id = ?').run(id);\n }\n\n deleteSuite(id: string): void {\n this.#stmt('DELETE FROM suites WHERE id = ?').run(id);\n }\n}\n", "PRAGMA journal_mode = WAL;\nPRAGMA synchronous = NORMAL;\nPRAGMA foreign_keys = ON;\n\nCREATE TABLE IF NOT EXISTS suites (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000)\n);\n\nCREATE TABLE IF NOT EXISTS runs (\n id TEXT PRIMARY KEY,\n suite_id TEXT NOT NULL,\n name TEXT NOT NULL,\n model TEXT NOT NULL,\n config TEXT,\n started_at INTEGER NOT NULL,\n finished_at INTEGER,\n status TEXT NOT NULL DEFAULT 'running' CHECK(status IN ('running', 'completed', 'failed')),\n summary TEXT,\n FOREIGN KEY (suite_id) REFERENCES suites(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_runs_suite_id ON runs(suite_id);\nCREATE INDEX IF NOT EXISTS idx_runs_started_at ON runs(started_at);\n\nCREATE TABLE IF NOT EXISTS cases (\n id TEXT PRIMARY KEY,\n run_id TEXT NOT NULL,\n idx INTEGER NOT NULL,\n input TEXT NOT NULL,\n output TEXT,\n expected TEXT,\n latency_ms INTEGER,\n tokens_in INTEGER,\n tokens_out INTEGER,\n error TEXT,\n FOREIGN KEY (run_id) REFERENCES runs(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_cases_run_id ON cases(run_id);\n\nCREATE TABLE IF NOT EXISTS scores (\n id TEXT PRIMARY KEY,\n case_id TEXT NOT NULL,\n scorer_name TEXT NOT NULL,\n score REAL NOT NULL,\n reason TEXT,\n FOREIGN KEY (case_id) REFERENCES cases(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_scores_case_id ON scores(case_id);\n\nCREATE TABLE IF NOT EXISTS prompts (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL UNIQUE,\n content TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000)\n);\n\nCREATE INDEX IF NOT EXISTS idx_prompts_created_at ON prompts(created_at);\n"],
5
- "mappings": ";AAAA,SAAS,iBAAiB;AAC1B,SAAS,eAAe;AACxB;AAAA,EACE;AAAA,OAGK;;;ACNP;;;AD4FA,IAAM,eAAe,CAAC,WAAW,aAAa,QAAQ;AAItD,SAAS,YAAY,OAAmC;AACtD,SAAO,aAAa,KAAK,CAAC,WAAW,WAAW,KAAK;AACvD;AAkBA,SAAS,cACP,OACA,MACS;AACT,UAAQ,MAAM;AAAA,IACZ,KAAK;AACH,aAAO,OAAO,UAAU;AAAA,IAC1B,KAAK;AACH,aAAO,UAAU,QAAQ,OAAO,UAAU;AAAA,IAC5C,KAAK;AACH,aAAO,OAAO,UAAU;AAAA,IAC1B,KAAK;AACH,aAAO,UAAU,QAAQ,OAAO,UAAU;AAAA,EAC9C;AACF;AAEA,SAAS,kBAAkB,KAAa,SAA4B;AAClE,SAAO,OAAO,QAAQ,OAAO,EAC1B,OAAO,CAAC,CAAC,MAAM,IAAI,MAAM,CAAC,cAAc,IAAI,IAAI,GAAG,IAAI,CAAC,EACxD,IAAI,CAAC,CAAC,IAAI,MAAM,IAAI;AACzB;AAEA,SAAS,WACP,KACA,SAC0B;AAC1B,SAAO,kBAAkB,KAAK,OAAO,EAAE,WAAW;AACpD;AAEA,SAAS,QACP,KACA,SACA,KACU;AACV,MAAI,WAAW,KAAK,OAAO,EAAG,QAAO;AACrC,QAAM,IAAI;AAAA,IACR,gCAAgC,kBAAkB,KAAK,OAAO,EAAE,KAAK,IAAI,CAAC,mBAAmB,QAAQ,GAAG,CAAC;AAAA,EAC3G;AACF;AAEA,SAAS,QAAQ,KAAqB;AACpC,SAAO,IAAI,QAAQ,QAAQ,GAAG,EAAE,KAAK;AACvC;AAEA,SAAS,SAAS,OAAkD;AAClE,SAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAC5E;AAEA,SAAS,aAAa,OAAqC;AACzD,SACE,SAAS,KAAK,KACd,OAAO,MAAM,eAAe,YAC5B,OAAO,MAAM,cAAc,YAC3B,OAAO,MAAM,cAAc,YAC3B,OAAO,MAAM,mBAAmB,YAChC,OAAO,MAAM,kBAAkB,YAC/B,OAAO,MAAM,mBAAmB,YAChC,SAAS,MAAM,UAAU,KACzB,OAAO,OAAO,MAAM,UAAU,EAAE,MAAM,CAAC,UAAU,OAAO,UAAU,QAAQ;AAE9E;AAEA,SAAS,UACP,MACA,OACA,QACG;AACH,QAAM,QAAiB,KAAK,MAAM,IAAI;AACtC,MAAI,MAAM,KAAK,EAAG,QAAO;AACzB,QAAM,IAAI,UAAU,sBAAsB,MAAM,EAAE;AACpD;AAEA,IAAM,eAAe;AAAA,EACnB,IAAI;AAAA,EACJ,MAAM;AAAA,EACN,YAAY;AACd;AAEA,IAAM,aAAa;AAAA,EACjB,IAAI;AAAA,EACJ,UAAU;AAAA,EACV,MAAM;AAAA,EACN,OAAO;AAAA,EACP,QAAQ;AAAA,EACR,YAAY;AAAA,EACZ,aAAa;AAAA,EACb,QAAQ;AAAA,EACR,SAAS;AACX;AAEA,IAAM,cAAc;AAAA,EAClB,IAAI;AAAA,EACJ,QAAQ;AAAA,EACR,KAAK;AAAA,EACL,OAAO;AAAA,EACP,QAAQ;AAAA,EACR,UAAU;AAAA,EACV,YAAY;AAAA,EACZ,WAAW;AAAA,EACX,YAAY;AAAA,EACZ,OAAO;AACT;AAEA,IAAM,gBAAgB;AAAA,EACpB,IAAI;AAAA,EACJ,MAAM;AAAA,EACN,SAAS;AAAA,EACT,SAAS;AAAA,EACT,YAAY;AACd;AAEA,SAAS,QAAQ,KAA2C;AAC1D,SAAO,EAAE,IAAI,IAAI,IAAI,MAAM,IAAI,MAAM,YAAY,IAAI,WAAW;AAClE;AAEA,SAAS,MAAM,KAAuC;AACpD,MAAI,CAAC,YAAY,IAAI,MAAM,GAAG;AAC5B,UAAM,IAAI,UAAU,sBAAsB,IAAI,MAAM,YAAY,IAAI,EAAE,EAAE;AAAA,EAC1E;AACA,SAAO;AAAA,IACL,IAAI,IAAI;AAAA,IACR,UAAU,IAAI;AAAA,IACd,MAAM,IAAI;AAAA,IACV,OAAO,IAAI;AAAA,IACX,QAAQ,IAAI,SACR,UAAU,IAAI,QAAQ,UAAU,iBAAiB,IAAI,EAAE,EAAE,IACzD;AAAA,IACJ,YAAY,IAAI;AAAA,IAChB,aAAa,IAAI;AAAA,IACjB,QAAQ,IAAI;AAAA,IACZ,SAAS,IAAI,UACT,UAAU,IAAI,SAAS,cAAc,kBAAkB,IAAI,EAAE,EAAE,IAC/D;AAAA,EACN;AACF;AAEA,SAAS,OAAO,KAAyC;AACvD,QAAM,QAAiB,KAAK,MAAM,IAAI,KAAK;AAC3C,QAAM,WAAoB,IAAI,WAAW,KAAK,MAAM,IAAI,QAAQ,IAAI;AACpE,SAAO;AAAA,IACL,IAAI,IAAI;AAAA,IACR,QAAQ,IAAI;AAAA,IACZ,KAAK,IAAI;AAAA,IACT;AAAA,IACA,QAAQ,IAAI;AAAA,IACZ;AAAA,IACA,YAAY,IAAI;AAAA,IAChB,WAAW,IAAI;AAAA,IACf,YAAY,IAAI;AAAA,IAChB,OAAO,IAAI;AAAA,EACb;AACF;AAEA,SAAS,SAAS,KAA6C;AAC7D,SAAO;AAAA,IACL,IAAI,IAAI;AAAA,IACR,MAAM,IAAI;AAAA,IACV,SAAS,IAAI;AAAA,IACb,SAAS,IAAI;AAAA,IACb,YAAY,IAAI;AAAA,EAClB;AACF;AAEO,IAAM,WAAN,MAAe;AAAA,EACpB;AAAA,EACA,cAAc,oBAAI,IAAiD;AAAA,EAEnE,MAAM,KAAkD;AACtD,QAAI,OAAO,KAAK,YAAY,IAAI,GAAG;AACnC,QAAI,CAAC,MAAM;AACT,aAAO,KAAK,IAAI,QAAQ,GAAG;AAC3B,WAAK,YAAY,IAAI,KAAK,IAAI;AAAA,IAChC;AACA,WAAO;AAAA,EACT;AAAA,EAEA,KACE,SACA,QACG,QACmB;AACtB,UAAM,MAAM,KAAK,MAAM,GAAG,EAAE,IAAI,GAAG,MAAM;AACzC,WAAO,QAAQ,SAAY,SAAY,QAAQ,KAAK,SAAS,GAAG;AAAA,EAClE;AAAA;AAAA,EAGA,KACE,SACA,QACG,QACO;AACV,UAAM,MAAM,KAAK,KAAK,SAAS,KAAK,GAAG,MAAM;AAC7C,QAAI,QAAQ,QAAW;AACrB,YAAM,IAAI,MAAM,0BAA0B,QAAQ,GAAG,CAAC,EAAE;AAAA,IAC1D;AACA,WAAO;AAAA,EACT;AAAA,EAEA,KACE,SACA,QACG,QACS;AACZ,WAAO,KAAK,MAAM,GAAG,EAClB,IAAI,GAAG,MAAM,EACb,IAAI,CAAC,QAAQ,QAAQ,KAAK,SAAS,GAAG,CAAC;AAAA,EAC5C;AAAA,EAEA,aAAgB,IAAgB;AAC9B,SAAK,IAAI,KAAK,mBAAmB;AACjC,QAAI;AACF,YAAM,SAAS,GAAG;AAClB,WAAK,IAAI,KAAK,QAAQ;AACtB,aAAO;AAAA,IACT,SAAS,OAAO;AACd,UAAI;AACF,aAAK,IAAI,KAAK,UAAU;AAAA,MAC1B,QAAQ;AAAA,MAER;AACA,YAAM;AAAA,IACR;AAAA,EACF;AAAA,EAEA,YAAY,UAAkC;AAC5C,QAAI,oBAAoB,cAAc;AACpC,WAAK,MAAM;AAAA,IACb,OAAO;AACL,YAAM,SAAS,YAAY;AAC3B,gBAAU,QAAQ,MAAM,GAAG,EAAE,WAAW,KAAK,CAAC;AAC9C,WAAK,MAAM,IAAI,aAAa,MAAM;AAAA,IACpC;AACA,SAAK,IAAI,KAAK,kBAAG;AACjB,SAAK,iCAAiC;AACtC,SAAK,6BAA6B;AAClC,SAAK,IAAI;AAAA,MACP;AAAA,IACF;AAAA,EACF;AAAA,EAEA,+BAAqC;AACnC,UAAM,UAAU,KAAK,KAAK,EAAE,MAAM,OAAO,GAAG,4BAA4B;AAExE,QAAI,QAAQ,WAAW,EAAG;AAC1B,QAAI,QAAQ,KAAK,CAAC,WAAW,OAAO,SAAS,SAAS,EAAG;AAEzD,SAAK,aAAa,MAAM;AACtB,WAAK,IAAI,KAAK,8CAA8C;AAC5D,WAAK,IAAI,KAAK;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,OASb;AACD,WAAK,IAAI,KAAK;AAAA;AAAA;AAAA;AAAA,OAIb;AACD,WAAK,IAAI,KAAK,2BAA2B;AACzC,WAAK,IAAI;AAAA,QACP;AAAA,MACF;AACA,WAAK,IAAI;AAAA,QACP;AAAA,MACF;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEA,mCAAyC;AACvC,UAAM,UAAU,KAAK;AAAA,MACnB,EAAE,MAAM,QAAQ,SAAS,SAAS;AAAA,MAClC;AAAA,IACF;AAEA,QAAI,QAAQ,WAAW,EAAG;AAE1B,UAAM,cAAc,QAAQ,KAAK,CAAC,WAAW,OAAO,SAAS,UAAU;AACvE,UAAM,kBAAkB,aAAa,YAAY;AAEjD,UAAM,iBAAiB,KAAK;AAAA,MAC1B,EAAE,MAAM,QAAQ,WAAW,QAAQ,OAAO,OAAO;AAAA,MACjD;AAAA,IACF;AACA,UAAM,kBAAkB,eAAe;AAAA,MACrC,CAAC,OAAO,GAAG,SAAS,cAAc,GAAG,UAAU;AAAA,IACjD;AACA,UAAM,mBAAmB,iBAAiB,cAAc;AAExD,QAAI,mBAAmB,iBAAkB;AAEzC,SAAK,YAAY,MAAM;AACvB,SAAK,aAAa,MAAM;AACtB,WAAK,IAAI,KAAK;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,OAab;AAGD,WAAK,IAAI,KAAK,yCAAyC;AAEvD,WAAK,IAAI,KAAK;AAAA;AAAA;AAAA;AAAA;AAAA,OAKb;AAED,WAAK,IAAI,KAAK,iBAAiB;AAC/B,WAAK,IAAI,KAAK,sCAAsC;AACpD,WAAK,IAAI;AAAA,QACP;AAAA,MACF;AACA,WAAK,IAAI;AAAA,QACP;AAAA,MACF;AAAA,IACF,CAAC;AACD,SAAK,YAAY,MAAM;AAAA,EACzB;AAAA,EAEA,YAAY,MAAwB;AAClC,UAAM,KAAK,OAAO,WAAW;AAC7B,UAAM,MAAM,KAAK,IAAI;AACrB,SAAK;AAAA,MACH;AAAA,IACF,EAAE,IAAI,IAAI,MAAM,GAAG;AACnB,WAAO,EAAE,IAAI,MAAM,YAAY,IAAI;AAAA,EACrC;AAAA,EAEA,SAAS,IAAkC;AACzC,UAAM,MAAM,KAAK;AAAA,MACf;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO,MAAM,QAAQ,GAAG,IAAI;AAAA,EAC9B;AAAA,EAEA,YAAY,IAAY,MAAoB;AAC1C,SAAK,MAAM,yCAAyC,EAAE,IAAI,MAAM,EAAE;AAAA,EACpE;AAAA,EAEA,UAAU,IAAY,MAAoB;AACxC,SAAK,MAAM,uCAAuC,EAAE,IAAI,MAAM,EAAE;AAAA,EAClE;AAAA,EAEA,UAAU,KAKC;AACT,UAAM,KAAK,OAAO,WAAW;AAC7B,UAAM,MAAM,KAAK,IAAI;AACrB,SAAK;AAAA,MACH;AAAA,IACF,EAAE;AAAA,MACA;AAAA,MACA,IAAI;AAAA,MACJ,IAAI;AAAA,MACJ,IAAI;AAAA,MACJ,IAAI,SAAS,KAAK,UAAU,IAAI,MAAM,IAAI;AAAA,MAC1C;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAAA,EAEA,UACE,OACA,QACA,SACM;AACN,SAAK;AAAA,MACH;AAAA,IACF,EAAE,IAAI,KAAK,IAAI,GAAG,QAAQ,UAAU,KAAK,UAAU,OAAO,IAAI,MAAM,KAAK;AAAA,EAC3E;AAAA,EAEA,UAAU,OAAyB;AACjC,SAAK,aAAa,MAAM;AACtB,YAAM,OAAO,KAAK;AAAA,QAChB;AAAA,MACF;AACA,iBAAW,KAAK,OAAO;AACrB,aAAK;AAAA,UACH,EAAE;AAAA,UACF,EAAE;AAAA,UACF,EAAE;AAAA,UACF,KAAK,UAAU,EAAE,KAAK;AAAA,UACtB,EAAE;AAAA,UACF,EAAE,YAAY,OAAO,KAAK,UAAU,EAAE,QAAQ,IAAI;AAAA,UAClD,EAAE;AAAA,UACF,EAAE;AAAA,UACF,EAAE;AAAA,UACF,EAAE,SAAS;AAAA,QACb;AAAA,MACF;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEA,WAAW,QAA2B;AACpC,SAAK,aAAa,MAAM;AACtB,YAAM,OAAO,KAAK;AAAA,QAChB;AAAA,MACF;AACA,iBAAW,KAAK,QAAQ;AACtB,aAAK,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU,IAAI;AAAA,MACpE;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEA,OAAO,OAAmC;AACxC,UAAM,MAAM,KAAK,KAAK,YAAY,mCAAmC,KAAK;AAC1E,WAAO,MAAM,MAAM,GAAG,IAAI;AAAA,EAC5B;AAAA,EAEA,SAAS,SAA4B;AACnC,UAAM,OAAO,UACT,KAAK;AAAA,MACH;AAAA,MACA;AAAA,MACA;AAAA,IACF,IACA,KAAK,KAAK,YAAY,wCAAwC;AAClE,WAAO,KAAK,IAAI,KAAK;AAAA,EACvB;AAAA,EAEA,SAAS,OAA0B;AACjC,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,MACA;AAAA,IACF,EAAE,IAAI,MAAM;AAAA,EACd;AAAA,EAEA,gBAAgB,OAAe,YAAY,KAAuB;AAChE,UAAM,OAAO,KAAK;AAAA,MAChB;AAAA,QACE,GAAG;AAAA,QACH,aAAa;AAAA,QACb,OAAO;AAAA,QACP,cAAc;AAAA,MAChB;AAAA,MACA;AAAA;AAAA;AAAA;AAAA;AAAA,MAKA;AAAA,MACA;AAAA,IACF;AAEA,UAAM,UAAU,oBAAI,IAA4B;AAChD,eAAW,OAAO,MAAM;AACtB,UAAI,IAAI,QAAQ,IAAI,IAAI,EAAE;AAC1B,UAAI,CAAC,GAAG;AACN,YAAI,EAAE,GAAG,OAAO,GAAG,GAAG,QAAQ,CAAC,EAAE;AACjC,gBAAQ,IAAI,IAAI,IAAI,CAAC;AAAA,MACvB;AACA,QAAE,OAAO,KAAK;AAAA,QACZ,aAAa,IAAI;AAAA,QACjB,OAAO,IAAI;AAAA,QACX,QAAQ,IAAI;AAAA,MACd,CAAC;AAAA,IACH;AACA,WAAO,MAAM,KAAK,QAAQ,OAAO,CAAC;AAAA,EACpC;AAAA,EAEA,cAAc,OAAe,YAAY,KAAiB;AACxD,UAAM,SAAS,KAAK;AAAA,MAClB;AAAA,QACE,YAAY;AAAA,QACZ,gBAAgB;AAAA,QAChB,eAAe;AAAA,QACf,gBAAgB;AAAA,MAClB;AAAA,MACA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAMA;AAAA,IACF;AAEA,UAAM,cAAc,KAAK;AAAA,MACvB,EAAE,aAAa,QAAQ,WAAW,SAAS;AAAA,MAC3C;AAAA;AAAA;AAAA;AAAA;AAAA,MAKA;AAAA,IACF;AAEA,UAAM,aAAqC,CAAC;AAC5C,eAAW,OAAO,aAAa;AAC7B,iBAAW,IAAI,WAAW,IAAI,IAAI;AAAA,IACpC;AAEA,UAAM,WAAW,KAAK;AAAA,MACpB,EAAE,IAAI,QAAQ,UAAU,SAAS;AAAA,MACjC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAMA;AAAA,IACF;AAEA,QAAI,YAAY;AAChB,QAAI,YAAY;AAChB,eAAW,OAAO,UAAU;AAC1B,UAAI,IAAI,YAAY,UAAW;AAAA,UAC1B;AAAA,IACP;AAEA,WAAO;AAAA,MACL,YAAY,OAAO;AAAA,MACnB;AAAA,MACA;AAAA,MACA;AAAA,MACA,gBAAgB,OAAO;AAAA,MACvB,eAAe,OAAO;AAAA,MACtB,gBAAgB,OAAO;AAAA,IACzB;AAAA,EACF;AAAA,EAEA,gBAAgB,MAAoC;AAClD,UAAM,MAAM,KAAK;AAAA,MACf;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO,MAAM,QAAQ,GAAG,IAAI;AAAA,EAC9B;AAAA,EAEA,sBAAsB,SAAiB,OAAoC;AACzE,UAAM,MAAM,QACR,KAAK;AAAA,MACH;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IACA,KAAK;AAAA,MACH;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACJ,WAAO,MAAM,MAAM,GAAG,IAAI;AAAA,EAC5B;AAAA,EAEA,aAAyB;AACvB,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,IACF,EAAE,IAAI,OAAO;AAAA,EACf;AAAA,EAEA,aAAa,MAAc,SAA4B;AACrD,UAAM,KAAK,OAAO,WAAW;AAC7B,UAAM,MAAM,KAAK,IAAI;AAErB,UAAM,EAAE,cAAc,IAAI,KAAK;AAAA,MAC7B,EAAE,eAAe,UAAU;AAAA,MAC3B;AAAA,MACA;AAAA,IACF;AACA,UAAM,WAAW,iBAAiB,KAAK;AAEvC,SAAK;AAAA,MACH;AAAA,IACF,EAAE,IAAI,IAAI,MAAM,SAAS,SAAS,GAAG;AACrC,WAAO,EAAE,IAAI,MAAM,SAAS,SAAS,YAAY,IAAI;AAAA,EACvD;AAAA,EAEA,cAA2B;AACzB,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,IACF,EAAE,IAAI,QAAQ;AAAA,EAChB;AAAA,EAEA,UAAU,IAAmC;AAC3C,UAAM,MAAM,KAAK;AAAA,MACf;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO,MAAM,SAAS,GAAG,IAAI;AAAA,EAC/B;AAAA,EAEA,aAAa,IAAkB;AAC7B,SAAK,MAAM,kCAAkC,EAAE,IAAI,EAAE;AAAA,EACvD;AAAA,EAEA,SAAS,IAAkB;AACzB,SAAK,aAAa,MAAM;AACtB,WAAK,MAAM,oCAAoC,EAAE,IAAI,EAAE;AACvD,WAAK;AAAA,QACH;AAAA,MACF,EAAE,IAAI,WAAW,KAAK,IAAI,GAAG,EAAE;AAAA,IACjC,CAAC;AAAA,EACH;AAAA,EAEA,UAAU,IAAkB;AAC1B,SAAK,MAAM,+BAA+B,EAAE,IAAI,EAAE;AAAA,EACpD;AAAA,EAEA,YAAY,IAAkB;AAC5B,SAAK,MAAM,iCAAiC,EAAE,IAAI,EAAE;AAAA,EACtD;AACF;",
4
+ "sourcesContent": ["import { Declaration } from '@zukhruf/sqlite-drift';\nimport { transaction } from '@zukhruf/sqlite-transaction';\nimport { mkdirSync } from 'node:fs';\nimport { dirname } from 'node:path';\nimport {\n DatabaseSync,\n type SQLInputValue,\n type SQLOutputValue,\n} from 'node:sqlite';\n\nimport SCHEMA from './ddl.sqlite.sql';\n\nconst schema = new Declaration([SCHEMA], { rows: 'keep' });\n\nexport interface SuiteRow {\n id: string;\n name: string;\n created_at: number;\n}\n\nexport interface RunRow {\n id: string;\n suite_id: string;\n name: string;\n model: string;\n config: Record<string, unknown> | null;\n started_at: number;\n finished_at: number | null;\n status: RunStatus;\n summary: RunSummary | null;\n}\n\nexport interface CaseRow {\n id: string;\n run_id: string;\n idx: number;\n input: unknown;\n output: string | null;\n expected: unknown | null;\n latency_ms: number | null;\n tokens_in: number | null;\n tokens_out: number | null;\n error: string | null;\n}\n\nexport interface CaseWithScores extends CaseRow {\n scores: Array<{ scorer_name: string; score: number; reason: string | null }>;\n}\n\nexport interface ScoreRow {\n id: string;\n case_id: string;\n scorer_name: string;\n score: number;\n reason: string | null;\n}\n\nexport interface RunSummary {\n totalCases: number;\n passCount: number;\n failCount: number;\n meanScores: Record<string, number>;\n totalLatencyMs: number;\n totalTokensIn: number;\n totalTokensOut: number;\n}\n\nexport interface PromptRow {\n id: string;\n name: string;\n version: number;\n content: string;\n created_at: number;\n}\n\nexport interface CaseData {\n id: string;\n run_id: string;\n idx: number;\n input: unknown;\n output: string | null;\n expected?: unknown;\n latency_ms: number;\n tokens_in: number;\n tokens_out: number;\n error?: string;\n}\n\nexport interface ScoreData {\n id: string;\n case_id: string;\n scorer_name: string;\n score: number;\n reason?: string;\n}\n\nconst RUN_STATUSES = ['running', 'completed', 'failed'] as const;\n\ntype RunStatus = (typeof RUN_STATUSES)[number];\n\nfunction isRunStatus(value: string): value is RunStatus {\n return RUN_STATUSES.some((status) => status === value);\n}\n\ntype SqlRow = Record<string, SQLOutputValue>;\n\n/** What node:sqlite returns for each kind of column the store reads. */\ninterface ColumnTypes {\n text: string;\n 'text?': string | null;\n number: number;\n 'number?': number | null;\n}\n\ntype ColumnType = keyof ColumnTypes;\n\ntype Columns = Readonly<Record<string, ColumnType>>;\n\ntype RowOf<C extends Columns> = { -readonly [K in keyof C]: ColumnTypes[C[K]] };\n\nfunction matchesColumn(\n value: SQLOutputValue | undefined,\n type: ColumnType,\n): boolean {\n switch (type) {\n case 'text':\n return typeof value === 'string';\n case 'text?':\n return value === null || typeof value === 'string';\n case 'number':\n return typeof value === 'number';\n case 'number?':\n return value === null || typeof value === 'number';\n }\n}\n\nfunction mismatchedColumns(row: SqlRow, columns: Columns): string[] {\n return Object.entries(columns)\n .filter(([name, type]) => !matchesColumn(row[name], type))\n .map(([name]) => name);\n}\n\nfunction hasColumns<C extends Columns>(\n row: SqlRow,\n columns: C,\n): row is SqlRow & RowOf<C> {\n return mismatchedColumns(row, columns).length === 0;\n}\n\nfunction readRow<C extends Columns>(\n row: SqlRow,\n columns: C,\n sql: string,\n): RowOf<C> {\n if (hasColumns(row, columns)) return row;\n throw new TypeError(\n `Unexpected values in columns ${mismatchedColumns(row, columns).join(', ')} of a row from: ${oneLine(sql)}`,\n );\n}\n\nfunction oneLine(sql: string): string {\n return sql.replace(/\\s+/g, ' ').trim();\n}\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n return typeof value === 'object' && value !== null && !Array.isArray(value);\n}\n\nfunction isRunSummary(value: unknown): value is RunSummary {\n return (\n isRecord(value) &&\n typeof value.totalCases === 'number' &&\n typeof value.passCount === 'number' &&\n typeof value.failCount === 'number' &&\n typeof value.totalLatencyMs === 'number' &&\n typeof value.totalTokensIn === 'number' &&\n typeof value.totalTokensOut === 'number' &&\n isRecord(value.meanScores) &&\n Object.values(value.meanScores).every((score) => typeof score === 'number')\n );\n}\n\nfunction parseJson<T>(\n text: string,\n guard: (value: unknown) => value is T,\n source: string,\n): T {\n const value: unknown = JSON.parse(text);\n if (guard(value)) return value;\n throw new TypeError(`Unexpected JSON in ${source}`);\n}\n\nconst suiteColumns = {\n id: 'text',\n name: 'text',\n created_at: 'number',\n} as const;\n\nconst runColumns = {\n id: 'text',\n suite_id: 'text',\n name: 'text',\n model: 'text',\n config: 'text?',\n started_at: 'number',\n finished_at: 'number?',\n status: 'text',\n summary: 'text?',\n} as const;\n\nconst caseColumns = {\n id: 'text',\n run_id: 'text',\n idx: 'number',\n input: 'text',\n output: 'text?',\n expected: 'text?',\n latency_ms: 'number?',\n tokens_in: 'number?',\n tokens_out: 'number?',\n error: 'text?',\n} as const;\n\nconst promptColumns = {\n id: 'text',\n name: 'text',\n version: 'number',\n content: 'text',\n created_at: 'number',\n} as const;\n\nfunction toSuite(row: RowOf<typeof suiteColumns>): SuiteRow {\n return { id: row.id, name: row.name, created_at: row.created_at };\n}\n\nfunction toRun(row: RowOf<typeof runColumns>): RunRow {\n if (!isRunStatus(row.status)) {\n throw new TypeError(`Unexpected status \"${row.status}\" of run ${row.id}`);\n }\n return {\n id: row.id,\n suite_id: row.suite_id,\n name: row.name,\n model: row.model,\n config: row.config\n ? parseJson(row.config, isRecord, `config of run ${row.id}`)\n : null,\n started_at: row.started_at,\n finished_at: row.finished_at,\n status: row.status,\n summary: row.summary\n ? parseJson(row.summary, isRunSummary, `summary of run ${row.id}`)\n : null,\n };\n}\n\nfunction toCase(row: RowOf<typeof caseColumns>): CaseRow {\n const input: unknown = JSON.parse(row.input);\n const expected: unknown = row.expected ? JSON.parse(row.expected) : null;\n return {\n id: row.id,\n run_id: row.run_id,\n idx: row.idx,\n input,\n output: row.output,\n expected,\n latency_ms: row.latency_ms,\n tokens_in: row.tokens_in,\n tokens_out: row.tokens_out,\n error: row.error,\n };\n}\n\nfunction toPrompt(row: RowOf<typeof promptColumns>): PromptRow {\n return {\n id: row.id,\n name: row.name,\n version: row.version,\n content: row.content,\n created_at: row.created_at,\n };\n}\n\nexport class RunStore {\n #db: DatabaseSync;\n #statements = new Map<string, ReturnType<DatabaseSync['prepare']>>();\n\n #stmt(sql: string): ReturnType<DatabaseSync['prepare']> {\n let stmt = this.#statements.get(sql);\n if (!stmt) {\n stmt = this.#db.prepare(sql);\n this.#statements.set(sql, stmt);\n }\n return stmt;\n }\n\n #get<const C extends Columns>(\n columns: C,\n sql: string,\n ...params: SQLInputValue[]\n ): RowOf<C> | undefined {\n const row = this.#stmt(sql).get(...params);\n return row === undefined ? undefined : readRow(row, columns, sql);\n }\n\n /** Reads an aggregate query, which returns exactly one row. */\n #one<const C extends Columns>(\n columns: C,\n sql: string,\n ...params: SQLInputValue[]\n ): RowOf<C> {\n const row = this.#get(columns, sql, ...params);\n if (row === undefined) {\n throw new Error(`Expected one row from: ${oneLine(sql)}`);\n }\n return row;\n }\n\n #all<const C extends Columns>(\n columns: C,\n sql: string,\n ...params: SQLInputValue[]\n ): RowOf<C>[] {\n return this.#stmt(sql)\n .all(...params)\n .map((row) => readRow(row, columns, sql));\n }\n\n constructor(pathOrDb?: string | DatabaseSync) {\n if (pathOrDb instanceof DatabaseSync) {\n this.#db = pathOrDb;\n } else {\n const dbPath = pathOrDb ?? '.evals/store.db';\n mkdirSync(dirname(dbPath), { recursive: true });\n this.#db = new DatabaseSync(dbPath);\n }\n this.#db.exec(`\n PRAGMA journal_mode = WAL;\n PRAGMA synchronous = NORMAL;\n PRAGMA foreign_keys = ON;\n PRAGMA busy_timeout = 5000;\n `);\n const runsTable = this.#get(\n { name: 'text' },\n \"SELECT name FROM sqlite_schema WHERE type = 'table' AND name = 'runs'\",\n );\n if (runsTable !== undefined) {\n // Runs from layouts before suite_id was required have no suite. With\n // foreign keys on, deleting them also deletes their cases and scores.\n this.#db.exec(\n 'DELETE FROM runs WHERE suite_id IS NULL OR NOT EXISTS (SELECT 1 FROM suites WHERE suites.id = runs.suite_id)',\n );\n }\n schema.apply(this.#db);\n }\n\n createSuite(name: string): SuiteRow {\n const id = crypto.randomUUID();\n const now = Date.now();\n this.#stmt(\n 'INSERT INTO suites (id, name, created_at) VALUES (?, ?, ?)',\n ).run(id, name, now);\n return { id, name, created_at: now };\n }\n\n getSuite(id: string): SuiteRow | undefined {\n const row = this.#get(\n suiteColumns,\n 'SELECT * FROM suites WHERE id = ?',\n id,\n );\n return row ? toSuite(row) : undefined;\n }\n\n renameSuite(id: string, name: string): void {\n this.#stmt('UPDATE suites SET name = ? WHERE id = ?').run(name, id);\n }\n\n renameRun(id: string, name: string): void {\n this.#stmt('UPDATE runs SET name = ? WHERE id = ?').run(name, id);\n }\n\n createRun(run: {\n suite_id: string;\n name: string;\n model: string;\n config?: Record<string, unknown>;\n }): string {\n const id = crypto.randomUUID();\n const now = Date.now();\n this.#stmt(\n 'INSERT INTO runs (id, suite_id, name, model, config, started_at) VALUES (?, ?, ?, ?, ?, ?)',\n ).run(\n id,\n run.suite_id,\n run.name,\n run.model,\n run.config ? JSON.stringify(run.config) : null,\n now,\n );\n return id;\n }\n\n finishRun(\n runId: string,\n status: 'completed' | 'failed',\n summary?: RunSummary,\n ): void {\n this.#stmt(\n 'UPDATE runs SET finished_at = ?, status = ?, summary = ? WHERE id = ?',\n ).run(Date.now(), status, summary ? JSON.stringify(summary) : null, runId);\n }\n\n saveCases(cases: CaseData[]): void {\n transaction(\n this.#db,\n () => {\n const stmt = this.#stmt(\n 'INSERT INTO cases (id, run_id, idx, input, output, expected, latency_ms, tokens_in, tokens_out, error) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)',\n );\n for (const c of cases) {\n stmt.run(\n c.id,\n c.run_id,\n c.idx,\n JSON.stringify(c.input),\n c.output,\n c.expected != null ? JSON.stringify(c.expected) : null,\n c.latency_ms,\n c.tokens_in,\n c.tokens_out,\n c.error ?? null,\n );\n }\n },\n { mode: 'deferred' },\n );\n }\n\n saveScores(scores: ScoreData[]): void {\n transaction(\n this.#db,\n () => {\n const stmt = this.#stmt(\n 'INSERT INTO scores (id, case_id, scorer_name, score, reason) VALUES (?, ?, ?, ?, ?)',\n );\n for (const s of scores) {\n stmt.run(s.id, s.case_id, s.scorer_name, s.score, s.reason ?? null);\n }\n },\n { mode: 'deferred' },\n );\n }\n\n getRun(runId: string): RunRow | undefined {\n const row = this.#get(runColumns, 'SELECT * FROM runs WHERE id = ?', runId);\n return row ? toRun(row) : undefined;\n }\n\n listRuns(suiteId?: string): RunRow[] {\n const rows = suiteId\n ? this.#all(\n runColumns,\n 'SELECT * FROM runs WHERE suite_id = ? ORDER BY started_at',\n suiteId,\n )\n : this.#all(runColumns, 'SELECT * FROM runs ORDER BY started_at');\n return rows.map(toRun);\n }\n\n getCases(runId: string): CaseRow[] {\n return this.#all(\n caseColumns,\n 'SELECT * FROM cases WHERE run_id = ? ORDER BY idx',\n runId,\n ).map(toCase);\n }\n\n getFailingCases(runId: string, threshold = 0.5): CaseWithScores[] {\n const rows = this.#all(\n {\n ...caseColumns,\n scorer_name: 'text',\n score: 'number',\n score_reason: 'text?',\n },\n `SELECT c.*, s.scorer_name, s.score, s.reason as score_reason\n FROM cases c\n JOIN scores s ON s.case_id = c.id\n WHERE c.run_id = ? AND s.score < ?\n ORDER BY c.idx`,\n runId,\n threshold,\n );\n\n const caseMap = new Map<string, CaseWithScores>();\n for (const row of rows) {\n let c = caseMap.get(row.id);\n if (!c) {\n c = { ...toCase(row), scores: [] };\n caseMap.set(row.id, c);\n }\n c.scores.push({\n scorer_name: row.scorer_name,\n score: row.score,\n reason: row.score_reason,\n });\n }\n return Array.from(caseMap.values());\n }\n\n getRunSummary(runId: string, threshold = 0.5): RunSummary {\n const totals = this.#one(\n {\n totalCases: 'number',\n totalLatencyMs: 'number',\n totalTokensIn: 'number',\n totalTokensOut: 'number',\n },\n `SELECT\n COUNT(DISTINCT c.id) as totalCases,\n COALESCE(SUM(c.latency_ms), 0) as totalLatencyMs,\n COALESCE(SUM(c.tokens_in), 0) as totalTokensIn,\n COALESCE(SUM(c.tokens_out), 0) as totalTokensOut\n FROM cases c WHERE c.run_id = ?`,\n runId,\n );\n\n const scorerMeans = this.#all(\n { scorer_name: 'text', meanScore: 'number' },\n `SELECT s.scorer_name, AVG(s.score) as meanScore\n FROM scores s\n JOIN cases c ON c.id = s.case_id\n WHERE c.run_id = ?\n GROUP BY s.scorer_name`,\n runId,\n );\n\n const meanScores: Record<string, number> = {};\n for (const row of scorerMeans) {\n meanScores[row.scorer_name] = row.meanScore;\n }\n\n const passFail = this.#all(\n { id: 'text', minScore: 'number' },\n `SELECT c.id,\n MIN(s.score) as minScore\n FROM cases c\n JOIN scores s ON s.case_id = c.id\n WHERE c.run_id = ?\n GROUP BY c.id`,\n runId,\n );\n\n let passCount = 0;\n let failCount = 0;\n for (const row of passFail) {\n if (row.minScore >= threshold) passCount++;\n else failCount++;\n }\n\n return {\n totalCases: totals.totalCases,\n passCount,\n failCount,\n meanScores,\n totalLatencyMs: totals.totalLatencyMs,\n totalTokensIn: totals.totalTokensIn,\n totalTokensOut: totals.totalTokensOut,\n };\n }\n\n findSuiteByName(name: string): SuiteRow | undefined {\n const row = this.#get(\n suiteColumns,\n 'SELECT * FROM suites WHERE name = ? ORDER BY created_at DESC LIMIT 1',\n name,\n );\n return row ? toSuite(row) : undefined;\n }\n\n getLatestCompletedRun(suiteId: string, model?: string): RunRow | undefined {\n const row = model\n ? this.#get(\n runColumns,\n 'SELECT * FROM runs WHERE suite_id = ? AND status = ? AND model = ? ORDER BY started_at DESC LIMIT 1',\n suiteId,\n 'completed',\n model,\n )\n : this.#get(\n runColumns,\n 'SELECT * FROM runs WHERE suite_id = ? AND status = ? ORDER BY started_at DESC LIMIT 1',\n suiteId,\n 'completed',\n );\n return row ? toRun(row) : undefined;\n }\n\n listSuites(): SuiteRow[] {\n return this.#all(\n suiteColumns,\n 'SELECT * FROM suites ORDER BY created_at DESC',\n ).map(toSuite);\n }\n\n createPrompt(name: string, content: string): PromptRow {\n const id = crypto.randomUUID();\n const now = Date.now();\n\n const { latestVersion } = this.#one(\n { latestVersion: 'number?' },\n 'SELECT MAX(version) as latestVersion FROM prompts WHERE name = ?',\n name,\n );\n const version = (latestVersion ?? 0) + 1;\n\n this.#stmt(\n 'INSERT INTO prompts (id, name, version, content, created_at) VALUES (?, ?, ?, ?, ?)',\n ).run(id, name, version, content, now);\n return { id, name, version, content, created_at: now };\n }\n\n listPrompts(): PromptRow[] {\n return this.#all(\n promptColumns,\n 'SELECT * FROM prompts ORDER BY name COLLATE NOCASE ASC, version DESC',\n ).map(toPrompt);\n }\n\n getPrompt(id: string): PromptRow | undefined {\n const row = this.#get(\n promptColumns,\n 'SELECT * FROM prompts WHERE id = ?',\n id,\n );\n return row ? toPrompt(row) : undefined;\n }\n\n deletePrompt(id: string): void {\n this.#stmt('DELETE FROM prompts WHERE id = ?').run(id);\n }\n\n resetRun(id: string): void {\n transaction(\n this.#db,\n () => {\n this.#stmt('DELETE FROM cases WHERE run_id = ?').run(id);\n this.#stmt(\n 'UPDATE runs SET status = ?, started_at = ?, finished_at = NULL, summary = NULL WHERE id = ?',\n ).run('running', Date.now(), id);\n },\n { mode: 'deferred' },\n );\n }\n\n deleteRun(id: string): void {\n this.#stmt('DELETE FROM runs WHERE id = ?').run(id);\n }\n\n deleteSuite(id: string): void {\n this.#stmt('DELETE FROM suites WHERE id = ?').run(id);\n }\n}\n", "CREATE TABLE IF NOT EXISTS suites (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000)\n);\n\nCREATE TABLE IF NOT EXISTS runs (\n id TEXT PRIMARY KEY,\n suite_id TEXT NOT NULL,\n name TEXT NOT NULL,\n model TEXT NOT NULL,\n config TEXT,\n started_at INTEGER NOT NULL,\n finished_at INTEGER,\n status TEXT NOT NULL DEFAULT 'running' CHECK(status IN ('running', 'completed', 'failed')),\n summary TEXT,\n FOREIGN KEY (suite_id) REFERENCES suites(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_runs_suite_id ON runs(suite_id);\nCREATE INDEX IF NOT EXISTS idx_runs_started_at ON runs(started_at);\n\nCREATE TABLE IF NOT EXISTS cases (\n id TEXT PRIMARY KEY,\n run_id TEXT NOT NULL,\n idx INTEGER NOT NULL,\n input TEXT NOT NULL,\n output TEXT,\n expected TEXT,\n latency_ms INTEGER,\n tokens_in INTEGER,\n tokens_out INTEGER,\n error TEXT,\n FOREIGN KEY (run_id) REFERENCES runs(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_cases_run_id ON cases(run_id);\n\nCREATE TABLE IF NOT EXISTS scores (\n id TEXT PRIMARY KEY,\n case_id TEXT NOT NULL,\n scorer_name TEXT NOT NULL,\n score REAL NOT NULL,\n reason TEXT,\n FOREIGN KEY (case_id) REFERENCES cases(id) ON DELETE CASCADE\n);\n\nCREATE INDEX IF NOT EXISTS idx_scores_case_id ON scores(case_id);\n\nCREATE TABLE IF NOT EXISTS prompts (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n version INTEGER NOT NULL DEFAULT 1,\n content TEXT NOT NULL,\n created_at INTEGER NOT NULL DEFAULT (unixepoch() * 1000),\n UNIQUE(name, version)\n);\n\nCREATE INDEX IF NOT EXISTS idx_prompts_created_at ON prompts(created_at);\nCREATE INDEX IF NOT EXISTS idx_prompts_name_version ON prompts(name, version DESC);\n"],
5
+ "mappings": ";AAAA,SAAS,mBAAmB;AAC5B,SAAS,mBAAmB;AAC5B,SAAS,iBAAiB;AAC1B,SAAS,eAAe;AACxB;AAAA,EACE;AAAA,OAGK;;;ACRP;;;ADYA,IAAM,SAAS,IAAI,YAAY,CAAC,kBAAM,GAAG,EAAE,MAAM,OAAO,CAAC;AAoFzD,IAAM,eAAe,CAAC,WAAW,aAAa,QAAQ;AAItD,SAAS,YAAY,OAAmC;AACtD,SAAO,aAAa,KAAK,CAAC,WAAW,WAAW,KAAK;AACvD;AAkBA,SAAS,cACP,OACA,MACS;AACT,UAAQ,MAAM;AAAA,IACZ,KAAK;AACH,aAAO,OAAO,UAAU;AAAA,IAC1B,KAAK;AACH,aAAO,UAAU,QAAQ,OAAO,UAAU;AAAA,IAC5C,KAAK;AACH,aAAO,OAAO,UAAU;AAAA,IAC1B,KAAK;AACH,aAAO,UAAU,QAAQ,OAAO,UAAU;AAAA,EAC9C;AACF;AAEA,SAAS,kBAAkB,KAAa,SAA4B;AAClE,SAAO,OAAO,QAAQ,OAAO,EAC1B,OAAO,CAAC,CAAC,MAAM,IAAI,MAAM,CAAC,cAAc,IAAI,IAAI,GAAG,IAAI,CAAC,EACxD,IAAI,CAAC,CAAC,IAAI,MAAM,IAAI;AACzB;AAEA,SAAS,WACP,KACA,SAC0B;AAC1B,SAAO,kBAAkB,KAAK,OAAO,EAAE,WAAW;AACpD;AAEA,SAAS,QACP,KACA,SACA,KACU;AACV,MAAI,WAAW,KAAK,OAAO,EAAG,QAAO;AACrC,QAAM,IAAI;AAAA,IACR,gCAAgC,kBAAkB,KAAK,OAAO,EAAE,KAAK,IAAI,CAAC,mBAAmB,QAAQ,GAAG,CAAC;AAAA,EAC3G;AACF;AAEA,SAAS,QAAQ,KAAqB;AACpC,SAAO,IAAI,QAAQ,QAAQ,GAAG,EAAE,KAAK;AACvC;AAEA,SAAS,SAAS,OAAkD;AAClE,SAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAC5E;AAEA,SAAS,aAAa,OAAqC;AACzD,SACE,SAAS,KAAK,KACd,OAAO,MAAM,eAAe,YAC5B,OAAO,MAAM,cAAc,YAC3B,OAAO,MAAM,cAAc,YAC3B,OAAO,MAAM,mBAAmB,YAChC,OAAO,MAAM,kBAAkB,YAC/B,OAAO,MAAM,mBAAmB,YAChC,SAAS,MAAM,UAAU,KACzB,OAAO,OAAO,MAAM,UAAU,EAAE,MAAM,CAAC,UAAU,OAAO,UAAU,QAAQ;AAE9E;AAEA,SAAS,UACP,MACA,OACA,QACG;AACH,QAAM,QAAiB,KAAK,MAAM,IAAI;AACtC,MAAI,MAAM,KAAK,EAAG,QAAO;AACzB,QAAM,IAAI,UAAU,sBAAsB,MAAM,EAAE;AACpD;AAEA,IAAM,eAAe;AAAA,EACnB,IAAI;AAAA,EACJ,MAAM;AAAA,EACN,YAAY;AACd;AAEA,IAAM,aAAa;AAAA,EACjB,IAAI;AAAA,EACJ,UAAU;AAAA,EACV,MAAM;AAAA,EACN,OAAO;AAAA,EACP,QAAQ;AAAA,EACR,YAAY;AAAA,EACZ,aAAa;AAAA,EACb,QAAQ;AAAA,EACR,SAAS;AACX;AAEA,IAAM,cAAc;AAAA,EAClB,IAAI;AAAA,EACJ,QAAQ;AAAA,EACR,KAAK;AAAA,EACL,OAAO;AAAA,EACP,QAAQ;AAAA,EACR,UAAU;AAAA,EACV,YAAY;AAAA,EACZ,WAAW;AAAA,EACX,YAAY;AAAA,EACZ,OAAO;AACT;AAEA,IAAM,gBAAgB;AAAA,EACpB,IAAI;AAAA,EACJ,MAAM;AAAA,EACN,SAAS;AAAA,EACT,SAAS;AAAA,EACT,YAAY;AACd;AAEA,SAAS,QAAQ,KAA2C;AAC1D,SAAO,EAAE,IAAI,IAAI,IAAI,MAAM,IAAI,MAAM,YAAY,IAAI,WAAW;AAClE;AAEA,SAAS,MAAM,KAAuC;AACpD,MAAI,CAAC,YAAY,IAAI,MAAM,GAAG;AAC5B,UAAM,IAAI,UAAU,sBAAsB,IAAI,MAAM,YAAY,IAAI,EAAE,EAAE;AAAA,EAC1E;AACA,SAAO;AAAA,IACL,IAAI,IAAI;AAAA,IACR,UAAU,IAAI;AAAA,IACd,MAAM,IAAI;AAAA,IACV,OAAO,IAAI;AAAA,IACX,QAAQ,IAAI,SACR,UAAU,IAAI,QAAQ,UAAU,iBAAiB,IAAI,EAAE,EAAE,IACzD;AAAA,IACJ,YAAY,IAAI;AAAA,IAChB,aAAa,IAAI;AAAA,IACjB,QAAQ,IAAI;AAAA,IACZ,SAAS,IAAI,UACT,UAAU,IAAI,SAAS,cAAc,kBAAkB,IAAI,EAAE,EAAE,IAC/D;AAAA,EACN;AACF;AAEA,SAAS,OAAO,KAAyC;AACvD,QAAM,QAAiB,KAAK,MAAM,IAAI,KAAK;AAC3C,QAAM,WAAoB,IAAI,WAAW,KAAK,MAAM,IAAI,QAAQ,IAAI;AACpE,SAAO;AAAA,IACL,IAAI,IAAI;AAAA,IACR,QAAQ,IAAI;AAAA,IACZ,KAAK,IAAI;AAAA,IACT;AAAA,IACA,QAAQ,IAAI;AAAA,IACZ;AAAA,IACA,YAAY,IAAI;AAAA,IAChB,WAAW,IAAI;AAAA,IACf,YAAY,IAAI;AAAA,IAChB,OAAO,IAAI;AAAA,EACb;AACF;AAEA,SAAS,SAAS,KAA6C;AAC7D,SAAO;AAAA,IACL,IAAI,IAAI;AAAA,IACR,MAAM,IAAI;AAAA,IACV,SAAS,IAAI;AAAA,IACb,SAAS,IAAI;AAAA,IACb,YAAY,IAAI;AAAA,EAClB;AACF;AAEO,IAAM,WAAN,MAAe;AAAA,EACpB;AAAA,EACA,cAAc,oBAAI,IAAiD;AAAA,EAEnE,MAAM,KAAkD;AACtD,QAAI,OAAO,KAAK,YAAY,IAAI,GAAG;AACnC,QAAI,CAAC,MAAM;AACT,aAAO,KAAK,IAAI,QAAQ,GAAG;AAC3B,WAAK,YAAY,IAAI,KAAK,IAAI;AAAA,IAChC;AACA,WAAO;AAAA,EACT;AAAA,EAEA,KACE,SACA,QACG,QACmB;AACtB,UAAM,MAAM,KAAK,MAAM,GAAG,EAAE,IAAI,GAAG,MAAM;AACzC,WAAO,QAAQ,SAAY,SAAY,QAAQ,KAAK,SAAS,GAAG;AAAA,EAClE;AAAA;AAAA,EAGA,KACE,SACA,QACG,QACO;AACV,UAAM,MAAM,KAAK,KAAK,SAAS,KAAK,GAAG,MAAM;AAC7C,QAAI,QAAQ,QAAW;AACrB,YAAM,IAAI,MAAM,0BAA0B,QAAQ,GAAG,CAAC,EAAE;AAAA,IAC1D;AACA,WAAO;AAAA,EACT;AAAA,EAEA,KACE,SACA,QACG,QACS;AACZ,WAAO,KAAK,MAAM,GAAG,EAClB,IAAI,GAAG,MAAM,EACb,IAAI,CAAC,QAAQ,QAAQ,KAAK,SAAS,GAAG,CAAC;AAAA,EAC5C;AAAA,EAEA,YAAY,UAAkC;AAC5C,QAAI,oBAAoB,cAAc;AACpC,WAAK,MAAM;AAAA,IACb,OAAO;AACL,YAAM,SAAS,YAAY;AAC3B,gBAAU,QAAQ,MAAM,GAAG,EAAE,WAAW,KAAK,CAAC;AAC9C,WAAK,MAAM,IAAI,aAAa,MAAM;AAAA,IACpC;AACA,SAAK,IAAI,KAAK;AAAA;AAAA;AAAA;AAAA;AAAA,KAKb;AACD,UAAM,YAAY,KAAK;AAAA,MACrB,EAAE,MAAM,OAAO;AAAA,MACf;AAAA,IACF;AACA,QAAI,cAAc,QAAW;AAG3B,WAAK,IAAI;AAAA,QACP;AAAA,MACF;AAAA,IACF;AACA,WAAO,MAAM,KAAK,GAAG;AAAA,EACvB;AAAA,EAEA,YAAY,MAAwB;AAClC,UAAM,KAAK,OAAO,WAAW;AAC7B,UAAM,MAAM,KAAK,IAAI;AACrB,SAAK;AAAA,MACH;AAAA,IACF,EAAE,IAAI,IAAI,MAAM,GAAG;AACnB,WAAO,EAAE,IAAI,MAAM,YAAY,IAAI;AAAA,EACrC;AAAA,EAEA,SAAS,IAAkC;AACzC,UAAM,MAAM,KAAK;AAAA,MACf;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO,MAAM,QAAQ,GAAG,IAAI;AAAA,EAC9B;AAAA,EAEA,YAAY,IAAY,MAAoB;AAC1C,SAAK,MAAM,yCAAyC,EAAE,IAAI,MAAM,EAAE;AAAA,EACpE;AAAA,EAEA,UAAU,IAAY,MAAoB;AACxC,SAAK,MAAM,uCAAuC,EAAE,IAAI,MAAM,EAAE;AAAA,EAClE;AAAA,EAEA,UAAU,KAKC;AACT,UAAM,KAAK,OAAO,WAAW;AAC7B,UAAM,MAAM,KAAK,IAAI;AACrB,SAAK;AAAA,MACH;AAAA,IACF,EAAE;AAAA,MACA;AAAA,MACA,IAAI;AAAA,MACJ,IAAI;AAAA,MACJ,IAAI;AAAA,MACJ,IAAI,SAAS,KAAK,UAAU,IAAI,MAAM,IAAI;AAAA,MAC1C;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAAA,EAEA,UACE,OACA,QACA,SACM;AACN,SAAK;AAAA,MACH;AAAA,IACF,EAAE,IAAI,KAAK,IAAI,GAAG,QAAQ,UAAU,KAAK,UAAU,OAAO,IAAI,MAAM,KAAK;AAAA,EAC3E;AAAA,EAEA,UAAU,OAAyB;AACjC;AAAA,MACE,KAAK;AAAA,MACL,MAAM;AACJ,cAAM,OAAO,KAAK;AAAA,UAChB;AAAA,QACF;AACA,mBAAW,KAAK,OAAO;AACrB,eAAK;AAAA,YACH,EAAE;AAAA,YACF,EAAE;AAAA,YACF,EAAE;AAAA,YACF,KAAK,UAAU,EAAE,KAAK;AAAA,YACtB,EAAE;AAAA,YACF,EAAE,YAAY,OAAO,KAAK,UAAU,EAAE,QAAQ,IAAI;AAAA,YAClD,EAAE;AAAA,YACF,EAAE;AAAA,YACF,EAAE;AAAA,YACF,EAAE,SAAS;AAAA,UACb;AAAA,QACF;AAAA,MACF;AAAA,MACA,EAAE,MAAM,WAAW;AAAA,IACrB;AAAA,EACF;AAAA,EAEA,WAAW,QAA2B;AACpC;AAAA,MACE,KAAK;AAAA,MACL,MAAM;AACJ,cAAM,OAAO,KAAK;AAAA,UAChB;AAAA,QACF;AACA,mBAAW,KAAK,QAAQ;AACtB,eAAK,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU,IAAI;AAAA,QACpE;AAAA,MACF;AAAA,MACA,EAAE,MAAM,WAAW;AAAA,IACrB;AAAA,EACF;AAAA,EAEA,OAAO,OAAmC;AACxC,UAAM,MAAM,KAAK,KAAK,YAAY,mCAAmC,KAAK;AAC1E,WAAO,MAAM,MAAM,GAAG,IAAI;AAAA,EAC5B;AAAA,EAEA,SAAS,SAA4B;AACnC,UAAM,OAAO,UACT,KAAK;AAAA,MACH;AAAA,MACA;AAAA,MACA;AAAA,IACF,IACA,KAAK,KAAK,YAAY,wCAAwC;AAClE,WAAO,KAAK,IAAI,KAAK;AAAA,EACvB;AAAA,EAEA,SAAS,OAA0B;AACjC,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,MACA;AAAA,IACF,EAAE,IAAI,MAAM;AAAA,EACd;AAAA,EAEA,gBAAgB,OAAe,YAAY,KAAuB;AAChE,UAAM,OAAO,KAAK;AAAA,MAChB;AAAA,QACE,GAAG;AAAA,QACH,aAAa;AAAA,QACb,OAAO;AAAA,QACP,cAAc;AAAA,MAChB;AAAA,MACA;AAAA;AAAA;AAAA;AAAA;AAAA,MAKA;AAAA,MACA;AAAA,IACF;AAEA,UAAM,UAAU,oBAAI,IAA4B;AAChD,eAAW,OAAO,MAAM;AACtB,UAAI,IAAI,QAAQ,IAAI,IAAI,EAAE;AAC1B,UAAI,CAAC,GAAG;AACN,YAAI,EAAE,GAAG,OAAO,GAAG,GAAG,QAAQ,CAAC,EAAE;AACjC,gBAAQ,IAAI,IAAI,IAAI,CAAC;AAAA,MACvB;AACA,QAAE,OAAO,KAAK;AAAA,QACZ,aAAa,IAAI;AAAA,QACjB,OAAO,IAAI;AAAA,QACX,QAAQ,IAAI;AAAA,MACd,CAAC;AAAA,IACH;AACA,WAAO,MAAM,KAAK,QAAQ,OAAO,CAAC;AAAA,EACpC;AAAA,EAEA,cAAc,OAAe,YAAY,KAAiB;AACxD,UAAM,SAAS,KAAK;AAAA,MAClB;AAAA,QACE,YAAY;AAAA,QACZ,gBAAgB;AAAA,QAChB,eAAe;AAAA,QACf,gBAAgB;AAAA,MAClB;AAAA,MACA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAMA;AAAA,IACF;AAEA,UAAM,cAAc,KAAK;AAAA,MACvB,EAAE,aAAa,QAAQ,WAAW,SAAS;AAAA,MAC3C;AAAA;AAAA;AAAA;AAAA;AAAA,MAKA;AAAA,IACF;AAEA,UAAM,aAAqC,CAAC;AAC5C,eAAW,OAAO,aAAa;AAC7B,iBAAW,IAAI,WAAW,IAAI,IAAI;AAAA,IACpC;AAEA,UAAM,WAAW,KAAK;AAAA,MACpB,EAAE,IAAI,QAAQ,UAAU,SAAS;AAAA,MACjC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,MAMA;AAAA,IACF;AAEA,QAAI,YAAY;AAChB,QAAI,YAAY;AAChB,eAAW,OAAO,UAAU;AAC1B,UAAI,IAAI,YAAY,UAAW;AAAA,UAC1B;AAAA,IACP;AAEA,WAAO;AAAA,MACL,YAAY,OAAO;AAAA,MACnB;AAAA,MACA;AAAA,MACA;AAAA,MACA,gBAAgB,OAAO;AAAA,MACvB,eAAe,OAAO;AAAA,MACtB,gBAAgB,OAAO;AAAA,IACzB;AAAA,EACF;AAAA,EAEA,gBAAgB,MAAoC;AAClD,UAAM,MAAM,KAAK;AAAA,MACf;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO,MAAM,QAAQ,GAAG,IAAI;AAAA,EAC9B;AAAA,EAEA,sBAAsB,SAAiB,OAAoC;AACzE,UAAM,MAAM,QACR,KAAK;AAAA,MACH;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IACA,KAAK;AAAA,MACH;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACJ,WAAO,MAAM,MAAM,GAAG,IAAI;AAAA,EAC5B;AAAA,EAEA,aAAyB;AACvB,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,IACF,EAAE,IAAI,OAAO;AAAA,EACf;AAAA,EAEA,aAAa,MAAc,SAA4B;AACrD,UAAM,KAAK,OAAO,WAAW;AAC7B,UAAM,MAAM,KAAK,IAAI;AAErB,UAAM,EAAE,cAAc,IAAI,KAAK;AAAA,MAC7B,EAAE,eAAe,UAAU;AAAA,MAC3B;AAAA,MACA;AAAA,IACF;AACA,UAAM,WAAW,iBAAiB,KAAK;AAEvC,SAAK;AAAA,MACH;AAAA,IACF,EAAE,IAAI,IAAI,MAAM,SAAS,SAAS,GAAG;AACrC,WAAO,EAAE,IAAI,MAAM,SAAS,SAAS,YAAY,IAAI;AAAA,EACvD;AAAA,EAEA,cAA2B;AACzB,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,IACF,EAAE,IAAI,QAAQ;AAAA,EAChB;AAAA,EAEA,UAAU,IAAmC;AAC3C,UAAM,MAAM,KAAK;AAAA,MACf;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO,MAAM,SAAS,GAAG,IAAI;AAAA,EAC/B;AAAA,EAEA,aAAa,IAAkB;AAC7B,SAAK,MAAM,kCAAkC,EAAE,IAAI,EAAE;AAAA,EACvD;AAAA,EAEA,SAAS,IAAkB;AACzB;AAAA,MACE,KAAK;AAAA,MACL,MAAM;AACJ,aAAK,MAAM,oCAAoC,EAAE,IAAI,EAAE;AACvD,aAAK;AAAA,UACH;AAAA,QACF,EAAE,IAAI,WAAW,KAAK,IAAI,GAAG,EAAE;AAAA,MACjC;AAAA,MACA,EAAE,MAAM,WAAW;AAAA,IACrB;AAAA,EACF;AAAA,EAEA,UAAU,IAAkB;AAC1B,SAAK,MAAM,+BAA+B,EAAE,IAAI,EAAE;AAAA,EACpD;AAAA,EAEA,YAAY,IAAkB;AAC5B,SAAK,MAAM,iCAAiC,EAAE,IAAI,EAAE;AAAA,EACtD;AACF;",
6
6
  "names": []
7
7
  }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@deepagents/evals",
3
- "version": "7.1.2",
3
+ "version": "7.1.4",
4
4
  "license": "MIT",
5
5
  "repository": {
6
6
  "type": "git",
@@ -76,6 +76,8 @@
76
76
  "access": "public"
77
77
  },
78
78
  "dependencies": {
79
+ "@zukhruf/sqlite-drift": "^0.3.33",
80
+ "@zukhruf/sqlite-transaction": "^0.3.33",
79
81
  "chalk": "^5.6.2",
80
82
  "openai": "^6.48.0",
81
83
  "p-memoize": "^8.0.0"