88e86b3faccd
Rebuild benchmarks as a clean single-root reposi
2 months ago
| 1 | #!/usr/bin/env python3
|
| 2 | """Emit and validate Oak agent workflow benchmark result row schema.
|
| 3 |
|
| 4 | The schema is JSON-schema-like but intentionally dependency-free. Running this
|
| 5 | script with no arguments prints the schema as JSON. Use --validate-jsonl to check
|
| 6 | JSON or JSONL result rows against the subset of schema rules implemented here.
|
| 7 | """
|
| 8 |
|
| 9 | from __future__ import annotations
|
| 10 |
|
| 11 | import argparse
|
| 12 | import json
|
| 13 | from pathlib import Path
|
| 14 | from typing import Any, Iterable, List
|
| 15 |
|
| 16 | SCHEMA_VERSION = 2
|
| 17 |
|
| 18 | VCS_MODES = [
|
| 19 | "git_single_checkout",
|
| 20 | "git_worktree_per_task",
|
| 21 | "oak_single_mount",
|
| 22 | "oak_mount_per_task",
|
| 23 | "oak_space_per_task",
|
| 24 | ]
|
| 25 |
|
| 26 | SUBJECT_KINDS = ["git", "oak", "other"]
|
| 27 | OUTCOMES = ["pass", "fail", "timeout", "error", "abandoned"]
|
| 28 | TOKEN_SOURCES = ["provider_reported", "adapter_estimated", "char_only", "mixed", "unknown"]
|
| 29 | TURN_SOURCES = ["adapter_stream", "estimated", "unmeasured"]
|
| 30 | INSTRUCTION_LEVELS = ["zero-shot", "cheat-sheet", "full-docs", "adapter-default"]
|
| 31 |
|
| 32 | # Schema v2 measurement contract: a field that a runner cannot actually measure
|
| 33 | # MUST be null, never a fabricated zero. Dashboards must render null as
|
| 34 | # "unmeasured", not as a good result. Each metric group carries an optional
|
| 35 | # measurement_source describing how its values were obtained.
|
| 36 |
|
| 37 |
|
| 38 | def string(description: str = "") -> dict[str, Any]:
|
| 39 | schema: dict[str, Any] = {"type": "string"}
|
| 40 | if description:
|
| 41 | schema["description"] = description
|
| 42 | return schema
|
| 43 |
|
| 44 |
|
| 45 | def optional_string(description: str = "") -> dict[str, Any]:
|
| 46 | schema: dict[str, Any] = {"type": ["string", "null"]}
|
| 47 | if description:
|
| 48 | schema["description"] = description
|
| 49 | return schema
|
| 50 |
|
| 51 |
|
| 52 | def non_negative_integer(description: str = "") -> dict[str, Any]:
|
| 53 | schema: dict[str, Any] = {"type": "integer", "minimum": 0}
|
| 54 | if description:
|
| 55 | schema["description"] = description
|
| 56 | return schema
|
| 57 |
|
| 58 |
|
| 59 | def optional_non_negative_integer(description: str = "") -> dict[str, Any]:
|
| 60 | schema: dict[str, Any] = {"type": ["integer", "null"], "minimum": 0}
|
| 61 | if description:
|
| 62 | schema["description"] = description
|
| 63 | return schema
|
| 64 |
|
| 65 |
|
| 66 | def non_negative_number(description: str = "") -> dict[str, Any]:
|
| 67 | schema: dict[str, Any] = {"type": "number", "minimum": 0}
|
| 68 | if description:
|
| 69 | schema["description"] = description
|
| 70 | return schema
|
| 71 |
|
| 72 |
|
| 73 | def optional_non_negative_number(description: str = "") -> dict[str, Any]:
|
| 74 | schema: dict[str, Any] = {"type": ["number", "null"], "minimum": 0}
|
| 75 | if description:
|
| 76 | schema["description"] = description
|
| 77 | return schema
|
| 78 |
|
| 79 |
|
| 80 | def boolean(description: str = "") -> dict[str, Any]:
|
| 81 | schema: dict[str, Any] = {"type": "boolean"}
|
| 82 | if description:
|
| 83 | schema["description"] = description
|
| 84 | return schema
|
| 85 |
|
| 86 |
|
| 87 | def optional_boolean(description: str = "") -> dict[str, Any]:
|
| 88 | schema: dict[str, Any] = {"type": ["boolean", "null"]}
|
| 89 | if description:
|
| 90 | schema["description"] = description
|
| 91 | return schema
|
| 92 |
|
| 93 |
|
| 94 | def string_array(description: str = "") -> dict[str, Any]:
|
| 95 | schema: dict[str, Any] = {"type": "array", "items": {"type": "string"}}
|
| 96 | if description:
|
| 97 | schema["description"] = description
|
| 98 | return schema
|
| 99 |
|
| 100 |
|
| 101 | TOKEN_METRICS_SCHEMA: dict[str, Any] = {
|
| 102 | "type": "object",
|
| 103 | "additionalProperties": False,
|
| 104 | "required": [
|
| 105 | "token_source",
|
| 106 | "input_tokens_reported",
|
| 107 | "output_tokens_reported",
|
| 108 | "reasoning_tokens_reported",
|
| 109 | "cache_read_tokens_reported",
|
| 110 | "cache_write_tokens_reported",
|
| 111 | "total_tokens_reported",
|
| 112 | "normalized_input_chars",
|
| 113 | "normalized_output_chars",
|
| 114 | "transcript_chars",
|
| 115 | "context_truncation_events",
|
| 116 | "token_budget",
|
| 117 | "cost_weighted_total",
|
| 118 | ],
|
| 119 | "properties": {
|
| 120 | "token_source": {"type": "string", "enum": TOKEN_SOURCES},
|
| 121 | "tokenizer": optional_string("Tokenizer or accounting method, when known."),
|
| 122 | "input_tokens_reported": optional_non_negative_integer("Provider-reported input tokens."),
|
| 123 | "output_tokens_reported": optional_non_negative_integer("Provider-reported output tokens."),
|
| 124 | "reasoning_tokens_reported": optional_non_negative_integer("Provider-reported reasoning tokens."),
|
| 125 | "cache_read_tokens_reported": optional_non_negative_integer("Provider-reported cache-read tokens."),
|
| 126 | "cache_write_tokens_reported": optional_non_negative_integer("Provider-reported cache-write tokens."),
|
| 127 | "total_tokens_reported": optional_non_negative_integer("Provider-reported total tokens."),
|
| 128 | "normalized_input_chars": non_negative_integer("Prompt and admitted tool-result characters."),
|
| 129 | "normalized_output_chars": non_negative_integer("Assistant output characters."),
|
| 130 | "transcript_chars": non_negative_integer("Full normalized transcript characters."),
|
| 131 | "context_truncation_events": optional_non_negative_integer(
|
| 132 | "Times context was compacted or truncated. Null when the adapter does not expose compaction events."
|
| 133 | ),
|
| 134 | "token_budget": optional_non_negative_integer("Configured token budget, if any."),
|
| 135 | "cost_weighted_total": optional_non_negative_number(
|
| 136 | "Billing-direction weighted token cost. Agent-emitted text (commands, assistant output) is "
|
| 137 | "model OUTPUT and weighted higher than agent-ingested tool results (model INPUT)."
|
| 138 | ),
|
| 139 | "cost_weights": optional_string("Weights used for cost_weighted_total, e.g. 'emitted=5,ingested=1'."),
|
| 140 | },
|
| 141 | }
|
| 142 |
|
| 143 | TOOL_CALL_METRICS_SCHEMA: dict[str, Any] = {
|
| 144 | "type": "object",
|
| 145 | "additionalProperties": False,
|
| 146 | "required": [
|
| 147 | "tool_calls_total",
|
| 148 | "tool_calls_by_kind",
|
| 149 | "terminal_commands_total",
|
| 150 | "file_read_calls",
|
| 151 | "file_write_calls",
|
| 152 | "search_calls",
|
| 153 | "edit_calls",
|
| 154 | "vcs_commands_total",
|
| 155 | "test_commands_total",
|
| 156 | "network_calls_total",
|
| 157 | "privileged_calls_total",
|
| 158 | "help_calls_total",
|
| 159 | "unknown_command_failures_total",
|
| 160 | "parallel_batches_total",
|
| 161 | "max_concurrency",
|
| 162 | ],
|
| 163 | "properties": {
|
| 164 | "tool_calls_total": non_negative_integer(),
|
| 165 | "tool_calls_by_kind": {
|
| 166 | "type": "object",
|
| 167 | "additionalProperties": non_negative_integer("Count by adapter-defined tool kind."),
|
| 168 | "properties": {},
|
| 169 | },
|
| 170 | "terminal_commands_total": non_negative_integer(),
|
| 171 | "file_read_calls": non_negative_integer(),
|
| 172 | "file_write_calls": non_negative_integer(),
|
| 173 | "search_calls": non_negative_integer(),
|
| 174 | "edit_calls": non_negative_integer(),
|
| 175 | "vcs_commands_total": non_negative_integer(),
|
| 176 | "test_commands_total": non_negative_integer(),
|
| 177 | "network_calls_total": non_negative_integer(),
|
| 178 | "privileged_calls_total": non_negative_integer(),
|
| 179 | "help_calls_total": non_negative_integer(
|
| 180 | "Discovery calls: --help/-h flags, help subcommands, and man pages. "
|
| 181 | "High counts signal the agent does not know the tool zero-shot."
|
| 182 | ),
|
| 183 | "unknown_command_failures_total": optional_non_negative_integer(
|
| 184 | "Commands rejected as unknown subcommand/flag (hallucinated usage). Null when exit "
|
| 185 | "status or error text is not adapter-visible."
|
| 186 | ),
|
| 187 | "vcs_info_followup_calls_total": optional_non_negative_integer(
|
| 188 | "Successful read-only VCS info commands immediately re-asked in the same family "
|
| 189 | "(e.g. status, then status with more flags). Each one means the first output did "
|
| 190 | "not carry enough information to act on — the hidden cost of compact output. "
|
| 191 | "Null when per-call outcomes are not adapter-visible."
|
| 192 | ),
|
| 193 | "parallel_batches_total": optional_non_negative_integer(
|
| 194 | "Tool batches issued in parallel. Null when the adapter does not expose batching."
|
| 195 | ),
|
| 196 | "max_concurrency": optional_non_negative_integer(
|
| 197 | "Peak concurrent tool calls. Null when the adapter does not expose concurrency."
|
| 198 | ),
|
| 199 | "measurement_source": optional_string("How tool calls were observed, e.g. adapter_stream_events."),
|
| 200 | },
|
| 201 | }
|
| 202 |
|
| 203 | FAILURE_RETRY_METRICS_SCHEMA: dict[str, Any] = {
|
| 204 | "type": "object",
|
| 205 | "additionalProperties": False,
|
| 206 | "required": [
|
| 207 | "failed_tool_calls_total",
|
| 208 | "failed_commands_total",
|
| 209 | "failed_vcs_commands_total",
|
| 210 | "failed_test_commands_total",
|
| 211 | "retry_attempts_total",
|
| 212 | "retry_chains_total",
|
| 213 | "same_command_retries_total",
|
| 214 | "retry_after_failure_ms",
|
| 215 | "recoverable_failures_total",
|
| 216 | "unrecoverable_failures_total",
|
| 217 | "turns_to_recovery",
|
| 218 | "tokens_to_recovery",
|
| 219 | ],
|
| 220 | "properties": {
|
| 221 | "failed_tool_calls_total": optional_non_negative_integer(
|
| 222 | "Null when per-call success is not adapter-visible."
|
| 223 | ),
|
| 224 | "failed_commands_total": optional_non_negative_integer(),
|
| 225 | "failed_vcs_commands_total": optional_non_negative_integer(),
|
| 226 | "failed_test_commands_total": optional_non_negative_integer(),
|
| 227 | "retry_attempts_total": optional_non_negative_integer(),
|
| 228 | "retry_chains_total": optional_non_negative_integer(),
|
| 229 | "same_command_retries_total": optional_non_negative_integer(),
|
| 230 | "retry_after_failure_ms": optional_non_negative_number(),
|
| 231 | "recoverable_failures_total": optional_non_negative_integer(),
|
| 232 | "unrecoverable_failures_total": optional_non_negative_integer(),
|
| 233 | "turns_to_recovery": optional_non_negative_integer(
|
| 234 | "Assistant turns between the first failed tool call and the next successful tool call. "
|
| 235 | "Measures error-message ergonomics: a good error resolves in one turn."
|
| 236 | ),
|
| 237 | "tokens_to_recovery": optional_non_negative_integer(
|
| 238 | "Model output tokens spent in the recovery window, when per-turn usage is available."
|
| 239 | ),
|
| 240 | "failed_command_examples": string_array("Small redacted examples for debugging."),
|
| 241 | "measurement_source": optional_string("How failures/retries were observed."),
|
| 242 | },
|
| 243 | }
|
| 244 |
|
| 245 | WALL_CLOCK_METRICS_SCHEMA: dict[str, Any] = {
|
| 246 | "type": "object",
|
| 247 | "additionalProperties": False,
|
| 248 | "required": [
|
| 249 | "total_ms",
|
| 250 | "setup_ms",
|
| 251 | "agent_active_ms",
|
| 252 | "tool_wait_ms",
|
| 253 | "vcs_ms",
|
| 254 | "test_ms",
|
| 255 | "merge_ms",
|
| 256 | "cleanup_ms",
|
| 257 | "idle_ms",
|
| 258 | "timeout_ms",
|
| 259 | ],
|
| 260 | "properties": {
|
| 261 | "total_ms": non_negative_number(),
|
| 262 | "setup_ms": non_negative_number(),
|
| 263 | "agent_active_ms": non_negative_number(),
|
| 264 | "tool_wait_ms": optional_non_negative_number(
|
| 265 | "Null when the adapter does not separate tool wait from model time."
|
| 266 | ),
|
| 267 | "vcs_ms": non_negative_number(),
|
| 268 | "test_ms": non_negative_number(),
|
| 269 | "merge_ms": optional_non_negative_number("Null when the scenario has no merge phase."),
|
| 270 | "cleanup_ms": optional_non_negative_number("Null when cleanup is not separately timed."),
|
| 271 | "idle_ms": optional_non_negative_number("Null when harness queueing is not separately timed."),
|
| 272 | "timeout_ms": optional_non_negative_number(),
|
| 273 | "measurement_source": optional_string("How phases were attributed."),
|
| 274 | },
|
| 275 | }
|
| 276 |
|
| 277 | BYTES_METRICS_SCHEMA: dict[str, Any] = {
|
| 278 | "type": "object",
|
| 279 | "additionalProperties": False,
|
| 280 | "required": [
|
| 281 | "workspace_logical_bytes",
|
| 282 | "workspace_materialized_bytes",
|
| 283 | "bytes_read_by_agent",
|
| 284 | "bytes_written_by_agent",
|
| 285 | "bytes_hydrated",
|
| 286 | "files_hydrated",
|
| 287 | "hydration_events",
|
| 288 | "large_file_bytes_scanned",
|
| 289 | "binary_bytes_touched",
|
| 290 | "command_output_bytes",
|
| 291 | ],
|
| 292 | "properties": {
|
| 293 | "workspace_logical_bytes": non_negative_integer("Fixture size if fully materialized."),
|
| 294 | "workspace_materialized_bytes": non_negative_integer("Bytes materialized before timed work."),
|
| 295 | "bytes_read_by_agent": optional_non_negative_integer(
|
| 296 | "File bytes read through agent-visible read tools. Null when read-tool payloads are "
|
| 297 | "not adapter-visible — never transcript output bytes, which live in command_output_bytes. "
|
| 298 | "Hydration/read-efficiency claims must not cite a proxy here."
|
| 299 | ),
|
| 300 | "bytes_written_by_agent": non_negative_integer("File bytes written through agent-visible tools."),
|
| 301 | "bytes_hydrated": optional_non_negative_integer(
|
| 302 | "Bytes loaded lazily by VCS or filesystem. Null when no hydration probe ran."
|
| 303 | ),
|
| 304 | "files_hydrated": optional_non_negative_integer(
|
| 305 | "Files loaded lazily by VCS or filesystem. Null when no hydration probe ran."
|
| 306 | ),
|
| 307 | "hydration_events": optional_non_negative_integer(
|
| 308 | "Hydration events observed by the harness. Null when no hydration probe ran."
|
| 309 | ),
|
| 310 | "large_file_bytes_scanned": optional_non_negative_integer(
|
| 311 | "Bytes scanned in known large-file fixtures. Null when not instrumented."
|
| 312 | ),
|
| 313 | "binary_bytes_touched": optional_non_negative_integer(
|
| 314 | "Binary bytes read or written. Null when not instrumented."
|
| 315 | ),
|
| 316 | "command_output_bytes": non_negative_integer("Terminal output bytes admitted to transcript."),
|
| 317 | "measurement_source": optional_string("Adapter, filesystem probe, VCS trace, or estimate."),
|
| 318 | },
|
| 319 | }
|
| 320 |
|
| 321 | VCS_METRICS_SCHEMA: dict[str, Any] = {
|
| 322 | "type": "object",
|
| 323 | "additionalProperties": False,
|
| 324 | "required": [
|
| 325 | "commands_total",
|
| 326 | "status_commands",
|
| 327 | "diff_commands",
|
| 328 | "snapshot_commands",
|
| 329 | "branch_commands",
|
| 330 | "merge_commands",
|
| 331 | "cleanup_commands",
|
| 332 | "commits_created",
|
| 333 | "branches_created",
|
| 334 | "conflicts_total",
|
| 335 | "conflicts_resolved",
|
| 336 | "dirty_files_at_end",
|
| 337 | "cleanup_success",
|
| 338 | ],
|
| 339 | "properties": {
|
| 340 | "commands_total": optional_non_negative_integer(
|
| 341 | "Null when the agent transcript does not expose VCS command invocations."
|
| 342 | ),
|
| 343 | "status_commands": optional_non_negative_integer(),
|
| 344 | "diff_commands": optional_non_negative_integer(),
|
| 345 | "snapshot_commands": optional_non_negative_integer(),
|
| 346 | "branch_commands": optional_non_negative_integer(),
|
| 347 | "merge_commands": optional_non_negative_integer(),
|
| 348 | "cleanup_commands": optional_non_negative_integer(),
|
| 349 | "commits_created": non_negative_integer(),
|
| 350 | "branches_created": optional_non_negative_integer(),
|
| 351 | "conflicts_total": optional_non_negative_integer(
|
| 352 | "Null when conflict detection is not instrumented for the scenario."
|
| 353 | ),
|
| 354 | "conflicts_resolved": optional_non_negative_integer(),
|
f86d1109f5df
Review benchmark validity and add upstream-inspi
11 days ago
| 355 | "dirty_files_at_end": optional_non_negative_integer(
|
| 356 | "Null when the final status command fails or its output cannot be interpreted."
|
| 357 | ),
|
88e86b3faccd
Rebuild benchmarks as a clean single-root reposi
2 months ago
| 358 | "cleanup_success": boolean(),
|
| 359 | "measurement_source": optional_string("How VCS command outcomes were observed."),
|
| 360 | # PATH-shim attribution fields (optional, additive). Null when the
|
| 361 | # --vcs-shim instrument was off or recorded nothing: unmeasured, not
|
| 362 | # zero (ADR-0002).
|
| 363 | "shim_sidecar_path": optional_string(
|
| 364 | "Artifact path of the shim sidecar JSONL ({command, argv, start_ms, end_ms, "
|
| 365 | "elapsed_ms, returncode, shim} per VCS call). Null when the shim was off."
|
| 366 | ),
|
| 367 | "shim_sidecar_path_unavailable_reason": optional_string(
|
| 368 | "Reason shim_sidecar_path is null."
|
| 369 | ),
|
| 370 | "shim_overhead": optional_non_negative_number(
|
| 371 | "Requested alias for shim_overhead_ms: median extra ms per call added by the PATH shim."
|
| 372 | ),
|
| 373 | "shim_overhead_ms": optional_non_negative_number(
|
| 374 | "Median extra ms per call added by the PATH shim, measured per run via a "
|
| 375 | "calibration pass. Recorded for interpretation, never subtracted from measurements."
|
| 376 | ),
|
| 377 | "shim_overhead_unavailable_reason": optional_string(
|
| 378 | "Reason shim_overhead/shim_overhead_ms is null."
|
| 379 | ),
|
| 380 | "vcs_ms": optional_non_negative_number(
|
| 381 | "Sum of shim-recorded VCS call wall time (includes shim overhead)."
|
| 382 | ),
|
| 383 | "vcs_ms_unavailable_reason": optional_string("Reason shim-recorded vcs_ms is null."),
|
| 384 | "vcs_call_count_shim": optional_non_negative_integer(
|
| 385 | "Number of VCS binary invocations recorded by the PATH shim."
|
| 386 | ),
|
| 387 | "vcs_call_count_shim_unavailable_reason": optional_string(
|
| 388 | "Reason vcs_call_count_shim is null."
|
| 389 | ),
|
| 390 | "agent_blocked_on_vcs_ms": optional_non_negative_number(
|
| 391 | "oakbench.thrash.agent_blocked_on_vcs_ms over the shim sidecar rows."
|
| 392 | ),
|
| 393 | "agent_blocked_on_vcs_ms_unavailable_reason": optional_string(
|
| 394 | "Reason agent_blocked_on_vcs_ms is null."
|
| 395 | ),
|
| 396 | "vcs_share_of_task_wall": optional_non_negative_number(
|
| 397 | "agent_blocked_on_vcs_ms divided by agent wall time."
|
| 398 | ),
|
| 399 | "vcs_share_of_task_wall_unavailable_reason": optional_string(
|
| 400 | "Reason vcs_share_of_task_wall is null."
|
| 401 | ),
|
| 402 | "thrash_events_count": optional_non_negative_integer(
|
| 403 | "Count of oakbench.thrash.thrash_events detected in the shim sidecar rows."
|
| 404 | ),
|
| 405 | "thrash_events_count_unavailable_reason": optional_string(
|
| 406 | "Reason thrash_events_count is null."
|
| 407 | ),
|
| 408 | "polling_loop_count": optional_non_negative_integer(
|
| 409 | "Count of shim-sidecar thrash events with type vcs_polling_loop."
|
| 410 | ),
|
| 411 | "polling_loop_count_unavailable_reason": optional_string(
|
| 412 | "Reason polling_loop_count is null."
|
| 413 | ),
|
| 414 | },
|
| 415 | }
|
| 416 |
|
| 417 | TEST_METRICS_SCHEMA: dict[str, Any] = {
|
| 418 | "type": "object",
|
| 419 | "additionalProperties": False,
|
| 420 | "required": [
|
| 421 | "commands_total",
|
| 422 | "passed_commands",
|
| 423 | "failed_commands",
|
| 424 | "assertions_total",
|
| 425 | "assertions_failed",
|
| 426 | "required_checks_passed",
|
| 427 | ],
|
| 428 | "properties": {
|
| 429 | "commands_total": non_negative_integer(),
|
e9b767a51662
Preserve measured agent failures and validate ca
9 days ago
| 430 | "passed_commands": optional_non_negative_integer(),
|
| 431 | "failed_commands": optional_non_negative_integer(),
|
88e86b3faccd
Rebuild benchmarks as a clean single-root reposi
2 months ago
| 432 | "assertions_total": optional_non_negative_integer(),
|
| 433 | "assertions_failed": optional_non_negative_integer(),
|
e9b767a51662
Preserve measured agent failures and validate ca
9 days ago
| 434 | "required_checks_passed": optional_boolean(),
|
88e86b3faccd
Rebuild benchmarks as a clean single-root reposi
2 months ago
| 435 | },
|
| 436 | }
|
| 437 |
|
| 438 | PARALLEL_METRICS_SCHEMA: dict[str, Any] = {
|
| 439 | "type": "object",
|
| 440 | "additionalProperties": False,
|
| 441 | "required": [
|
| 442 | "tasks_total",
|
| 443 | "tasks_completed",
|
| 444 | "max_parallel_tasks",
|
| 445 | "overlapping_files_count",
|
| 446 | "lost_updates_detected",
|
| 447 | ],
|
| 448 | "properties": {
|
| 449 | "tasks_total": non_negative_integer(),
|
| 450 | "tasks_completed": non_negative_integer(),
|
| 451 | "max_parallel_tasks": non_negative_integer(),
|
| 452 | "overlapping_files_count": optional_non_negative_integer(
|
| 453 | "Null when overlap detection did not run (e.g. single-task scenarios)."
|
| 454 | ),
|
| 455 | "lost_updates_detected": optional_non_negative_integer(
|
| 456 | "Null when no integrity check ran. Zero means the check ran and found nothing lost."
|
| 457 | ),
|
| 458 | "lock_wait_ms": optional_non_negative_number(
|
| 459 | "Time spent blocked on or retrying VCS locks under contention."
|
| 460 | ),
|
| 461 | "commit_throughput_per_s": optional_non_negative_number(
|
| 462 | "Successful snapshots per second across all parallel workers."
|
| 463 | ),
|
| 464 | "integrity_check_passed": optional_boolean(
|
| 465 | "fsck-equivalent result after the parallel run. Null when no check ran."
|
| 466 | ),
|
| 467 | "measurement_source": optional_string("How parallel behavior was observed."),
|
| 468 | },
|
| 469 | }
|
| 470 |
|
| 471 | TURN_METRICS_SCHEMA: dict[str, Any] = {
|
| 472 | "type": "object",
|
| 473 | "additionalProperties": False,
|
| 474 | "required": [
|
| 475 | "turn_source",
|
| 476 | "assistant_turns_total",
|
| 477 | "tool_calls_per_turn_avg",
|
| 478 | "max_tool_calls_per_turn",
|
| 479 | "cumulative_input_tokens_reported",
|
| 480 | "peak_input_tokens_reported",
|
| 481 | ],
|
| 482 | "properties": {
|
| 483 | "turn_source": {"type": "string", "enum": TURN_SOURCES},
|
| 484 | "assistant_turns_total": optional_non_negative_integer(
|
| 485 | "Model inference round-trips. Each turn re-pays the transcript as input tokens, so "
|
| 486 | "turns dominate agent cost; tool-call counts alone hide this."
|
| 487 | ),
|
| 488 | "tool_calls_per_turn_avg": optional_non_negative_number(),
|
| 489 | "max_tool_calls_per_turn": optional_non_negative_integer(
|
| 490 | "Peak batching within one turn. Batched calls are far cheaper than sequential turns."
|
| 491 | ),
|
| 492 | "cumulative_input_tokens_reported": optional_non_negative_integer(
|
| 493 | "Sum of provider-reported input tokens (including cache reads) across all turns. "
|
| 494 | "Captures context-residency cost: early verbose output is re-paid every later turn."
|
| 495 | ),
|
| 496 | "peak_input_tokens_reported": optional_non_negative_integer(
|
| 497 | "Largest single-turn input, a proxy for peak context size."
|
| 498 | ),
|
| 499 | "turn_timeline_path": optional_string(
|
| 500 | "Artifact path of per-turn JSONL: turn index, input/output/cache tokens, tool calls."
|
| 501 | ),
|
| 502 | },
|
| 503 | }
|
| 504 |
|
| 505 | SCHEMA: dict[str, Any] = {
|
| 506 | "$schema": "https://json-schema.org/draft/2020-12/schema",
|
| 507 | "title": "Oak agent workflow benchmark result row",
|
| 508 | "schema_version": SCHEMA_VERSION,
|
| 509 | "type": "object",
|
| 510 | "additionalProperties": False,
|
| 511 | "required": [
|
| 512 | "schema_version",
|
| 513 | "bench_id",
|
| 514 | "timestamp_utc",
|
| 515 | "scenario",
|
| 516 | "run",
|
| 517 | "subject",
|
| 518 | "subject_kind",
|
| 519 | "vcs_mode",
|
| 520 | "agent",
|
| 521 | "fixture",
|
| 522 | "outcome",
|
| 523 | "success",
|
| 524 | "success_criteria",
|
| 525 | "token_metrics",
|
| 526 | "turn_metrics",
|
| 527 | "tool_call_metrics",
|
| 528 | "failed_command_retry_metrics",
|
| 529 | "wall_clock_metrics",
|
| 530 | "bytes_metrics",
|
| 531 | "vcs_metrics",
|
| 532 | "test_metrics",
|
| 533 | "parallel_metrics",
|
| 534 | ],
|
| 535 | "properties": {
|
| 536 | "schema_version": {"type": "integer", "const": SCHEMA_VERSION},
|
| 537 | "bench_id": string("Stable identifier for a benchmark invocation."),
|
| 538 | "timestamp_utc": string("UTC ISO-8601 timestamp for row completion."),
|
| 539 | "profile": optional_string("Run profile such as smoke, standard, or large."),
|
| 540 | "host": optional_string("Host name used for run provenance."),
|
| 541 | "platform": optional_string("Platform string used for run provenance."),
|
| 542 | "machine": optional_string("Machine architecture used for run provenance."),
|
| 543 | "scenario": string("Scenario id from scenarios/agent.yaml."),
|
| 544 | "scenario_version": optional_string("Fixture or prompt version for this scenario."),
|
e9b767a51662
Preserve measured agent failures and validate ca
9 days ago
| 545 | "skipped": boolean("Whether this trial lacks a required measurement instrument."),
|
| 546 | "runner_id": optional_string(),
|
| 547 | "runner_class": optional_string(),
|
| 548 | "runner_profile": {"type": "object"},
|
| 549 | "cache_state": optional_string(),
|
| 550 | "cache_state_reason": optional_string(),
|
| 551 | "calibration_version": optional_string(),
|
| 552 | "calibration_unmeasured_reason": optional_string(),
|
| 553 | "workflow_integrity_evidence_class": optional_string(),
|
| 554 | "workflow_subject_binary_sha256": optional_string(),
|
| 555 | "env_isolation_version": optional_string("Version of environment isolation policy."),
|
| 556 | "oracle_version": optional_string("Version of task correctness evidence."),
|
| 557 | "workflow_integrity_source": optional_string("Actual worktree/committed-tree evidence source."),
|
| 558 | "workflow_integrity_passed": optional_boolean("Null when required correctness instrument is unavailable."),
|
| 559 | "workflow_integrity_failure_reason": optional_string("Observed incorrect task state."),
|
| 560 | "workflow_integrity_unmeasured_reason": optional_string("Why required correctness evidence is unavailable."),
|
| 561 | "skip_reason": optional_string("Explicit missing measurement capability."),
|
| 562 | "git_head_instrument_binary": optional_string("Resolved Git object-reading instrument."),
|
| 563 | "git_head_instrument_sha256": optional_string("Git object reader SHA256."),
|
| 564 | "git_head_instrument_version": optional_string("Git object reader reported version."),
|
| 565 | "oak_export_instrument_binary": optional_string("Subject-produced export binary."),
|
| 566 | "oak_export_instrument_sha256": optional_string("Subject-produced export binary SHA256."),
|
| 567 | "oak_export_head": optional_string("Oak HEAD checked unchanged across export."),
|
88e86b3faccd
Rebuild benchmarks as a clean single-root reposi
2 months ago
| 568 | "operation": optional_string("Optional report grouping operation, e.g. agent.workflow."),
|
| 569 | "benchmark_track": optional_string("Command semantics track such as agent-default or core-equivalent."),
|
| 570 | "command_semantics_version": optional_string("Version of the command semantics contract."),
|
| 571 | "elapsed_ms": optional_non_negative_number("Compatibility mirror of wall_clock_metrics.total_ms."),
|
| 572 | "returncode": optional_non_negative_integer("Compatibility mirror: 0 for success, non-zero for failure."),
|
| 573 | "raw_output_bytes": optional_non_negative_integer("Compatibility mirror of captured agent stdout+stderr bytes."),
|
| 574 | "stdout_bytes": optional_non_negative_integer("Captured agent stdout bytes."),
|
| 575 | "stderr_bytes": optional_non_negative_integer("Captured agent stderr bytes."),
|
| 576 | "output_truncated": optional_boolean("Whether captured output was truncated before artifact storage."),
|
| 577 | "task_id": optional_string("Subtask id for parallel scenarios, if this row is per subtask."),
|
| 578 | "task_prompt_id": optional_string("Stable prompt/template id."),
|
| 579 | "run": non_negative_integer("Zero-based repetition index."),
|
| 580 | "subject": string("Benchmark subject such as git, oak_installed, or oak_local."),
|
| 581 | "subject_kind": {"type": "string", "enum": SUBJECT_KINDS},
|
| 582 | "subject_label": optional_string("Human label for reports."),
|
| 583 | "vcs_mode": {"type": "string", "enum": VCS_MODES},
|
| 584 | "agent": {
|
| 585 | "type": "object",
|
| 586 | "additionalProperties": False,
|
| 587 | "required": ["name", "adapter"],
|
| 588 | "properties": {
|
| 589 | "name": string("Agent or harness adapter name."),
|
| 590 | "adapter": string("Runner integration name."),
|
| 591 | "provider": optional_string("Provider name, if applicable."),
|
| 592 | "model": optional_string("Model id, if applicable."),
|
| 593 | "version": optional_string("Agent binary, prompt, or adapter version."),
|
| 594 | "environment": optional_string("Agent environment mode, such as minimal or local-default."),
|
| 595 | "instruction_level": {
|
| 596 | "type": ["string", "null"],
|
| 597 | "enum": INSTRUCTION_LEVELS + [None],
|
| 598 | "description": (
|
| 599 | "How much VCS-specific instruction the prompt carried. zero-shot measures the "
|
| 600 | "model-familiarity tax: Git is in pretraining data, a new VCS is not."
|
| 601 | ),
|
| 602 | },
|
| 603 | "temperature": {"type": ["number", "null"]},
|
| 604 | "max_output_tokens": optional_non_negative_integer(),
|
| 605 | },
|
| 606 | },
|
| 607 | "fixture": {
|
| 608 | "type": "object",
|
| 609 | "additionalProperties": False,
|
| 610 | "required": ["id", "repo_shape", "fixture_version"],
|
| 611 | "properties": {
|
| 612 | "id": string("Fixture id used by the runner."),
|
| 613 | "repo_shape": string("Human-readable fixture shape."),
|
| 614 | "fixture_version": string("Stable fixture content version or hash."),
|
| 615 | "logical_file_count": optional_non_negative_integer(),
|
| 616 | "logical_bytes": optional_non_negative_integer(),
|
| 617 | },
|
| 618 | },
|
| 619 | "source": {
|
| 620 | "type": "object",
|
| 621 | "additionalProperties": True,
|
| 622 | "properties": {
|
| 623 | "oak_hash": optional_string(),
|
| 624 | "oak_status": optional_string(),
|
| 625 | "benchmark_repo_hash": optional_string(),
|
| 626 | },
|
| 627 | },
|
| 628 | "outcome": {"type": "string", "enum": OUTCOMES},
|
| 629 | "success": boolean("True only when deterministic scenario criteria passed."),
|
| 630 | "failure_reason": optional_string("Short stable failure reason for dashboards."),
|
| 631 | "success_criteria": {
|
| 632 | "type": "object",
|
| 633 | "additionalProperties": False,
|
| 634 | "required": ["passed", "checks"],
|
| 635 | "properties": {
|
| 636 | "passed": boolean(),
|
| 637 | "checks": {
|
| 638 | "type": "object",
|
e9b767a51662
Preserve measured agent failures and validate ca
9 days ago
| 639 | "additionalProperties": optional_boolean("Individual deterministic check result; null if unmeasured."),
|
88e86b3faccd
Rebuild benchmarks as a clean single-root reposi
2 months ago
| 640 | "properties": {},
|
| 641 | },
|
| 642 | "notes": optional_string(),
|
| 643 | },
|
| 644 | },
|
| 645 | "token_metrics": TOKEN_METRICS_SCHEMA,
|
| 646 | "turn_metrics": TURN_METRICS_SCHEMA,
|
| 647 | "tool_call_metrics": TOOL_CALL_METRICS_SCHEMA,
|
| 648 | "failed_command_retry_metrics": FAILURE_RETRY_METRICS_SCHEMA,
|
| 649 | "wall_clock_metrics": WALL_CLOCK_METRICS_SCHEMA,
|
| 650 | "bytes_metrics": BYTES_METRICS_SCHEMA,
|
| 651 | "vcs_metrics": VCS_METRICS_SCHEMA,
|
| 652 | "test_metrics": TEST_METRICS_SCHEMA,
|
| 653 | "parallel_metrics": PARALLEL_METRICS_SCHEMA,
|
| 654 | "artifacts": {
|
| 655 | "type": "object",
|
| 656 | "additionalProperties": False,
|
| 657 | "properties": {
|
| 658 | "transcript_path": optional_string(),
|
| 659 | "command_log_path": optional_string(),
|
| 660 | "diff_path": optional_string(),
|
| 661 | "workspace_path": optional_string(),
|
| 662 | "extra_paths": string_array(),
|
| 663 | },
|
| 664 | },
|
| 665 | "notes": optional_string(),
|
| 666 | },
|
| 667 | }
|
| 668 |
|
| 669 |
|
| 670 | def type_matches(value: Any, expected_type: Any) -> bool:
|
| 671 | if isinstance(expected_type, list):
|
| 672 | return any(type_matches(value, item) for item in expected_type)
|
| 673 | if expected_type == "null":
|
| 674 | return value is None
|
| 675 | if expected_type == "boolean":
|
| 676 | return isinstance(value, bool)
|
| 677 | if expected_type == "integer":
|
| 678 | return isinstance(value, int) and not isinstance(value, bool)
|
| 679 | if expected_type == "number":
|
| 680 | return (isinstance(value, int) or isinstance(value, float)) and not isinstance(value, bool)
|
| 681 | if expected_type == "string":
|
| 682 | return isinstance(value, str)
|
| 683 | if expected_type == "object":
|
| 684 | return isinstance(value, dict)
|
| 685 | if expected_type == "array":
|
| 686 | return isinstance(value, list)
|
| 687 | return True
|
| 688 |
|
| 689 |
|
| 690 | def display_path(path: str, key: str) -> str:
|
| 691 | if path == "$":
|
| 692 | return "$" + "." + key
|
| 693 | return path + "." + key
|
| 694 |
|
| 695 |
|
| 696 | def validate_value(value: Any, schema: dict[str, Any], path: str) -> List[str]:
|
| 697 | errors: List[str] = []
|
| 698 | expected_type = schema.get("type")
|
| 699 | if expected_type is not None and not type_matches(value, expected_type):
|
| 700 | errors.append(f"{path}: expected {expected_type}, got {type(value).__name__}")
|
| 701 | return errors
|
| 702 |
|
| 703 | if "const" in schema and value != schema["const"]:
|
| 704 | errors.append(f"{path}: expected constant {schema['const']!r}, got {value!r}")
|
| 705 | if "enum" in schema and value not in schema["enum"]:
|
| 706 | errors.append(f"{path}: expected one of {schema['enum']!r}, got {value!r}")
|
| 707 | if "minimum" in schema and isinstance(value, (int, float)) and not isinstance(value, bool):
|
| 708 | if value < schema["minimum"]:
|
| 709 | errors.append(f"{path}: expected >= {schema['minimum']}, got {value!r}")
|
| 710 |
|
| 711 | if isinstance(value, dict):
|
| 712 | required = schema.get("required", [])
|
| 713 | for key in required:
|
| 714 | if key not in value:
|
| 715 | errors.append(f"{path}: missing required property {key!r}")
|
| 716 | properties = schema.get("properties", {})
|
| 717 | additional = schema.get("additionalProperties", True)
|
| 718 | for key, item in value.items():
|
| 719 | child_path = display_path(path, key)
|
| 720 | if key in properties:
|
| 721 | errors.extend(validate_value(item, properties[key], child_path))
|
| 722 | elif additional is False:
|
| 723 | errors.append(f"{child_path}: unexpected property")
|
| 724 | elif isinstance(additional, dict):
|
| 725 | errors.extend(validate_value(item, additional, child_path))
|
| 726 | elif isinstance(value, list) and "items" in schema:
|
| 727 | item_schema = schema["items"]
|
| 728 | for index, item in enumerate(value):
|
| 729 | errors.extend(validate_value(item, item_schema, f"{path}[{index}]"))
|
| 730 |
|
| 731 | return errors
|
| 732 |
|
| 733 |
|
| 734 | def validate_schema_shape(schema: dict[str, Any]) -> List[str]:
|
| 735 | errors: List[str] = []
|
| 736 | for key in ["title", "type", "required", "properties"]:
|
| 737 | if key not in schema:
|
| 738 | errors.append(f"schema missing {key!r}")
|
| 739 | if schema.get("type") != "object":
|
| 740 | errors.append("schema root type must be object")
|
| 741 | properties = schema.get("properties", {})
|
| 742 | for required_key in schema.get("required", []):
|
| 743 | if required_key not in properties:
|
| 744 | errors.append(f"schema requires {required_key!r} but has no property definition")
|
| 745 | return errors
|
| 746 |
|
| 747 |
|
| 748 | def validate_row(row: Any) -> List[str]:
|
| 749 | if not isinstance(row, dict):
|
| 750 | return [f"$: expected object row, got {type(row).__name__}"]
|
| 751 | return validate_value(row, SCHEMA, "$")
|
| 752 |
|
| 753 |
|
| 754 | def iter_rows(path: Path) -> Iterable[Any]:
|
| 755 | text = path.read_text(encoding="utf-8")
|
| 756 | stripped = text.strip()
|
| 757 | if not stripped:
|
| 758 | return []
|
| 759 | if path.suffix == ".json" or stripped[0] == "[":
|
| 760 | data = json.loads(stripped)
|
| 761 | if isinstance(data, list):
|
| 762 | return data
|
| 763 | return [data]
|
| 764 | rows: List[Any] = []
|
| 765 | for line_number, line in enumerate(text.splitlines(), 1):
|
| 766 | if not line.strip():
|
| 767 | continue
|
| 768 | try:
|
| 769 | rows.append(json.loads(line))
|
| 770 | except json.JSONDecodeError as exc:
|
| 771 | rows.append({"__json_error__": f"line {line_number}: {exc}"})
|
| 772 | return rows
|
| 773 |
|
| 774 |
|
| 775 | def validate_file(path: Path) -> dict[str, Any]:
|
| 776 | errors: List[str] = []
|
| 777 | rows = list(iter_rows(path))
|
| 778 | for index, row in enumerate(rows, 1):
|
| 779 | if isinstance(row, dict) and "__json_error__" in row:
|
| 780 | errors.append(row["__json_error__"])
|
| 781 | continue
|
| 782 | for error in validate_row(row):
|
| 783 | errors.append(f"row {index}: {error}")
|
| 784 | return {"ok": not errors, "rows": len(rows), "errors": errors}
|
| 785 |
|
| 786 |
|
| 787 | def example_row() -> dict[str, Any]:
|
| 788 | return {
|
| 789 | "schema_version": SCHEMA_VERSION,
|
| 790 | "bench_id": "20260609T000000Z",
|
| 791 | "timestamp_utc": "2026-06-09T00:00:00Z",
|
| 792 | "profile": "smoke",
|
| 793 | "scenario": "one_line_bug_fix",
|
| 794 | "scenario_version": "v1",
|
| 795 | "task_id": None,
|
| 796 | "task_prompt_id": "agent-one-line-bug-fix-v1",
|
| 797 | "run": 0,
|
| 798 | "subject": "oak_local",
|
| 799 | "subject_kind": "oak",
|
| 800 | "subject_label": "Oak local",
|
| 801 | "vcs_mode": "oak_space_per_task",
|
| 802 | "agent": {
|
| 803 | "name": "example-agent",
|
| 804 | "adapter": "example-adapter",
|
| 805 | "provider": None,
|
| 806 | "model": None,
|
| 807 | "version": None,
|
| 808 | "environment": "minimal",
|
| 809 | "instruction_level": "cheat-sheet",
|
| 810 | "temperature": None,
|
| 811 | "max_output_tokens": None,
|
| 812 | },
|
| 813 | "fixture": {
|
| 814 | "id": "agent_one_line_bug_fix",
|
| 815 | "repo_shape": "small text application",
|
| 816 | "fixture_version": "v1",
|
| 817 | "logical_file_count": 42,
|
| 818 | "logical_bytes": 65536,
|
| 819 | },
|
| 820 | "source": {"oak_hash": None, "oak_status": None, "benchmark_repo_hash": None},
|
| 821 | "outcome": "pass",
|
| 822 | "success": True,
|
| 823 | "failure_reason": None,
|
| 824 | "success_criteria": {
|
| 825 | "passed": True,
|
| 826 | "checks": {"tests_passed": True, "expected_files_changed": True},
|
| 827 | "notes": None,
|
| 828 | },
|
| 829 | "token_metrics": {
|
| 830 | "token_source": "char_only",
|
| 831 | "tokenizer": None,
|
| 832 | "input_tokens_reported": None,
|
| 833 | "output_tokens_reported": None,
|
| 834 | "reasoning_tokens_reported": None,
|
| 835 | "cache_read_tokens_reported": None,
|
| 836 | "cache_write_tokens_reported": None,
|
| 837 | "total_tokens_reported": None,
|
| 838 | "normalized_input_chars": 12000,
|
| 839 | "normalized_output_chars": 1200,
|
| 840 | "transcript_chars": 13200,
|
| 841 | "context_truncation_events": None,
|
| 842 | "token_budget": None,
|
| 843 | "cost_weighted_total": 4500.0,
|
| 844 | "cost_weights": "emitted=5,ingested=1",
|
| 845 | },
|
| 846 | "turn_metrics": {
|
| 847 | "turn_source": "adapter_stream",
|
| 848 | "assistant_turns_total": 6,
|
| 849 | "tool_calls_per_turn_avg": 1.33,
|
| 850 | "max_tool_calls_per_turn": 2,
|
| 851 | "cumulative_input_tokens_reported": 48000,
|
| 852 | "peak_input_tokens_reported": 11000,
|
| 853 | "turn_timeline_path": None,
|
| 854 | },
|
| 855 | "tool_call_metrics": {
|
| 856 | "tool_calls_total": 8,
|
| 857 | "tool_calls_by_kind": {"terminal": 3, "read": 3, "edit": 1, "search": 1},
|
| 858 | "terminal_commands_total": 3,
|
| 859 | "file_read_calls": 3,
|
| 860 | "file_write_calls": 1,
|
| 861 | "search_calls": 1,
|
| 862 | "edit_calls": 1,
|
| 863 | "vcs_commands_total": 2,
|
| 864 | "test_commands_total": 1,
|
| 865 | "network_calls_total": 0,
|
| 866 | "privileged_calls_total": 0,
|
| 867 | "help_calls_total": 0,
|
| 868 | "unknown_command_failures_total": None,
|
| 869 | "parallel_batches_total": None,
|
| 870 | "max_concurrency": None,
|
| 871 | "measurement_source": "adapter_stream_events",
|
| 872 | },
|
| 873 | "failed_command_retry_metrics": {
|
| 874 | "failed_tool_calls_total": 0,
|
| 875 | "failed_commands_total": 0,
|
| 876 | "failed_vcs_commands_total": 0,
|
| 877 | "failed_test_commands_total": 0,
|
| 878 | "retry_attempts_total": None,
|
| 879 | "retry_chains_total": None,
|
| 880 | "same_command_retries_total": None,
|
| 881 | "retry_after_failure_ms": None,
|
| 882 | "recoverable_failures_total": None,
|
| 883 | "unrecoverable_failures_total": None,
|
| 884 | "turns_to_recovery": None,
|
| 885 | "tokens_to_recovery": None,
|
| 886 | "failed_command_examples": [],
|
| 887 | "measurement_source": "adapter_stream_events",
|
| 888 | },
|
| 889 | "wall_clock_metrics": {
|
| 890 | "total_ms": 15000.0,
|
| 891 | "setup_ms": 800.0,
|
| 892 | "agent_active_ms": 7000.0,
|
| 893 | "tool_wait_ms": None,
|
| 894 | "vcs_ms": 900.0,
|
| 895 | "test_ms": 2100.0,
|
| 896 | "merge_ms": None,
|
| 897 | "cleanup_ms": None,
|
| 898 | "idle_ms": None,
|
| 899 | "timeout_ms": None,
|
| 900 | "measurement_source": "outer_cli_wall_clock",
|
| 901 | },
|
| 902 | "bytes_metrics": {
|
| 903 | "workspace_logical_bytes": 65536,
|
| 904 | "workspace_materialized_bytes": 32768,
|
| 905 | "bytes_read_by_agent": 18000,
|
| 906 | "bytes_written_by_agent": 140,
|
| 907 | "bytes_hydrated": None,
|
| 908 | "files_hydrated": None,
|
| 909 | "hydration_events": None,
|
| 910 | "large_file_bytes_scanned": None,
|
| 911 | "binary_bytes_touched": None,
|
| 912 | "command_output_bytes": 3000,
|
| 913 | "measurement_source": "example-adapter",
|
| 914 | },
|
| 915 | "vcs_metrics": {
|
| 916 | "commands_total": 2,
|
| 917 | "status_commands": 1,
|
| 918 | "diff_commands": 0,
|
| 919 | "snapshot_commands": 1,
|
| 920 | "branch_commands": 0,
|
| 921 | "merge_commands": 0,
|
| 922 | "cleanup_commands": 0,
|
| 923 | "commits_created": 1,
|
| 924 | "branches_created": 1,
|
| 925 | "conflicts_total": None,
|
| 926 | "conflicts_resolved": None,
|
| 927 | "dirty_files_at_end": 0,
|
| 928 | "cleanup_success": True,
|
| 929 | "measurement_source": "transcript_command_classification",
|
| 930 | },
|
| 931 | "test_metrics": {
|
| 932 | "commands_total": 1,
|
| 933 | "passed_commands": 1,
|
| 934 | "failed_commands": 0,
|
| 935 | "assertions_total": None,
|
| 936 | "assertions_failed": None,
|
| 937 | "required_checks_passed": True,
|
| 938 | },
|
| 939 | "parallel_metrics": {
|
| 940 | "tasks_total": 1,
|
| 941 | "tasks_completed": 1,
|
| 942 | "max_parallel_tasks": 1,
|
| 943 | "overlapping_files_count": None,
|
| 944 | "lost_updates_detected": None,
|
| 945 | "lock_wait_ms": None,
|
| 946 | "commit_throughput_per_s": None,
|
| 947 | "integrity_check_passed": None,
|
| 948 | "measurement_source": "single_task_run",
|
| 949 | },
|
| 950 | "artifacts": {
|
| 951 | "transcript_path": None,
|
| 952 | "command_log_path": None,
|
| 953 | "diff_path": None,
|
| 954 | "workspace_path": None,
|
| 955 | "extra_paths": [],
|
| 956 | },
|
| 957 | "notes": None,
|
| 958 | }
|
| 959 |
|
| 960 |
|
| 961 | def parse_args() -> argparse.Namespace:
|
| 962 | parser = argparse.ArgumentParser(description=__doc__)
|
| 963 | parser.add_argument("--validate-jsonl", type=Path, help="Validate a JSON or JSONL result file.")
|
| 964 | parser.add_argument("--example", action="store_true", help="Print an example valid result row.")
|
| 965 | parser.add_argument("--compact", action="store_true", help="Print compact JSON.")
|
| 966 | return parser.parse_args()
|
| 967 |
|
| 968 |
|
| 969 | def print_json(value: Any, compact: bool) -> None:
|
| 970 | if compact:
|
| 971 | print(json.dumps(value, sort_keys=True, separators=(",", ":")))
|
| 972 | else:
|
| 973 | print(json.dumps(value, indent=2, sort_keys=True))
|
| 974 |
|
| 975 |
|
| 976 | def main() -> int:
|
| 977 | args = parse_args()
|
| 978 | schema_errors = validate_schema_shape(SCHEMA)
|
| 979 | if schema_errors:
|
| 980 | print_json({"ok": False, "schema_errors": schema_errors}, args.compact)
|
| 981 | return 1
|
| 982 |
|
| 983 | if args.example:
|
| 984 | row = example_row()
|
| 985 | errors = validate_row(row)
|
| 986 | if errors:
|
| 987 | print_json({"ok": False, "example_errors": errors}, args.compact)
|
| 988 | return 1
|
| 989 | print_json(row, args.compact)
|
| 990 | return 0
|
| 991 |
|
| 992 | if args.validate_jsonl:
|
| 993 | result = validate_file(args.validate_jsonl)
|
| 994 | print_json(result, args.compact)
|
| 995 | return 0 if result["ok"] else 1
|
| 996 |
|
| 997 | print_json(SCHEMA, args.compact)
|
| 998 | return 0
|
| 999 |
|
| 1000 |
|
| 1001 | if __name__ == "__main__":
|
| 1002 | raise SystemExit(main())
|