failproofai 1.0.3 → 1.0.4-beta.2
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.
- package/.next/standalone/.next/BUILD_ID +1 -1
- package/.next/standalone/.next/build-manifest.json +5 -5
- package/.next/standalone/.next/prerender-manifest.json +5 -5
- package/.next/standalone/.next/required-server-files.json +1 -1
- package/.next/standalone/.next/server/app/_global-error/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/_global-error/page/server-reference-manifest.json +1 -1
- package/.next/standalone/.next/server/app/_global-error/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/_global-error/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/_global-error.html +1 -1
- package/.next/standalone/.next/server/app/_global-error.rsc +7 -7
- package/.next/standalone/.next/server/app/_global-error.segments/__PAGE__.segment.rsc +6 -6
- package/.next/standalone/.next/server/app/_global-error.segments/_full.segment.rsc +7 -7
- package/.next/standalone/.next/server/app/_global-error.segments/_tree.segment.rsc +1 -1
- package/.next/standalone/.next/server/app/_not-found/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/_not-found/page/server-reference-manifest.json +1 -1
- package/.next/standalone/.next/server/app/_not-found/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/_not-found/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/_not-found.html +1 -1
- package/.next/standalone/.next/server/app/_not-found.rsc +15 -15
- package/.next/standalone/.next/server/app/_not-found.segments/_full.segment.rsc +15 -15
- package/.next/standalone/.next/server/app/_not-found.segments/_not-found/__PAGE__.segment.rsc +14 -14
- package/.next/standalone/.next/server/app/_not-found.segments/_tree.segment.rsc +2 -2
- package/.next/standalone/.next/server/app/api/audit/invite/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/audit/invite/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/audit/run/route.js +4 -4
- package/.next/standalone/.next/server/app/api/audit/run/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/audit/run/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/audit/status/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/audit/status/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/auth/login-request/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/auth/login-request/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/auth/login-verify/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/auth/login-verify/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/auth/logout/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/auth/logout/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/auth/status/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/auth/status/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/api/download/[project]/[session]/route.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/api/download/[project]/[session]/route_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/audit/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/audit/page/server-reference-manifest.json +37 -4
- package/.next/standalone/.next/server/app/audit/page.js +2 -2
- package/.next/standalone/.next/server/app/audit/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/audit/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/index.html +1 -1
- package/.next/standalone/.next/server/app/index.rsc +15 -15
- package/.next/standalone/.next/server/app/index.segments/__PAGE__.segment.rsc +14 -14
- package/.next/standalone/.next/server/app/index.segments/_full.segment.rsc +15 -15
- package/.next/standalone/.next/server/app/index.segments/_tree.segment.rsc +2 -2
- package/.next/standalone/.next/server/app/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/page/server-reference-manifest.json +1 -1
- package/.next/standalone/.next/server/app/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/policies/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/policies/page/server-reference-manifest.json +14 -14
- package/.next/standalone/.next/server/app/policies/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/policies/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/project/[name]/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/project/[name]/page/server-reference-manifest.json +1 -1
- package/.next/standalone/.next/server/app/project/[name]/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/project/[name]/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/project/[name]/session/[sessionId]/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/project/[name]/session/[sessionId]/page/react-loadable-manifest.json +2 -2
- package/.next/standalone/.next/server/app/project/[name]/session/[sessionId]/page/server-reference-manifest.json +2 -2
- package/.next/standalone/.next/server/app/project/[name]/session/[sessionId]/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/project/[name]/session/[sessionId]/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/projects/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/projects/page/server-reference-manifest.json +1 -1
- package/.next/standalone/.next/server/app/projects/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/projects/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/app/settings/page/build-manifest.json +2 -2
- package/.next/standalone/.next/server/app/settings/page/server-reference-manifest.json +4 -4
- package/.next/standalone/.next/server/app/settings/page.js.nft.json +1 -1
- package/.next/standalone/.next/server/app/settings/page_client-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/chunks/{[externals]__0-9ti7z._.js → [externals]__1lh7m5d._.js} +1 -1
- package/.next/standalone/.next/server/chunks/{[externals]__0u3wrvq._.js → [externals]__1rqkg_y._.js} +1 -1
- package/.next/standalone/.next/server/chunks/[root-of-the-server]__0o07qi9._.js +1 -1
- package/.next/standalone/.next/server/chunks/[root-of-the-server]__1adacul._.js +1 -1
- package/.next/standalone/.next/server/chunks/{_1-akg6w._.js → _08w6xzm._.js} +2 -2
- package/.next/standalone/.next/server/chunks/_0tovk6q._.js +1 -1
- package/.next/standalone/.next/server/chunks/_0trp3yc._.js +1 -1
- package/.next/standalone/.next/server/chunks/_1ek68ln._.js +4 -4
- package/.next/standalone/.next/server/chunks/lib_telemetry_ts_0dctyyw._.js +1 -1
- package/.next/standalone/.next/server/chunks/node_modules_posthog-node_dist_entrypoints_index_node_mjs_01r25oi._.js +2 -2
- package/.next/standalone/.next/server/chunks/node_modules_posthog-node_dist_entrypoints_index_node_mjs_09z9-p7._.js +2 -2
- package/.next/standalone/.next/server/chunks/package_json_[json]_cjs_1nxcc4v._.js +1 -1
- package/.next/standalone/.next/server/chunks/src_hooks_fp-home_ts_09kv0bn._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__013jr2b._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__01wy8d-._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__02npjtd._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__0da85px._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__0ftmoxc._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__0oa1lav._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__0p-5p8u._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__0rgu2r3._.js +3 -0
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__0s740oi._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__1p2otjt._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/_0-oij9d._.js +23 -0
- package/.next/standalone/.next/server/chunks/ssr/_08x1r5t._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/_0l2hi_d._.js +3 -0
- package/.next/standalone/.next/server/chunks/ssr/_0oq1dh6._.js +3 -0
- package/.next/standalone/.next/server/chunks/ssr/_1es2j7i._.js +47 -13
- package/.next/standalone/.next/server/chunks/ssr/_1u8-lu2._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/_next-internal_server_app_policies_page_actions_1sp2-yo.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/app_audit__components_audit-dashboard_tsx_0p9ud47._.js +1 -69
- package/.next/standalone/.next/server/chunks/ssr/app_audit__components_rerun-button_tsx_0blihgw._.js +3 -0
- package/.next/standalone/.next/server/chunks/ssr/app_global-error_tsx_1kp6l3x._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/app_policies_hooks-client_tsx_19dqvpc._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/app_settings_settings-client_tsx_20lq-mq._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/node_modules_13n822a._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/node_modules_posthog-node_dist_entrypoints_index_node_mjs_11bnuzn._.js +2 -2
- package/.next/standalone/.next/server/chunks/ssr/src_hooks_builtin-policies_ts_09j2ndl._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/src_hooks_fp-config_ts_04t589g._.js +1 -1
- package/.next/standalone/.next/server/chunks/ssr/src_hooks_fp-home_ts_0je3xkv._.js +1 -1
- package/.next/standalone/.next/server/middleware-build-manifest.js +5 -5
- package/.next/standalone/.next/server/pages/404.html +1 -1
- package/.next/standalone/.next/server/pages/500.html +1 -1
- package/.next/standalone/.next/server/server-reference-manifest.js +1 -1
- package/.next/standalone/.next/server/server-reference-manifest.json +54 -21
- package/.next/standalone/.next/static/chunks/010bv1w6j171t.js +1 -0
- package/.next/standalone/.next/static/chunks/04r6ch8uf_n8m.js +1 -0
- package/.next/standalone/.next/static/chunks/{29-iqnp9v_axt.css → 0o-hh5_turzlz.css} +1 -1
- package/.next/standalone/.next/static/chunks/{40nr6ehj4p3cw.css → 0qmsunv04x4na.css} +1 -1
- package/.next/standalone/.next/static/chunks/{0ollg-nnu3t3e.js → 0wz8yftk18ts2.js} +1 -1
- package/.next/standalone/.next/static/chunks/0zebh1n9jkfbt.js +1 -0
- package/.next/standalone/.next/static/chunks/1eyl7fzkmnkf_.js +1 -0
- package/.next/standalone/.next/static/chunks/1pb1oztsbwcss.js +1 -0
- package/.next/standalone/.next/static/chunks/{1-_zkmnm-jwei.js → 2aquitk72k2op.js} +1 -1
- package/.next/standalone/.next/static/chunks/2bi_1y0a_smt7.js +6 -0
- package/.next/standalone/.next/static/chunks/{42yfkhekx440s.js → 2ej3b8gk5ittu.js} +1 -1
- package/.next/standalone/.next/static/chunks/{354-n7p7labpn.js → 2zafgs90r_leq.js} +1 -1
- package/.next/standalone/.next/static/chunks/{2xm5mj1d21b4n.js → 32spub4wqjem-.js} +1 -1
- package/.next/standalone/.next/static/chunks/{1a6u7l3giejch.js → 3m4upvybtrexd.js} +1 -1
- package/.next/standalone/.next/static/chunks/{2k9f4tyv04809.css → 3uyhmm01g64k4.css} +1 -0
- package/.next/standalone/.next/static/chunks/{turbopack-2x-o099q64i8h.js → turbopack-3s3c-u_u_l0u9.js} +1 -1
- package/.next/standalone/.opencode/plugins/failproofai.mjs +51 -27
- package/.next/standalone/app/actions/get-leaks.ts +100 -0
- package/.next/standalone/app/audit/_components/audit-dashboard.tsx +112 -33
- package/.next/standalone/app/audit/_components/audit-poster.tsx +29 -10
- package/.next/standalone/app/audit/_components/come-back-better-section.tsx +6 -1
- package/.next/standalone/app/audit/_components/empty-state.tsx +4 -1
- package/.next/standalone/app/audit/_components/how-to-improve-section.tsx +19 -5
- package/.next/standalone/app/audit/_components/leak-section.tsx +179 -0
- package/.next/standalone/app/audit/_components/share-templates.ts +110 -44
- package/.next/standalone/app/audit/audit-styles.css +114 -1
- package/.next/standalone/fp-cloud-cli/CHANGELOG.md +26 -3
- package/.next/standalone/fp-cloud-cli/fp_cli/_click_compat.py +31 -8
- package/.next/standalone/fp-cloud-cli/fp_cli/permissions.py +1 -0
- package/.next/standalone/fp-cloud-cli/pyproject.toml +9 -2
- package/.next/standalone/fp-cloud-cli/tests/test_click_compat.py +54 -0
- package/.next/standalone/fp-cloud-cli/uv.lock +9 -12
- package/.next/standalone/lib/auth/api-server-client.ts +28 -0
- package/.next/standalone/lib/claude-sessions.ts +101 -27
- package/.next/standalone/node_modules/@next/env/package.json +1 -1
- package/.next/standalone/node_modules/next/dist/build/swc/index.js +1 -1
- package/.next/standalone/node_modules/next/dist/compiled/next-server/app-page-turbo-experimental.runtime.prod.js +1 -1
- package/.next/standalone/node_modules/next/dist/compiled/next-server/app-page-turbo.runtime.prod.js +1 -1
- package/.next/standalone/node_modules/next/dist/compiled/next-server/app-route-turbo.runtime.prod.js +1 -1
- package/.next/standalone/node_modules/next/dist/compiled/next-server/pages-turbo.runtime.prod.js +5 -5
- package/.next/standalone/node_modules/next/dist/experimental/testmode/fetch.js +10 -2
- package/.next/standalone/node_modules/next/dist/lib/patch-incorrect-lockfile.js +3 -3
- package/.next/standalone/node_modules/next/dist/lib/typescript/runTypeScriptCli.js +2 -2
- package/.next/standalone/node_modules/next/dist/lib/verify-typescript-setup.js +3 -1
- package/.next/standalone/node_modules/next/dist/server/config.js +1 -1
- package/.next/standalone/node_modules/next/dist/server/dev/hot-reloader-turbopack.js +2 -2
- package/.next/standalone/node_modules/next/dist/server/dev/hot-reloader-webpack.js +1 -1
- package/.next/standalone/node_modules/next/dist/server/lib/app-info-log.js +1 -1
- package/.next/standalone/node_modules/next/dist/server/lib/incremental-cache/file-system-cache.js +17 -4
- package/.next/standalone/node_modules/next/dist/server/lib/start-server.js +1 -1
- package/.next/standalone/node_modules/next/dist/shared/lib/router/utils/escape-path-delimiters.js +1 -1
- package/.next/standalone/node_modules/next/dist/telemetry/anonymous-meta.js +1 -1
- package/.next/standalone/node_modules/next/dist/telemetry/events/swc-load-failure.js +1 -1
- package/.next/standalone/node_modules/next/dist/telemetry/events/version.js +2 -2
- package/.next/standalone/node_modules/next/package.json +11 -11
- package/.next/standalone/package.json +14 -13
- package/.next/standalone/sdk/python/CHANGELOG.md +155 -0
- package/.next/standalone/sdk/python/README.md +8 -0
- package/.next/standalone/sdk/python/examples/evaluator_worker.py +121 -0
- package/.next/standalone/sdk/python/failproofai_sdk/_version.py +1 -1
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/__init__.py +101 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/__main__.py +49 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/_sandbox_runner.py +63 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/authoring.py +404 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/client.py +299 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/protocol.py +754 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/runtime.py +936 -0
- package/.next/standalone/sdk/python/failproofai_sdk/evaluator/source.py +684 -0
- package/.next/standalone/sdk/python/tests/fixtures/evaluator_v2/README.md +28 -0
- package/.next/standalone/sdk/python/tests/fixtures/evaluator_v2/contract.json +252 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_authoring.py +124 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_client.py +253 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_example.py +35 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_http_e2e.py +636 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_main.py +47 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_protocol.py +246 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_review_fixes.py +164 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_runtime.py +1103 -0
- package/.next/standalone/sdk/python/tests/test_evaluator_source.py +430 -0
- package/.next/standalone/sdk/python/tests/test_zero_dependencies.py +17 -0
- package/.next/standalone/sdk/python/uv.lock +29 -29
- package/.next/standalone/server.js +1 -1
- package/README.md +13 -58
- package/dist/cli.mjs +5368 -8756
- package/dist/worker.mjs +539 -312
- package/lib/auth/api-server-client.ts +28 -0
- package/lib/claude-sessions.ts +101 -27
- package/package.json +14 -13
- package/pi-extension/index.ts +42 -4
- package/scripts/changelog-open.py +115 -0
- package/src/audit/cli.ts +134 -0
- package/src/audit/desktop-notify.ts +420 -0
- package/src/audit/harm-report.ts +101 -0
- package/src/audit/index.ts +141 -13
- package/src/audit/leak-fingerprint.ts +200 -0
- package/src/audit/leak-notice.ts +161 -0
- package/src/audit/leak-record.ts +232 -0
- package/src/audit/leak-scan.ts +292 -0
- package/src/audit/leak-store.ts +217 -0
- package/src/audit/macos-notifier.ts +310 -0
- package/src/audit/redact-example.ts +266 -14
- package/src/audit/report-harm.ts +12 -1
- package/src/audit/report.ts +6 -3
- package/src/audit/schedule-cli.ts +37 -0
- package/src/audit/scoring.ts +49 -0
- package/src/audit/types.ts +41 -0
- package/src/hooks/builtin-policies.ts +67 -2
- package/src/hooks/configure-wizard.ts +20 -0
- package/src/hooks/fp-config.ts +71 -11
- package/src/hooks/fp-home.ts +27 -0
- package/src/hooks/handler.ts +59 -1
- package/src/hooks/integrations.ts +47 -13
- package/src/hooks/notice.ts +155 -0
- package/src/hooks/uninstall-cli.ts +15 -0
- package/.next/standalone/.next/server/chunks/ssr/[root-of-the-server]__1b5jzxr._.js +0 -3
- package/.next/standalone/.next/server/chunks/ssr/_0j6gd_k._.js +0 -23
- package/.next/standalone/.next/server/chunks/ssr/_0v8eebd._.js +0 -3
- package/.next/standalone/.next/server/chunks/ssr/_1ylae7o._.js +0 -3
- package/.next/standalone/.next/server/chunks/ssr/_1zopuov._.js +0 -3
- package/.next/standalone/.next/server/chunks/ssr/node_modules_html-to-image_es_index_1ao30b1.js +0 -3
- package/.next/standalone/.next/static/chunks/03slu0c8_71y7.js +0 -1
- package/.next/standalone/.next/static/chunks/0nm11ysn0153h.js +0 -1
- package/.next/standalone/.next/static/chunks/0u64yey446k4f.css +0 -1
- package/.next/standalone/.next/static/chunks/1r9676mkvihto.js +0 -1
- package/.next/standalone/.next/static/chunks/30pct5s734jua.js +0 -6
- package/.next/standalone/.next/static/chunks/3pibxnbhjza7n.js +0 -69
- package/.next/standalone/.next/static/chunks/3qfoeqhy5l10x.js +0 -1
- package/.next/standalone/.next/static/chunks/3zkg2s2vzxc3d.js +0 -1
- package/.next/standalone/.next/static/chunks/41b5ptaocvi_l.js +0 -1
- /package/.next/standalone/.next/static/{JBYgkXkA1PS48m9P_hKNw → aKNv4Kmv98Xpwns0Xa-C-}/_buildManifest.js +0 -0
- /package/.next/standalone/.next/static/{JBYgkXkA1PS48m9P_hKNw → aKNv4Kmv98Xpwns0Xa-C-}/_clientMiddlewareManifest.js +0 -0
- /package/.next/standalone/.next/static/{JBYgkXkA1PS48m9P_hKNw → aKNv4Kmv98Xpwns0Xa-C-}/_ssgManifest.js +0 -0
|
@@ -0,0 +1,754 @@
|
|
|
1
|
+
"""Dependency-free wire models for the outbound Evaluator v2 protocol."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
import math
|
|
6
|
+
from collections.abc import Mapping
|
|
7
|
+
from dataclasses import asdict, dataclass
|
|
8
|
+
from enum import Enum
|
|
9
|
+
from typing import Any, TypeVar
|
|
10
|
+
|
|
11
|
+
PROTOCOL_VERSION = "2"
|
|
12
|
+
TRANSCRIPT_SCHEMA_VERSION = "2"
|
|
13
|
+
RESULT_SCHEMA_VERSION = "2"
|
|
14
|
+
|
|
15
|
+
REGISTER_PATH = "/v1/evaluator/workers/register"
|
|
16
|
+
CLAIM_PATH = "/v1/evaluator/assignments/claim"
|
|
17
|
+
TRANSCRIPT_PATH = "/v1/evaluator/assignments/{assignment_id}/transcript"
|
|
18
|
+
DEFINITIONS_PATH = "/v1/evaluator/assignments/{assignment_id}/definitions"
|
|
19
|
+
PLAN_PATH = "/v1/evaluator/assignments/{assignment_id}/plan"
|
|
20
|
+
HEARTBEAT_PATH = "/v1/evaluator/runs/heartbeat"
|
|
21
|
+
RESULT_PATH = "/v1/evaluator/runs/{evaluation_run_id}/result"
|
|
22
|
+
WORKER_ID_HEADER = "X-FailproofAI-Worker-Id"
|
|
23
|
+
LEASE_GENERATION_HEADER = "X-FailproofAI-Lease-Generation"
|
|
24
|
+
|
|
25
|
+
HEARTBEAT_INTERVAL_SECONDS = 30
|
|
26
|
+
LEASE_DURATION_SECONDS = 120
|
|
27
|
+
# Fallback poll cadence if the register response omits poll_interval_seconds. The
|
|
28
|
+
# worker prefers the server-advertised value; claims are normal short polls, never
|
|
29
|
+
# long-polls, so this only bounds idle latency, not connection lifetime.
|
|
30
|
+
DEFAULT_POLL_INTERVAL_SECONDS = 10
|
|
31
|
+
MAX_ATTEMPTS = 5
|
|
32
|
+
|
|
33
|
+
MAX_CATALOG_DEFINITIONS = 100
|
|
34
|
+
MAX_CLAIM_CAPACITY = 32
|
|
35
|
+
MAX_TRANSCRIPT_BYTES = 25 * 1024 * 1024
|
|
36
|
+
MAX_RESULTS_PER_RUN = 25
|
|
37
|
+
MAX_EVAL_KEY_BYTES = 128
|
|
38
|
+
MAX_DISPLAY_NAME_BYTES = 128
|
|
39
|
+
MAX_VERSION_BYTES = 128
|
|
40
|
+
MAX_WORKER_ID_BYTES = 128
|
|
41
|
+
MAX_LABEL_BYTES = 64
|
|
42
|
+
MAX_LABELS_PER_RESULT = 20
|
|
43
|
+
MAX_SUMMARY_BYTES = 4 * 1024
|
|
44
|
+
MAX_REASONING_BYTES = 16 * 1024
|
|
45
|
+
MAX_UNIT_BYTES = 64
|
|
46
|
+
MAX_DISPLAY_VALUE_BYTES = 256
|
|
47
|
+
MAX_DESCRIPTION_BYTES = 1_000
|
|
48
|
+
MAX_ERROR_CODE_BYTES = 64
|
|
49
|
+
MAX_ERROR_MESSAGE_BYTES = 4 * 1024
|
|
50
|
+
|
|
51
|
+
ERROR_SPECS = {
|
|
52
|
+
"invalid_credentials": {"http_status": 401, "retryable": False},
|
|
53
|
+
"instance_disabled": {"http_status": 403, "retryable": False},
|
|
54
|
+
"insufficient_permissions": {"http_status": 403, "retryable": False},
|
|
55
|
+
"assignment_not_found": {"http_status": 404, "retryable": False},
|
|
56
|
+
"run_not_found": {"http_status": 404, "retryable": False},
|
|
57
|
+
"catalog_mismatch": {"http_status": 409, "retryable": False},
|
|
58
|
+
"lease_lost": {"http_status": 409, "retryable": False},
|
|
59
|
+
"plan_conflict": {"http_status": 409, "retryable": False},
|
|
60
|
+
"submission_conflict": {"http_status": 409, "retryable": False},
|
|
61
|
+
"retry_budget_exhausted": {"http_status": 409, "retryable": False},
|
|
62
|
+
"transcript_too_large": {"http_status": 413, "retryable": False},
|
|
63
|
+
"invalid_request": {"http_status": 422, "retryable": False},
|
|
64
|
+
"invalid_catalog": {"http_status": 422, "retryable": False},
|
|
65
|
+
"incomplete_plan": {"http_status": 422, "retryable": False},
|
|
66
|
+
"unsupported_protocol_version": {"http_status": 426, "retryable": False},
|
|
67
|
+
"internal_error": {"http_status": 500, "retryable": True},
|
|
68
|
+
}
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
class ProtocolError(ValueError):
|
|
72
|
+
"""A local or remote evaluator protocol contract violation."""
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
class UnsupportedProtocolVersion(ProtocolError):
|
|
76
|
+
def __init__(self, received: str) -> None:
|
|
77
|
+
super().__init__(
|
|
78
|
+
f"unsupported evaluator protocol version {received!r}; "
|
|
79
|
+
f"supported major version is {PROTOCOL_VERSION}"
|
|
80
|
+
)
|
|
81
|
+
self.received = received
|
|
82
|
+
|
|
83
|
+
|
|
84
|
+
def validate_protocol_version(version: str) -> None:
|
|
85
|
+
if version != PROTOCOL_VERSION:
|
|
86
|
+
raise UnsupportedProtocolVersion(version)
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
class EvaluatorKind(str, Enum):
|
|
90
|
+
MANAGED = "managed"
|
|
91
|
+
CUSTOMER = "customer"
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
class ResultKind(str, Enum):
|
|
95
|
+
SCORE = "score"
|
|
96
|
+
METRIC = "metric"
|
|
97
|
+
ASSERTION = "assertion"
|
|
98
|
+
|
|
99
|
+
|
|
100
|
+
class ExecutionMode(str, Enum):
|
|
101
|
+
LOCAL = "local"
|
|
102
|
+
PYTHON = "python"
|
|
103
|
+
|
|
104
|
+
|
|
105
|
+
class TerminalRunStatus(str, Enum):
|
|
106
|
+
SUCCEEDED = "succeeded"
|
|
107
|
+
FAILED = "failed"
|
|
108
|
+
TIMED_OUT = "timed_out"
|
|
109
|
+
CANCELLED = "cancelled"
|
|
110
|
+
|
|
111
|
+
|
|
112
|
+
_EnumT = TypeVar("_EnumT", bound=Enum)
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
def _wire(value: Any) -> Any:
|
|
116
|
+
if isinstance(value, Enum):
|
|
117
|
+
return value.value
|
|
118
|
+
if hasattr(value, "__dataclass_fields__"):
|
|
119
|
+
return {key: _wire(item) for key, item in asdict(value).items()}
|
|
120
|
+
if isinstance(value, (list, tuple)):
|
|
121
|
+
return [_wire(item) for item in value]
|
|
122
|
+
if isinstance(value, dict):
|
|
123
|
+
return {key: _wire(item) for key, item in value.items()}
|
|
124
|
+
return value
|
|
125
|
+
|
|
126
|
+
|
|
127
|
+
class WireModel:
|
|
128
|
+
def to_wire(self) -> dict[str, Any]:
|
|
129
|
+
return _wire(self)
|
|
130
|
+
|
|
131
|
+
|
|
132
|
+
def _string(data: Mapping[str, Any], key: str) -> str:
|
|
133
|
+
value = data.get(key)
|
|
134
|
+
if not isinstance(value, str):
|
|
135
|
+
raise ProtocolError(f"{key} must be a string")
|
|
136
|
+
return value
|
|
137
|
+
|
|
138
|
+
|
|
139
|
+
def _integer(data: Mapping[str, Any], key: str) -> int:
|
|
140
|
+
value = data.get(key)
|
|
141
|
+
if isinstance(value, bool) or not isinstance(value, int):
|
|
142
|
+
raise ProtocolError(f"{key} must be an integer")
|
|
143
|
+
return value
|
|
144
|
+
|
|
145
|
+
|
|
146
|
+
def _list(data: Mapping[str, Any], key: str) -> list[Any]:
|
|
147
|
+
value = data.get(key)
|
|
148
|
+
if not isinstance(value, list):
|
|
149
|
+
raise ProtocolError(f"{key} must be an array")
|
|
150
|
+
return value
|
|
151
|
+
|
|
152
|
+
|
|
153
|
+
def _object(value: Any, field_name: str) -> Mapping[str, Any]:
|
|
154
|
+
if not isinstance(value, Mapping):
|
|
155
|
+
raise ProtocolError(f"{field_name} must be an object")
|
|
156
|
+
return value
|
|
157
|
+
|
|
158
|
+
|
|
159
|
+
def _object_list(data: Mapping[str, Any], key: str) -> tuple[Mapping[str, Any], ...]:
|
|
160
|
+
return tuple(
|
|
161
|
+
_object(value, f"{key}[{index}]")
|
|
162
|
+
for index, value in enumerate(_list(data, key))
|
|
163
|
+
)
|
|
164
|
+
|
|
165
|
+
|
|
166
|
+
def _string_list(data: Mapping[str, Any], key: str) -> tuple[str, ...]:
|
|
167
|
+
values = _list(data, key)
|
|
168
|
+
for index, value in enumerate(values):
|
|
169
|
+
if not isinstance(value, str):
|
|
170
|
+
raise ProtocolError(f"{key}[{index}] must be a string")
|
|
171
|
+
return tuple(values)
|
|
172
|
+
|
|
173
|
+
|
|
174
|
+
def _enum(enum_type: type[_EnumT], data: Mapping[str, Any], key: str) -> _EnumT:
|
|
175
|
+
value = _string(data, key)
|
|
176
|
+
try:
|
|
177
|
+
return enum_type(value)
|
|
178
|
+
except ValueError as error:
|
|
179
|
+
allowed = ", ".join(repr(item.value) for item in enum_type)
|
|
180
|
+
raise ProtocolError(f"{key} must be one of {allowed}") from error
|
|
181
|
+
|
|
182
|
+
|
|
183
|
+
def _positive_integer(data: Mapping[str, Any], key: str) -> int:
|
|
184
|
+
value = _integer(data, key)
|
|
185
|
+
if value <= 0:
|
|
186
|
+
raise ProtocolError(f"{key} must be greater than zero")
|
|
187
|
+
return value
|
|
188
|
+
|
|
189
|
+
|
|
190
|
+
def _nonnegative_integer(data: Mapping[str, Any], key: str) -> int:
|
|
191
|
+
value = _integer(data, key)
|
|
192
|
+
if value < 0:
|
|
193
|
+
raise ProtocolError(f"{key} must not be negative")
|
|
194
|
+
return value
|
|
195
|
+
|
|
196
|
+
|
|
197
|
+
def _optional_string(data: Mapping[str, Any], key: str) -> str | None:
|
|
198
|
+
value = data.get(key)
|
|
199
|
+
if value is not None and not isinstance(value, str):
|
|
200
|
+
raise ProtocolError(f"{key} must be a string or null")
|
|
201
|
+
return value
|
|
202
|
+
|
|
203
|
+
|
|
204
|
+
@dataclass(frozen=True)
|
|
205
|
+
class CatalogDefinition(WireModel):
|
|
206
|
+
eval_key: str
|
|
207
|
+
display_name: str
|
|
208
|
+
eval_version: str
|
|
209
|
+
result_kind: ResultKind
|
|
210
|
+
labels: tuple[str, ...] = ()
|
|
211
|
+
|
|
212
|
+
@classmethod
|
|
213
|
+
def from_wire(cls, data: Mapping[str, Any]) -> CatalogDefinition:
|
|
214
|
+
return cls(
|
|
215
|
+
eval_key=_string(data, "eval_key"),
|
|
216
|
+
display_name=_string(data, "display_name"),
|
|
217
|
+
eval_version=_string(data, "eval_version"),
|
|
218
|
+
result_kind=_enum(ResultKind, data, "result_kind"),
|
|
219
|
+
labels=_string_list(data, "labels"),
|
|
220
|
+
)
|
|
221
|
+
|
|
222
|
+
|
|
223
|
+
@dataclass(frozen=True)
|
|
224
|
+
class RegisterRequest(WireModel):
|
|
225
|
+
worker_id: str
|
|
226
|
+
sdk_version: str
|
|
227
|
+
catalog_revision: str
|
|
228
|
+
max_concurrency: int
|
|
229
|
+
definitions: tuple[CatalogDefinition, ...]
|
|
230
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
231
|
+
|
|
232
|
+
@classmethod
|
|
233
|
+
def from_wire(cls, data: Mapping[str, Any]) -> RegisterRequest:
|
|
234
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
235
|
+
return cls(
|
|
236
|
+
worker_id=_string(data, "worker_id"),
|
|
237
|
+
sdk_version=_string(data, "sdk_version"),
|
|
238
|
+
catalog_revision=_string(data, "catalog_revision"),
|
|
239
|
+
max_concurrency=_integer(data, "max_concurrency"),
|
|
240
|
+
definitions=tuple(
|
|
241
|
+
CatalogDefinition.from_wire(item)
|
|
242
|
+
for item in _object_list(data, "definitions")
|
|
243
|
+
),
|
|
244
|
+
)
|
|
245
|
+
|
|
246
|
+
|
|
247
|
+
@dataclass(frozen=True)
|
|
248
|
+
class RegisterResponse(WireModel):
|
|
249
|
+
evaluator_instance_id: str
|
|
250
|
+
evaluator_kind: EvaluatorKind
|
|
251
|
+
heartbeat_interval_seconds: int
|
|
252
|
+
lease_duration_seconds: int
|
|
253
|
+
poll_interval_seconds: int
|
|
254
|
+
claim_limit: int
|
|
255
|
+
disabled_definitions: tuple[str, ...] = ()
|
|
256
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
257
|
+
|
|
258
|
+
@classmethod
|
|
259
|
+
def from_wire(cls, data: Mapping[str, Any]) -> RegisterResponse:
|
|
260
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
261
|
+
return cls(
|
|
262
|
+
evaluator_instance_id=_string(data, "evaluator_instance_id"),
|
|
263
|
+
evaluator_kind=_enum(EvaluatorKind, data, "evaluator_kind"),
|
|
264
|
+
heartbeat_interval_seconds=_integer(data, "heartbeat_interval_seconds"),
|
|
265
|
+
lease_duration_seconds=_integer(data, "lease_duration_seconds"),
|
|
266
|
+
poll_interval_seconds=_integer(data, "poll_interval_seconds"),
|
|
267
|
+
claim_limit=_integer(data, "claim_limit"),
|
|
268
|
+
disabled_definitions=_string_list(data, "disabled_definitions"),
|
|
269
|
+
)
|
|
270
|
+
|
|
271
|
+
|
|
272
|
+
@dataclass(frozen=True)
|
|
273
|
+
class ClaimRequest(WireModel):
|
|
274
|
+
worker_id: str
|
|
275
|
+
catalog_revision: str
|
|
276
|
+
capacity: int
|
|
277
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
278
|
+
|
|
279
|
+
@classmethod
|
|
280
|
+
def from_wire(cls, data: Mapping[str, Any]) -> ClaimRequest:
|
|
281
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
282
|
+
return cls(
|
|
283
|
+
worker_id=_string(data, "worker_id"),
|
|
284
|
+
catalog_revision=_string(data, "catalog_revision"),
|
|
285
|
+
capacity=_integer(data, "capacity"),
|
|
286
|
+
)
|
|
287
|
+
|
|
288
|
+
|
|
289
|
+
@dataclass(frozen=True)
|
|
290
|
+
class Assignment(WireModel):
|
|
291
|
+
assignment_id: str
|
|
292
|
+
lease_generation: int
|
|
293
|
+
lease_expires_at: str
|
|
294
|
+
session_id: str
|
|
295
|
+
session_revision_id: str
|
|
296
|
+
agent_id: str
|
|
297
|
+
environment: str
|
|
298
|
+
trigger_reason: str
|
|
299
|
+
event_count: int
|
|
300
|
+
transcript_url: str
|
|
301
|
+
definitions_url: str = ""
|
|
302
|
+
|
|
303
|
+
@classmethod
|
|
304
|
+
def from_wire(cls, data: Mapping[str, Any]) -> Assignment:
|
|
305
|
+
return cls(
|
|
306
|
+
assignment_id=_string(data, "assignment_id"),
|
|
307
|
+
lease_generation=_positive_integer(data, "lease_generation"),
|
|
308
|
+
lease_expires_at=_string(data, "lease_expires_at"),
|
|
309
|
+
session_id=_string(data, "session_id"),
|
|
310
|
+
session_revision_id=_string(data, "session_revision_id"),
|
|
311
|
+
agent_id=_string(data, "agent_id"),
|
|
312
|
+
environment=_string(data, "environment"),
|
|
313
|
+
trigger_reason=_string(data, "trigger_reason"),
|
|
314
|
+
event_count=_nonnegative_integer(data, "event_count"),
|
|
315
|
+
transcript_url=_string(data, "transcript_url"),
|
|
316
|
+
definitions_url=str(data.get("definitions_url") or ""),
|
|
317
|
+
)
|
|
318
|
+
|
|
319
|
+
|
|
320
|
+
@dataclass(frozen=True)
|
|
321
|
+
class AssignmentDefinition(WireModel):
|
|
322
|
+
eval_key: str
|
|
323
|
+
display_name: str
|
|
324
|
+
eval_version: str
|
|
325
|
+
result_kind: ResultKind
|
|
326
|
+
labels: tuple[str, ...] = ()
|
|
327
|
+
execution_mode: ExecutionMode = ExecutionMode.LOCAL
|
|
328
|
+
condition_source: str | None = None
|
|
329
|
+
source_checksum: str | None = None
|
|
330
|
+
timeout_seconds: float | None = None
|
|
331
|
+
|
|
332
|
+
@classmethod
|
|
333
|
+
def from_wire(cls, data: Mapping[str, Any]) -> AssignmentDefinition:
|
|
334
|
+
timeout = data.get("timeout_seconds")
|
|
335
|
+
if timeout is not None:
|
|
336
|
+
if isinstance(timeout, bool) or not isinstance(timeout, (int, float)):
|
|
337
|
+
raise ProtocolError("timeout_seconds must be a number or null")
|
|
338
|
+
timeout = float(timeout)
|
|
339
|
+
if not math.isfinite(timeout) or timeout <= 0:
|
|
340
|
+
raise ProtocolError("timeout_seconds must be finite and greater than zero")
|
|
341
|
+
return cls(
|
|
342
|
+
eval_key=_string(data, "eval_key"),
|
|
343
|
+
display_name=_string(data, "display_name"),
|
|
344
|
+
eval_version=_string(data, "eval_version"),
|
|
345
|
+
result_kind=_enum(ResultKind, data, "result_kind"),
|
|
346
|
+
labels=_string_list(data, "labels"),
|
|
347
|
+
# Require execution_mode explicitly. Coercing a falsy/missing value to
|
|
348
|
+
# 'local' silently ran a server-authored ('python') definition down the
|
|
349
|
+
# customer-local path (or vice-versa); a malformed wire value is a
|
|
350
|
+
# protocol error, not a default (F2).
|
|
351
|
+
execution_mode=_enum(ExecutionMode, data, "execution_mode"),
|
|
352
|
+
condition_source=_optional_string(data, "condition_source"),
|
|
353
|
+
source_checksum=_optional_string(data, "source_checksum"),
|
|
354
|
+
timeout_seconds=timeout,
|
|
355
|
+
)
|
|
356
|
+
|
|
357
|
+
|
|
358
|
+
@dataclass(frozen=True)
|
|
359
|
+
class DefinitionsResponse(WireModel):
|
|
360
|
+
assignment_id: str
|
|
361
|
+
catalog_revision: str
|
|
362
|
+
definitions: tuple[AssignmentDefinition, ...]
|
|
363
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
364
|
+
|
|
365
|
+
@classmethod
|
|
366
|
+
def from_wire(cls, data: Mapping[str, Any]) -> DefinitionsResponse:
|
|
367
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
368
|
+
return cls(
|
|
369
|
+
assignment_id=_string(data, "assignment_id"),
|
|
370
|
+
catalog_revision=_string(data, "catalog_revision"),
|
|
371
|
+
definitions=tuple(
|
|
372
|
+
AssignmentDefinition.from_wire(item)
|
|
373
|
+
for item in _object_list(data, "definitions")
|
|
374
|
+
),
|
|
375
|
+
)
|
|
376
|
+
|
|
377
|
+
|
|
378
|
+
@dataclass(frozen=True)
|
|
379
|
+
class ClaimResponse(WireModel):
|
|
380
|
+
assignments: tuple[Assignment, ...]
|
|
381
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
382
|
+
|
|
383
|
+
@classmethod
|
|
384
|
+
def from_wire(cls, data: Mapping[str, Any]) -> ClaimResponse:
|
|
385
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
386
|
+
return cls(
|
|
387
|
+
tuple(
|
|
388
|
+
Assignment.from_wire(item) for item in _object_list(data, "assignments")
|
|
389
|
+
)
|
|
390
|
+
)
|
|
391
|
+
|
|
392
|
+
|
|
393
|
+
@dataclass(frozen=True)
|
|
394
|
+
class TranscriptEvent(WireModel):
|
|
395
|
+
id: str
|
|
396
|
+
ts: str
|
|
397
|
+
event_type: str
|
|
398
|
+
payload: Mapping[str, Any]
|
|
399
|
+
|
|
400
|
+
@classmethod
|
|
401
|
+
def from_wire(cls, data: Mapping[str, Any]) -> TranscriptEvent:
|
|
402
|
+
payload = data.get("payload")
|
|
403
|
+
if not isinstance(payload, Mapping):
|
|
404
|
+
raise ProtocolError("payload must be an object")
|
|
405
|
+
return cls(
|
|
406
|
+
id=_string(data, "id"),
|
|
407
|
+
ts=_string(data, "ts"),
|
|
408
|
+
event_type=_string(data, "event_type"),
|
|
409
|
+
payload=dict(payload),
|
|
410
|
+
)
|
|
411
|
+
|
|
412
|
+
|
|
413
|
+
@dataclass(frozen=True)
|
|
414
|
+
class SessionTranscript(WireModel):
|
|
415
|
+
assignment_id: str
|
|
416
|
+
session_id: str
|
|
417
|
+
session_revision_id: str
|
|
418
|
+
agent_id: str
|
|
419
|
+
environment: str
|
|
420
|
+
started_at: str
|
|
421
|
+
ended_at: str
|
|
422
|
+
event_count: int
|
|
423
|
+
events: tuple[TranscriptEvent, ...]
|
|
424
|
+
schema_version: str = TRANSCRIPT_SCHEMA_VERSION
|
|
425
|
+
|
|
426
|
+
@classmethod
|
|
427
|
+
def from_wire(cls, data: Mapping[str, Any]) -> SessionTranscript:
|
|
428
|
+
version = _string(data, "schema_version")
|
|
429
|
+
if version != TRANSCRIPT_SCHEMA_VERSION:
|
|
430
|
+
raise ProtocolError(f"unsupported transcript schema version {version!r}")
|
|
431
|
+
events = tuple(
|
|
432
|
+
TranscriptEvent.from_wire(item) for item in _object_list(data, "events")
|
|
433
|
+
)
|
|
434
|
+
event_count = _nonnegative_integer(data, "event_count")
|
|
435
|
+
if event_count != len(events):
|
|
436
|
+
raise ProtocolError(
|
|
437
|
+
f"event_count is {event_count}, but transcript contains {len(events)} events"
|
|
438
|
+
)
|
|
439
|
+
return cls(
|
|
440
|
+
assignment_id=_string(data, "assignment_id"),
|
|
441
|
+
session_id=_string(data, "session_id"),
|
|
442
|
+
session_revision_id=_string(data, "session_revision_id"),
|
|
443
|
+
agent_id=_string(data, "agent_id"),
|
|
444
|
+
environment=_string(data, "environment"),
|
|
445
|
+
started_at=_string(data, "started_at"),
|
|
446
|
+
ended_at=_string(data, "ended_at"),
|
|
447
|
+
event_count=event_count,
|
|
448
|
+
events=events,
|
|
449
|
+
)
|
|
450
|
+
|
|
451
|
+
def events_of_type(self, event_type: str) -> tuple[TranscriptEvent, ...]:
|
|
452
|
+
return tuple(event for event in self.events if event.event_type == event_type)
|
|
453
|
+
|
|
454
|
+
def count(self, event_type: str) -> int:
|
|
455
|
+
return sum(event.event_type == event_type for event in self.events)
|
|
456
|
+
|
|
457
|
+
|
|
458
|
+
@dataclass(frozen=True)
|
|
459
|
+
class EvalSelection(WireModel):
|
|
460
|
+
eval_key: str
|
|
461
|
+
eval_version: str
|
|
462
|
+
|
|
463
|
+
@classmethod
|
|
464
|
+
def from_wire(cls, data: Mapping[str, Any]) -> EvalSelection:
|
|
465
|
+
return cls(_string(data, "eval_key"), _string(data, "eval_version"))
|
|
466
|
+
|
|
467
|
+
|
|
468
|
+
@dataclass(frozen=True)
|
|
469
|
+
class SkippedEval(WireModel):
|
|
470
|
+
eval_key: str
|
|
471
|
+
eval_version: str
|
|
472
|
+
reason_code: str
|
|
473
|
+
|
|
474
|
+
@classmethod
|
|
475
|
+
def from_wire(cls, data: Mapping[str, Any]) -> SkippedEval:
|
|
476
|
+
return cls(
|
|
477
|
+
_string(data, "eval_key"),
|
|
478
|
+
_string(data, "eval_version"),
|
|
479
|
+
_string(data, "reason_code"),
|
|
480
|
+
)
|
|
481
|
+
|
|
482
|
+
|
|
483
|
+
@dataclass(frozen=True)
|
|
484
|
+
class PlanRequest(WireModel):
|
|
485
|
+
worker_id: str
|
|
486
|
+
lease_generation: int
|
|
487
|
+
selected: tuple[EvalSelection, ...] = ()
|
|
488
|
+
skipped: tuple[SkippedEval, ...] = ()
|
|
489
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
490
|
+
|
|
491
|
+
@classmethod
|
|
492
|
+
def from_wire(cls, data: Mapping[str, Any]) -> PlanRequest:
|
|
493
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
494
|
+
return cls(
|
|
495
|
+
worker_id=_string(data, "worker_id"),
|
|
496
|
+
lease_generation=_positive_integer(data, "lease_generation"),
|
|
497
|
+
selected=tuple(
|
|
498
|
+
EvalSelection.from_wire(item) for item in _object_list(data, "selected")
|
|
499
|
+
),
|
|
500
|
+
skipped=tuple(
|
|
501
|
+
SkippedEval.from_wire(item) for item in _object_list(data, "skipped")
|
|
502
|
+
),
|
|
503
|
+
)
|
|
504
|
+
|
|
505
|
+
|
|
506
|
+
@dataclass(frozen=True)
|
|
507
|
+
class PlannedRun(WireModel):
|
|
508
|
+
evaluation_run_id: str
|
|
509
|
+
eval_key: str
|
|
510
|
+
eval_version: str
|
|
511
|
+
execution_mode: ExecutionMode = ExecutionMode.LOCAL
|
|
512
|
+
evaluator_source: str | None = None
|
|
513
|
+
source_checksum: str | None = None
|
|
514
|
+
timeout_seconds: float | None = None
|
|
515
|
+
|
|
516
|
+
@classmethod
|
|
517
|
+
def from_wire(cls, data: Mapping[str, Any]) -> PlannedRun:
|
|
518
|
+
timeout = data.get("timeout_seconds")
|
|
519
|
+
if timeout is not None:
|
|
520
|
+
if isinstance(timeout, bool) or not isinstance(timeout, (int, float)):
|
|
521
|
+
raise ProtocolError("timeout_seconds must be a number or null")
|
|
522
|
+
timeout = float(timeout)
|
|
523
|
+
if not math.isfinite(timeout) or timeout <= 0:
|
|
524
|
+
raise ProtocolError("timeout_seconds must be finite and greater than zero")
|
|
525
|
+
return cls(
|
|
526
|
+
evaluation_run_id=_string(data, "evaluation_run_id"),
|
|
527
|
+
eval_key=_string(data, "eval_key"),
|
|
528
|
+
eval_version=_string(data, "eval_version"),
|
|
529
|
+
# Require execution_mode explicitly. Coercing a falsy/missing value to
|
|
530
|
+
# 'local' silently ran a server-authored ('python') definition down the
|
|
531
|
+
# customer-local path (or vice-versa); a malformed wire value is a
|
|
532
|
+
# protocol error, not a default (F2).
|
|
533
|
+
execution_mode=_enum(ExecutionMode, data, "execution_mode"),
|
|
534
|
+
evaluator_source=_optional_string(data, "evaluator_source"),
|
|
535
|
+
source_checksum=_optional_string(data, "source_checksum"),
|
|
536
|
+
timeout_seconds=timeout,
|
|
537
|
+
)
|
|
538
|
+
|
|
539
|
+
|
|
540
|
+
@dataclass(frozen=True)
|
|
541
|
+
class PlanResponse(WireModel):
|
|
542
|
+
assignment_id: str
|
|
543
|
+
assignment_status: str
|
|
544
|
+
runs: tuple[PlannedRun, ...]
|
|
545
|
+
idempotent_replay: bool = False
|
|
546
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
547
|
+
|
|
548
|
+
@classmethod
|
|
549
|
+
def from_wire(cls, data: Mapping[str, Any]) -> PlanResponse:
|
|
550
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
551
|
+
replay = data.get("idempotent_replay", False)
|
|
552
|
+
if not isinstance(replay, bool):
|
|
553
|
+
raise ProtocolError("idempotent_replay must be a boolean")
|
|
554
|
+
return cls(
|
|
555
|
+
assignment_id=_string(data, "assignment_id"),
|
|
556
|
+
assignment_status=_string(data, "assignment_status"),
|
|
557
|
+
runs=tuple(
|
|
558
|
+
PlannedRun.from_wire(item) for item in _object_list(data, "runs")
|
|
559
|
+
),
|
|
560
|
+
idempotent_replay=replay,
|
|
561
|
+
)
|
|
562
|
+
|
|
563
|
+
|
|
564
|
+
@dataclass(frozen=True)
|
|
565
|
+
class HeartbeatRun(WireModel):
|
|
566
|
+
evaluation_run_id: str
|
|
567
|
+
state: str
|
|
568
|
+
progress: float | None = None
|
|
569
|
+
|
|
570
|
+
@classmethod
|
|
571
|
+
def from_wire(cls, data: Mapping[str, Any]) -> HeartbeatRun:
|
|
572
|
+
progress = data.get("progress")
|
|
573
|
+
if progress is not None:
|
|
574
|
+
if isinstance(progress, bool) or not isinstance(progress, (int, float)):
|
|
575
|
+
raise ProtocolError("progress must be a number or null")
|
|
576
|
+
progress = float(progress)
|
|
577
|
+
if not math.isfinite(progress) or not 0 <= progress <= 1:
|
|
578
|
+
raise ProtocolError("progress must be finite and between 0 and 1")
|
|
579
|
+
return cls(
|
|
580
|
+
evaluation_run_id=_string(data, "evaluation_run_id"),
|
|
581
|
+
state=_string(data, "state"),
|
|
582
|
+
progress=progress,
|
|
583
|
+
)
|
|
584
|
+
|
|
585
|
+
|
|
586
|
+
@dataclass(frozen=True)
|
|
587
|
+
class HeartbeatRequest(WireModel):
|
|
588
|
+
worker_id: str
|
|
589
|
+
lease_generation: int
|
|
590
|
+
runs: tuple[HeartbeatRun, ...]
|
|
591
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
592
|
+
|
|
593
|
+
@classmethod
|
|
594
|
+
def from_wire(cls, data: Mapping[str, Any]) -> HeartbeatRequest:
|
|
595
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
596
|
+
return cls(
|
|
597
|
+
worker_id=_string(data, "worker_id"),
|
|
598
|
+
lease_generation=_positive_integer(data, "lease_generation"),
|
|
599
|
+
runs=tuple(
|
|
600
|
+
HeartbeatRun.from_wire(item) for item in _object_list(data, "runs")
|
|
601
|
+
),
|
|
602
|
+
)
|
|
603
|
+
|
|
604
|
+
|
|
605
|
+
@dataclass(frozen=True)
|
|
606
|
+
class HeartbeatResponse(WireModel):
|
|
607
|
+
lease_expires_at: str
|
|
608
|
+
accepted_run_ids: tuple[str, ...]
|
|
609
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
610
|
+
|
|
611
|
+
@classmethod
|
|
612
|
+
def from_wire(cls, data: Mapping[str, Any]) -> HeartbeatResponse:
|
|
613
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
614
|
+
return cls(
|
|
615
|
+
lease_expires_at=_string(data, "lease_expires_at"),
|
|
616
|
+
accepted_run_ids=_string_list(data, "accepted_run_ids"),
|
|
617
|
+
)
|
|
618
|
+
|
|
619
|
+
|
|
620
|
+
@dataclass(frozen=True)
|
|
621
|
+
class ResultItem(WireModel):
|
|
622
|
+
result_key: str
|
|
623
|
+
result_kind: ResultKind
|
|
624
|
+
numeric_value: float | None = None
|
|
625
|
+
bool_value: bool | None = None
|
|
626
|
+
text_value: str | None = None
|
|
627
|
+
unit: str = ""
|
|
628
|
+
display_value: str | None = None
|
|
629
|
+
description: str | None = None
|
|
630
|
+
reasoning: str | None = None
|
|
631
|
+
labels: tuple[str, ...] = ()
|
|
632
|
+
|
|
633
|
+
@classmethod
|
|
634
|
+
def from_wire(cls, data: Mapping[str, Any]) -> ResultItem:
|
|
635
|
+
numeric = data.get("numeric_value")
|
|
636
|
+
if numeric is not None:
|
|
637
|
+
if isinstance(numeric, bool) or not isinstance(numeric, (int, float)):
|
|
638
|
+
raise ProtocolError("numeric_value must be a number or null")
|
|
639
|
+
numeric = float(numeric)
|
|
640
|
+
if not math.isfinite(numeric):
|
|
641
|
+
raise ProtocolError("numeric_value must be finite")
|
|
642
|
+
boolean = data.get("bool_value")
|
|
643
|
+
if boolean is not None and not isinstance(boolean, bool):
|
|
644
|
+
raise ProtocolError("bool_value must be a boolean or null")
|
|
645
|
+
return cls(
|
|
646
|
+
result_key=_string(data, "result_key"),
|
|
647
|
+
result_kind=_enum(ResultKind, data, "result_kind"),
|
|
648
|
+
numeric_value=numeric,
|
|
649
|
+
bool_value=boolean,
|
|
650
|
+
text_value=_optional_string(data, "text_value"),
|
|
651
|
+
unit=_string(data, "unit"),
|
|
652
|
+
display_value=_optional_string(data, "display_value"),
|
|
653
|
+
description=_optional_string(data, "description"),
|
|
654
|
+
reasoning=_optional_string(data, "reasoning"),
|
|
655
|
+
labels=_string_list(data, "labels"),
|
|
656
|
+
)
|
|
657
|
+
|
|
658
|
+
|
|
659
|
+
@dataclass(frozen=True)
|
|
660
|
+
class ResultRequest(WireModel):
|
|
661
|
+
submission_id: str
|
|
662
|
+
worker_id: str
|
|
663
|
+
lease_generation: int
|
|
664
|
+
status: TerminalRunStatus
|
|
665
|
+
started_at: str
|
|
666
|
+
finished_at: str
|
|
667
|
+
duration_ms: int
|
|
668
|
+
summary: str | None
|
|
669
|
+
results: tuple[ResultItem, ...]
|
|
670
|
+
error_code: str | None
|
|
671
|
+
error_message: str | None
|
|
672
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
673
|
+
result_schema_version: str = RESULT_SCHEMA_VERSION
|
|
674
|
+
|
|
675
|
+
@classmethod
|
|
676
|
+
def from_wire(cls, data: Mapping[str, Any]) -> ResultRequest:
|
|
677
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
678
|
+
result_schema_version = _string(data, "result_schema_version")
|
|
679
|
+
if result_schema_version != RESULT_SCHEMA_VERSION:
|
|
680
|
+
raise ProtocolError(
|
|
681
|
+
f"unsupported result schema version {result_schema_version!r}"
|
|
682
|
+
)
|
|
683
|
+
return cls(
|
|
684
|
+
submission_id=_string(data, "submission_id"),
|
|
685
|
+
worker_id=_string(data, "worker_id"),
|
|
686
|
+
lease_generation=_positive_integer(data, "lease_generation"),
|
|
687
|
+
status=_enum(TerminalRunStatus, data, "status"),
|
|
688
|
+
started_at=_string(data, "started_at"),
|
|
689
|
+
finished_at=_string(data, "finished_at"),
|
|
690
|
+
duration_ms=_nonnegative_integer(data, "duration_ms"),
|
|
691
|
+
summary=_optional_string(data, "summary"),
|
|
692
|
+
results=tuple(
|
|
693
|
+
ResultItem.from_wire(item) for item in _object_list(data, "results")
|
|
694
|
+
),
|
|
695
|
+
error_code=_optional_string(data, "error_code"),
|
|
696
|
+
error_message=_optional_string(data, "error_message"),
|
|
697
|
+
)
|
|
698
|
+
|
|
699
|
+
|
|
700
|
+
@dataclass(frozen=True)
|
|
701
|
+
class ResultResponse(WireModel):
|
|
702
|
+
evaluation_run_id: str
|
|
703
|
+
submission_id: str
|
|
704
|
+
status: str
|
|
705
|
+
idempotent_replay: bool
|
|
706
|
+
result_count: int
|
|
707
|
+
result_checksum: str
|
|
708
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
709
|
+
|
|
710
|
+
@classmethod
|
|
711
|
+
def from_wire(cls, data: Mapping[str, Any]) -> ResultResponse:
|
|
712
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
713
|
+
replay = data.get("idempotent_replay")
|
|
714
|
+
if not isinstance(replay, bool):
|
|
715
|
+
raise ProtocolError("idempotent_replay must be a boolean")
|
|
716
|
+
return cls(
|
|
717
|
+
evaluation_run_id=_string(data, "evaluation_run_id"),
|
|
718
|
+
submission_id=_string(data, "submission_id"),
|
|
719
|
+
status=_string(data, "status"),
|
|
720
|
+
idempotent_replay=replay,
|
|
721
|
+
result_count=_nonnegative_integer(data, "result_count"),
|
|
722
|
+
result_checksum=_string(data, "result_checksum"),
|
|
723
|
+
)
|
|
724
|
+
|
|
725
|
+
|
|
726
|
+
@dataclass(frozen=True)
|
|
727
|
+
class RemoteError(WireModel):
|
|
728
|
+
code: str
|
|
729
|
+
message: str
|
|
730
|
+
retryable: bool
|
|
731
|
+
request_id: str
|
|
732
|
+
|
|
733
|
+
@classmethod
|
|
734
|
+
def from_wire(cls, data: Mapping[str, Any]) -> RemoteError:
|
|
735
|
+
retryable = data.get("retryable")
|
|
736
|
+
if not isinstance(retryable, bool):
|
|
737
|
+
raise ProtocolError("retryable must be a boolean")
|
|
738
|
+
return cls(
|
|
739
|
+
code=_string(data, "code"),
|
|
740
|
+
message=_string(data, "message"),
|
|
741
|
+
retryable=retryable,
|
|
742
|
+
request_id=_string(data, "request_id"),
|
|
743
|
+
)
|
|
744
|
+
|
|
745
|
+
|
|
746
|
+
@dataclass(frozen=True)
|
|
747
|
+
class ErrorResponse(WireModel):
|
|
748
|
+
error: RemoteError
|
|
749
|
+
protocol_version: str = PROTOCOL_VERSION
|
|
750
|
+
|
|
751
|
+
@classmethod
|
|
752
|
+
def from_wire(cls, data: Mapping[str, Any]) -> ErrorResponse:
|
|
753
|
+
validate_protocol_version(_string(data, "protocol_version"))
|
|
754
|
+
return cls(RemoteError.from_wire(_object(data.get("error"), "error")))
|