nullius-eval 0.1.0__py3-none-any.whl

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.
Files changed (291) hide show
  1. _nullius_app/.env.example +55 -0
  2. _nullius_app/app.py +96 -0
  3. _nullius_app/requirements.txt +1 -0
  4. _nullius_app/static/brand/README.md +20 -0
  5. _nullius_app/static/brand/docker.svg +1 -0
  6. _nullius_app/static/brand/favicon.svg +1 -0
  7. _nullius_app/static/brand/huggingface.svg +37 -0
  8. _nullius_app/static/brand/mark-dark.svg +1 -0
  9. _nullius_app/static/brand/marks.svg +34 -0
  10. _nullius_app/static/brand/tokens.json +51 -0
  11. _nullius_app/static/brand/typesafe.png +0 -0
  12. _nullius_app/static/brand/wordmark-dark.svg +1 -0
  13. _nullius_app/static/builder/command-builder-core.js +763 -0
  14. _nullius_app/static/builder/command-builder.css +248 -0
  15. _nullius_app/static/builder/command-builder.js +997 -0
  16. _nullius_app/static/builder/model-picker.js +113 -0
  17. _nullius_app/static/catalogue/add.js +46 -0
  18. _nullius_app/static/catalogue/agents/README.md +19 -0
  19. _nullius_app/static/catalogue/agents/goose.svg +1 -0
  20. _nullius_app/static/catalogue/agents/jiuwenswarm.svg +33 -0
  21. _nullius_app/static/catalogue/agents/marks.json +8 -0
  22. _nullius_app/static/catalogue/agents/mini-swe-agent.svg +7 -0
  23. _nullius_app/static/catalogue/agents/opencode.svg +7 -0
  24. _nullius_app/static/catalogue/catalogue.css +436 -0
  25. _nullius_app/static/catalogue/catalogue.js +87 -0
  26. _nullius_app/static/catalogue/marks.svg +29 -0
  27. _nullius_app/static/catalogue/new.js +78 -0
  28. _nullius_app/static/catalogue/versions.js +27 -0
  29. _nullius_app/static/compare/comparison.css +117 -0
  30. _nullius_app/static/controls.js +117 -0
  31. _nullius_app/static/disclosures.js +70 -0
  32. _nullius_app/static/fonts/0569c822cab6c510.ttf +0 -0
  33. _nullius_app/static/fonts/DMSans-Medium.woff2 +0 -0
  34. _nullius_app/static/fonts/DMSans-Regular.woff2 +0 -0
  35. _nullius_app/static/fonts/IBMPlexMono-Medium.woff2 +0 -0
  36. _nullius_app/static/fonts/IBMPlexMono-Regular.woff2 +0 -0
  37. _nullius_app/static/fonts/InstrumentSerif-Italic.woff2 +0 -0
  38. _nullius_app/static/fonts/InstrumentSerif-Regular.woff2 +0 -0
  39. _nullius_app/static/fonts/README.md +11 -0
  40. _nullius_app/static/fonts/c1e7453ea26a1d1e.ttf +0 -0
  41. _nullius_app/static/fonts/dmsans-OFL.txt +93 -0
  42. _nullius_app/static/fonts/ibmplexmono-OFL.txt +93 -0
  43. _nullius_app/static/fonts/instrumentserif-OFL.txt +93 -0
  44. _nullius_app/static/fonts.css +49 -0
  45. _nullius_app/static/notifications.js +49 -0
  46. _nullius_app/static/onboarding.css +78 -0
  47. _nullius_app/static/onboarding.js +158 -0
  48. _nullius_app/static/runs/board.css +358 -0
  49. _nullius_app/static/runs/board.js +127 -0
  50. _nullius_app/static/runs/launches.css +88 -0
  51. _nullius_app/static/runs/live-board.css +186 -0
  52. _nullius_app/static/runs/live-board.js +394 -0
  53. _nullius_app/static/runs/path-suggestions.js +119 -0
  54. _nullius_app/static/runs/start.css +38 -0
  55. _nullius_app/static/settings/autosave.js +162 -0
  56. _nullius_app/static/settings/custom-keys.js +100 -0
  57. _nullius_app/static/settings/data-providers.js +88 -0
  58. _nullius_app/static/settings/judges.js +61 -0
  59. _nullius_app/static/settings/modal.css +32 -0
  60. _nullius_app/static/settings/modal.js +91 -0
  61. _nullius_app/static/settings/navigation.js +55 -0
  62. _nullius_app/static/settings/prices.js +68 -0
  63. _nullius_app/static/settings/providers.js +134 -0
  64. _nullius_app/static/settings/sandbox.js +20 -0
  65. _nullius_app/static/settings/settings.css +288 -0
  66. _nullius_app/static/style.css +1752 -0
  67. _nullius_app/static/traces/conversation.css +49 -0
  68. _nullius_app/static/traces/conversation.js +215 -0
  69. _nullius_app/templates/agents/entry.html +145 -0
  70. _nullius_app/templates/agents/kits.html +55 -0
  71. _nullius_app/templates/agents/standard.html +58 -0
  72. _nullius_app/templates/base.html +479 -0
  73. _nullius_app/templates/builder/new_run.html +55 -0
  74. _nullius_app/templates/catalogue/add_dialog.html +32 -0
  75. _nullius_app/templates/catalogue/benchmark.html +270 -0
  76. _nullius_app/templates/catalogue/benchmarks.html +91 -0
  77. _nullius_app/templates/catalogue/new.html +141 -0
  78. _nullius_app/templates/catalogue/use_bundle.html +11 -0
  79. _nullius_app/templates/compare/_detail.html +37 -0
  80. _nullius_app/templates/compare/compare.html +298 -0
  81. _nullius_app/templates/compare/compare_picker.html +51 -0
  82. _nullius_app/templates/compare/compare_task.html +329 -0
  83. _nullius_app/templates/runs/_execution.html +131 -0
  84. _nullius_app/templates/runs/_runtime.html +5 -0
  85. _nullius_app/templates/runs/ascii-mark.html +485 -0
  86. _nullius_app/templates/runs/feasibility.html +127 -0
  87. _nullius_app/templates/runs/index.html +609 -0
  88. _nullius_app/templates/runs/job.html +460 -0
  89. _nullius_app/templates/runs/preparing.html +14 -0
  90. _nullius_app/templates/runs/reconcile.html +124 -0
  91. _nullius_app/templates/runs/setup.html +272 -0
  92. _nullius_app/templates/runs/start.html +51 -0
  93. _nullius_app/templates/settings/custom_keys.html +82 -0
  94. _nullius_app/templates/settings/data_providers.html +27 -0
  95. _nullius_app/templates/settings/settings.html +564 -0
  96. _nullius_app/templates/traces/_charts.html +165 -0
  97. _nullius_app/templates/traces/_conversation.html +37 -0
  98. _nullius_app/templates/traces/_judgement.html +72 -0
  99. _nullius_app/templates/traces/artifact.html +15 -0
  100. _nullius_app/templates/traces/trial.html +838 -0
  101. _nullius_app/templates/workspaces/workspaces.html +54 -0
  102. _nullius_app/web/__init__.py +1 -0
  103. _nullius_app/web/agents.py +230 -0
  104. _nullius_app/web/builder.py +406 -0
  105. _nullius_app/web/catalogue.py +370 -0
  106. _nullius_app/web/compare.py +489 -0
  107. _nullius_app/web/core.py +17 -0
  108. _nullius_app/web/execute.py +223 -0
  109. _nullius_app/web/feasibility.py +60 -0
  110. _nullius_app/web/filters.py +231 -0
  111. _nullius_app/web/judge.py +623 -0
  112. _nullius_app/web/judge_tasks.py +33 -0
  113. _nullius_app/web/onboarding.py +20 -0
  114. _nullius_app/web/reconcile.py +47 -0
  115. _nullius_app/web/runs.py +464 -0
  116. _nullius_app/web/settings.py +324 -0
  117. _nullius_app/web/shared.py +507 -0
  118. _nullius_app/web/traces.py +306 -0
  119. _nullius_app/web/workspaces.py +84 -0
  120. nullius/__init__.py +32 -0
  121. nullius/__main__.py +5 -0
  122. nullius/catalogue/__init__.py +0 -0
  123. nullius/catalogue/agents/codex/agent.json +222 -0
  124. nullius/catalogue/agents/codex/config.toml +25 -0
  125. nullius/catalogue/agents/goose/agent.json +131 -0
  126. nullius/catalogue/agents/goose/config.yaml +20 -0
  127. nullius/catalogue/agents/jiuwenswarm/agent.json +197 -0
  128. nullius/catalogue/agents/jiuwenswarm/agent.py +355 -0
  129. nullius/catalogue/agents/mini-swe-agent/agent.json +68 -0
  130. nullius/catalogue/agents/mini-swe-agent/run.py +37 -0
  131. nullius/catalogue/agents/opencode/agent.json +135 -0
  132. nullius/catalogue/agents/opencode/opencode.json +32 -0
  133. nullius/catalogue/agents/reference/agent.json +89 -0
  134. nullius/catalogue/agents/reference/agent.py +209 -0
  135. nullius/catalogue/agents/terminus-2/agent.json +64 -0
  136. nullius/catalogue/agents/terminus-2/run.py +139 -0
  137. nullius/catalogue/benchmarks/gaia/benchmark.json +72 -0
  138. nullius/catalogue/benchmarks/gaia/data-source.json +29 -0
  139. nullius/catalogue/benchmarks/gaia/driver.py +188 -0
  140. nullius/catalogue/benchmarks/gaia/env/data.Dockerfile +1 -0
  141. nullius/catalogue/benchmarks/gaia/standard.json +278 -0
  142. nullius/catalogue/benchmarks/gaia/upstream-settings.json +275 -0
  143. nullius/catalogue/benchmarks/gsm8k/benchmark.json +34 -0
  144. nullius/catalogue/benchmarks/gsm8k/driver.py +101 -0
  145. nullius/catalogue/benchmarks/gsm8k/standard.json +124 -0
  146. nullius/catalogue/benchmarks/gsm8k/upstream-settings.json +86 -0
  147. nullius/catalogue/benchmarks/humaneval/benchmark.json +36 -0
  148. nullius/catalogue/benchmarks/humaneval/driver.py +243 -0
  149. nullius/catalogue/benchmarks/humaneval/standard.json +115 -0
  150. nullius/catalogue/benchmarks/humaneval/upstream-settings.json +72 -0
  151. nullius/catalogue/benchmarks/locomo/benchmark.json +43 -0
  152. nullius/catalogue/benchmarks/locomo/driver.py +270 -0
  153. nullius/catalogue/benchmarks/locomo/standard.json +265 -0
  154. nullius/catalogue/benchmarks/locomo/upstream-settings.json +112 -0
  155. nullius/catalogue/benchmarks/longmemeval/benchmark.json +65 -0
  156. nullius/catalogue/benchmarks/longmemeval/driver.py +379 -0
  157. nullius/catalogue/benchmarks/longmemeval/standard.json +733 -0
  158. nullius/catalogue/benchmarks/longmemeval/upstream-settings.json +76 -0
  159. nullius/catalogue/benchmarks/swe-bench-verified/benchmark.json +87 -0
  160. nullius/catalogue/benchmarks/swe-bench-verified/driver.py +116 -0
  161. nullius/catalogue/benchmarks/swe-bench-verified/images.json +502 -0
  162. nullius/catalogue/benchmarks/swe-bench-verified/standard.json +108 -0
  163. nullius/catalogue/benchmarks/swe-bench-verified/upstream-settings.json +32 -0
  164. nullius/catalogue/benchmarks/tau3-bench/benchmark.json +49 -0
  165. nullius/catalogue/benchmarks/tau3-bench/driver.py +608 -0
  166. nullius/catalogue/benchmarks/tau3-bench/mcp_server.py +114 -0
  167. nullius/catalogue/benchmarks/tau3-bench/search_relay.py +19 -0
  168. nullius/catalogue/benchmarks/tau3-bench/standard.json +547 -0
  169. nullius/catalogue/benchmarks/tau3-bench/upstream-settings.json +639 -0
  170. nullius/catalogue/benchmarks/terminal-bench/benchmark.json +69 -0
  171. nullius/catalogue/benchmarks/terminal-bench/driver.py +267 -0
  172. nullius/catalogue/benchmarks/terminal-bench/standard.json +166 -0
  173. nullius/catalogue/benchmarks/terminal-bench/upstream-settings.json +36 -0
  174. nullius/catalogue/capabilities/keyword-memory/capability.json +33 -0
  175. nullius/catalogue/capabilities/keyword-memory/capability.py +55 -0
  176. nullius/catalogue/capabilities/mem0/capability.json +135 -0
  177. nullius/catalogue/capabilities/mem0/capability.py +173 -0
  178. nullius/catalogue/catalog.py +426 -0
  179. nullius/catalogue/citation.py +116 -0
  180. nullius/catalogue/harbor_entry.py +367 -0
  181. nullius/catalogue/sources.py +107 -0
  182. nullius/catalogue/storage.py +125 -0
  183. nullius/cli.py +85 -0
  184. nullius/compare/__init__.py +0 -0
  185. nullius/compare/cli.py +265 -0
  186. nullius/compare/comparison.py +184 -0
  187. nullius/execution.py +1653 -0
  188. nullius/gitrefs.py +128 -0
  189. nullius/harbor/__init__.py +0 -0
  190. nullius/harbor/feed.py +228 -0
  191. nullius/harbor/source.py +582 -0
  192. nullius/hosts.py +368 -0
  193. nullius/importers/__init__.py +140 -0
  194. nullius/importers/atif.py +387 -0
  195. nullius/importers/cli.py +135 -0
  196. nullius/importers/otel.py +274 -0
  197. nullius/judge/__init__.py +14 -0
  198. nullius/judge/batch.py +122 -0
  199. nullius/judge/client.py +433 -0
  200. nullius/judge/floors.json +86 -0
  201. nullius/judge/lanes.py +89 -0
  202. nullius/judge/questions.py +309 -0
  203. nullius/judge/reconcile.py +176 -0
  204. nullius/judge/state.py +224 -0
  205. nullius/judge/store.py +167 -0
  206. nullius/migrate.py +268 -0
  207. nullius/onboarding.py +116 -0
  208. nullius/paths.py +89 -0
  209. nullius/prerequisites.py +129 -0
  210. nullius/prices.py +283 -0
  211. nullius/progress.py +294 -0
  212. nullius/providers.py +335 -0
  213. nullius/recipes/__init__.py +0 -0
  214. nullius/recipes/launch.py +78 -0
  215. nullius/redact.py +72 -0
  216. nullius/runs/__init__.py +0 -0
  217. nullius/runs/artifacts.py +67 -0
  218. nullius/runs/classify.py +216 -0
  219. nullius/runs/cli.py +316 -0
  220. nullius/runs/feasibility.py +442 -0
  221. nullius/runs/launches.py +149 -0
  222. nullius/runs/live.py +715 -0
  223. nullius/runs/parse.py +1016 -0
  224. nullius/runs/verifier.py +241 -0
  225. nullius/sandbox.py +109 -0
  226. nullius/secrets.py +218 -0
  227. nullius/settings.py +544 -0
  228. nullius/standard/__init__.py +15 -0
  229. nullius/standard/acp.py +305 -0
  230. nullius/standard/cancellation.py +19 -0
  231. nullius/standard/cli.py +703 -0
  232. nullius/standard/conform.py +149 -0
  233. nullius/standard/container.py +1870 -0
  234. nullius/standard/definitions.py +1651 -0
  235. nullius/standard/egress.py +192 -0
  236. nullius/standard/fidelity.py +256 -0
  237. nullius/standard/forward.py +139 -0
  238. nullius/standard/hostreq.py +105 -0
  239. nullius/standard/howto.py +220 -0
  240. nullius/standard/install.py +428 -0
  241. nullius/standard/job.py +551 -0
  242. nullius/standard/jsonl.py +129 -0
  243. nullius/standard/kits/agent-conversation-v1/driver.py +27 -0
  244. nullius/standard/kits/agent-conversation-v1/standard.json +26 -0
  245. nullius/standard/kits/agent-search-v1/driver.py +171 -0
  246. nullius/standard/kits/agent-search-v1/standard.json +99 -0
  247. nullius/standard/kits/agent-task-v1/driver.py +25 -0
  248. nullius/standard/kits/agent-task-v1/standard.json +100 -0
  249. nullius/standard/kits/agent-tools-v1/driver.py +206 -0
  250. nullius/standard/kits/agent-tools-v1/mcp_kit.py +209 -0
  251. nullius/standard/kits/agent-tools-v1/standard.json +119 -0
  252. nullius/standard/kits/agent-workspace-v1/driver.py +65 -0
  253. nullius/standard/kits/agent-workspace-v1/standard.json +95 -0
  254. nullius/standard/kits/cap-memory-v1/driver.py +39 -0
  255. nullius/standard/kits/cap-memory-v1/standard.json +91 -0
  256. nullius/standard/ledger.py +326 -0
  257. nullius/standard/match.py +66 -0
  258. nullius/standard/mcp_bridge.py +62 -0
  259. nullius/standard/mcp_machine.py +193 -0
  260. nullius/standard/merge.py +239 -0
  261. nullius/standard/names.py +37 -0
  262. nullius/standard/network_requirements.py +44 -0
  263. nullius/standard/pace.py +53 -0
  264. nullius/standard/payload.py +278 -0
  265. nullius/standard/pins.py +130 -0
  266. nullius/standard/ports.py +70 -0
  267. nullius/standard/prove.py +373 -0
  268. nullius/standard/recipe.py +746 -0
  269. nullius/standard/relay.py +785 -0
  270. nullius/standard/run.py +1781 -0
  271. nullius/standard/setup_container.py +208 -0
  272. nullius/standard/shard.py +71 -0
  273. nullius/standard/web_services.py +30 -0
  274. nullius/traces/__init__.py +0 -0
  275. nullius/traces/behavior.py +131 -0
  276. nullius/traces/charts.py +898 -0
  277. nullius/traces/conversation.py +159 -0
  278. nullius/traces/otel.py +608 -0
  279. nullius/traces/trace.py +1477 -0
  280. nullius/traces/trajectory.py +586 -0
  281. nullius/workspace.py +417 -0
  282. nullius/workspace_cli.py +119 -0
  283. nullius_eval-0.1.0.dist-info/METADATA +93 -0
  284. nullius_eval-0.1.0.dist-info/RECORD +291 -0
  285. nullius_eval-0.1.0.dist-info/WHEEL +5 -0
  286. nullius_eval-0.1.0.dist-info/entry_points.txt +2 -0
  287. nullius_eval-0.1.0.dist-info/licenses/LICENSE +202 -0
  288. nullius_eval-0.1.0.dist-info/licenses/app/static/fonts/dmsans-OFL.txt +93 -0
  289. nullius_eval-0.1.0.dist-info/licenses/app/static/fonts/ibmplexmono-OFL.txt +93 -0
  290. nullius_eval-0.1.0.dist-info/licenses/app/static/fonts/instrumentserif-OFL.txt +93 -0
  291. nullius_eval-0.1.0.dist-info/top_level.txt +2 -0
@@ -0,0 +1,460 @@
1
+ {% from 'runs/_execution.html' import execution_strip %}
2
+ {% extends "base.html" %}
3
+ {% block title %}{{ job.job_id }} — Nullius{% endblock %}
4
+ {% block crumb %}/ {{ job.job_id }}{% endblock %}
5
+ {% block head %}<link rel="stylesheet" href="{{ url_for('static', filename='runs/live-board.css') }}">
6
+ <link rel="stylesheet" href="{{ url_for('static', filename='runs/board.css') }}">{% endblock %}
7
+ {% block body %}
8
+
9
+ <header class="run-header">
10
+ <div class="page-title"><h1>{{ job.job_id }}</h1></div>
11
+ <div class="run-meta" aria-label="Run details">
12
+ {% if job.agent_label %}<span>{{ job.agent_label }}{% if job.agent_version %} <span class="mono faint" title="Resolved agent version, recorded in the trial's agent_info">v{{ job.agent_version }}</span>{% endif %}</span>{% endif %}
13
+ <span class="mono">{{ job.model_name }}</span>
14
+ <span title="{{ job.dataset }}">{{ job.dataset|dsname }}</span>
15
+ {% if control %}<span class="chip-control" title="Harbor control: never calls a model">control</span>
16
+ <span class="control-verdict {{ control.state }}" id="control-verdict" data-state="{{ control.state }}"
17
+ title="{{ control.control }} must be {{ control.expected }} on every task; this is a check on the tasks, not a score">{{ control.label }}</span>{% endif %}
18
+ {% if job.runner %}<span title="What ran: the runner, and the upstream commit its install resolved">{{ job.runner }}{% if job.runner_commit %} <span class="mono">{{ job.runner_commit }}</span>{% endif %}</span>
19
+ {% else %}<span title="What ran, from the job\'s lock.json">harbor {% if job.harbor_label %}<span class="mono">{{ job.harbor_label }}</span>{% else %}<span class="faint">unknown</span>{% endif %}</span>{% endif %}
20
+ <span>{{ job.trials|length }}{% if job.n_planned and job.n_planned != job.trials|length %} of {{ job.n_planned }} planned{% endif %} trials</span>
21
+ {% if job.n_retries %}<span class="warn" title="Trials Harbor restarted after an error">{{ job.n_retries }} retried</span>{% endif %}
22
+ {% if job.duration_min %}<span>{{ job.duration_min|runtime }}</span>{% endif %}
23
+ </div>
24
+ <div class="run-actions" role="group" aria-label="Run actions">
25
+ {% if execution and execution.plan.recipe %}<a class="btn sm ghost" href="{{ url_for('new_run', run=job.job_id) }}"
26
+ title="Open the run builder with the recipe this run was started from, every setting as it was; Run gives it a new name">New run like this</a>
27
+ {% else %}<a class="btn sm ghost" href="{{ url_for('new_run', like=job.job_id, workspace=request.args.get('workspace') or none) }}"
28
+ title="Open the run builder with what this run recorded: its agents and their parameters, model, dataset and task filter, limits, timeouts and Harbor; anything a recipe cannot say is named there">New run like this</a>{% endif %}
29
+ {% set controls = feasibility_controls(job) %}{% if controls %}<a class="btn sm ghost" href="{{ url_for('feasibility_page', job_id=job.job_id) }}" title="What the {{ controls|join(' and ') }} control{{ 's' if controls|length > 1 else '' }} said about each task: can an empty solution fail and a reference solution pass">Feasibility · {{ controls|join(' + ') }}</a>{% endif %}
30
+ {% if job.harbor_command %}<button class="btn sm ghost" type="button" data-copy="{{ job.harbor_command }}"
31
+ title="{{ job.harbor_command }}">Copy command</button>{% endif %}
32
+ </div>
33
+ </header>
34
+
35
+ {% if execution %}{{ execution_strip(execution, csrf, own=own if execution.plan.steps()|length > 1 else none) }}{% endif %}
36
+ {% if job.is_preparing and not live_board and not ended %}
37
+ {# Launched, not started: Harbor has written the configuration and nothing
38
+ else. There is no result record and no feed to fold yet, so this page
39
+ waits for the feed and reloads itself when it appears. #}
40
+ <div class="alert warn" id="preparing" data-live="{{ url_for('job_live', job_id=job.job_id) }}">
41
+ <strong>Preparing</strong> — this run was launched and {{ job.runner or 'Harbor' }} has written its configuration; no trial has started yet.
42
+ The live board appears here as soon as the first event lands. This page refreshes itself.
43
+ </div>
44
+ {% elif (job.is_running or job.is_preparing) and ended %}
45
+ {# Harbor's record says trials are running; the runner that started it says
46
+ the run ended. The runner is right: nothing will finish those trials. #}
47
+ <div class="alert" id="ended-by-runner">
48
+ <strong>This run was {{ ended[0] }}</strong> at {{ ended[1].strftime('%H:%M:%S') }}Z, before Harbor closed the job:
49
+ {{ job.n_running_trials }} trial{{ '' if job.n_running_trials == 1 else 's' }} cut short, {{ job.n_pending_trials }} never started.
50
+ The numbers on this page are what was recorded up to then.
51
+ </div>
52
+ {% elif job.is_running and not view %}
53
+ <div class="alert warn">
54
+ <strong>This run is still going</strong> — {{ job.n_running_trials }} trial{{ '' if job.n_running_trials == 1 else 's' }} running,
55
+ {{ job.n_pending_trials }} pending. Every number on this page is partial.
56
+ </div>
57
+ {% endif %}
58
+
59
+ {% if job.budget and job.budget.stopped and job.budget.max_usd %}
60
+ {# The run's own spending cap ended it (a Nullius standard run's --max-usd): nothing failed, so the notice is neutral. #}
61
+ <div class="alert" id="cap-stopped">
62
+ <strong>{% if job.budget.shards %}{{ job.budget.shards_stopped }} of {{ job.budget.shards }} shards stopped at their share of the spending cap{% else %}Stopped at its spending cap{% endif %}</strong> of {{ job.budget.max_usd|cap }}{% if job.budget.spent_usd is not none %}, with {{ job.budget.spent_usd|cap }} spent{% endif %}:
63
+ {{ job.budget.cut_off|length }} cut off, {{ job.budget.not_run|length }} not run.
64
+ </div>
65
+ {% endif %}
66
+
67
+ {% if live_board and live_board.error %}
68
+ <div class="alert warn"><strong>The live feed could not be read</strong> — <span class="mono">{{ live_board.error }}</span>. The sections below come from the trial results and are unaffected.</div>
69
+ {% endif %}
70
+ {% if view %}
71
+ {# Two views of one page. Live: the job's events.jsonl folded every second while it runs, and kept once it ended as the
72
+ record of how it ran. Results: the page rendered from the job's files when this page loaded. ?view= picks; without
73
+ it, Live while the feed says the job is going. At job.end Live stops asking and says Results is ready; no reload. #}
74
+ <div class="view-tabs" role="tablist" aria-label="Views of this run">
75
+ <button type="button" role="tab" id="tab-live" data-view-tab="live" aria-controls="view-live" aria-selected="{{ 'true' if view == 'live' else 'false' }}"{% if view != 'live' %} tabindex="-1"{% endif %}><span class="live-dot{{ ' on' if live_board.running }}" id="live-dot" aria-hidden="true"></span>Live</button>
76
+ <button type="button" role="tab" id="tab-results" data-view-tab="results" aria-controls="view-results" aria-selected="{{ 'true' if view == 'results' else 'false' }}"{% if view != 'results' %} tabindex="-1"{% endif %}>Results <span class="results-badge" id="results-badge" data-trials="{{ job.trials|length }}">{{ 'as of load' if live_board.running else '' }}</span></button>
77
+ </div>
78
+ <div role="tabpanel" id="view-live" aria-labelledby="tab-live"{% if view != 'live' %} hidden{% endif %}>
79
+ <section class="lb" id="live-board" aria-label="Live board" data-state="{{ live_board.initial.state }}"
80
+ data-endpoint="{{ url_for('job_live', job_id=job.job_id) }}"
81
+ data-trial-base="{{ '' if job.runner and not job.standard and not job.trials else url_for('job', job_id=job.job_id) }}"
82
+ data-noun="{{ job.runner_items[0] if job.runner_items else '' }}">
83
+ <div class="lb-headline">
84
+ <span class="lb-state"><i aria-hidden="true"></i><span data-lb-state>{{ live_board.initial.state|capitalize }}</span></span>
85
+ <span class="lb-clock" data-lb-head></span>
86
+ <span class="lb-meta" data-lb-meta></span>
87
+ </div>
88
+ <div class="lb-stats" data-lb-stats></div>
89
+ <div class="lb-mix">
90
+ <div class="lb-mixbar" data-lb-mixbar role="img" aria-label="Outcome mix of finished trials, as the feed reports them"></div>
91
+ <ul class="lb-legend" data-lb-legend aria-label="Filter trials by outcome"></ul>
92
+ <div class="lb-lanes" data-lb-lanes hidden></div>
93
+ </div>
94
+ <div class="lb-body">
95
+ <div class="lb-main">
96
+ <div class="lb-tiles-wrap"><span data-lb-tiles-label>All trials</span><div class="lb-tiles" data-lb-tiles></div></div>
97
+ <div data-lb-flight></div>
98
+ </div>
99
+ <aside class="lb-drawer" data-lb-drawer aria-label="Trial detail" hidden></aside>
100
+ </div>
101
+ <p class="lb-foot" data-lb-foot hidden><span data-lb-status></span></p>
102
+ </section>
103
+ <script id="live-board-initial" type="application/json">{{ live_board.initial | tojson }}</script>
104
+ </div>
105
+ <div role="tabpanel" id="view-results" aria-labelledby="tab-results"{% if view != 'results' %} hidden{% endif %}>
106
+ {% if job.is_running and not ended %}
107
+ <div class="alert warn">
108
+ <strong>Results as of this page's load</strong> — the run is still going: {{ job.n_running_trials }} trial{{ '' if job.n_running_trials == 1 else 's' }} running,
109
+ {{ job.n_pending_trials }} pending. Every number here is partial; Live follows it as it goes.
110
+ </div>
111
+ {% endif %}
112
+ {% endif %}
113
+ {% if job.is_preparing and not live_board and not ended %}
114
+ <script>
115
+ // Ask for the feed every second while preparing. Reload when it appears or
116
+ // the runner reports an end; the settled page no longer starts this poll.
117
+ (() => {
118
+ const url = document.getElementById('preparing').dataset.live;
119
+ const tick = async () => { if (document.hidden) return; try { const r = await nulliusFetch(url, {cache: 'no-store'}); if (r.ok) location.reload(); } catch {} };
120
+ setInterval(tick, 1000);
121
+ })();
122
+ </script>
123
+ {% endif %}
124
+
125
+ {#- The run page below its header: the scoreboard, the trials it scopes, and two disclosures.
126
+ Board counts are whole-run and never change under a filter; the trial rows are rendered here, and job.js only filters,
127
+ sorts and links them. Every lane and tally count is nullius.judge.lanes, which is reconcile's. -#}
128
+ {% macro slots(jr, extra) -%}
129
+ <span class="slots{{ extra }}" role="group" aria-label="{% if jr.flags %}Flags raised: {{ jr.flags|join(', ') }}{% else %}No flags raised{% endif %}">
130
+ {%- for f in judgement.flag_ids %}{% if f in jr.flags %}<span class="pip on" data-n="{{ loop.index }}" data-f="{{ f }}" data-tip="{{ loop.index }} · {{ f }} · raised" aria-label="{{ loop.index }} {{ f }}"></span>
131
+ {%- else %}<span class="pip" data-n="{{ loop.index }}" data-f="{{ f }}" data-tip="{{ loop.index }} · {{ f }} · not raised" aria-hidden="true"></span>{% endif %}{% endfor -%}
132
+ </span>
133
+ {%- endmacro %}
134
+ {% set vocab = job.vocab %}
135
+ {% set lanes = judgement.lanes if judgement else none %}
136
+ {% set show_judge = judgement and not control %}
137
+ {% set show_ins = job.trials and not control %}
138
+ {% set tq = task_quality_summary(job) if show_ins else none %}
139
+ {% set task_labels = task_labels_for(job) %}
140
+ {% set nlines = {} %}{% if lanes %}{% for r in lanes.rows %}{% set _ = nlines.update({r.outcome: r.lines|length}) %}{% endfor %}{% endif %}
141
+ {% set big = namespace(tok=0, tot=0) %}{% for t in job.trials %}{% set big.tok = [big.tok, (t.n_input_tokens or 0) + (t.n_output_tokens or 0)]|max %}{% set big.tot = [big.tot, t.dur_total or 0]|max %}{% endfor %}
142
+ {% set ANCHOR = {'disagree': 'reconcile-disagree', 'other': 'reconcile-other', 'agree': 'reconcile-agree', 'suggestion': 'reconcile-suggestions',
143
+ 'gated': 'reconcile-gated', 'not_comparable': 'reconcile-not-comparable', 'not_judged': 'reconcile-not-judged'} %}
144
+ {% set KEY = {'jog': '<path d="M0 1.5H5L10.5 7H16"/>', 'straight': '<path d="M0 5H16"/>', 'dotted': '<path d="M0 5H9" class="dt"/>',
145
+ 'open': '<path d="M0 5H7"/><circle cx="11" cy="5" r="3" class="op"/>', 'ring': '<path d="M0 5H7"/><circle cx="11" cy="5" r="3" class="rg"/>'} %}
146
+ {% set STUB = {'not_judged': 'ring', 'gated': 'dotted', 'suggestion': 'open', 'not_comparable': 'open'} %}
147
+ {% set GLYPH = {'agree': '<path d="M1 7H13"/><circle cx="17" cy="7" r="3.2"/>', 'disagree': '<path d="M1 2.5H6L10.5 7H13"/><circle cx="17" cy="7" r="3.2"/>',
148
+ 'other': '<path d="M1 2.5H6L10.5 7H13"/><circle cx="17" cy="7" r="3.2"/>', 'gated': '<path d="M1 7H13" class="dt"/><circle cx="17" cy="7" r="3.2"/>',
149
+ 'suggestion': '<path d="M1 7H13"/><circle cx="17" cy="7" r="3.2"/>', 'not_comparable': '<path d="M1 7H13"/><circle cx="17" cy="7" r="3.2"/>'} %}
150
+
151
+ <div class="rp" id="run-page">
152
+ {% if job.n_regressions %}
153
+ <div class="alert error">
154
+ <strong>{{ job.n_regressions }} trial{{ '' if job.n_regressions == 1 else 's' }} broke previously-passing tests.</strong>
155
+ A patch that fails {{ vocab.held_term or vocab.held or 'the regression guard' }} is worse than one that changes nothing — it introduced a regression.
156
+ </div>
157
+ {% endif %}
158
+
159
+ <div class="bw">
160
+ <section class="rp-card board{{ '' if show_judge else ' no-judge' }}" id="board" aria-label="Run summary">
161
+
162
+ {# the hero: the adjusted rate; a control's headline is its verdict, and its rates are shown as absence #}
163
+ {% if control %}
164
+ <div class="blk b-rate control-verdict {{ control.state }}" id="run-hero" title="{{ control.control }} must be {{ control.expected }} on every task; this is a check on the tasks, not a score">
165
+ <div class="lab">{{ control.control }} · {{ control.expected }}</div>
166
+ <div class="hero{{ ' long' if (control.n_behaved ~ '/' ~ control.n)|length > 4 }}">{{ control.n_behaved }}<small>/{{ control.n }}</small></div>
167
+ <div class="hk"><dl><dt>raw rate</dt><dd>—</dd><dt>adjusted rate</dt><dd>—</dd></dl></div>
168
+ </div>
169
+ {% elif job.scored_only %}
170
+ {#- a benchmark that gives a score and no pass or fail: the mean score is the headline, and the rates are shown as absence #}
171
+ <div class="blk b-rate" id="run-hero">
172
+ <div class="lab"><span class="def" data-tip="This benchmark gives each trial a score and no pass or fail, so there is no solve rate: the mean score{% if job.n_scored_parts %}, over every one of the {{ job.n_scored_parts }} scored{% endif %}, stands here.">Mean score</span></div>
173
+ <div class="hero{{ ' long' if ('%.2f'|format(job.mean_score))|length > 3 }}" id="adjusted-rate">{{ '%.2f'|format(job.mean_score) }}</div>
174
+ <div class="hk" aria-label="Health">
175
+ <dl>
176
+ <dt>raw rate</dt><dd id="raw-rate" class="em" data-tip="No pass or fail in this benchmark: no rate">—</dd>
177
+ <dt>infra</dt><dd{% if job.n_infra %} class="warn"{% endif %} data-tip="Trials that never got a fair attempt — rate limits, API errors, setup failures and spending caps">{{ job.n_infra }}</dd>
178
+ </dl>
179
+ </div>
180
+ </div>
181
+ {% else %}
182
+ <div class="blk b-rate" id="run-hero">
183
+ <div class="lab"><span class="def" data-tip="Solves divided by the trials that actually ran: infra outcomes, which failed before the agent could be judged, are left out. Harbor's raw mean counts them as misses.">Adjusted rate</span></div>
184
+ {% if job.has_verdict %}
185
+ {% set rate = job.adjusted_solve_rate|pct %}
186
+ {#- the column holds three characters at full size; "100%" is set smaller rather than cross into Outcomes #}
187
+ <div class="hero{{ ' long' if rate|length > 3 }}" id="adjusted-rate">{{ rate }}</div>
188
+ {% set ran = job.trials|length - job.n_infra %}
189
+ <div class="meter" role="img" aria-label="{{ ran }} trials ran, {{ job.n_resolved }} solved, {{ job.n_infra }} infra"
190
+ data-tip="Solved {{ job.n_resolved }} of the {{ ran }} trial{{ '' if ran == 1 else 's' }} that ran{% if job.n_infra %}; {{ job.n_infra }} infra, hollow and apart, left out of the adjusted rate{% endif %}.">
191
+ {% if job.trials|length <= 12 %}
192
+ <span class="mu" style="--n:{{ job.trials|length }}">
193
+ {%- for t in job.trials if t.outcome == 'resolved' %}<i class="s hl" data-t="{{ t.trial_name }}"></i>{% endfor -%}
194
+ {%- for t in job.trials if t.outcome != 'resolved' and not t.is_infra %}<i class="hl" data-t="{{ t.trial_name }}"></i>{% endfor -%}
195
+ {%- if job.n_infra %}<span class="gap"></span>{% for t in job.trials if t.is_infra %}<i class="x hl" data-t="{{ t.trial_name }}"></i>{% endfor %}{% endif -%}
196
+ </span>
197
+ {% else %}
198
+ <span class="bar1"><span style="flex:{{ ran }}"><i style="width:{{ (100 * job.n_resolved / ran) if ran else 0 }}%"></i></span>{% if job.n_infra %}<span class="x" style="flex:{{ job.n_infra }}"></span>{% endif %}</span>
199
+ {% endif %}
200
+ <b>{{ job.n_resolved }}/{{ ran }}</b>
201
+ </div>
202
+ {% else %}
203
+ <div class="hero none" id="adjusted-rate" data-tip="No trial in this run carries a verdict — nothing was scored">—</div>
204
+ {% endif %}
205
+ <div class="hk" aria-label="Health">
206
+ {% if job.has_verdict and job.raw_solve_rate == job.adjusted_solve_rate %}<p id="raw-rate" data-tip="No outcome in this run is infra, so the raw rate is also {{ job.raw_solve_rate|pct }}">raw = adjusted</p>{% endif %}
207
+ <dl>
208
+ {% if not job.has_verdict or job.raw_solve_rate != job.adjusted_solve_rate %}<dt>raw rate</dt><dd id="raw-rate" class="em" data-tip="Harbor's mean: resolved over every trial, infra included">{{ job.raw_solve_rate|pct }}</dd>{% endif %}
209
+ <dt>infra</dt><dd{% if job.n_infra %} class="warn"{% endif %} data-tip="Trials that never got a fair attempt — rate limits, API errors, setup failures and spending caps">{{ job.n_infra }}</dd>
210
+ <dt>broke tests</dt><dd{% if job.n_regressions %} class="bad"{% endif %} data-tip="Trials whose patch broke tests that passed before it">{{ job.n_regressions }}</dd>
211
+ </dl>
212
+ </div>
213
+ </div>
214
+ {% endif %}
215
+
216
+ {# the record: the outcome bar, and its legend rows - each filters the trials and starts its own lanes #}
217
+ <div class="blk b-out">
218
+ <div class="lab">Outcomes</div>
219
+ <div class="slot">
220
+ <div class="mix run-mix" role="group" aria-label="Outcome mix: each segment filters the trials">
221
+ {% for k in ORDER %}{% if job.counts[k] %}<a class="mseg fx hl oc-{{ k }}" style="flex:{{ job.counts[k] }}" data-o="{{ k }}" href="?outcome={{ k }}" tabindex="-1"
222
+ aria-label="{{ LABELS[k] }}: {{ job.counts[k] }}" data-tip="{{ LABELS[k] }}: {{ job.counts[k] }}"></a>{% endif %}{% endfor %}
223
+ </div>
224
+ </div>
225
+ <div class="run-outcome-filters" role="group" aria-label="Filter trials by outcome">
226
+ {% for k in ORDER %}{% if job.counts[k] %}{% set h = 45 + 30 * ((nlines.get(k) or 1) - 1) %}
227
+ <a class="row lg fx hl oc-{{ k }}" data-o="{{ k }}" href="?outcome={{ k }}"{% if h != 45 %} style="height:{{ h }}px"{% endif %}
228
+ data-tip="{{ DESCRIPTIONS[k] }}"><i class="sw"></i><b>{{ job.counts[k] }}</b><span>{{ LABELS[k] }}{% if k in INFRA %}<span class="infra">infra</span>{% endif %}</span></a>
229
+ {% endif %}{% endfor %}
230
+ </div>
231
+ <div class="row tot"><b>{{ job.trials|length }}</b><span>Trials{% if job.n_planned and job.n_planned != job.trials|length %} of {{ job.n_planned }} planned{% endif %}</span></div>
232
+ </div>
233
+
234
+ {% if show_judge %}
235
+ {# the judge: one advisory group. The lanes sit on the outcomes' row lines; the flags beside them #}
236
+ <div class="blk b-judge" id="judgement">
237
+ <div class="lab">{{ 'Trajectory judge' if job.runner else 'Judge' }} <span class="adv" data-tip="A judge model's reading of each transcript. Experimental: the recorded outcomes stand, and the rates count only them.">experimental</span>
238
+ <span class="aside">
239
+ <span data-tip="Trials the judge has read: each unresolved one for its cause and flags, a resolved one for flags">judged <b>{{ judgement.judged }}/{{ judgement.trials }}</b></span>
240
+ {% if judgement.truncation and judgement.truncation.truncated %}<span class="def" tabindex="0" data-tip="The trial state given to the judge was cut in {{ judgement.truncation.truncated }} of {{ judgement.truncation.records }} records: {% for kind, k in judgement.truncation.kinds.items() %}{{ kind }}: {{ k.truncated }} of {{ k.calls }} call{{ '' if k.calls == 1 else 's' }} cut to {{ judgement.truncation.limit }} characters (largest {{ k.max_chars }}){{ '; ' if not loop.last }}{% endfor %}{% if judgement.truncation.untyped %}; {{ judgement.truncation.untyped }} earlier record{{ '' if judgement.truncation.untyped == 1 else 's' }} cut, state kind not recorded{% endif %}">state truncated <b>{{ judgement.truncation.truncated }}/{{ judgement.truncation.records }}</b></span>{% endif %}
241
+ {% if judgement.judged %}<a class="xl" href="{{ url_for('reconcile_page', job_id=job.job_id) }}" data-tip="Trial by trial: the recorded outcome, the judge's cause, its confidence and status, with the confidence gate applied">Classifier vs judge →</a>{% endif %}
242
+ </span></div>
243
+ {% if job.runner %}<p class="small muted runner-judge" id="runner-judge">The outcomes are {{ job.runner }}'s own judge{% if job.runner_judge %}: {{ job.runner_judge }}{% endif %}.
244
+ This one is Nullius's, asked on a click: it reads each conversation for a cause and flags, and changes no outcome.</p>{% endif %}
245
+ <div class="jcols">
246
+ {% if lanes %}
247
+ <figure class="lanes" aria-label="Recorded outcome against the judge's reading">
248
+ <div class="lh" aria-hidden="true"><span></span>{% if lanes.has_agree %}<span class="ah">agrees</span>{% endif %}{% if lanes.has_disagree %}<span class="dh">disagrees</span>{% endif %}</div>
249
+ {% for r in lanes.rows %}
250
+ <div class="lr hl oc-{{ r.outcome }}" data-o="{{ r.outcome }}">
251
+ {% for g in r.lines %}
252
+ {% if g.lane == 'gutter' %}
253
+ {% set tip = {'not_asked': r.label ~ ': never asked a failure cause, so nothing to reconcile.',
254
+ 'not_judged': g.n ~ ' ' ~ r.label ~ ' not judged yet: no record.',
255
+ 'gated': g.n ~ ' ' ~ r.label ~ ' read as ' ~ (g.cause or '') ~ ', below the ' ~ lanes.threshold ~ ' confidence gate: noise, counted apart and never ranked.',
256
+ 'suggestion': g.n ~ ' ' ~ r.label ~ ' read as suggesting ' ~ (LABELS.get(g.suggests, g.suggests) or '') ~ ': no claim to test; experimental, never written back.',
257
+ 'not_comparable': g.n ~ ' ' ~ r.label ~ ': no claim to test, and the pick names no label.'}[g.status] %}
258
+ <div class="ll"><span class="na" data-tip="{{ tip }}">{% if STUB.get(g.status) %}<svg class="key" viewBox="0 0 16 10" aria-hidden="true">{{ KEY[STUB[g.status]]|safe }}</svg>{% endif %}{{ g.word }}{% if r.lines|length > 1 %}<b>{{ g.n }}</b>{% endif %}</span></div>
259
+ {% else %}
260
+ {% set side = 'a' if g.lane == 'agree' else 'd' %}
261
+ {% set tip = g.n ~ ' ' ~ r.label ~ ' read as ' ~ g.cause ~ ' · ' ~ g.status_label ~ '. Filters to ' ~ ('that trial.' if g.n == 1 else 'those trials.') %}
262
+ <div class="ll hl" data-o="{{ r.outcome }}" data-c="{{ g.cause }}" data-s="{{ g.status }}"><span class="cx {{ side }}{{ ' jog' if side == 'd' }}" aria-hidden="true"></span><a
263
+ class="cn {{ side }} fx hl" href="?outcome={{ r.outcome }}&amp;cause={{ g.cause }}&amp;status={{ g.status }}" data-o="{{ r.outcome }}" data-c="{{ g.cause }}" data-s="{{ g.status }}"
264
+ data-sl="{{ g.status_label }}" aria-label="{{ tip }}" data-tip="{{ tip }}"><i class="rg" aria-hidden="true"></i><span>{{ g.cause }}</span><b>{{ g.n }}</b></a></div>
265
+ {% endif %}
266
+ {% endfor %}
267
+ </div>
268
+ {% endfor %}
269
+ <div class="tally" aria-label="Reconcile counts">
270
+ {% for grp in lanes.tally %}<span class="tg"><svg class="key" viewBox="0 0 16 10" aria-hidden="true">{{ KEY[grp.shape]|safe }}</svg>
271
+ {%- for s, label, n in grp['items'] %}<a class="tli{{ ' lead' if s == 'disagree' and n }}" href="{{ url_for('reconcile_page', job_id=job.job_id) }}#{{ ANCHOR[s] }}"
272
+ data-tip="{{ label }}: {{ n }}. Opens Classifier vs judge at {{ label }}.">{{ label }}<b>{{ n }}</b>{% if s in ('disagree', 'agree') and lanes.of %}<small>/{{ lanes.of }}</small>{% endif %}</a>{% endfor %}</span>
273
+ {% endfor %}
274
+ </div>
275
+ </figure>
276
+ {% endif %}
277
+ {% if judgement.flags %}
278
+ <div class="b-flags" role="group" aria-label="Flags raised: each filters the trials">
279
+ {% set never = judgement.flag_ids|reject('in', judgement.flags|map('first')|list)|list %}
280
+ <div class="slot"><span>Flags raised</span><span class="end"{% if never %} data-tip="Never raised: {{ never|join(', ') }}"{% endif %}>of {{ judgement.judged }} judged</span></div>
281
+ {% for f, n in judgement.flags %}{% set num = judgement.flag_ids.index(f) + 1 %}
282
+ <a class="fr fx hl" data-f="{{ f }}" href="?flag={{ f }}" data-tip="{{ num }} · {{ f }}: raised on {{ n }} of {{ judgement.judged }} judged trials. Filters to them.">
283
+ <span class="pip on key-n" data-n="{{ num }}" aria-hidden="true"></span><span class="fn">{{ f }}</span><span class="fb"><i style="width:{{ (100 * n / judgement.judged)|round(1) }}%"></i></span><span class="fc">{{ n }}</span></a>
284
+ {% endfor %}
285
+ </div>
286
+ {% endif %}
287
+ </div>
288
+ {% if (judgement.unjudged) or (judgement.llm.unjudged) or (tq and tq.unjudged) %}
289
+ <div class="jb">
290
+ {# one button per judge: each judges apart, so a person asks for the one they want #}
291
+ {% for lane, name, n, keyed in [('system-one', 'System One', judgement.unjudged, judgement.keyed), ('llm', 'LLM judge', judgement.llm.unjudged, judgement.llm.keyed)] if n %}
292
+ {% if keyed %}
293
+ <form method="post" action="{{ url_for('judge_job_route', job_id=job.job_id) }}" class="header-form" id="judge-all-{{ lane }}">
294
+ <input type="hidden" name="csrf" value="{{ csrf }}"><input type="hidden" name="lane" value="{{ lane }}">
295
+ <button class="btn sm ghost" type="submit" title="{{ name }}: judge every trial it has not judged">{{ name }} · judge {{ n }}</button>
296
+ </form>
297
+ {% else %}<a class="btn sm ghost" href="{{ url_for('settings_page') }}#judge" title="{{ name }} is not configured: set it in Settings, Judge">{{ name }} · set up</a>{% endif %}
298
+ {% endfor %}
299
+ {% if tq and tq.unjudged and tq.keyed %}
300
+ <form method="post" action="{{ url_for('judge_tasks_route', job_id=job.job_id) }}" class="header-form" id="judge-tasks">
301
+ <input type="hidden" name="csrf" value="{{ csrf }}">
302
+ <button class="btn sm ghost" type="submit" title="Score every unscored task instruction: one call per task over its text alone">Score {{ tq.unjudged }} instruction{{ '' if tq.unjudged == 1 else 's' }}</button>
303
+ </form>
304
+ {% endif %}
305
+ </div>
306
+ {% endif %}
307
+ </div>
308
+ {% endif %}
309
+
310
+ <div class="blk b-tot">
311
+ <div class="lab">Run totals</div>
312
+ <div class="kv" data-tip="Recorded input + output tokens; cached input is already part of input"><span class="k">Tokens</span><span class="v">{{ (job.total_input_tokens + job.total_output_tokens)|tokens(job.kind) }}</span></div>
313
+ <div class="sp"><details class="pop"><summary aria-label="Token split: input, cached, output"><span class="tw"></span>in · cached · out</summary>
314
+ <div class="popm" role="group" aria-label="Token split"><span>input</span><b>{{ '{:,}'.format(job.total_input_tokens) }}</b><span>cached (part of input)</span><b>{{ '{:,}'.format(job.total_cache_tokens) }}</b><span>output</span><b>{{ '{:,}'.format(job.total_output_tokens) }}</b></div></details></div>
315
+ <div class="kv" data-tip="{{ 'A control never calls a model, so this is a real zero' if job.is_control else 'Only recorded by agents that report it' }}"><span class="k">Cost</span><span class="v">{{ job.total_cost_usd|usd(job.kind) }}</span></div>
316
+ {% if job.budget and job.budget.max_usd %}<div class="kv" id="run-cap" data-tip="Its relay refuses a call that does not fit under it{% if job.budget.unsettled_usd %}. {{ job.budget.unsettled_usd|usd }} stays booked for calls whose cost was never reported{% endif %}"><span class="k">Spending cap</span><span class="v">{{ job.budget.max_usd|cap }}</span></div>
317
+ {% if job.budget.not_run %}<div class="sp"><details class="pop"><summary aria-label="The units not run"><span class="tw"></span>{{ job.budget.not_run|length }} not run</summary>
318
+ <div class="popm cap-units" role="group" aria-label="The units not run">{% for u in job.budget.not_run %}<span>{{ u }}</span>{% endfor %}</div></details></div>{% endif %}{% endif %}
319
+ {% if job.budget and job.budget.unit_max_usd %}<div class="kv" id="unit-cap" data-tip="For each {{ job.runner_items[0] if job.runner_items else 'trial' }}, its judge's calls apart: one that reaches it is cut off there and counts against the agent, and the run goes on"><span class="k">Cap per {{ job.runner_items[0] if job.runner_items else 'trial' }}</span><span class="v">{{ job.budget.unit_max_usd|cap }}</span></div>
320
+ {% if job.budget.over_unit %}<div class="sp"><details class="pop"><summary aria-label="The units that reached it"><span class="tw"></span>{{ job.budget.over_unit|length }} reached it</summary>
321
+ <div class="popm cap-units" role="group" aria-label="The units that reached it">{% for u in job.budget.over_unit %}<span>{{ u }}</span>{% endfor %}</div></details></div>{% endif %}{% endif %}
322
+ {% if job.mean_score is not none %}<div class="kv" data-tip="{% if job.n_scored_parts %}The mean score over every one of the {{ job.n_scored_parts }} scored, whichever trial it is under{% else %}The mean of each item's graded score (an F1, partial credit){% endif %}{% if not job.scored_only %}; the rates count an item resolved at its benchmark's pass mark{% endif %}"><span class="k">Mean score</span><span class="v">{{ '%.2f'|format(job.mean_score) }}</span></div>{% endif %}
323
+ {% for g in job.score_by_group %}<div class="kv by-group" data-tip="The mean score of the {{ g.n }} scored with {{ job.parts_by }} {{ g.group }}"><span class="k">{{ job.parts_by|capitalize }} {{ g.group }}</span><span class="v">{{ '%.2f'|format(g.mean) }} <small class="faint">{{ g.n }}</small></span></div>{% endfor %}
324
+ {% if job.duration_min %}<div class="kv" data-tip="Start to finish, the whole run"><span class="k">Duration</span><span class="v">{{ job.duration_min|runtime }}</span></div>{% endif %}
325
+ </div>
326
+ </section>
327
+ </div>
328
+
329
+ {# the trials: one filter row names every active filter, whichever mark set it #}
330
+ <section class="rp-card trials" aria-label="Trials in this run">
331
+ <div class="tbar" role="group" aria-label="Filter the trials">
332
+ <h2 id="trials-heading" tabindex="-1">Trials</h2><span class="cnt" id="trial-count">{{ job.trials|length }}</span>
333
+ <div class="facet" data-k="o"><details><summary class="fsum"><span class="k">Outcome</span><span class="v">All</span><span class="car">▾</span></summary>
334
+ <div class="fmenu">
335
+ {% for k in ORDER %}{% if job.counts[k] %}<a class="fx" data-o="{{ k }}" href="?outcome={{ k }}"><i class="sw oc-{{ k }}"></i><span>{{ LABELS[k] }}</span><b>{{ job.counts[k] }}</b></a>{% endif %}{% endfor %}
336
+ </div></details><button class="x plain" type="button" hidden aria-label="Remove the outcome filter">✕</button></div>
337
+ {% if judgement and judgement.causes %}
338
+ <div class="facet j" data-k="c"><details><summary class="fsum"><span class="k">Judge cause</span><span class="v">All</span><span class="car">▾</span></summary>
339
+ <div class="fmenu" id="judge-filters">
340
+ {% for cause, n in judgement.causes %}<a class="fx" data-c="{{ cause }}" href="?cause={{ cause }}"><span class="mono">{{ cause }}</span><b>{{ n }}</b></a>{% endfor %}
341
+ </div></details><button class="x plain" type="button" hidden aria-label="Remove the judge cause filter">✕</button></div>
342
+ {% endif %}
343
+ {% if judgement and judgement.flags %}
344
+ <div class="facet j" data-k="f"><details><summary class="fsum"><span class="k">Flag</span><span class="v">All</span><span class="car">▾</span></summary>
345
+ <div class="fmenu">
346
+ {% for f, n in judgement.flags %}<a class="fx" data-f="{{ f }}" href="?flag={{ f }}"><span class="pip on key-n" data-n="{{ judgement.flag_ids.index(f) + 1 }}" aria-hidden="true"></span><span class="mono">{{ f }}</span><b>{{ n }}</b></a>{% endfor %}
347
+ </div></details><button class="x plain" type="button" hidden aria-label="Remove the flag filter">✕</button></div>
348
+ {% endif %}
349
+ <button class="fclear plain" id="clear-filters" type="button" hidden>Show all {{ job.trials|length }}</button>
350
+ </div>
351
+ <p class="sr" id="trials-live" aria-live="polite"></p>
352
+ {% set show_cause = show_judge %}
353
+ {% set show_flags = show_judge and judgement.judged %}
354
+ {% set n_judge = (1 if show_cause else 0) + (1 if show_flags else 0) + (1 if show_ins else 0) %}
355
+ <div class="tscroll" role="region" aria-label="Trials in this run" tabindex="0">
356
+ <table class="tt">
357
+ <colgroup>
358
+ <col><col class="w-out">{% if show_cause %}<col class="w-cause">{% endif %}{% if show_flags %}<col class="w-flags">{% endif %}{% if show_ins %}<col class="w-ins">{% endif %}
359
+ {% if vocab.gained %}<col class="w-chk">{% endif %}{% if vocab.held %}<col class="w-chk">{% endif %}
360
+ <col class="w-tok"><col class="w-tm"><col class="w-env"><col class="w-ag"><col class="w-ver"><col class="w-go">
361
+ </colgroup>
362
+ <thead>
363
+ <tr class="grp">
364
+ <th colspan="2"></th>
365
+ {% if n_judge %}<th colspan="{{ n_judge }}"><span class="gl"><span data-tip="A judge model's reading, experimental as on the board: the recorded outcomes stand.">Judge</span>{% if tq %}<span class="gx" id="task-quality" data-tip="Task instructions the judge has scored, once per dataset and task">instructions scored <b>{{ tq.judged }}/{{ tq.tasks }}</b></span>{% endif %}</span></th>{% endif %}
366
+ {% if vocab.gained or vocab.held %}<th colspan="{{ (1 if vocab.gained else 0) + (1 if vocab.held else 0) }}"><span class="gl">Tests</span></th>{% endif %}
367
+ <th></th>
368
+ <th colspan="4"><span class="gl">Time</span></th>
369
+ <th></th>
370
+ </tr>
371
+ <tr class="cols">
372
+ <th scope="col">Task</th>
373
+ <th scope="col">Outcome</th>
374
+ {% if show_cause %}<th scope="col"><span data-tip="The judge's failure cause, drawn as on the board: straight when the recorded outcome allows it, jogged when it does not, ending in the judge's hollow ring. A resolved trial is not asked.">Cause<span class="nfh">{% if show_flags %} · flags{% endif %}</span></span></th>{% endif %}
375
+ {% if show_flags %}<th scope="col" class="c-flags"><span data-tip="One slot per flag, numbered in the vocabulary's order: {% for f in judgement.flag_ids %}{{ loop.index }} {{ f }}{{ ' · ' if not loop.last }}{% endfor %}. Filled: raised on this trial.">Flags</span></th>{% endif %}
376
+ {% if show_ins %}<th scope="col" class="num"><span data-tip="A judge model's score of the task's instruction text alone: about the task, not this trial">Instruction</span></th>{% endif %}
377
+ {% if vocab.gained %}<th scope="col" class="num" title="{% if vocab.gained_term %}{{ vocab.gained_term }} {% endif %}pass / fail">{{ vocab.gained_abbr or vocab.gained }}</th>{% endif %}
378
+ {% if vocab.held %}<th scope="col" class="num" title="{% if vocab.held_term %}{{ vocab.held_term }} {% endif %}pass / fail">{{ vocab.held_abbr or vocab.held }}</th>{% endif %}
379
+ <th scope="col" class="num"><a class="sort" href="?sort=tokens" data-sort="tokens" data-tip="Recorded input + output tokens; cached input is already included. Sorts.">Tokens</a></th>
380
+ <th scope="col" class="num"><a class="sort" href="?sort=total" data-sort="total" data-tip="Recorded start-to-finish trial duration, including setup and overhead. Sorts.">Total</a></th>
381
+ <th scope="col" class="num"><span data-tip="The environment's setup, before the agent starts">Env</span></th>
382
+ <th scope="col" class="num"><span data-tip="The agent's own run">Agent</span></th>
383
+ <th scope="col" class="num"><span data-tip="The verifier; — when it did not run">Verify</span></th>
384
+ <th scope="col"><span class="sr">Open</span></th>
385
+ </tr>
386
+ </thead>
387
+ <tbody class="rows" id="trial-rows">
388
+ {% for t in job.trials %}
389
+ {%- set jr = judgement.by_trial.get(t.trial_name) if judgement else none %}
390
+ {%- set status = lanes.status.get(t.trial_name) if lanes else '' %}
391
+ {%- set used = (t.n_input_tokens or 0) + (t.n_output_tokens or 0) %}
392
+ {%- set url = url_for('trial', job_id=job.job_id, trial_name=t.trial_name) %}
393
+ <tr data-outcome="{{ t.outcome }}" data-cause="{{ (jr.cause or '') if jr else '' }}" data-status="{{ status or '' }}" data-flags="{{ jr.flags|join(' ') if jr else '' }}"
394
+ data-trial="{{ t.trial_name }}" data-tokens="{{ used }}" data-total="{{ t.dur_total or 0 }}" data-i="{{ loop.index0 }}">
395
+ <td><a class="task" href="{{ url }}" title="{{ t.task_name }}">{{ task_labels[t.trial_name] }}</a></td>
396
+ <td><span class="oc oc-{{ t.outcome }}"><i></i>{{ t.outcome_label }}</span>
397
+ {%- if t.score is not none %} <span class="badge trial-score" title="{% if t.n_parts %}The mean score of its {{ t.n_parts }} parts{% else %}Its graded score{% endif %}, as the benchmark's own scorer gave it">{{ '%.2f'|format(t.score) }}</span>{% endif %}
398
+ {%- if t.n_rate_limited_retries %} <span class="badge rate-limits" title="{{ t.n_rate_limited_retries }} rate-limited call{{ '' if t.n_rate_limited_retries == 1 else 's' }} retried by the agent's SDK (agent/gateway.log) — the run survived them, so the outcome does not show them">429 ×{{ t.n_rate_limited_retries }}</span>{% endif -%}
399
+ {%- if t.exception_type %}
400
+ {%- if t.exception_message %}<details class="exd"><summary><span class="tw"></span>{{ t.exception_type }}</summary><code>{{ t.exception_message }}</code></details>
401
+ {%- else %}<span class="exn">{{ t.exception_type }}</span>{% endif %}{% endif %}</td>
402
+ {% if show_cause %}<td>
403
+ {%- if t.outcome == 'resolved' %}<span class="rc" data-tip="A resolved trial is never asked a failure cause"><span class="na">not asked</span></span>
404
+ {%- elif jr and jr.cause %}<span class="rc" data-tip="Recorded {{ t.outcome_label }}; the judge reads {{ jr.cause }}{% if status in GLYPH %} · {{ {'agree': 'Agree', 'disagree': 'Disagree', 'other': 'Cause outside the list', 'gated': 'Below threshold', 'suggestion': 'Suggested', 'not_comparable': 'Not comparable'}[status] }}{% endif %}"><svg class="kg" viewBox="0 0 22 14" aria-hidden="true">{{ (GLYPH.get(status) or GLYPH['agree'])|safe }}</svg><span class="ctag">{{ jr.cause }}</span></span>
405
+ {%- else %}<span class="rc"><span class="na">not judged</span></span>{% endif %}
406
+ {%- if show_flags and jr %}{{ slots(jr, ' nf') }}{% endif %}</td>{% endif %}
407
+ {% if show_flags %}<td class="c-flags">{% if jr %}{{ slots(jr, '') }}{% endif %}</td>{% endif %}
408
+ {% if show_ins %}<td class="num">{%- set q = task_quality_of(job, t) %}{% if q %}<span class="ins task-quality" data-task-quality="{{ q.level }}" data-tip="Instruction quality {{ '%.1f'|format(q.score) }} of {{ q.top }}: {{ q.label }}{% if q.confidence is not none %} · confidence {{ '%.0f'|format(q.confidence * 100) }}%{% endif %} · {{ q.model }} · experimental, about the task text, not this trial">{{ '%.1f'|format(q.score) }}<small>/{{ q.top }}</small></span>{% else %}<span class="faint">—</span>{% endif %}</td>{% endif %}
409
+ {% if vocab.gained %}<td class="num chk">{% if t.f2p_pass or t.f2p_fail %}<span class="ok">{{ t.f2p_pass }}</span><span class="faint">/</span><span class="bad">{{ t.f2p_fail }}</span>{% else %}<span class="faint">—</span>{% endif %}</td>{% endif %}
410
+ {% if vocab.held %}<td class="num chk">
411
+ {%- if t.p2p_pass or t.p2p_fail -%}
412
+ <span class="ok">{{ t.p2p_pass }}</span><span class="faint">/</span><span class="bad">{{ t.p2p_fail }}</span>
413
+ {%- if t.is_regression %} <span class="badge danger" title="Broke {{ t.p2p_fail }} previously-passing test(s)">⚠</span>{% endif -%}
414
+ {%- else %}<span class="faint">—</span>{% endif -%}
415
+ </td>{% endif %}
416
+ <td class="num run-token-count"><span class="mg" style="--v:{{ '%.3f'|format(used / big.tok if big.tok else 0) }}" data-tip="{% if used or job.is_control %}Input: {{ '{:,}'.format(t.n_input_tokens or 0) }} · Output: {{ '{:,}'.format(t.n_output_tokens or 0) }} · Cached input: {{ '{:,}'.format(t.n_cache_tokens or 0) }} (included in input){% else %}Token usage not recorded{% endif %}">{{ used|tokens(job.kind) }}</span></td>
417
+ <td class="num tot run-total-time"><span class="mg" style="--v:{{ '%.3f'|format((t.dur_total or 0) / big.tot if big.tot else 0) }}">{{ t.dur_total|secs }}</span></td>
418
+ <td class="num phase">{{ t.dur_env|secs }}</td>
419
+ <td class="num phase">{{ t.dur_agent|secs }}</td>
420
+ <td class="num phase">{{ t.dur_verifier|secs }}</td>
421
+ <td class="go"><a href="{{ url }}" aria-label="Open {{ t.task_base }}" tabindex="-1">→</a></td>
422
+ </tr>
423
+ {% endfor %}
424
+ <tr class="empty" id="no-trial-matches" hidden><td colspan="20">No trial matches <button class="fclear plain" type="button" data-clear>Show all {{ job.trials|length }}</button></td></tr>
425
+ </tbody>
426
+ </table>
427
+ </div>
428
+ </section>
429
+
430
+ {# behind a click: what the run recorded about its install, and the page's one block of prose #}
431
+ <div class="more">
432
+ {% if job.harbor_command %}
433
+ <details id="configuration">
434
+ <summary title="The command this run ran"><span class="tw"></span>Configuration<span class="end"></span></summary>
435
+ {% if job.harbor_command %}<div class="rcmd"><code>{{ job.harbor_command }}</code><button class="btn sm ghost" type="button" data-copy="{{ job.harbor_command }}">Copy</button></div>{% endif %}
436
+ </details>
437
+ {% endif %}
438
+ <details id="definitions">
439
+ <summary><span class="tw"></span>Definitions<span class="end"><span>outcomes · rates · judge</span></span></summary>
440
+ <dl class="gl">
441
+ {% for k in ORDER %}{% if job.counts[k] %}<dt class="oc-{{ k }}"><i class="sw"></i>{{ LABELS[k] }}{% if k in INFRA %} <span class="infra">infra</span>{% endif %}</dt><dd>{{ DESCRIPTIONS[k] }}</dd>{% endif %}{% endfor %}
442
+ <dt class="gs">Adjusted rate</dt><dd>Solves divided by the trials that actually ran: it drops the outcomes marked <span class="infra">infra</span>, which failed before the agent could be judged. The raw rate is Harbor's mean over every trial, shown beside it only when the two differ. A run in which no trial carries a verdict shows —, never 0%.</dd>
443
+ {% if not control %}<dt>Solved meter</dt><dd>One unit per trial that ran, solves filled; an infra trial is a hollow unit after a gap. Past 12 trials the units become one bar. A hovered trial row lights its unit.</dd>{% endif %}
444
+ <dt>Tokens · Total</dt><dd>Under each trial's value a thin bar from zero, on one scale for the column: the largest trial fills its track. Total is the trial start to finish and includes overhead that Env, Agent and Verify do not.</dd>
445
+ {% if show_judge %}
446
+ <dt class="gs">Marks</dt><dd><span class="mk"><i class="sw oc-agent_error"></i> solid = recorded</span><span class="mk"><svg class="key" viewBox="0 0 16 10" aria-hidden="true"><path d="M0 6H9.5"/><circle cx="12.5" cy="6" r="2.5" class="op"/></svg> hollow ring = judge reading</span><span class="mk"><svg class="key" viewBox="0 0 16 10" aria-hidden="true"><path d="M0 5H9" class="dt"/></svg> dotted = below the confidence gate</span></dd>
447
+ <dt><span class="adv">experimental</span></dt><dd>A judge model's reading of each transcript (cause, flags) or of each task's instruction text (Instruction). Shown beside the recorded outcomes, never replacing them; the rates count only the record.</dd>
448
+ <dt>Lanes</dt><dd>Each outcome row starts its own readings. A line runs <b>straight</b> into <i>agrees</i> when the recorded outcome allows the judge's cause, and <b>jogs</b> into <i>disagrees</i> when it does not. What is not a finding about the claim stays in the gutter: <i>not asked</i> (a resolved trial is never asked its cause), <i>not judged</i>, <i>below gate</i> (under {{ lanes.threshold if lanes else '0.5' }} confidence), and an Unknown outcome's reading. A cause filters to exactly its trials.</dd>
449
+ <dt>Tally</dt><dd>Under the lanes: the Classifier vs judge page's own counts and words. Disagree and Agree always show, over the comparable judged trials; the others only when non-zero. Each opens that page at its status.</dd>
450
+ <dt>Flags</dt><dd>One judgment per behaviour, asked of every judged trial. Each flag keeps a fixed number, its place in the question vocabulary. The board lists the flags raised, most raised first; each trial row keeps one slot per flag in number order, filled where the flag was raised.</dd>
451
+ {% endif %}
452
+ {% if show_ins %}<dt>Instruction</dt><dd>A judge model's score of each task's instruction text alone, once per dataset and task. An unclear instruction caps every agent on that task. Experimental.</dd>{% endif %}
453
+ </dl>
454
+ </details>
455
+ </div>
456
+ </div>
457
+ {% if view %}</div>{% endif %}
458
+ {% endblock %}
459
+ {% block scripts %}<script src="{{ url_for('static', filename='runs/live-board.js') }}"></script>
460
+ <script src="{{ url_for('static', filename='runs/board.js') }}"></script>{% endblock %}
@@ -0,0 +1,14 @@
1
+ {% from 'runs/_execution.html' import execution_strip %}
2
+ {% extends "base.html" %}
3
+ {% block title %}{{ job_id }} — Nullius{% endblock %}
4
+ {% block crumb %}/ {{ job_id }}{% endblock %}
5
+ {% block head %}<link rel="stylesheet" href="{{ url_for('static', filename='runs/live-board.css') }}">{% endblock %}
6
+ {% block body %}
7
+ <header class="run-header">
8
+ <div class="page-title"><h1>{{ job_id }}</h1></div>
9
+ {% if ex.plan.runner == 'harbor' %}<p class="subtitle">Started from the builder. This page becomes the live board as soon as Harbor writes the job.</p>
10
+ {% elif ex.plan.runner == 'standard' %}<p class="subtitle">Started from the builder with Nullius. This page becomes the live board as soon as the install is done and the run starts.</p>
11
+ {% else %}<p class="subtitle">Started from the builder with the {{ ex.plan.runner }} runner. This page becomes the live board as soon as this arm's questions are chosen, after the install.</p>{% endif %}
12
+ </header>
13
+ {{ execution_strip(ex, csrf, waiting=true, own=own) }}
14
+ {% endblock %}