Spaces:
Running on Zero
Running on Zero
File size: 38,147 Bytes
849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b ef0cd9b 8457788 ef0cd9b 849ee7b 8457788 ef0cd9b 8457788 ef0cd9b 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 849ee7b 8457788 ef0cd9b 8457788 849ee7b 8457788 ef0cd9b 8457788 849ee7b 8457788 c8055f7 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b c8055f7 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 ef0cd9b 8457788 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b fbb7c0c 849ee7b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 | """Deterministic codebook analysis for coding-agent narrative traces."""
from __future__ import annotations
import re
import time
from collections import Counter
from datetime import datetime, timezone
from pathlib import Path
from typing import Iterable
from model_runtime import MODEL_CHOICES, run_model_analysis
from parser import parse_trace
from profiling import Profiler, get_logger
from redaction import redact_text
from schemas import (
APPRAISALS,
DETOUR_TYPES,
DIFFICULTY_TYPES,
OUTCOME_CLAIMS,
RECOVERY_PATTERNS,
RESOLUTION_MODES,
AnalysisResult,
DifficultyEpisode,
MessageSpan,
NarrativeMessage,
)
logger = get_logger()
ANALYSIS_SCOPE = (
"assistant narrative messages only, with user prompts included only as optional context; "
"raw tool-call contents are ignored by default"
)
DIFFICULTY_SIGNALS = {
"error",
"failed",
"failure",
"fails",
"problem",
"issue",
"bug",
"blocked",
"blocker",
"cannot",
"can't",
"could not",
"unclear",
"ambiguous",
"not sure",
"risk",
"regression",
"compatibility",
"however",
"but",
"unfortunately",
"missing",
"incomplete",
"permission",
"auth",
"timeout",
"dependency",
"conflict",
"mismatch",
"unexpected",
"verify",
"verification",
"test failing",
}
INTENTION_SIGNALS = {
"i will",
"i'll",
"i am going to",
"i'm going to",
"next",
"plan",
"goal",
"need to",
"i need",
"i should",
"let me",
"i'm checking",
"i'm going",
"i will inspect",
"i'll inspect",
}
SHIFT_SIGNALS = {
"instead",
"alternative",
"switch",
"change approach",
"narrow",
"smaller",
"decompose",
"split",
"break down",
"roll back",
"revert",
"try another",
"workaround",
"safer",
"different route",
"verify with",
}
OUTCOME_SIGNALS = {
"done",
"fixed",
"resolved",
"complete",
"implemented",
"verified",
"passes",
"works",
"unable",
"could not",
"still failing",
"needs verification",
"not verified",
"caveat",
"partial",
"partially",
}
PROBLEM_EVIDENCE_SIGNALS = {
"failed",
"failure",
"fails",
"still failing",
"test failing",
"issue",
"bug",
"blocked",
"blocker",
"cannot",
"can't",
"could not",
"unclear",
"ambiguous",
"not sure",
"risk",
"regression",
"unfortunately",
"missing",
"incomplete",
"permission",
"dependency",
"conflict",
"mismatch",
"unexpected",
}
ANALYSIS_STEPS = ("extract", "redact", "chart", "classify", "synthesize")
def _accumulate_notes(counter: Counter[str], notes: Iterable[str]) -> None:
"""Fold ``"label: count"`` note strings into a running counter."""
for note in notes:
label, _, count = note.partition(": ")
counter[label] += int(count or 0)
def stream_deterministic_analysis(
path: str | Path,
*,
include_user_context: bool = True,
redact_secrets: bool = True,
ignore_tool_calls: bool = True,
model_redact=None,
profiler: Profiler | None = None,
stream_redact_progress: bool = False,
):
"""Run the deterministic pipeline as a generator.
Yields ``("progress", info)`` after each real stage completes — ``info`` has
a ``stage`` name (one of :data:`ANALYSIS_STEPS`) and the running ``messages``
count — then a final ``("result", (AnalysisResult, str))``. Callers that
don't care about progress can just drain it for the tuple.
``model_redact`` is an optional ``(list[str]) -> list[RedactionResult]``
callable applied on top of regex redaction; the Server injects a GPU- or
CPU-bound ``openai/privacy-filter`` pass. It is absent locally and in tests,
so redaction falls back to regex only. ``profiler`` collects per-stage
timings; one is created if not supplied.
"""
prof = profiler or Profiler("deterministic")
_started = time.perf_counter()
parsed_messages, agent_type = parse_trace(
path,
include_user_context=include_user_context,
ignore_tool_calls=ignore_tool_calls,
)
prof.record("extract", time.perf_counter() - _started)
message_count = len(parsed_messages)
prof.mark(messages=message_count, agent=agent_type)
logger.info("parsed %d narrative messages (agent=%s)", message_count, agent_type)
yield ("progress", {"stage": "extract", "messages": message_count})
redaction_count = 0
privacy_notes = [
"Uploaded traces are processed for this request only; the app exports a redacted narrative text file.",
"The analysis uses visible messages and does not inspect hidden reasoning.",
]
if ignore_tool_calls:
privacy_notes.append("Tool-call contents were ignored before analysis.")
_redact_started = time.perf_counter()
messages = parsed_messages
if redact_secrets:
all_notes: Counter[str] = Counter()
redacted_messages: list[NarrativeMessage] = []
model_used = False
model_failed = False
# Process in chunks so slow (CPU) runs can stream per-message progress.
# Without streaming (ZeroGPU) it is a single chunk = one GPU allocation;
# with streaming the update count is capped at ~30 regardless of size.
if stream_redact_progress and message_count:
chunk = max(1, (message_count + 29) // 30)
else:
chunk = message_count or 1
for start in range(0, message_count, chunk):
chunk_messages = parsed_messages[start : start + chunk]
# Pass 1: deterministic regex redaction (always available).
regex_results = [redact_text(message.text) for message in chunk_messages]
texts = [red.text for red in regex_results]
# Pass 2: optional model PII pass on top. The Server injects a GPU- or
# CPU-bound openai/privacy-filter pass; it is absent locally and in
# tests, so regex-only redaction is used. Once it is unavailable we
# stop retrying it for the rest of the trace.
model_results = None
if model_redact is not None and not model_failed:
try:
model_results = model_redact(texts)
model_used = True
except Exception as exc: # noqa: BLE001 - graceful degradation
privacy_notes.append(
"AI privacy filter was unavailable "
f"({type(exc).__name__}); regex redaction was applied."
)
model_failed = True
model_results = None
for i, message in enumerate(chunk_messages):
text = texts[i]
redaction_count += regex_results[i].count
_accumulate_notes(all_notes, regex_results[i].notes)
if model_results is not None:
text = model_results[i].text
redaction_count += model_results[i].count
_accumulate_notes(all_notes, model_results[i].notes)
redacted_messages.append(
NarrativeMessage(
index=message.index,
role=message.role,
text=text,
timestamp=message.timestamp,
source=message.source,
)
)
yield (
"progress",
{
"stage": "redact",
"processed": min(start + chunk, message_count),
"total": message_count,
},
)
messages = redacted_messages
if model_used:
privacy_notes.append(
"AI privacy filter (openai/privacy-filter) screened for names, "
"contacts, and other personal data."
)
if all_notes:
privacy_notes.append(
"Redactions applied: "
+ ", ".join(f"{label} ({count})" for label, count in sorted(all_notes.items()))
+ "."
)
else:
privacy_notes.append("No likely secrets matched the redaction patterns.")
else:
privacy_notes.append("Secret redaction was disabled by the user.")
prof.record("redact", time.perf_counter() - _redact_started)
prof.mark(redactions=redaction_count)
if not redact_secrets or message_count == 0:
# No chunk loop ran (redaction disabled or empty trace) — still advance.
yield ("progress", {"stage": "redact", "processed": message_count, "total": message_count})
_chart_started = time.perf_counter()
episodes = identify_episodes(messages)
prof.record("chart", time.perf_counter() - _chart_started)
prof.mark(episodes=len(episodes))
yield ("progress", {"stage": "chart", "messages": message_count})
_classify_started = time.perf_counter()
result = AnalysisResult(
trace_title=derive_trace_title(path, agent_type),
agent_type_guess=agent_type,
analysis_scope=ANALYSIS_SCOPE,
privacy_notes=privacy_notes,
episodes=episodes,
overall_patterns=summarize_patterns(episodes, messages),
narrative_message_count=len(messages),
redaction_count=redaction_count,
engine="deterministic-codebook",
)
prof.record("classify", time.perf_counter() - _classify_started)
yield ("progress", {"stage": "classify", "messages": message_count})
_synth_started = time.perf_counter()
narrative_text = render_redacted_narrative(messages)
prof.record("synthesize", time.perf_counter() - _synth_started)
yield ("progress", {"stage": "synthesize", "messages": message_count})
yield ("result", (result, narrative_text, messages))
_PRODUCTIVE_VALUES = {"yes", "no", "mixed", "unknown"}
_VALID_TONES = {"stable", "iterative", "detour", "partial", "risk", "unknown"}
_VALID_HONESTY = {"candid", "mixed", "overclaimed"}
def build_numbered_narrative(
messages: list[NarrativeMessage], *, char_budget: int = 16000, per_message: int = 320
) -> str:
"""Number the (redacted) messages by real index for the model.
Long traces are sampled evenly across the session (keeping the first and last)
so the model sees the whole timeline within its context budget; each line keeps
the message's real index and timestamp so the model can cite spans.
"""
if not messages:
return ""
max_messages = max(1, char_budget // per_message)
if len(messages) <= max_messages:
chosen = messages
else:
stride = len(messages) / max_messages
picks = sorted({0, len(messages) - 1, *(int(i * stride) for i in range(max_messages))})
chosen = [messages[i] for i in picks if 0 <= i < len(messages)]
lines = []
for message in chosen:
snippet = " ".join(message.text.split())[:per_message]
lines.append(f"[{message.index}] {message.role} {message.timestamp or ''}: {snippet}")
return "\n".join(lines)
def build_codebook_hint(episodes: list[DifficultyEpisode]) -> str:
if not episodes:
return "(none)"
return "; ".join(
f"{ep.episode_id} msgs {ep.message_span.start_index}-{ep.message_span.end_index}"
for ep in episodes[:12]
)
def _coerce_code(value: object, vocab: dict[str, str]) -> str:
code = str(value or "").strip()
return code if code in vocab else "unknown"
# Weak models sometimes echo the schema placeholders verbatim; drop those.
_PLACEHOLDER_RE = re.compile(
r"^\s*(<.*>|<=.*|\d+(\s*-\s*\d+)?\s+sentences?.*|one key.*|short verbatim.*|up to \d+.*|a message index.*)\s*$",
re.IGNORECASE,
)
def _clean_text(value: object) -> str:
text = str(value or "").strip()
if not text or _PLACEHOLDER_RE.match(text):
return ""
return text
def _clean_verdict(verdict: dict) -> dict[str, str]:
tone = str(verdict.get("tone", "")).strip().lower()
honesty = str(verdict.get("honesty", "")).strip().lower()
return {
"tone": tone if tone in _VALID_TONES else "unknown",
"headline": _clean_text(verdict.get("headline")) or "Session analyzed by the model.",
"detail": _clean_text(verdict.get("detail")),
"honesty": honesty if honesty in _VALID_HONESTY else "mixed",
}
def _episode_from_model(
raw: dict, ordinal: int, index_to_timestamp: dict[int, str | None], max_index: int
) -> DifficultyEpisode:
def clamp(value: object) -> int:
try:
return max(0, min(int(value), max_index))
except (TypeError, ValueError):
return 0
start = clamp(raw.get("start_index", 0))
end = clamp(raw.get("end_index", start))
if end < start:
start, end = end, start
start_time = index_to_timestamp.get(start)
end_time = index_to_timestamp.get(end)
span = MessageSpan(
start_index=start,
end_index=end,
start_time=start_time,
end_time=end_time,
duration_label=duration_label(start_time, end_time) if start_time and end_time else "unknown",
)
productive = str(raw.get("productive_detour", "unknown")).strip().lower()
quotes = [cleaned for q in (raw.get("evidence_quotes") or []) if (cleaned := _clean_text(q))][:3]
difficulty = _clean_text(raw.get("reported_difficulty"))
title = _clean_text(raw.get("title")) or (difficulty[:60] if difficulty else "Difficulty episode")
return DifficultyEpisode(
episode_id=f"E{ordinal:02d}",
title=title,
message_span=span,
initial_intention=_clean_text(raw.get("initial_intention")),
reported_difficulty=difficulty,
difficulty_type=_coerce_code(raw.get("difficulty_type"), DIFFICULTY_TYPES),
appraisal=_coerce_code(raw.get("appraisal"), APPRAISALS),
strategy_before=_clean_text(raw.get("strategy_before")),
strategy_after=_clean_text(raw.get("strategy_after")),
detour_type=_coerce_code(raw.get("detour_type"), DETOUR_TYPES),
resolution_mode=_coerce_code(raw.get("resolution_mode"), RESOLUTION_MODES),
recovery_pattern=_coerce_code(raw.get("recovery_pattern"), RECOVERY_PATTERNS),
outcome_claim=_coerce_code(raw.get("outcome_claim"), OUTCOME_CLAIMS),
productive_detour=productive if productive in _PRODUCTIVE_VALUES else "unknown",
evidence_quotes=quotes,
analyst_memo=_clean_text(raw.get("analyst_memo")),
)
def apply_model_analysis(
result: AnalysisResult,
messages: list[NarrativeMessage],
analysis_engine: str,
*,
run=None,
) -> None:
"""Replace the deterministic analysis with a model-produced one (codebook is the fallback).
``run`` defaults to :func:`run_model_analysis` (resolved at call time so tests
can monkeypatch it); the Server passes a GPU- or CPU-bound runner. On success
the model's episodes, overall patterns, and verdict replace the rule-based
ones. On any failure the deterministic codebook result is kept and the reason
recorded in ``model_notes``.
"""
if analysis_engine == "deterministic":
return
if analysis_engine not in MODEL_CHOICES:
result.model_notes.append(
f"Unknown analysis engine {analysis_engine!r}; rule-based analysis was returned."
)
return
runner = run or run_model_analysis
numbered_narrative = build_numbered_narrative(messages)
codebook_hint = build_codebook_hint(result.episodes)
try:
produced = runner(
engine=analysis_engine,
numbered_narrative=numbered_narrative,
agent_type=result.agent_type_guess,
codebook_hint=codebook_hint,
)
except Exception as exc:
error_message = str(exc).strip().rstrip(".")
result.model_notes.append(
"Model analysis was requested but unavailable: "
f"{type(exc).__name__}: {error_message}. "
"Rule-based analysis was returned."
)
return
analysis = produced.analysis
index_to_timestamp = {message.index: message.timestamp for message in messages}
max_index = (len(messages) - 1) if messages else 0
episodes = [
_episode_from_model(raw, ordinal + 1, index_to_timestamp, max_index)
for ordinal, raw in enumerate(analysis.get("episodes", []))
]
result.episodes = episodes
patterns = analysis.get("overall_patterns")
if isinstance(patterns, dict) and patterns:
result.overall_patterns = {key: str(value) for key, value in patterns.items()}
else:
result.overall_patterns = summarize_patterns(episodes, messages)
verdict = analysis.get("verdict")
if isinstance(verdict, dict) and verdict:
result.session_verdict = _clean_verdict(verdict)
result.engine = produced.model_id
result.model_notes.append(produced.note)
def analyze_trace_file(
path: str | Path,
*,
include_user_context: bool = True,
redact_secrets: bool = True,
ignore_tool_calls: bool = True,
report_style: str = "field_notes",
analysis_engine: str = "deterministic",
) -> tuple[AnalysisResult, str]:
"""Parse, optionally redact, and analyze an uploaded trace file."""
result: AnalysisResult | None = None
narrative_text = ""
messages: list[NarrativeMessage] = []
for kind, payload in stream_deterministic_analysis(
path,
include_user_context=include_user_context,
redact_secrets=redact_secrets,
ignore_tool_calls=ignore_tool_calls,
):
if kind == "result":
result, narrative_text, messages = payload
assert result is not None
apply_model_analysis(result, messages, analysis_engine)
return result, narrative_text
def derive_trace_title(path: str | Path, agent_type: str) -> str:
stem = Path(path).stem if path else "uploaded trace"
readable_agent = {
"codex": "Codex",
"claude_code": "Claude Code",
"pi": "Pi Agent",
"unknown": "Agent",
}.get(agent_type, "Agent")
return f"{readable_agent} trace: {stem}"
def identify_episodes(messages: list[NarrativeMessage]) -> list[DifficultyEpisode]:
assistant_indexes = [message.index for message in messages if message.role == "assistant"]
if not assistant_indexes:
return []
candidate_spans: list[tuple[int, int]] = []
for index, message in enumerate(messages):
if message.role != "assistant":
continue
score = signal_score(message.text, DIFFICULTY_SIGNALS)
score += 1 if signal_score(message.text, SHIFT_SIGNALS) else 0
if score < 2:
continue
start = previous_assistant_index(messages, index, max_distance=2)
end = next_episode_end(messages, index, max_distance=3)
candidate_spans.append((start, end))
if not candidate_spans:
return []
merged_spans = merge_spans(candidate_spans)
episodes: list[DifficultyEpisode] = []
for episode_number, (start, end) in enumerate(merged_spans[:12], start=1):
span_messages = [
message
for message in messages
if start <= message.index <= end and message.role == "assistant"
]
if not span_messages:
continue
episodes.append(build_episode(episode_number, start, end, span_messages))
return episodes
def previous_assistant_index(
messages: list[NarrativeMessage],
index: int,
*,
max_distance: int,
) -> int:
start = messages[index].index
for position in range(index - 1, max(-1, index - max_distance - 1), -1):
if messages[position].role == "assistant" and signal_score(messages[position].text, INTENTION_SIGNALS):
start = messages[position].index
break
return start
def next_episode_end(
messages: list[NarrativeMessage],
index: int,
*,
max_distance: int,
) -> int:
end = messages[index].index
for position in range(index, min(len(messages), index + max_distance + 1)):
if messages[position].role != "assistant":
continue
end = messages[position].index
if position > index and signal_score(messages[position].text, OUTCOME_SIGNALS):
break
return end
def merge_spans(spans: Iterable[tuple[int, int]]) -> list[tuple[int, int]]:
ordered = sorted(spans)
merged: list[tuple[int, int]] = []
for start, end in ordered:
if not merged or start > merged[-1][1] + 1:
merged.append((start, end))
else:
prev_start, prev_end = merged[-1]
merged[-1] = (prev_start, max(prev_end, end))
return merged
def build_episode(
episode_number: int,
start: int,
end: int,
span_messages: list[NarrativeMessage],
) -> DifficultyEpisode:
combined = "\n\n".join(message.text for message in span_messages)
difficulty_sentence = first_difficulty_sentence(combined) or first_sentence(combined)
intention = first_sentence_with(combined, INTENTION_SIGNALS) or first_sentence(combined)
shift_sentence = first_sentence_after_with(
combined,
SHIFT_SIGNALS,
after_sentence=difficulty_sentence,
)
outcome_sentence = last_sentence_with(combined, OUTCOME_SIGNALS)
difficulty_type = classify_difficulty(difficulty_sentence)
if difficulty_type == "unknown":
difficulty_type = classify_difficulty(combined)
appraisal = classify_appraisal(combined)
detour_type = classify_detour(combined)
resolution_mode = classify_resolution(combined)
outcome_claim = classify_outcome(combined)
recovery_pattern = classify_recovery(combined, detour_type, outcome_claim)
productive_detour = classify_productive_detour(detour_type, outcome_claim, recovery_pattern)
title = make_episode_title(difficulty_type, difficulty_sentence)
evidence = compact_quotes([difficulty_sentence, shift_sentence, outcome_sentence])
return DifficultyEpisode(
episode_id=f"E{episode_number:02d}",
title=title,
message_span=MessageSpan(
start_index=start,
end_index=end,
start_time=span_messages[0].timestamp,
end_time=span_messages[-1].timestamp,
duration_label=duration_label(span_messages[0].timestamp, span_messages[-1].timestamp),
),
initial_intention=trim_sentence(intention, max_words=36),
reported_difficulty=trim_sentence(difficulty_sentence, max_words=40),
difficulty_type=difficulty_type,
appraisal=appraisal,
strategy_before=infer_strategy_before(combined),
strategy_after=trim_sentence(shift_sentence or outcome_sentence or "No explicit strategy shift was visible.", max_words=36),
detour_type=detour_type,
resolution_mode=resolution_mode,
recovery_pattern=recovery_pattern,
outcome_claim=outcome_claim,
productive_detour=productive_detour,
evidence_quotes=evidence,
analyst_memo=make_analyst_memo(
difficulty_type,
appraisal,
detour_type,
recovery_pattern,
outcome_claim,
),
)
def signal_score(text: str, signals: set[str]) -> int:
return sum(1 for signal in signals if contains_signal(text, signal))
def contains_signal(text: str, signal: str) -> bool:
"""Match a codebook signal as a token or phrase, never as an arbitrary substring."""
needle = signal.strip().lower()
if not needle:
return False
pattern = re.escape(needle).replace(r"\ ", r"\s+")
if needle[0].isalnum():
pattern = rf"(?<![a-z0-9]){pattern}"
if needle[-1].isalnum():
pattern = rf"{pattern}(?![a-z0-9])"
return re.search(pattern, text.lower()) is not None
def contains_any(text: str, needles: Iterable[str]) -> bool:
return any(contains_signal(text, needle) for needle in needles)
def first_sentence(text: str) -> str:
return split_sentences(text)[0] if split_sentences(text) else ""
def first_sentence_with(text: str, signals: set[str]) -> str:
for sentence in split_sentences(text):
if signal_score(sentence, signals):
return sentence
return ""
def last_sentence_with(text: str, signals: set[str]) -> str:
for sentence in reversed(split_sentences(text)):
if signal_score(sentence, signals):
return sentence
return ""
def first_sentence_after_with(
text: str,
signals: set[str],
*,
after_sentence: str,
) -> str:
sentences = split_sentences(text)
start = 0
if after_sentence in sentences:
start = sentences.index(after_sentence) + 1
for sentence in sentences[start:]:
if signal_score(sentence, signals):
return sentence
return first_sentence_with(text, signals)
def first_difficulty_sentence(text: str) -> str:
signaled = [
sentence
for sentence in split_sentences(text)
if signal_score(sentence, DIFFICULTY_SIGNALS)
]
if not signaled:
return ""
for sentence in signaled:
if not signal_score(sentence, INTENTION_SIGNALS) or contains_any(
sentence,
PROBLEM_EVIDENCE_SIGNALS,
):
return sentence
return signaled[0]
def split_sentences(text: str) -> list[str]:
normalized = re.sub(r"\s+", " ", text).strip()
if not normalized:
return []
parts = re.split(r"(?<=[.!?])\s+|\n+", normalized)
return [part.strip(" -") for part in parts if part.strip(" -")]
def classify_difficulty(text: str) -> str:
lowered = text.lower()
checks = [
("verification_difficulty", ("verify", "verification", "test", "reproduce", "confirmed", "validate", "cannot run", "not able to run")),
("requirement_uncertainty", ("requirement", "spec", "unclear", "ambiguous", "user intent", "not specified", "scope unclear")),
("environment_blocker", ("dependency", "install", "permission", "auth", "network", "timeout", "sandbox", "environment", "ci", "build fail")),
("compatibility_risk", ("regression", "break", "compatibility", "existing behavior", "side effect", "risk", "backward")),
("localization_difficulty", ("where", "locate", "which file", "module", "root cause", "grep", "search", "trace through")),
("architecture_complexity", ("architecture", "dependency", "shared", "coupling", "system structure", "cross-module", "data flow")),
("implementation_difficulty", ("implement", "tricky", "complex", "not sure how", "hard to", "edge case")),
("insufficient_context", ("more context", "missing context", "need context", "cannot inspect", "not enough information")),
("conflicting_assumptions", ("assumption", "expected", "actually", "mismatch", "conflict", "turns out")),
]
return first_matching_code(lowered, checks)
def classify_appraisal(text: str) -> str:
lowered = text.lower()
checks = [
("cannot_reliably_verify", ("cannot verify", "can't verify", "not verified", "need to verify", "cannot run", "unable to run")),
("needs_more_context", ("need more context", "need to inspect", "need more information", "missing context")),
("initial_hypothesis_wrong", ("hypothesis", "assumption", "i thought", "turns out", "actually")),
("risk_is_higher_than_expected", ("risk", "regression", "side effect", "break existing", "higher than expected")),
("scope_too_large", ("too large", "scope", "narrow", "smaller", "limit this")),
("needs_alternative_path", ("alternative", "instead", "different approach", "try another", "workaround")),
("task_boundary_unclear", ("boundary", "unclear", "ambiguous", "not specified")),
("local_fix_possible", ("local", "small patch", "focused", "straightforward", "fix")),
]
return first_matching_code(lowered, checks)
def classify_detour(text: str) -> str:
lowered = text.lower()
checks = [
("premature_closure", ("done", "complete", "fixed", "should work")),
("rollback_or_reversal", ("roll back", "rollback the", "revert", "abandon", "undo")),
("verification_shift", ("verify with", "instead test", "different verification", "check by", "validate by")),
("hypothesis_switch", ("new hypothesis", "different hypothesis", "assumption was", "turns out")),
("workaround", ("workaround", "bypass", "skip the issue", "without fixing", "temporary fix")),
("scope_narrowing", ("narrow", "smaller", "limit", "focus only", "minimal")),
("decomposition", ("decompose", "break down", "split", "step by step")),
("alternative_path", ("alternative", "instead", "switch", "try another", "different approach")),
]
code = first_matching_code(lowered, checks)
if code == "premature_closure" and signal_score(text, DIFFICULTY_SIGNALS) >= 2:
return "premature_closure"
if code == "premature_closure":
return "direct_continuation"
return code if code != "unknown" else "direct_continuation"
def classify_resolution(text: str) -> str:
lowered = text.lower()
checks = [
("explicit_limitation", ("could not", "unable", "limitation", "caveat", "not verified")),
("goal_reduction", ("partial", "partially", "narrow", "smaller scope", "only")),
("structural_change", ("refactor", "architecture", "new module", "extract", "centralize", "schema")),
("defensive_handling", ("guard", "validate", "fallback", "error handling", "defensive", "sanitize")),
("alternative_implementation", ("alternative implementation", "different implementation", "switch to", "instead")),
("problem_reframing", ("reframe", "actually", "not a", "instead of treating")),
("information_gathering", ("inspect", "search", "read", "looked at", "context")),
("minimal_patch", ("small patch", "focused change", "minimal", "fix")),
]
return first_matching_code(lowered, checks)
def classify_outcome(text: str) -> str:
lowered = text.lower()
success_claim = contains_any(
lowered,
("done", "fixed", "implemented", "resolved", "complete", "verified", "passes"),
)
unresolved_evidence = contains_any(
lowered,
(
"still failing",
"still fails",
"skipped",
"skip the issue",
"workaround",
"without fixing",
"should work",
),
)
if success_claim and unresolved_evidence:
return "premature_success_claim"
if contains_any(lowered, ("not resolved", "still failing", "could not", "unable to")):
return "not_resolved"
if contains_any(lowered, ("need to verify", "needs verification", "not verified", "cannot verify", "can't verify")):
return "needs_verification"
if contains_any(lowered, ("partial", "partially", "some of", "subset")):
return "partially_resolved"
if contains_any(lowered, ("caveat", "assuming", "should", "likely", "not run")) and contains_any(
lowered,
("done", "fixed", "implemented", "resolved", "complete"),
):
return "resolved_with_caveat"
if success_claim:
if signal_score(text, DIFFICULTY_SIGNALS) >= 3 and "verified" not in lowered and "passes" not in lowered:
return "premature_success_claim"
return "resolved_with_confidence"
if contains_any(lowered, ("uncertain", "not sure", "proceed")):
return "uncertain_but_proceeding"
return "unknown"
def classify_recovery(text: str, detour_type: str, outcome_claim: str) -> str:
lowered = text.lower()
if outcome_claim in {"not_resolved", "needs_verification"}:
return "failed_recovery" if outcome_claim == "not_resolved" else "partial_recovery"
if outcome_claim == "premature_success_claim":
return "overconfident_recovery"
if any(token in lowered for token in ("assumption", "turns out", "actually", "hypothesis")):
return "reflective_recovery"
if detour_type in {"alternative_path", "workaround", "scope_narrowing", "verification_shift"}:
return "detour_recovery"
if any(token in lowered for token in ("retry", "again", "iterate", "second", "another attempt")):
return "iterative_recovery"
if outcome_claim in {"resolved_with_confidence", "resolved_with_caveat"}:
return "smooth_recovery"
return "unknown"
def classify_productive_detour(detour_type: str, outcome_claim: str, recovery_pattern: str) -> str:
if detour_type in {"direct_continuation", "unknown"}:
return "unknown"
if recovery_pattern in {"overconfident_recovery", "failed_recovery", "avoidant_recovery"}:
return "no"
if outcome_claim in {"partially_resolved", "needs_verification", "resolved_with_caveat"}:
return "mixed"
return "yes"
def first_matching_code(lowered_text: str, checks: list[tuple[str, tuple[str, ...]]]) -> str:
for code, needles in checks:
if contains_any(lowered_text, needles):
return code
return "unknown"
def infer_strategy_before(text: str) -> str:
sentence = first_sentence_with(text, INTENTION_SIGNALS)
if sentence:
return trim_sentence(sentence, max_words=36)
return "The agent appears to continue from the prior task context."
def make_episode_title(difficulty_type: str, sentence: str) -> str:
label = difficulty_type.replace("_", " ").title()
topic = trim_sentence(sentence, max_words=8)
return f"{label}: {topic}" if topic else label
def compact_quotes(sentences: list[str | None]) -> list[str]:
quotes: list[str] = []
seen: set[str] = set()
for sentence in sentences:
if not sentence:
continue
quote = trim_sentence(sentence, max_words=30)
if quote and quote not in seen:
quotes.append(quote)
seen.add(quote)
return quotes[:3]
def trim_sentence(text: str, *, max_words: int) -> str:
words = re.sub(r"\s+", " ", text or "").strip().split()
if len(words) <= max_words:
return " ".join(words)
return " ".join(words[:max_words]).rstrip(",.;:") + "..."
def make_analyst_memo(
difficulty_type: str,
appraisal: str,
detour_type: str,
recovery_pattern: str,
outcome_claim: str,
) -> str:
return (
f"The visible narrative frames this as {difficulty_type.replace('_', ' ')}; "
f"the appraisal is {appraisal.replace('_', ' ')}, with {detour_type.replace('_', ' ')} "
f"and {recovery_pattern.replace('_', ' ')}. The outcome claim reads as "
f"{outcome_claim.replace('_', ' ')}."
)
def summarize_patterns(
episodes: list[DifficultyEpisode],
messages: list[NarrativeMessage],
) -> dict[str, str]:
if not episodes:
return {
"difficulty_style": "No explicit difficulty episode was detected in the visible assistant narrative.",
"detour_style": "No strategy shift or detour was visible enough to classify.",
"recovery_style": "No recovery pattern can be inferred from the available narrative.",
"risk_or_caveat": "The analyzer only inspects visible narrative messages, so absence of evidence is not proof that the session was difficulty-free.",
}
difficulty_counts = Counter(episode.difficulty_type for episode in episodes)
detour_counts = Counter(episode.detour_type for episode in episodes)
recovery_counts = Counter(episode.recovery_pattern for episode in episodes)
outcome_counts = Counter(episode.outcome_claim for episode in episodes)
primary_difficulty = readable_count_summary(difficulty_counts)
primary_detour = readable_count_summary(detour_counts)
primary_recovery = readable_count_summary(recovery_counts)
risky = [
episode.episode_id
for episode in episodes
if episode.outcome_claim in {"needs_verification", "premature_success_claim", "not_resolved"}
]
caveat = (
f"Watch {', '.join(risky)}: these episodes end with unresolved, unverifiable, or overconfident claims."
if risky
else f"Outcome claims lean toward {readable_count_summary(outcome_counts)}."
)
return {
"difficulty_style": f"Main difficulty pattern: {primary_difficulty}.",
"detour_style": f"Main detour pattern: {primary_detour}.",
"recovery_style": f"Main recovery pattern: {primary_recovery}.",
"risk_or_caveat": caveat,
}
def readable_count_summary(counter: Counter[str]) -> str:
if not counter:
return "unknown"
return ", ".join(f"{code.replace('_', ' ')} ({count})" for code, count in counter.most_common(3))
def duration_label(start_time: str | None, end_time: str | None) -> str:
if not start_time or not end_time:
return "unknown"
start = parse_timestamp(start_time)
end = parse_timestamp(end_time)
if not start or not end or end < start:
return "unknown"
seconds = int((end - start).total_seconds())
if seconds < 60:
return f"{seconds}s"
minutes, secs = divmod(seconds, 60)
if minutes < 60:
return f"{minutes}m {secs}s"
hours, minutes = divmod(minutes, 60)
return f"{hours}h {minutes}m"
def parse_timestamp(value: str) -> datetime | None:
text = value.strip()
if text.endswith("Z"):
text = text[:-1] + "+00:00"
try:
parsed = datetime.fromisoformat(text)
except ValueError:
return None
if parsed.tzinfo is None:
return parsed.replace(tzinfo=timezone.utc)
return parsed
def render_redacted_narrative(messages: list[NarrativeMessage]) -> str:
blocks: list[str] = []
for message in messages:
timestamp = f" [{message.timestamp}]" if message.timestamp else ""
blocks.append(f"## {message.index:04d} {message.role}{timestamp}\n\n{message.text}")
return "\n\n".join(blocks).strip() + "\n"
|