okstra 0.186.5 → 0.186.6

package/runtime/validators/validate-run.py

@@ -80,7 +80,9 @@ from okstra_ctl.report_translation import (  # noqa: E402
 )
 from okstra_ctl.stage_citations import enumerated_stage_numbers  # noqa: E402
 from okstra_ctl.plan_items import (  # noqa: E402
+    CRITIC_WORKER_ID,
     advisory_plan_body_gating,
+    is_critic_worker,
     stage_scope_bucket as _item_stage_scope_bucket,
 )
 from okstra_ctl.incremental_scope import (  # noqa: E402
@@ -92,6 +94,7 @@ from okstra_ctl.clarification_items import (  # noqa: E402
     APPROVAL_BLOCKS,
     PROCEEDING_DISPOSITIONS,
     clarification_disposition,
+    incorporated_clarification_ids,
     progress_blocking_ids,
     row_blocks_progress,
 )
@@ -512,7 +515,7 @@ def _validate_agent_dispatch_contract(
 
     links = [
         row for row in (team_state.get("agentResultLinks") or [])
-        if isinstance(row, Mapping)
+        if isinstance(row, Mapping) and not row.get("supersededBy")
     ]
     paths: dict[str, str] = {}
     dispatch_link_counts: dict[str, int] = {}
@@ -3922,6 +3925,34 @@ def _single_vote_block_survives(item: dict, kinds: set[str]) -> bool:
     )
 
 
+def _critic_non_error_verdicts(item: dict) -> list[dict]:
+    return [
+        row
+        for row in (item.get("verdicts") or [])
+        if isinstance(row, dict)
+        and is_critic_worker(str(row.get("worker") or ""))
+        and str(row.get("verdict") or "").strip().upper()
+        not in ("", "VERIFICATION-ERROR")
+    ]
+
+
+def _tie_gate_class(item: dict, agree: list, disagree: list) -> str | None:
+    """분석자 동수면 critic 이 가르고, 없으면 재검증. 동수가 아니면 None."""
+    if not (len(disagree) == len(agree) and disagree):
+        return None
+    critic = _critic_non_error_verdicts(item)
+    if not critic:
+        return "needs-reverify"
+    if any(
+        str(row.get("verdict") or "").strip().upper() == "DISAGREE"
+        and str(row.get("breakageKind") or "").strip().lower()
+        not in _ADVISORY_ONLY_KINDS
+        for row in critic
+    ):
+        return "majority-disagree"
+    return "has-dissent"
+
+
 def _classify_plan_item_gate(item: dict) -> str:
     """Recompute one plan item's gate class from its per-worker verdicts,
     per `prompts/lead/plan-body-verification.md` "Round protocol". Returns one of
@@ -3930,7 +3961,8 @@ def _classify_plan_item_gate(item: dict) -> str:
     (``dissent-isolated`` / ``partial-consensus`` on ``b``/``c``/``e``) is
     ``majority-disagree`` so the user gate sees it. ``has-dissent`` remains
     advisory-only, rollback items, and a single-vote kind that lost its
-    reproduction.
+    reproduction. An analyser 1-1 is ``needs-reverify`` until ``critic-worker``
+    settles it.
     """
     tokens = [
         (
@@ -3939,6 +3971,7 @@ def _classify_plan_item_gate(item: dict) -> str:
         )
         for v in (item.get("verdicts") or [])
         if isinstance(v, dict)
+        and not is_critic_worker(str(v.get("worker") or ""))
     ]
     non_error = [(vd, bk) for (vd, bk) in tokens if vd and vd != "VERIFICATION-ERROR"]
     if not non_error:
@@ -3988,19 +4021,9 @@ def _classify_plan_item_gate(item: dict) -> str:
     # made the gate stricter than a healthy roster would.
     if len(non_error) >= 2 and len(blocking_disagree) > len(agree):
         return "majority-disagree"
-    # A tie is not consensus, and until now it read as one. The majority test is
-    # strict, so an even panel splitting 1-AGREE / 1-DISAGREE on a blocking kind
-    # fell through to `has-dissent` and the gate passed — the dissent recorded
-    # and never acted on. An even panel is not only the two-analyser roster: one
-    # UNVERIFIABLE or one lost dispatch turns any roster even for that item.
-    # Send the split back for a round; if it survives a round that judged the
-    # rewritten text, nothing further is going to settle it and the user decides.
-    # `_validate_unresolved_tie_was_reverified` is what makes the first branch
-    # more than a label — `needs-reverify` folds into `passed-with-dissent`.
-    if len(non_error) >= 2 and len(blocking_disagree) == len(agree):
-        if _max_verdict_round(item) >= _TIE_SETTLED_ROUND:
-            return "majority-disagree"
-        return "needs-reverify"
+    settled = _tie_gate_class(item, agree, blocking_disagree)
+    if settled is not None and len(non_error) >= 2:
+        return settled
     if (
         len(non_error) >= 2
         and blocking_disagree
@@ -4016,12 +4039,6 @@ def _classify_plan_item_gate(item: dict) -> str:
     return "has-dissent"
 
 
-# 동수를 한 번 재검증한 뒤에도 갈리면 그때는 사용자가 판단한다. 초기 검증이
-# 라운드 1이고 자가수정 뒤의 표적 재검증이 라운드 2이므로, 라운드 2 이상의
-# 판정이 붙은 동수는 이미 한 번 돌아온 것이다.
-_TIE_SETTLED_ROUND = 2
-
-
 def _max_verdict_round(item: dict) -> int:
     """이 항목의 판정이 붙은 가장 늦은 라운드. 스탬프가 없으면 1.
 
@@ -4038,18 +4055,25 @@ def _max_verdict_round(item: dict) -> int:
     return max(rounds, default=1)
 
 
+def _is_even_analyser_split(item: dict) -> bool:
+    tokens = [
+        str(row.get("verdict") or "").strip().upper()
+        for row in (item.get("verdicts") or [])
+        if isinstance(row, dict)
+        and not is_critic_worker(str(row.get("worker") or ""))
+        and str(row.get("verdict") or "").strip().upper()
+        not in ("", "VERIFICATION-ERROR")
+    ]
+    if len(tokens) < 2:
+        return False
+    disagree = sum(1 for token in tokens if token == "DISAGREE")
+    agree = sum(1 for token in tokens if token in {"AGREE", "SUPPLEMENT"})
+    return disagree == agree and disagree > 0
+
+
 def _is_unsettled_tie(item: dict) -> bool:
-    """아직 재검증되지 않은 동수 항목."""
-    return (
-        _classify_plan_item_gate(item) == "needs-reverify"
-        and _max_verdict_round(item) < _TIE_SETTLED_ROUND
-        and len([
-            verdict for verdict in (item.get("verdicts") or [])
-            if isinstance(verdict, dict)
-            and str(verdict.get("verdict") or "").strip().upper()
-            not in ("", "VERIFICATION-ERROR")
-        ]) >= 2
-    )
+    """분석자는 갈렸고 critic 표가 아직 없는 동수 항목."""
+    return _is_even_analyser_split(item) and not _critic_non_error_verdicts(item)
 
 
 def _disagree_breakage_kinds(item: dict) -> set[str]:
@@ -5353,6 +5377,7 @@ def _validate_approval_context(
     activities = _approval_activities_by_id(data)
     activity_timestamps = _canonical_activity_timestamps(run_manifest, report_path)
     report_approved = (data.get("frontmatter") or {}).get("approved") is True
+    incorporated = incorporated_clarification_ids(data)
     for row in data.get("clarificationItems") or []:
         if not isinstance(row, dict) or row.get("blocks") != "approval":
             continue
@@ -5434,7 +5459,9 @@ def _validate_approval_context(
             failures,
         )
         if report_approved and row_blocks_progress(
-            str(row.get("status") or ""), clarification_disposition(row)
+            str(row.get("status") or ""),
+            clarification_disposition(row),
+            incorporated=row_id in incorporated,
         ):
             failures.append(
                 f"final-report data.json: approval is true while clarification `{row_id}` "
@@ -5498,6 +5525,7 @@ def _validate_v3_approval_context(data: dict, failures: list[str]) -> None:
     activities = _approval_activities_by_id(data)
     _validate_v3_plan_backlinks(data, activities, failures)
     approved = (data.get("frontmatter") or {}).get("approved") is True
+    incorporated = incorporated_clarification_ids(data)
     for row in data.get("clarificationItems") or []:
         if not isinstance(row, dict) or row.get("blocks") != "approval":
             continue
@@ -5511,8 +5539,11 @@ def _validate_v3_approval_context(data: dict, failures: list[str]) -> None:
             row, context, failures, schema_version="3.0"
         )
         _validate_v3_resolution_links(row, activities, failures)
+        row_id = str(row.get("id") or "")
         if approved and row_blocks_progress(
-            str(row.get("status") or ""), clarification_disposition(row)
+            str(row.get("status") or ""),
+            clarification_disposition(row),
+            incorporated=row_id in incorporated,
         ):
             failures.append(
                 f"final-report data.json: approval is true while clarification "
@@ -5528,12 +5559,8 @@ def _validate_activity_contract_plan_limits(
     if not _is_activity_contract_v1_planning(run_manifest):
         return
     pbv = (data.get("implementationPlanning") or {}).get("planBodyVerification") or {}
-    rounds_applied = pbv.get("selfFixRoundsApplied", 0)
-    if isinstance(rounds_applied, int) and rounds_applied > 1:
-        failures.append(
-            "final-report data.json: activity contract v1 selfFixRoundsApplied "
-            "must be at most one automatic self-fix round"
-        )
+    # 리포트 칸은 이어진 런의 누적이다. 자동 자가수정 1회 상한은 이번 창의
+    # 상태 파일이 세고, 세션 적합성이 그 횟수와 `self-fix-applied` 를 맞춘다.
     if pbv.get("selfFixStopReason") == "cause-group-recurrence":
         failures.append(
             "final-report data.json: activity contract v1 cannot newly emit "
@@ -6287,7 +6314,11 @@ def _next_step_texts(steps: object) -> list[str]:
 
 def _has_unresolved_approval_blocker(data: dict) -> bool:
     return bool(
-        progress_blocking_ids(data.get("clarificationItems"), APPROVAL_BLOCKS)
+        progress_blocking_ids(
+            data.get("clarificationItems"),
+            APPROVAL_BLOCKS,
+            report_data=data,
+        )
     )
 
 
@@ -7320,21 +7351,12 @@ def _validate_unresolved_tie_was_reverified(
     data: dict,
     failures: list[str],
 ) -> None:
-    """A split panel is sent back once before the gate is declared.
+    """A split panel goes to critic-worker before the gate is declared.
 
     The gate needs a strict majority to block, so a panel splitting evenly on a
     blocking kind reaches neither consensus nor `majority-disagree`. That state
-    is classified `needs-reverify`, and `needs-reverify` folds into
-    `passed-with-dissent` — which is correct for the shape it was built for (a
-    peer that returned nothing) and wrong for this one: nothing failed here, two
-    verifiers read the same plan and disagreed, and passing on that records a
-    dissent nobody acted on.
-
-    So the round is not optional. Re-dispatch those items and record the votes
-    with `--round 2`; a split that survives becomes `majority-disagree` and the
-    user decides. This is satisfiable with the machinery the contract already
-    defines — it is the same targeted re-verification step 7 runs after a
-    self-fix, with the tied items added to that queue.
+    is classified `needs-reverify` until `critic-worker` settles it. Passing
+    without that vote records a dissent nobody acted on.
     """
     ip = data.get("implementationPlanning")
     if not isinstance(ip, dict):
@@ -7353,14 +7375,13 @@ def _validate_unresolved_tie_was_reverified(
     if not unsettled:
         return
     failures.append(
-        f"final-report data.json: plan item(s) {unsettled} carry an even split "
-        "on a blocking breakage kind and were never re-verified. A tie is not "
-        "consensus: the gate's majority test is strict, so this split neither "
-        "blocks nor resolves, and declaring the gate on it passes a dissent "
-        "nobody settled. Re-dispatch those items in a plan-body round and "
-        "record the votes with `okstra plan-items apply-verdicts --data "
-        "<data.json> --verdicts <verdicts.json> --round 2`. A split that "
-        "survives that round becomes `majority-disagree` and goes to the user."
+        "final-report data.json: plan item(s) "
+        f"{unsettled} carry an even split on a blocking breakage kind and "
+        f"have no `{CRITIC_WORKER_ID}` vote. A tie is not consensus. Dispatch "
+        f"`{CRITIC_WORKER_ID}` on those items only (`okstra plan-items "
+        "prepare --tie-vote`) and record the vote with `okstra plan-items "
+        "apply-verdicts --append --round 2`. Critic AGREE settles the split; "
+        "critic DISAGREE blocks."
     )
 
 
@@ -7368,7 +7389,7 @@ def _validate_tie_received_extra_vote(
     data: dict,
     failures: list[str],
 ) -> None:
-    """동수는 같은 둘을 다시 돌리는 것이 아니라 세 번째 표로 가른다."""
+    """동수는 같은 둘을 다시 돌리는 것이 아니라 critic 이 가른다."""
     ip = data.get("implementationPlanning")
     if not isinstance(ip, dict):
         return
@@ -7384,17 +7405,17 @@ def _validate_tie_received_extra_vote(
         and str(item.get("id") or "").strip() not in accepted
         and _stage_scope_bucket(item, pbv) == "in-scope"
         and _is_even_blocking_split(item)
-        and _distinct_verdict_workers(item) < 3
+        and not _critic_non_error_verdicts(item)
     })
     if not missing:
         return
     failures.append(
         f"final-report data.json: plan item(s) {missing} carry an even split "
-        "on a blocking breakage kind and have no third vote. Re-running the "
-        "original two does not settle a 1-1 split. Dispatch one extra analyser "
-        "whose prompt is those items only (`okstra plan-items prepare "
-        "--tie-vote`) and record the vote with `okstra plan-items "
-        "apply-verdicts --append --round 2`."
+        "on a blocking breakage kind and have no critic vote. Re-running the "
+        "original two does not settle a 1-1 split. Dispatch "
+        f"`{CRITIC_WORKER_ID}` whose prompt is those items only "
+        "(`okstra plan-items prepare --tie-vote`) and record the vote with "
+        "`okstra plan-items apply-verdicts --append --round 2`."
     )
 
 
@@ -7411,14 +7432,6 @@ def _is_even_blocking_split(item: dict) -> bool:
     return _is_unsettled_tie(forced)
 
 
-def _distinct_verdict_workers(item: dict) -> int:
-    return len({
-        str(row.get("worker") or "")
-        for row in (item.get("verdicts") or [])
-        if isinstance(row, dict) and str(row.get("worker") or "").strip()
-    })
-
-
 def _validate_advisory_plan_body_gating(data: dict, failures: list[str]) -> None:
     """gating=false 는 검출 표면 0 + 스테이지 1 일 때만 받는다."""
     ip = data.get("implementationPlanning")
@@ -9622,7 +9635,13 @@ def _validate_convergence_rounds_match_manifest(
 def _validate_convergence_states(
     run_dir, failures, run_manifest: dict | None = None, project_root: Path | None = None,
 ) -> None:
-    """이번 런이 가리키는 수렴 상태만 본다. 디렉터리의 옛 seq 파일은 건너뛴다."""
+    """이번 런이 가리키는 수렴 상태만 본다. 디렉터리의 옛 seq 파일은 건너뛴다.
+
+    매니페스트가 경로를 채워도 파일이 없으면 검사하지 않는다. render-only 와
+    workflow 픽스처는 수렴을 돌리지 않아 파일이 없고, 없는 파일을 실패로 치면
+    예전 glob 이 빈 결과를 내던 계약이 깨진다. 목적은 현재 런이 아닌 seq 를
+    보지 않는 것이다.
+    """
     from pathlib import Path as _Path
 
     declared = (run_manifest or {}).get("convergenceStatePath")
@@ -9632,7 +9651,7 @@ def _validate_convergence_states(
             if project_root is None:
                 return
             path = project_root / path
-        paths = [path]
+        paths = [path] if path.is_file() else []
     else:
         state_dir = _Path(run_dir) / "state"
         if not state_dir.is_dir():
@@ -9644,10 +9663,6 @@ def _validate_convergence_states(
         ]
     for state_path in paths:
         if not state_path.is_file():
-            if isinstance(declared, str) and declared.strip():
-                failures.append(
-                    f"convergence state {state_path.name}: missing declared artifact"
-                )
             continue
         try:
             state = json.loads(state_path.read_text(encoding="utf-8"))

package/runtime/validators/validate_session_conformance.py

@@ -47,6 +47,7 @@ from okstra_ctl.lead_events import (  # noqa: E402
 )
 from okstra_ctl.agent_activity import ACTIVITY_FIELDS  # noqa: E402
 from okstra_ctl.domain.host import HostNotRegistered  # noqa: E402
+from okstra_ctl.final_report_paths import final_report_data_path  # noqa: E402
 from okstra_ctl.registry.host_registry import default_host_registry  # noqa: E402
 from okstra_ctl.wrapper_status import read_wrapper_status  # noqa: E402
 
@@ -624,16 +625,7 @@ def _plan_body_state_path(run_dir: Path, suffix: str) -> Path:
 
 def _ids_reported_as_asked(report_path: Path) -> list[str]:
     """리포트가 "사용자에게 물었다"고 기록한 열린 승인 차단 행의 id."""
-    name = report_path.name
-    if not name.endswith(".md"):
-        return []
-    data_path = report_path.with_name(name.removesuffix(".md") + ".data.json")
-    try:
-        doc = json.loads(data_path.read_text())
-    except (OSError, json.JSONDecodeError):
-        return []
-    if not isinstance(doc, dict):
-        return []
+    doc = _read_report_data(report_path)
     asked: list[str] = []
     for row in doc.get("clarificationItems") or []:
         if not isinstance(row, dict):
@@ -666,12 +658,13 @@ def _activity_index(events: list[LeadEvent]) -> dict[str, list[LeadEvent]]:
 
 
 def _read_report_data(report_path: Path) -> Mapping[str, Any]:
-    name = report_path.name
-    if not name.endswith(".md"):
-        return {}
-    data_path = report_path.with_name(name.removesuffix(".md") + ".data.json")
+    """`--report` 가 data.json · markdown · html 이어도 같은 레코드를 연다.
+
+    Phase 7 는 data.json 을 넘긴다. `.md` 만 받던 동안 투영과 self-fix 횟수가
+    빈 객체에서 나와 `projected=<missing>` / `selfFixRoundsApplied=0` 이 됐다.
+    """
     try:
-        data = json.loads(data_path.read_text(encoding="utf-8"))
+        data = json.loads(final_report_data_path(report_path).read_text(encoding="utf-8"))
     except (OSError, json.JSONDecodeError):
         return {}
     return data if isinstance(data, Mapping) else {}
@@ -758,7 +751,16 @@ def _check_projected_agent_activity(
         {field: event.details.get(field) for field in ACTIVITY_FIELDS}
         for event in events
     ]
-    projected = report_data.get("agentActivity")
+    raw_projected = report_data.get("agentActivity")
+    projected = (
+        [
+            {field: row.get(field) for field in ACTIVITY_FIELDS}
+            for row in raw_projected
+            if isinstance(row, Mapping)
+        ]
+        if isinstance(raw_projected, list)
+        else raw_projected
+    )
     if projected == expected:
         return
     mismatch = "length"
@@ -850,6 +852,43 @@ def _allowed_automatic_rounds(self_fix_rounds: int, *, gating: bool = True) -> i
     return 2 + max(self_fix_rounds, 0)
 
 
+def _self_fix_rounds_from_state(run_dir: Path, suffix: str | None) -> int | None:
+    """이번 런 상태 파일의 자가수정 횟수. 없으면 None — 호출자가 리포트로 폴백.
+
+    리포트 칸은 이어진 seq 의 누적이라, 이번 창의 `self-fix-applied` 건수와
+    비교하면 어긋난다.
+    """
+    if not suffix:
+        return None
+    try:
+        doc = json.loads(_plan_body_state_path(run_dir, suffix).read_text())
+    except (OSError, json.JSONDecodeError):
+        return None
+    if not isinstance(doc, dict):
+        return None
+    projection = doc.get("planBodyVerification")
+    value = (
+        projection.get("selfFixRoundsApplied")
+        if isinstance(projection, dict)
+        else None
+    )
+    if isinstance(value, int) and value >= 0:
+        return value
+    value = doc.get("selfFixRoundsApplied")
+    return value if isinstance(value, int) and value >= 0 else None
+
+
+def _is_plan_body_verification_activity(event: LeadEvent) -> bool:
+    """`verification-round-completed` 가 계획 본문 배치인지.
+
+    같은 kind 로 적대 재검증 라운드도 남는다. 그 건을 본문 라운드에 넣으면
+    recorded 가 roundCount 보다 커진다. 요약이 `adversarial reverify` 이면
+    본문이 아니다. 픽스처 요약(`verification-round-completed for …`)은 본문이다.
+    """
+    summary = str(event.details.get("summary") or "").lower()
+    return "adversarial reverify" not in summary and "adversarial re-verify" not in summary
+
+
 def _plan_body_verification(report_data: Mapping[str, Any]) -> Mapping[str, Any] | None:
     planning = report_data.get("implementationPlanning")
     verification = (
@@ -970,14 +1009,23 @@ def _check_activity_round_counts(
     errors: list[str],
 ) -> None:
     verification_rounds = _plan_body_rounds_ran(run_dir, suffix)
-    recorded_verifications = len(indexed.get("verification-round-completed", []))
+    recorded_verifications = len([
+        event
+        for event in indexed.get("verification-round-completed", [])
+        if _is_plan_body_verification_activity(event)
+    ])
     user_reverification_rounds = _resolved_correctness_reverification_rounds(
         indexed,
         report_data,
         verification_rounds,
     )
     automatic_rounds = verification_rounds - len(user_reverification_rounds)
-    self_fix_rounds = _self_fix_rounds_applied(report_data)
+    state_self_fix = _self_fix_rounds_from_state(run_dir, suffix)
+    self_fix_rounds = (
+        state_self_fix
+        if state_self_fix is not None
+        else _self_fix_rounds_applied(report_data)
+    )
     allowed_rounds = _allowed_automatic_rounds(
         self_fix_rounds, gating=_plan_body_gating(report_data),
     )
@@ -1435,6 +1483,11 @@ def _check_cmux_adapter_read(
         return
     if evidence.sidecar_reads.get(CMUX_ADAPTER_BASENAME):
         return
+    dispatch_mode = str(team_state.get("dispatchMode", "")).strip()
+    if dispatch_mode in _WORKER_DISPATCH_MODES:
+        # grok 같은 artifact-only 호스트는 Read 도구 기록이 없다. 워커가
+        # okstra 디스패치로 나갔으면 어댑터가 가리키는 경로를 탄 것이다.
+        return
     errors.append(
         f"cmux adapter: no read of `{CMUX_ADAPTER_BASENAME}` (a `Read` call or a "
         f"shell command naming it) found in the "