@bitfab/sdk 0.74.2 → 0.74.4

package/dist/{registryCommand-BQcyBmhN.js → registryCommand-Bm2-Peq8.js}

@@ -644,6 +644,7 @@ const VALUE_FLAGS = /* @__PURE__ */ new Set([
 	"--dataset-id",
 	"--dataset-ids",
 	"--grader-ids",
+	"--category-ids",
 	"--mock",
 	"--param",
 	"--params",
@@ -688,6 +689,7 @@ const RESUME_CONFLICTING_FLAGS = [
 	["attempts", "--attempts"],
 	["graderIds", "--grader-ids"],
 	["onlyWithAssertions", "--only-with-assertions"],
+	["categoryIds", "--category-ids"],
 	["dryRun", "--dry-run"]
 ];
 const RESUME_CLEARED_OPTIONS = {
@@ -698,26 +700,30 @@ const RESUME_CLEARED_OPTIONS = {
 	attempts: void 0,
 	graderIds: void 0,
 	onlyWithAssertions: void 0,
+	categoryIds: void 0,
 	dryRun: void 0
 };
-function hasApprovedAssertion(assertions) {
-	return assertions.some((assertion) => assertion.approvalState === "approved");
+function narrowsToAssertions(narrowing) {
+	return narrowing.onlyWithAssertions || narrowing.categoryIds !== void 0;
 }
-async function boundTraceIds(client, traceIds, limit, onlyWithAssertions) {
+function hasApprovedAssertion(assertions, categoryIds) {
+	return assertions.some((assertion) => assertion.approvalState === "approved" && (categoryIds === void 0 || assertion.category_id !== null && categoryIds.includes(assertion.category_id)));
+}
+async function boundTraceIds(client, traceIds, limit, narrowing) {
 	if (!Number.isInteger(limit) || limit < 1) throw new BitfabError("Replay limit must be a positive integer.");
-	if (!onlyWithAssertions || limit >= traceIds.length) return traceIds.slice(0, limit);
+	if (!narrowsToAssertions(narrowing) || limit >= traceIds.length) return traceIds.slice(0, limit);
 	const selected = [];
 	for (const traceId of traceIds) {
 		const { assertions, inheritedFrom } = await client.traces.getAssertions(traceId);
-		if (inheritedFrom === null && hasApprovedAssertion(assertions)) {
+		if (inheritedFrom === null && hasApprovedAssertion(assertions, narrowing.categoryIds)) {
 			selected.push(traceId);
 			if (selected.length === limit) break;
 		}
 	}
-	if (selected.length === 0) throw new BitfabError("No traces with approved assertions matched this replay selection.");
+	if (selected.length === 0) throw new BitfabError(narrowing.categoryIds === void 0 ? "No traces with approved assertions matched this replay selection." : "No traces with an approved assertion in the given categories matched this replay selection.");
 	return selected;
 }
-async function pinnedDatasetMembers(client, datasetIds, limit, onlyWithAssertions = false) {
+async function pinnedDatasetMembers(client, datasetIds, limit, narrowing = { onlyWithAssertions: false }) {
 	const members = /* @__PURE__ */ new Set();
 	for (const datasetId of datasetIds) {
 		const { traceIds } = await client.datasets.listTraces(datasetId);
@@ -726,10 +732,10 @@ async function pinnedDatasetMembers(client, datasetIds, limit, onlyWithAssertion
 	const ordered = [...members].sort();
 	if (limit >= ordered.length) return;
 	if (limit > MAX_PINNED_TRACE_IDS) throw new BitfabError(`Bounding a dataset replay to ${limit} names one trace ID per replayed item, and a replay accepts at most ${MAX_PINNED_TRACE_IDS}. Drop the limit to replay every selected dataset in full.`);
-	return boundTraceIds(client, ordered, limit, onlyWithAssertions);
+	return boundTraceIds(client, ordered, limit, narrowing);
 }
 function usage(registry) {
-	return `Usage: bitfab-replay --registry <path> <${registry === void 0 ? "pipeline" : Object.keys(registry).join("|")}> [options]\n\nOptions:\n  --limit N (bounds a trace ID or dataset selection instead of replacing it)\n  --attempts N (replay each trace N times in this run, max ${MAX_ATTEMPTS})\n  --trace-ids id1,id2\n  --name NAME (what this run is testing, such as "baseline"; Bitfab already records the commit, branch, and datasets)\n  --notes TEXT (run conditions Bitfab cannot see, such as a forced feature flag)\n  --metadata key=value (repeatable; tags kept on the experiment, such as which schedule launched it, and filterable through client.experiments.list)\n  --concurrency N, --max-concurrency N\n  --code-change working-tree|last-branch|default-branch|ref|skip|manual (required: the code change this run tests; working-tree is the uncommitted changes vs HEAD, last-branch the diff against the branch this one was cut from, default-branch the diff against the default branch, ref the diff against --code-change-base, skip none, manual the edits in --code-change-file)\n  --code-change-base BRANCH|COMMIT (with --code-change ref: the branch or commit to diff against)\n  --code-change-file PATH (with --code-change manual: a JSON file of { description, files })\n  --experiment-group-id UUID (join this experiment group; a new UUID creates the group)\n  --experiment-group-name NAME (name the group when this run creates it or it has no name yet; requires --experiment-group-id)\n  --dataset-ids uuid1,uuid2 (--dataset-id is the same flag)\n  --grader-ids id1,id2\n  --only-with-assertions (replay only traces carrying an approved assertion)\n  --skip-assertion-judging (don't judge approved assertions on each replay; judging is on by default and costs model calls)\n  --judge-assertions (deprecated: judging is on by default)\n  --mock none|all|marked\n  --db-branch, --no-db-branch\n  --dry-run (resolve inputs, run nothing)\n  --fail-on-error (exit 1 after printing the result when any replayed item errored; under --dry-run, items whose inputs failed to resolve count)\n  --resume EXPERIMENT_ID (finish an experiment that stopped partway, running only the items it did not finish, with the traces, attempts, and graders it started with)\n  --force (with --resume, resume even when the experiment may still be running or the code changed since it started)\n  --params PATH\n  --param name=value (repeatable)\n  -h, --help`;
+	return `Usage: bitfab-replay --registry <path> <${registry === void 0 ? "pipeline" : Object.keys(registry).join("|")}> [options]\n\nOptions:\n  --limit N (bounds a trace ID or dataset selection instead of replacing it)\n  --attempts N (replay each trace N times in this run, max ${MAX_ATTEMPTS})\n  --trace-ids id1,id2\n  --name NAME (what this run is testing, such as "baseline"; Bitfab already records the commit, branch, and datasets)\n  --notes TEXT (run conditions Bitfab cannot see, such as a forced feature flag)\n  --metadata key=value (repeatable; tags kept on the experiment, such as which schedule launched it, and filterable through client.experiments.list)\n  --concurrency N, --max-concurrency N\n  --code-change working-tree|last-branch|default-branch|ref|skip|manual (required: the code change this run tests; working-tree is the uncommitted changes vs HEAD, last-branch the diff against the branch this one was cut from, default-branch the diff against the default branch, ref the diff against --code-change-base, skip none, manual the edits in --code-change-file)\n  --code-change-base BRANCH|COMMIT (with --code-change ref: the branch or commit to diff against)\n  --code-change-file PATH (with --code-change manual: a JSON file of { description, files })\n  --experiment-group-id UUID (join this experiment group; a new UUID creates the group)\n  --experiment-group-name NAME (name the group when this run creates it or it has no name yet; requires --experiment-group-id)\n  --dataset-ids uuid1,uuid2 (--dataset-id is the same flag)\n  --grader-ids id1,id2\n  --only-with-assertions (replay only traces carrying an approved assertion)\n  --category-ids id1,id2 (with --trace-ids or --dataset-ids: replay only the traces carrying an approved assertion in one of these categories; which assertions are judged stays the same)\n  --skip-assertion-judging (don't judge approved assertions on each replay; judging is on by default and costs model calls)\n  --judge-assertions (deprecated: judging is on by default)\n  --mock none|all|marked\n  --db-branch, --no-db-branch\n  --dry-run (resolve inputs, run nothing)\n  --fail-on-error (exit 1 after printing the result when any replayed item errored; under --dry-run, items whose inputs failed to resolve count)\n  --resume EXPERIMENT_ID (finish an experiment that stopped partway, running only the items it did not finish, with the traces, attempts, and graders it started with)\n  --force (with --resume, resume even when the experiment may still be running or the code changed since it started)\n  --params PATH\n  --param name=value (repeatable)\n  -h, --help`;
 }
 var ReplayCliHelp = class extends Error {};
 async function reseedFromRegistry(registry, pipeline, traceIds) {
@@ -939,6 +945,9 @@ function parseReplayCliArgs(registry, argv) {
 			case "--grader-ids":
 				parsed.graderIds = commaSeparated(flag, raw);
 				break;
+			case "--category-ids":
+				parsed.categoryIds = commaSeparated(flag, raw);
+				break;
 			case "--mock":
 				if (raw !== "none" && raw !== "all" && raw !== "marked") throw new BitfabError(`--mock must be one of: none, all, marked (received '${raw}').`);
 				parsed.mock = raw;
@@ -1164,6 +1173,7 @@ async function runReplayCli(registry, argv = typeof process === "undefined" ? []
 		datasetIds: selectedDatasetIds,
 		...args.graderIds === void 0 ? {} : { graderIds: args.graderIds },
 		...args.onlyWithAssertions === void 0 ? {} : { onlyWithAssertions: args.onlyWithAssertions },
+		...args.categoryIds === void 0 ? {} : { categoryIds: args.categoryIds },
 		...args.skipAssertionJudging === void 0 ? {} : { skipAssertionJudging: args.skipAssertionJudging },
 		...args.mock === void 0 ? {} : { mock: args.mock },
 		...args.dbBranch === void 0 ? {} : { dbBranch: args.dbBranch && registrationOptions.dbBranch !== void 0 && registrationOptions.dbBranch !== false ? registrationOptions.dbBranch : args.dbBranch },
@@ -1197,13 +1207,18 @@ async function runReplayCli(registry, argv = typeof process === "undefined" ? []
 			options.traceIds = args.traceIds;
 			options.datasetIds = void 0;
 		} else if (args.datasetIds !== void 0) options.traceIds = void 0;
+		if (options.categoryIds !== void 0 && options.traceIds === void 0 && options.datasetIds === void 0) throw new BitfabError("--category-ids narrows a trace ID or dataset selection. Pass it with --trace-ids or --dataset-ids.");
 		const bound = args.limit ?? registrationOptions.limit;
 		options.limit = void 0;
+		const narrowing = {
+			onlyWithAssertions: options.onlyWithAssertions === true,
+			categoryIds: options.categoryIds
+		};
 		if (options.traceIds !== void 0) {
-			if (bound !== void 0) options.traceIds = await boundTraceIds(registration.client, options.traceIds, bound, options.onlyWithAssertions === true);
+			if (bound !== void 0) options.traceIds = await boundTraceIds(registration.client, options.traceIds, bound, narrowing);
 		} else if (options.datasetIds !== void 0) {
 			if (bound !== void 0) {
-				const pinned = await pinnedDatasetMembers(registration.client, options.datasetIds, bound, options.onlyWithAssertions === true);
+				const pinned = await pinnedDatasetMembers(registration.client, options.datasetIds, bound, narrowing);
 				if (pinned !== void 0) options.traceIds = pinned;
 			}
 		} else options.limit = bound ?? 10;
@@ -1323,4 +1338,4 @@ function runIfEntrypoint(moduleUrl, main) {
 //#endregion
 export { runRegistryMain as a, runSeedCli as c, runIfEntrypoint as i, parseRegistryCommandArgs as n, checkReplayCliArgs as o, resolveRegistry as r, runReplayCli as s, importRegistry as t };
 
-//# sourceMappingURL=registryCommand-BQcyBmhN.js.map
+//# sourceMappingURL=registryCommand-Bm2-Peq8.js.map