feat: require validation cwd and HEAD proof in reviewer reports (Closes #398)

Rebased onto current prgs/master; merged with #396 validation failure
history by keeping both validator rules and workflow handoff fields.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

docs/gitea-dual-namespace-deployment.md

@@ -23,7 +23,7 @@ launched with exactly one static execution profile:
 |-----------------------------|----------------|-------------|
 | `gitea-author` | an author profile | implement issues, push branches, open PRs, comment |
 | `gitea-reviewer` | a reviewer profile | review, approve/request changes, merge |
-| `gitea-reconciler` | a reconciler profile | close already-landed open PRs after ancestry proof (#310) |
+| `gitea-reconciler` | a reconciler profile | close already-landed open PRs after ancestry proof (#304 profile; #310 close tool) |
 
 Properties:
 

docs/gitea-execution-profiles.md

@@ -226,11 +226,12 @@ explicit operator-directed closure of a contaminated PR). If `close_pr` ever
 resolves as unknown, agents must fail closed rather than fall back to the
 edit path.
 
-## Reconciler profile for already-landed open PRs (#310)
+## Reconciler profile for already-landed open PRs (#304 / #310)
 
-Normal author and reviewer profiles must not gain broad PR-close authority.
-Already-landed open PRs (head SHA is an ancestor of the target branch) need a
-dedicated reconciler profile such as `prgs-reconciler` with:
+Normal author and reviewer profiles must not gain broad `gitea.pr.close`
+authority. Already-landed open PRs (head SHA is an ancestor of the target
+branch) need a dedicated reconciler profile such as `prgs-reconciler` with a
+narrow operation set:
 
 - `gitea.read`
 - `gitea.pr.comment`
@@ -238,8 +239,14 @@ dedicated reconciler profile such as `prgs-reconciler` with:
 - `gitea.issue.close`
 - `gitea.pr.close`
 
-Use the `gitea-reconciler` MCP namespace (static profile launch) and the
-`gitea_reconcile_already_landed_pr` tool. The resolver task is
+Forbidden on reconciler profiles: `gitea.pr.approve`, `gitea.pr.merge`,
+`gitea.pr.review`, `gitea.pr.create`, `gitea.branch.push`, and
+`gitea.repo.commit`.
+
+Launch a static `gitea-reconciler` MCP namespace with
+`GITEA_MCP_PROFILE=prgs-reconciler`. Profile shape is validated by
+`reconciler_profile.assess_reconciler_profile` (#304). Use the
+`gitea_reconcile_already_landed_pr` tool (#310). The resolver task is
 `reconcile_already_landed_pr`. PR close is allowed only after live PR fetch,
 fresh target-branch fetch, recorded target SHA, and ancestor proof. PRs whose
 heads are not already landed cannot be closed through this path.

final_report_validator.py

@@ -456,6 +456,40 @@ def _rule_reviewer_git_fetch_readonly(
     return []
 
 
+def _rule_reviewer_validation_failure_history(
+    report_text: str,
+    *,
+    validation_session: dict | None = None,
+) -> list[dict[str, str]]:
+    from reviewer_validation_failure_history import (
+        assess_validation_failure_history_report,
+    )
+
+    session = validation_session or {}
+    observed = session.get("observed_failures") or []
+    if not observed:
+        return []
+
+    result = assess_validation_failure_history_report(
+        report_text,
+        validation_session=session,
+    )
+    if result.get("proven"):
+        return []
+    severity = "block" if result.get("violations") else "downgrade"
+    return [
+        validator_finding(
+            "reviewer.validation_failure_history",
+            severity,
+            "Validation failure history",
+            reason,
+            result.get("safe_next_action")
+            or "document every observed validation failure before claiming pass",
+        )
+        for reason in (result.get("violations") or result.get("reasons") or ["incomplete"])
+    ]
+
+
 def _rule_reviewer_validation_cwd_proof(
     report_text: str,
     *,
@@ -900,6 +934,7 @@ _RULES_BY_TASK: dict[str, list[Callable[..., list[dict[str, str]]]]] = {
         _rule_reviewer_mutation_categories,
         _rule_reviewer_git_fetch_readonly,
         _rule_reviewer_validation_command,
+        _rule_reviewer_validation_failure_history,
         _rule_reviewer_validation_cwd_proof,
         _rule_reviewer_validation_structured,
         _rule_reviewer_linked_issue,

gitea_mcp_server.py

@@ -504,6 +504,7 @@ import already_landed_reconcile  # noqa: E402
 import author_mutation_worktree  # noqa: E402
 import issue_claim_heartbeat  # noqa: E402
 import merged_cleanup_reconcile  # noqa: E402
+import reconciler_profile  # noqa: E402
 import reconciliation_workflow  # noqa: E402
 import review_merge_state_machine  # noqa: E402
 import native_mcp_preference  # noqa: E402
@@ -1244,7 +1245,7 @@ def gitea_create_pr(
     )
     if blocked:
         return blocked
-    verify_preflight_purity(remote)
+    verify_preflight_purity(remote, worktree_path=worktree_path)
     h, o, r = _resolve(remote, host, org, repo)
 
     # ── Issue Lock Validation (Issue #194 / #196) ──
@@ -3463,16 +3464,32 @@ def gitea_delete_branch(
         repo: Override the repository name.
 
     Returns:
-        dict with 'success' and 'message'.
+        dict with 'success' and 'message'; on a permission block,
+        'success'/'performed' False, 'reasons', and a structured
+        'permission_report' (#142) with no API call made.
     """
+    gate_reasons = _profile_operation_gate("gitea.branch.delete")
+    if gate_reasons:
+        return {
+            "success": False,
+            "performed": False,
+            "required_permission": "gitea.branch.delete",
+            "reasons": gate_reasons,
+            "permission_report": _permission_block_report("gitea.branch.delete"),
+        }
+
     verify_preflight_purity(remote)
     h, o, r = _resolve(remote, host, org, repo)
     auth = _auth(h)
     import urllib.parse
     encoded_branch = urllib.parse.quote(branch, safe="")
     url = f"{repo_api_url(h, o, r)}/branches/{encoded_branch}"
+    request_metadata = {
+        "branch": branch,
+        "required_permission": "gitea.branch.delete",
+    }
     with _audited("delete_branch", host=h, remote=remote, org=o, repo=r,
-                  target_branch=branch, request_metadata={"branch": branch}):
+                  target_branch=branch, request_metadata=request_metadata):
         api_request("DELETE", url, auth)
     return {"success": True, "message": f"Remote branch '{branch}' deleted."}
 
@@ -4472,12 +4489,15 @@ def gitea_create_issue_comment(
 def _role_kind(allowed, forbidden) -> str:
     """Classify the active profile from its normalized operations.
 
+    'reconciler' can close already-landed PRs without review/author powers;
     'author' can create PRs / push branches; 'reviewer' can approve/merge;
     'reconciler' can close already-landed PRs via ``gitea.pr.close`` without
     review/author powers; 'mixed' can do both (a config smell — the
     reviewer-deadlock invariant forbids it in v2 configs); 'limited' can do
     neither.
     """
+    if reconciler_profile.is_reconciler_profile(allowed, forbidden):
+        return "reconciler"
     def can(op):
         return gitea_config.check_operation(op, allowed, forbidden)[0]
     review = can("gitea.pr.approve") or can("gitea.pr.merge")
@@ -5646,6 +5666,10 @@ def gitea_get_runtime_context(
         "preflight_block_reasons": preflight["preflight_block_reasons"],
         "preflight_workspace": preflight.get("preflight_workspace"),
         "session_capabilities": session_capabilities,
+        "reconciler_profile_assessment": reconciler_profile.assess_reconciler_profile(
+            allowed, forbidden
+        ),
+        "role_kind": _role_kind(allowed, forbidden),
         "shell_health": native_mcp_preference.shell_health_status(),
     }
 

reconciler_profile.py

@@ -0,0 +1,94 @@
+"""Reconciler profile model for already-landed PR closure (#304).
+
+Defines the narrowly scoped operation set for a dedicated reconciler profile
+such as ``prgs-reconciler``. Close MCP tooling and ancestry gates ship in the
+#310 stack; this module validates profile shape only.
+"""
+
+from __future__ import annotations
+
+import gitea_config
+
+RECONCILER_REQUIRED_OPERATIONS = (
+    "gitea.read",
+    "gitea.pr.close",
+)
+
+RECONCILER_RECOMMENDED_OPERATIONS = (
+    "gitea.pr.comment",
+    "gitea.issue.comment",
+    "gitea.issue.close",
+)
+
+RECONCILER_FORBIDDEN_OPERATIONS = (
+    "gitea.pr.approve",
+    "gitea.pr.merge",
+    "gitea.pr.review",
+    "gitea.pr.create",
+    "gitea.branch.push",
+    "gitea.repo.commit",
+)
+
+
+def _normalized_allowed(allowed: list[str]) -> set[str]:
+    normalized: set[str] = set()
+    for entry in allowed or []:
+        try:
+            normalized.add(gitea_config.normalize_operation(entry))
+        except gitea_config.ConfigError:
+            continue
+    return normalized
+
+
+def _forbidden_in_allowed(allowed: list[str], ops: tuple[str, ...]) -> list[str]:
+    """Return forbidden ops that are explicitly listed in *allowed*."""
+    allowed_n = _normalized_allowed(allowed)
+    present: list[str] = []
+    for op in ops:
+        try:
+            if gitea_config.normalize_operation(op) in allowed_n:
+                present.append(op)
+        except gitea_config.ConfigError:
+            continue
+    return present
+
+
+def is_reconciler_profile(allowed: list[str], forbidden: list[str]) -> bool:
+    """Return True when *allowed*/*forbidden* describe a reconciler profile."""
+    def can(op: str) -> bool:
+        return gitea_config.check_operation(op, allowed, forbidden)[0]
+
+    if not can("gitea.pr.close"):
+        return False
+    if _forbidden_in_allowed(allowed, RECONCILER_FORBIDDEN_OPERATIONS):
+        return False
+    return True
+
+
+def assess_reconciler_profile(allowed: list[str], forbidden: list[str]) -> dict:
+    """Validate reconciler profile operations (read-only, fail closed)."""
+    allowed = list(allowed or [])
+    forbidden = list(forbidden or [])
+    reasons: list[str] = []
+
+    for op in RECONCILER_REQUIRED_OPERATIONS:
+        ok, _ = gitea_config.check_operation(op, allowed, forbidden)
+        if not ok:
+            reasons.append(f"missing required operation {op}")
+
+    for op in _forbidden_in_allowed(allowed, RECONCILER_FORBIDDEN_OPERATIONS):
+        reasons.append(f"forbidden operation must not be allowed: {op}")
+
+    missing_recommended = [
+        op for op in RECONCILER_RECOMMENDED_OPERATIONS
+        if not gitea_config.check_operation(op, allowed, forbidden)[0]
+    ]
+
+    return {
+        "is_reconciler_profile": is_reconciler_profile(allowed, forbidden),
+        "valid": not reasons and is_reconciler_profile(allowed, forbidden),
+        "reasons": reasons,
+        "missing_recommended_operations": missing_recommended,
+        "required_operations": list(RECONCILER_REQUIRED_OPERATIONS),
+        "forbidden_operations": list(RECONCILER_FORBIDDEN_OPERATIONS),
+    }

review_proofs.py

@@ -1735,6 +1735,11 @@ def build_final_report(checkout_proof, inventory, validation, contamination,
         if report_text
         else {"proven": True, "block": False, "reasons": []}
     )
+    proof_backed_handoff = (
+        assess_proof_backed_handoff_report(report_text)
+        if report_text and _PROOF_BACKED_HANDOFF_HINT.search(report_text)
+        else {"proven": True, "block": False, "reasons": []}
+    )
     if baseline_validation is None and report_text:
         baseline_validation = assess_reviewer_baseline_validation_proof(
             report_text=report_text,
@@ -1937,6 +1942,11 @@ def build_final_report(checkout_proof, inventory, validation, contamination,
             "reviewer baseline validation proof missing or failed (#325)"
         )
         downgrade_reasons.extend(baseline_validation.get("reasons", []))
+    if not proof_backed_handoff.get("proven"):
+        downgrade_reasons.append(
+            "proof-sensitive handoff claims lack command/tool evidence (#395)"
+        )
+        downgrade_reasons.extend(proof_backed_handoff.get("reasons", []))
     if not already_landed_classification.get("proven"):
         downgrade_reasons.append(
             "already-landed classification wording missing or invalid (#295)"
@@ -5088,6 +5098,12 @@ _PR_QUEUE_CLEANUP_REPORT_HINT = re.compile(
     re.I,
 )
 
+_PROOF_BACKED_HANDOFF_HINT = re.compile(
+    r"task:\s*review-merge-pr|task mode:\s*review-merge-pr|"
+    r"review-merge-pr|controller handoff",
+    re.I,
+)
+
 
 def assess_pr_queue_cleanup_report(report_text: str | None) -> dict:
     """#390: validate PR-only cleanup final reports (one PR per run)."""
@@ -5487,6 +5503,15 @@ def assess_validation_integrity_report(report_text, **kwargs):
     return _assess(report_text, **kwargs)
 
 
+def assess_validation_failure_history_report(report_text, **kwargs):
+    """#396: final reports must account for transient validation failures."""
+    from reviewer_validation_failure_history import (
+        assess_validation_failure_history_report as _assess,
+    )
+
+    return _assess(report_text, **kwargs)
+
+
 def assess_validation_cwd_proof_report(report_text, **kwargs):
     """#398: validation commands require explicit worktree cwd and HEAD proof."""
     from reviewer_validation_cwd_proof import assess_validation_cwd_proof_report as _assess
@@ -5573,3 +5598,10 @@ def assess_worktree_ownership_report(report_text, **kwargs):
     from reviewer_worktree_ownership import assess_worktree_ownership_report as _assess
 
     return _assess(report_text, **kwargs)
+
+
+def assess_proof_backed_handoff_report(report_text, **kwargs):
+    """#395: proof-sensitive review handoff claims require explicit evidence."""
+    from reviewer_proof_backed_handoff import assess_proof_backed_handoff_report as _assess
+
+    return _assess(report_text, **kwargs)

reviewer_proof_backed_handoff.py

@@ -0,0 +1,217 @@
+"""Proof-backed reviewer handoff claim verifier (#395)."""
+
+from __future__ import annotations
+
+import re
+from typing import Any
+
+_PAGINATION_FINALITY = re.compile(
+    r"has_more\s*:\s*false|is_final_page\s*:\s*true|"
+    r"inventory_complete\s*:\s*true|pages_fetched|total_count\s*:",
+    re.I,
+)
+_PAGE_SIZE_ONLY = re.compile(
+    r"(?:less|fewer) than (?:the )?(?:default )?page[- ]?size|"
+    r"under (?:the )?50[- ]?(?:item )?limit|"
+    r"returned \d+ (?:open )?prs?.*less than 50",
+    re.I,
+)
+_INVENTORY_COMPLETE_CLAIM = re.compile(
+    r"\b(?:inventory (?:is )?complete|inventory exhaustive|"
+    r"complete (?:pr )?inventory)\b",
+    re.I,
+)
+_SKIP_CLAIM = re.compile(
+    r"\b(?:earlier prs? skipped|skipped (?:earlier )?pr|"
+    r"skip(?:ped)? pr #?\d+|non-mergeable|prior request.changes)\b",
+    re.I,
+)
+_CONFLICT_PROOF = re.compile(
+    r"merge simulation|git merge --no-commit|conflicting files|"
+    r"conflict proof|merge_exit|non-mergeable",
+    re.I,
+)
+_BASELINE_CLAIM = re.compile(
+    r"\b(?:baseline (?:validation|worktree|comparison)|"
+    r"same as master|pre-existing (?:on )?master|"
+    r"failure signatures match)\b",
+    re.I,
+)
+_BASELINE_PATH = re.compile(r"baseline worktree path\s*:\s*(\S+)", re.I)
+_BASELINE_SHA = re.compile(r"baseline (?:target )?sha\s*:\s*([0-9a-f]{7,40})", re.I)
+_BASELINE_DIRTY_BEFORE = re.compile(
+    r"baseline.*dirty before|dirty before.*baseline", re.I
+)
+_BASELINE_DIRTY_AFTER = re.compile(
+    r"baseline.*dirty after|dirty after.*baseline", re.I
+)
+_BASELINE_COMMAND = re.compile(
+    r"baseline.*(?:validation )?command|pytest.*baseline", re.I
+)
+_BASELINE_RESULT = re.compile(
+    r"baseline.*(?:validation )?result|baseline_exit|baseline failures", re.I
+)
+_MASTER_INTEGRATION = re.compile(
+    r"\b(?:merged master into|merge(?:d)? (?:remote[- ]tracking )?branch.*master|"
+    r"master integration|integrated master|rebase.*master)\b",
+    re.I,
+)
+_CLEANUP_CLAIM = re.compile(
+    r"\b(?:worktree(?:s)? (?:were )?cleaned|cleanup (?:result|mutations)|"
+    r"removed (?:session[- ]owned )?worktree|git worktree remove)\b",
+    re.I,
+)
+_WORKTREE_LIST = re.compile(r"git worktree list|worktree list proof", re.I)
+_PROOF_SOURCE = re.compile(
+    r"proof source\s*:\s*(command|mcp metadata|prior blocker|not checked)",
+    re.I,
+)
+_HANDOFF_SECTION = re.compile(r"^##\s*Controller Handoff\s*$", re.I | re.M)
+
+
+def _handoff_fields(report_text: str) -> dict[str, str]:
+    text = report_text or ""
+    match = _HANDOFF_SECTION.search(text)
+    if not match:
+        return {}
+    fields: dict[str, str] = {}
+    for line in text[match.end() :].splitlines():
+        stripped = line.strip().lstrip("-*").strip()
+        if ":" not in stripped:
+            continue
+        key, value = stripped.split(":", 1)
+        fields[key.strip().lower()] = value.strip()
+    return fields
+
+
+def _command_text(entry: Any) -> str:
+    if isinstance(entry, dict):
+        return str(entry.get("command") or entry.get("tool") or "").strip()
+    return str(entry or "").strip()
+
+
+def _action_log_has(action_log: list | None, pattern: re.Pattern[str]) -> bool:
+    for entry in action_log or []:
+        blob = " ".join(
+            filter(
+                None,
+                [
+                    _command_text(entry),
+                    str(entry.get("result") or "") if isinstance(entry, dict) else "",
+                    str(entry.get("reason") or "") if isinstance(entry, dict) else "",
+                ],
+            )
+        )
+        if pattern.search(blob):
+            return True
+    return False
+
+
+def assess_proof_backed_handoff_report(
+    report_text: str,
+    *,
+    action_log: list | None = None,
+    inventory_session: dict | None = None,
+    baseline_session: dict | None = None,
+    skip_session: list[dict] | None = None,
+) -> dict[str, Any]:
+    """Require explicit command/tool evidence for proof-sensitive review claims (#395)."""
+    text = report_text or ""
+    fields = _handoff_fields(text)
+    reasons: list[str] = []
+    inventory = dict(inventory_session or {})
+    baseline = dict(baseline_session or {})
+    skips = list(skip_session or [])
+
+    pagination_field = fields.get("inventory pagination proof", "")
+    if _INVENTORY_COMPLETE_CLAIM.search(text) or _PAGE_SIZE_ONLY.search(text):
+        has_meta = bool(
+            inventory.get("inventory_complete")
+            or inventory.get("pagination_complete")
+            or _PAGINATION_FINALITY.search(pagination_field)
+            or _PAGINATION_FINALITY.search(text)
+            or _action_log_has(action_log, re.compile(r"gitea_list_prs", re.I))
+        )
+        if _PAGE_SIZE_ONLY.search(text) and not has_meta:
+            reasons.append(
+                "inventory completeness claimed from page-size assumption only; "
+                "require has_more=false, is_final_page=true, or inventory_complete=true"
+            )
+        elif _INVENTORY_COMPLETE_CLAIM.search(text) and not has_meta:
+            reasons.append(
+                "inventory complete claim lacks explicit pagination metadata proof"
+            )
+
+    if _SKIP_CLAIM.search(text) or fields.get("earlier prs skipped", "").lower() not in {
+        "",
+        "none",
+        "n/a",
+    }:
+        skip_proof = bool(
+            skips
+            or _CONFLICT_PROOF.search(text)
+            or _action_log_has(
+                action_log, re.compile(r"git merge --no-commit|gitea_get_pr_review_feedback", re.I)
+            )
+        )
+        if not skip_proof:
+            reasons.append(
+                "earlier PR skip claim lacks command/tool conflict or blocker proof"
+            )
+
+    if _BASELINE_CLAIM.search(text) or fields.get("baseline worktree used", "").lower() == "true":
+        baseline_ok = bool(
+            baseline.get("complete")
+            or (
+                _BASELINE_PATH.search(text)
+                and _BASELINE_SHA.search(text)
+                and (_BASELINE_DIRTY_BEFORE.search(text) or baseline.get("clean_before"))
+                and (_BASELINE_COMMAND.search(text) or baseline.get("command"))
+                and (_BASELINE_RESULT.search(text) or baseline.get("result"))
+                and (_BASELINE_DIRTY_AFTER.search(text) or baseline.get("clean_after"))
+            )
+        )
+        if not baseline_ok:
+            reasons.append(
+                "baseline validation claim missing worktree path, target SHA, "
+                "dirty-before/after status, command, or result proof"
+            )
+
+    if _MASTER_INTEGRATION.search(text):
+        if not _action_log_has(
+            action_log,
+            re.compile(r"git merge.*master|git rebase.*master", re.I),
+        ) and not re.search(r"merge_exit\s*=\s*\d+|merge simulation", text, re.I):
+            reasons.append(
+                "master integration claim lacks exact merge/rebase command and result"
+            )
+
+    if _CLEANUP_CLAIM.search(text) or fields.get("cleanup mutations", "").lower() not in {
+        "",
+        "none",
+        "n/a",
+    }:
+        if not (_WORKTREE_LIST.search(text) or _action_log_has(action_log, _WORKTREE_LIST)):
+            reasons.append(
+                "cleanup claim lacks final git worktree list or equivalent proof"
+            )
+
+    if re.search(r"\blive proof\b", text, re.I) and not _PROOF_SOURCE.search(text):
+        if not action_log:
+            reasons.append(
+                "live proof wording requires proof source classification "
+                "(command, MCP metadata, prior blocker, or not checked)"
+            )
+
+    proven = not reasons
+    return {
+        "proven": proven,
+        "block": not proven,
+        "reasons": reasons,
+        "safe_next_action": (
+            "cite explicit command/tool evidence or structured MCP pagination metadata "
+            "for each proof-sensitive claim"
+            if not proven
+            else "proceed"
+        ),
+    }

reviewer_validation_failure_history.py

@@ -0,0 +1,198 @@
+"""Transient validation failure history verifier (#396).
+
+Reviewer sessions may observe validation failures that later pass on rerun.
+Final reports must document every failure observed during the session, not
+only the last passing result.
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Any
+
+_SECTION_RE = re.compile(
+    r"validation failure history",
+    re.IGNORECASE,
+)
+_FAILURE_ENTRY_RE = re.compile(
+    r"(?:failure\s*(?:#|entry)?\s*\d+|validation failure)\s*:",
+    re.IGNORECASE,
+)
+_COMMAND_RE = re.compile(
+    r"(?:command|validation command)\s*:\s*(.+)",
+    re.IGNORECASE,
+)
+_FAILING_TEST_RE = re.compile(
+    r"(?:failing test|failure|error)\s*:\s*(.+)",
+    re.IGNORECASE,
+)
+_CAUSE_RE = re.compile(
+    r"(?:suspected cause|cause)\s*:\s*(.+)",
+    re.IGNORECASE,
+)
+_REPRODUCED_RE = re.compile(
+    r"(?:reproduced|reproduces)\s*:\s*(yes|no|unknown)",
+    re.IGNORECASE,
+)
+_BASELINE_RE = re.compile(
+    r"(?:on baseline master|baseline master|exists on baseline)\s*:\s*(yes|no|unknown|not checked)",
+    re.IGNORECASE,
+)
+_PR_CAUSED_RE = re.compile(
+    r"(?:pr[- ]caused|pr caused)\s*:\s*(yes|no|unknown|not proven)",
+    re.IGNORECASE,
+)
+_TRANSIENT_STATUS_RE = re.compile(
+    r"(?:passed after transient failure investigation|transient failure investigation)",
+    re.IGNORECASE,
+)
+_PLAIN_PASS_RE = re.compile(
+    r"validation\s*:\s*(?:pass|passed|strong|ok|green)\b",
+    re.IGNORECASE,
+)
+_ENV_CLEANUP_RE = re.compile(
+    r"(?:environmental cleanup|state cleaned|what changed between runs)\s*:",
+    re.IGNORECASE,
+)
+_UNKNOWN_CAUSE_RE = re.compile(
+    r"(?:suspected cause|cause)\s*:\s*(?:unknown|unexplained|not determined)",
+    re.IGNORECASE,
+)
+
+
+def _failure_documented_in_text(text: str, failure: dict[str, Any]) -> bool:
+    """Return True when *failure* appears documented in free-form report text."""
+    command = (failure.get("command") or "").strip()
+    failing = (failure.get("failing_test") or failure.get("error") or "").strip()
+    if command and command not in text:
+        return False
+    if failing and failing not in text:
+        return False
+    return bool(command or failing)
+
+
+def _section_has_structured_fields(text: str) -> bool:
+    if not _SECTION_RE.search(text):
+        return False
+    section_start = _SECTION_RE.search(text).start()
+    section = text[section_start:]
+    has_command = bool(_COMMAND_RE.search(section))
+    has_failure = bool(_FAILING_TEST_RE.search(section))
+    return has_command and has_failure
+
+
+def assess_validation_failure_history_report(
+    report_text: str,
+    *,
+    validation_session: dict | None = None,
+) -> dict[str, Any]:
+    """Require final reports to account for transient validation failures (#396)."""
+    text = report_text or ""
+    session = dict(validation_session or {})
+    reasons: list[str] = []
+    violations: list[str] = []
+
+    observed = list(session.get("observed_failures") or [])
+    final_status = (session.get("final_validation_status") or "").strip().lower()
+    cause_unknown = any(
+        (f.get("suspected_cause") or "").strip().lower() in {
+            "unknown", "unexplained", "not determined", ""
+        }
+        and not (f.get("pr_caused") or "").strip().lower() in {"yes", "no"}
+        for f in observed
+    ) or bool(session.get("cause_unknown"))
+
+    if not observed:
+        return {
+            "proven": True,
+            "block": False,
+            "reasons": [],
+            "violations": [],
+            "observed_failure_count": 0,
+            "safe_next_action": "proceed",
+        }
+
+    documented = _section_has_structured_fields(text)
+    if not documented:
+        for failure in observed:
+            if _failure_documented_in_text(text, failure):
+                documented = True
+                break
+
+    if not documented:
+        violations.append(
+            "session observed validation failure(s) but final report omits "
+            "Validation failure history"
+        )
+        reasons.append(
+            "every validation failure observed during the session must appear "
+            "in a Validation failure history section"
+        )
+
+    if documented and not _SECTION_RE.search(text):
+        reasons.append(
+            "failure details must appear under an explicit "
+            "'Validation failure history' heading"
+        )
+
+    for idx, failure in enumerate(observed, start=1):
+        prefix = f"failure #{idx}"
+        if not _failure_documented_in_text(text, failure) and documented:
+            reasons.append(
+                f"{prefix}: command and failing test/error not documented in report"
+            )
+        command = (failure.get("command") or "").strip()
+        failing = (failure.get("failing_test") or failure.get("error") or "").strip()
+        if not command:
+            reasons.append(f"{prefix}: missing command in session failure record")
+        if not failing:
+            reasons.append(
+                f"{prefix}: missing failing test or error in session failure record"
+            )
+
+    plain_pass = bool(_PLAIN_PASS_RE.search(text))
+    transient_wording = bool(_TRANSIENT_STATUS_RE.search(text))
+    final_passed = final_status in {
+        "passed",
+        "pass",
+        "passed_after_transient_failure_investigation",
+    }
+
+    if (final_passed or plain_pass) and observed:
+        if cause_unknown and not transient_wording:
+            violations.append(
+                "unknown transient failure cause cannot be erased as plain 'passed'"
+            )
+            reasons.append(
+                "when cause is unknown, use status "
+                "'passed after transient failure investigation'"
+            )
+        elif not transient_wording and final_status != "passed_after_transient_failure_investigation":
+            if not _ENV_CLEANUP_RE.search(text):
+                reasons.append(
+                    "later passing rerun must document what changed between runs "
+                    "or environmental cleanup performed"
+                )
+
+    if session.get("environmental_contamination") and not _ENV_CLEANUP_RE.search(text):
+        reasons.append(
+            "environmental /tmp state contamination must document cleanup or "
+            "what changed before the passing rerun"
+        )
+
+    proven = not reasons and not violations
+    return {
+        "proven": proven,
+        "block": bool(violations) or not proven,
+        "reasons": reasons,
+        "violations": violations,
+        "observed_failure_count": len(observed),
+        "documented": documented,
+        "safe_next_action": (
+            "add Validation failure history with command, failing test, cause, "
+            "reproduction, baseline comparison, and PR-caused evidence; use "
+            "'passed after transient failure investigation' when appropriate"
+            if not proven
+            else "proceed"
+        ),
+    }

skills/llm-project-workflow/schemas/review-merge-final-report.md

@@ -91,4 +91,23 @@ Verifier: `review_proofs.assess_queue_status_report()`.
 
 Narrative final report and controller handoff must agree on eligibility class,
 candidate/reviewed head SHA, mutation state, worktree usage, review decision,
-terminal review mutation, merge result, and linked issue status.
+terminal review mutation, merge result, and linked issue status.
+
+### Proof-backed claims (#395)
+
+Proof-sensitive claims must cite explicit command/tool evidence in the report
+or structured MCP metadata — not narrative alone:
+
+- **Inventory complete:** `has_more=false`, `is_final_page=true`,
+  `inventory_complete=true`, and/or `total_count` from `gitea_list_prs`.
+- **Skipped earlier PRs:** merge-simulation command output, conflict file list,
+  or live `gitea_get_pr_review_feedback` / mergeability fields.
+- **Baseline validation:** baseline worktree path, baseline target SHA,
+  dirty-before/after, exact command, exact result.
+- **Master integration:** exact `git merge` / `git rebase` command and exit status.
+- **Cleanup:** final `git worktree list` (or remove commands) proving session
+  worktrees were removed.
+
+When a claim relies on prior-session blocker state or MCP metadata only, label
+the proof source explicitly (`command`, `MCP metadata`, `prior blocker`,
+`not checked`). Do not use `live proof` without that classification.

skills/llm-project-workflow/workflows/review-merge-pr.md

@@ -538,6 +538,36 @@ Official validation integrity status must be one of:
 
 Do not invent a softer status.
 
+## 21A. Validation failure history rule
+
+Every validation failure observed during the review session must appear in the
+final report, even if a later rerun passes.
+
+Before claiming `Validation: passed`, list every failed validation command from
+the session under `Validation failure history`.
+
+For each failure, report:
+
+* command
+* failing test or error
+* suspected cause
+* whether it reproduced
+* whether it exists on baseline master
+* evidence for whether it is or is not PR-caused
+
+If a later rerun passes, also report:
+
+* what changed between runs
+* whether environmental state was cleaned (for example `/tmp` lock files)
+* why the pass is trustworthy
+
+If the cause is unknown, do not erase the earlier failure with plain
+`Validation: passed`. Use a status such as
+`passed after transient failure investigation`.
+
+`gitea_validate_review_final_report` rejects reports that omit known earlier
+validation failures when `validation_session.observed_failures` is supplied.
+
 ## 22. Baseline validation rule
 
 Do not run tests in the main checkout.
@@ -1118,6 +1148,7 @@ Controller Handoff:
 * Baseline worktree path:
 * Files reviewed:
 * Validation:
+* Validation failure history:
 * Validation cwd/HEAD proof:
 * Official validation integrity status:
 * Terminal review mutation:

tests/test_delete_branch_capability.py

@@ -0,0 +1,217 @@
+"""Tests for gitea_delete_branch capability gate (Issue #408).
+
+``gitea_delete_branch`` requires the exact ``gitea.branch.delete`` operation:
+without it the delete fails closed (no preflight, no auth lookup, no API call,
+structured permission report). With it, deletion proceeds through existing
+preflight and audit unchanged.
+"""
+import json
+import os
+import sys
+import tempfile
+import unittest
+from unittest.mock import patch
+
+sys.path.insert(0, str(__import__("pathlib").Path(__file__).resolve().parent.parent))
+
+import mcp_server
+from mcp_server import gitea_delete_branch
+
+FAKE_AUTH = "token fake"
+
+AUTHOR_NO_DELETE = {
+    "profile_name": "prgs-author",
+    "allowed_operations": [
+        "gitea.read", "gitea.pr.create", "gitea.pr.comment",
+        "gitea.branch.push", "gitea.issue.comment",
+    ],
+    "forbidden_operations": ["gitea.pr.approve", "gitea.pr.merge"],
+    "audit_label": "prgs-author",
+}
+
+AUTHOR_WITH_DELETE = {
+    "profile_name": "prgs-author-deleter",
+    "allowed_operations": AUTHOR_NO_DELETE["allowed_operations"] + [
+        "gitea.branch.delete",
+    ],
+    "forbidden_operations": ["gitea.pr.approve", "gitea.pr.merge"],
+    "audit_label": "prgs-author-deleter",
+}
+
+CONFIG = {
+    "version": 2,
+    "contexts": {
+        "ctx": {
+            "enabled": True,
+            "gitea": {"enabled": True, "base_url": "https://gitea.example.com"},
+        }
+    },
+    "profiles": {
+        "author-no-delete": {
+            "enabled": True,
+            "context": "ctx",
+            "role": "author",
+            "username": "author-user",
+            "auth": {"type": "env", "name": "GITEA_TOKEN_AUTHOR"},
+            "allowed_operations": AUTHOR_NO_DELETE["allowed_operations"],
+            "forbidden_operations": [],
+            "execution_profile": "author-no-delete",
+        },
+        "author-with-delete": {
+            "enabled": True,
+            "context": "ctx",
+            "role": "author",
+            "username": "deleter-user",
+            "auth": {"type": "env", "name": "GITEA_TOKEN_AUTHOR"},
+            "allowed_operations": AUTHOR_WITH_DELETE["allowed_operations"],
+            "forbidden_operations": [],
+            "execution_profile": "author-with-delete",
+        },
+        "reviewer-profile": {
+            "enabled": True,
+            "context": "ctx",
+            "role": "reviewer",
+            "username": "reviewer-user",
+            "auth": {"type": "env", "name": "GITEA_TOKEN_REVIEWER"},
+            "allowed_operations": [
+                "gitea.read", "gitea.pr.review", "gitea.pr.merge",
+            ],
+            "forbidden_operations": [
+                "gitea.branch.delete", "gitea.branch.push", "gitea.pr.create",
+            ],
+            "execution_profile": "reviewer-profile",
+        },
+    },
+    "rules": {"allow_runtime_switching": False},
+}
+
+
+class TestDeleteBranchToolGate(unittest.TestCase):
+
+    def setUp(self):
+        self._preflight_snapshot = (
+            mcp_server._preflight_whoami_called,
+            mcp_server._preflight_capability_called,
+        )
+        mcp_server._preflight_whoami_called = False
+        mcp_server._preflight_capability_called = False
+        self._remotes = patch.dict(mcp_server.REMOTES, {
+            "prgs": {"host": "gitea.example.com", "org": "Example-Org",
+                     "repo": "Example-Repo"},
+        })
+        self._remotes.start()
+        mcp_server._IDENTITY_CACHE.clear()
+        patch("gitea_config.load_config", return_value={}).start()
+        patch(
+            "gitea_config.is_runtime_switching_enabled", return_value=False
+        ).start()
+        patch("gitea_audit.audit_enabled", return_value=False).start()
+        self.mock_api = patch("mcp_server.api_request").start()
+        self.mock_auth = patch(
+            "mcp_server.get_auth_header", return_value=FAKE_AUTH
+        ).start()
+
+    def tearDown(self):
+        patch.stopall()
+        mcp_server._IDENTITY_CACHE.clear()
+        mcp_server._preflight_whoami_called, mcp_server._preflight_capability_called = (
+            self._preflight_snapshot
+        )
+
+    def _set_profile(self, profile):
+        patch("mcp_server.get_profile", return_value=profile).start()
+
+    def test_blocked_without_delete_capability(self):
+        self._set_profile(AUTHOR_NO_DELETE)
+        res = gitea_delete_branch(branch="feat/branch", remote="prgs")
+        self.assertFalse(res["success"])
+        self.assertFalse(res["performed"])
+        self.assertEqual(res["required_permission"], "gitea.branch.delete")
+        self.assertTrue(res["reasons"])
+        self.assertEqual(
+            res["permission_report"]["missing_permission"],
+            "gitea.branch.delete",
+        )
+        self.mock_api.assert_not_called()
+        self.mock_auth.assert_not_called()
+
+    def test_allowed_delete_proceeds(self):
+        self._set_profile(AUTHOR_WITH_DELETE)
+        mcp_server.record_preflight_check("whoami")
+        mcp_server.record_preflight_check("capability", resolved_role="author")
+        res = gitea_delete_branch(branch="feat/branch", remote="prgs")
+        self.assertTrue(res["success"])
+        self.assertIn("deleted", res["message"])
+        delete_calls = [
+            c for c in self.mock_api.call_args_list if c.args[0] == "DELETE"
+        ]
+        self.assertTrue(delete_calls)
+
+    def test_allowed_delete_audited_with_capability_proof(self):
+        self._set_profile(AUTHOR_WITH_DELETE)
+        patch("gitea_audit.audit_enabled", return_value=True).start()
+        mock_write = patch("gitea_audit.write_event").start()
+        mcp_server.record_preflight_check("whoami")
+        mcp_server.record_preflight_check("capability", resolved_role="author")
+        self.mock_api.return_value = {}
+        gitea_delete_branch(branch="feat/branch", remote="prgs")
+        mock_write.assert_called()
+        event = mock_write.call_args[0][0]
+        self.assertEqual(event["action"], "delete_branch")
+        self.assertEqual(
+            event["request_metadata"]["required_permission"],
+            "gitea.branch.delete",
+        )
+
+
+class TestDeleteBranchResolverParity(unittest.TestCase):
+    def setUp(self):
+        self._remotes = patch.dict(mcp_server.REMOTES, {
+            "prgs": {"host": "gitea.example.com", "org": "Example-Org",
+                     "repo": "Example-Repo"},
+        })
+        self._remotes.start()
+        mcp_server._IDENTITY_CACHE.clear()
+        self._dir = tempfile.TemporaryDirectory()
+        self.config_path = os.path.join(self._dir.name, "profiles.json")
+        with open(self.config_path, "w", encoding="utf-8") as fh:
+            fh.write(json.dumps(CONFIG))
+        patch(
+            "gitea_config.is_runtime_switching_enabled", return_value=False
+        ).start()
+        patch("gitea_audit.audit_enabled", return_value=False).start()
+        self.mock_api = patch("mcp_server.api_request").start()
+        patch("mcp_server.get_auth_header", return_value=FAKE_AUTH).start()
+
+    def tearDown(self):
+        patch.stopall()
+        mcp_server._IDENTITY_CACHE.clear()
+        self._dir.cleanup()
+
+    def _env(self, profile: str) -> dict:
+        return {
+            "GITEA_MCP_CONFIG": self.config_path,
+            "GITEA_MCP_PROFILE": profile,
+            "GITEA_TOKEN_AUTHOR": "author-pass",
+            "GITEA_TOKEN_REVIEWER": "reviewer-pass",
+        }
+
+    def test_reviewer_resolver_denial_blocks_raw_tool(self):
+        with patch.dict(os.environ, self._env("reviewer-profile"), clear=True):
+            resolve = mcp_server.gitea_resolve_task_capability(
+                task="delete_branch", remote="prgs")
+            self.assertFalse(resolve["allowed_in_current_session"])
+            res = gitea_delete_branch(branch="feat/branch", remote="prgs")
+        self.assertFalse(res["success"])
+        self.assertEqual(
+            res["permission_report"]["missing_permission"],
+            resolve["required_operation_permission"],
+        )
+        delete_calls = [
+            c for c in self.mock_api.call_args_list if c.args[0] == "DELETE"
+        ]
+        self.assertFalse(delete_calls)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_llm_workflow_split.py

@@ -207,6 +207,12 @@ def test_validation_integrity_verifier_exported():
     assert callable(assess_validation_integrity_report)
 
 
+def test_validation_failure_history_verifier_exported():
+    from review_proofs import assess_validation_failure_history_report
+
+    assert callable(assess_validation_failure_history_report)
+
+
 def test_validation_cwd_proof_verifier_exported():
     from review_proofs import assess_validation_cwd_proof_report
 
@@ -255,6 +261,12 @@ def test_inventory_worktree_verifier_exported():
     assert callable(assess_inventory_worktree_report)
 
 
+def test_proof_backed_handoff_verifier_exported():
+    from review_proofs import assess_proof_backed_handoff_report
+
+    assert callable(assess_proof_backed_handoff_report)
+
+
 def test_infra_stop_handoff_verifier_exported():
     from review_proofs import assess_infra_stop_handoff_report
 

tests/test_mcp_server.py

@@ -1005,9 +1005,19 @@ class TestReviewPR(unittest.TestCase):
 # ---------------------------------------------------------------------------
 class TestDeleteBranch(unittest.TestCase):
 
+    DELETE_PROFILE = {
+        "profile_name": "test-deleter",
+        "allowed_operations": ["gitea.read", "gitea.branch.delete"],
+        "forbidden_operations": [],
+        "audit_label": "test-deleter",
+    }
+
+    @patch("mcp_server.get_profile", return_value=DELETE_PROFILE)
     @patch("mcp_server.api_request")
     @patch("mcp_server.get_auth_header", return_value=FAKE_AUTH)
-    def test_delete_branch(self, _auth, mock_api):
+    def test_delete_branch(self, _auth, mock_api, _profile):
+        mcp_server.record_preflight_check("whoami")
+        mcp_server.record_preflight_check("capability", resolved_role="author")
         mock_api.return_value = {}
         result = gitea_delete_branch(branch="feat/branch")
         self.assertTrue(result["success"])

tests/test_reconciler_profile.py

@@ -0,0 +1,87 @@
+"""Tests for reconciler profile model (#304)."""
+
+import os
+import sys
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import gitea_mcp_server as mcp_server
+import migrate_profiles
+import reconciler_profile
+
+
+PRGS_RECONCILER_ALLOWED = [
+    "gitea.read",
+    "gitea.pr.comment",
+    "gitea.issue.comment",
+    "gitea.issue.close",
+    "gitea.pr.close",
+]
+PRGS_RECONCILER_FORBIDDEN = [
+    "gitea.pr.approve",
+    "gitea.pr.merge",
+    "gitea.pr.create",
+    "gitea.branch.push",
+]
+
+
+class TestReconcilerProfileModel(unittest.TestCase):
+    def test_prgs_reconciler_shape_valid(self):
+        result = reconciler_profile.assess_reconciler_profile(
+            PRGS_RECONCILER_ALLOWED,
+            PRGS_RECONCILER_FORBIDDEN,
+        )
+        self.assertTrue(result["is_reconciler_profile"])
+        self.assertTrue(result["valid"])
+        self.assertEqual(result["reasons"], [])
+
+    def test_author_profile_not_reconciler(self):
+        allowed = [
+            "gitea.read",
+            "gitea.pr.create",
+            "gitea.branch.push",
+            "gitea.issue.comment",
+        ]
+        forbidden = ["gitea.pr.approve", "gitea.pr.merge"]
+        self.assertFalse(
+            reconciler_profile.is_reconciler_profile(allowed, forbidden)
+        )
+
+    def test_reviewer_profile_not_reconciler(self):
+        allowed = [
+            "gitea.read",
+            "gitea.pr.review",
+            "gitea.pr.approve",
+            "gitea.pr.merge",
+        ]
+        forbidden = ["gitea.pr.create", "gitea.branch.push"]
+        self.assertFalse(
+            reconciler_profile.is_reconciler_profile(allowed, forbidden)
+        )
+
+    def test_broad_close_on_author_invalid(self):
+        allowed = PRGS_RECONCILER_ALLOWED + ["gitea.pr.create"]
+        result = reconciler_profile.assess_reconciler_profile(
+            allowed,
+            PRGS_RECONCILER_FORBIDDEN,
+        )
+        self.assertFalse(result["valid"])
+        self.assertFalse(result["is_reconciler_profile"])
+
+    def test_role_kind_detects_reconciler(self):
+        role = mcp_server._role_kind(
+            PRGS_RECONCILER_ALLOWED,
+            PRGS_RECONCILER_FORBIDDEN,
+        )
+        self.assertEqual(role, "reconciler")
+
+    def test_migrate_infer_role_reconciler(self):
+        self.assertEqual(
+            migrate_profiles.infer_role("prgs-reconciler", "prgs-reconciler"),
+            "reconciler",
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_reviewer_proof_backed_handoff.py

@@ -0,0 +1,117 @@
+"""Tests for proof-backed reviewer handoff verifier (#395)."""
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from reviewer_proof_backed_handoff import (  # noqa: E402
+    assess_proof_backed_handoff_report,
+)
+
+
+def _good_report() -> str:
+    return "\n".join([
+        "Inventory pagination proof: is_final_page: true, has_more: false, total_count: 11",
+        "Earlier PRs skipped: #372 non-mergeable — merge simulation conflicts in review_proofs.py",
+        "Baseline worktree path: branches/baseline-master-pr383",
+        "Baseline target SHA: 4243b60ca32d79f1ee5e6b0f10d7570e9dea2937",
+        "Baseline dirty before validation: false",
+        "Baseline validation command: venv/bin/python -m pytest tests/ -q",
+        "Baseline validation result: 1 failed (test_clean_reviewer_report_passes)",
+        "Baseline dirty after validation: false",
+        "Cleanup mutations: git worktree list showed session worktrees removed",
+        "## Controller Handoff",
+        "- Inventory pagination proof: inventory_complete: true, total_count: 11",
+        "- Earlier PRs skipped: #372 conflict proof via merge simulation",
+        "- Baseline worktree used: true",
+        "- Baseline worktree path: branches/baseline-master-pr383",
+        "- Cleanup mutations: git worktree remove branches/review-pr383",
+    ])
+
+
+class TestProofBackedHandoff(unittest.TestCase):
+    def test_fully_proof_backed_report_passes(self):
+        result = assess_proof_backed_handoff_report(
+            _good_report(),
+            action_log=[{"command": "gitea_list_prs", "result": "inventory_complete=true"}],
+            baseline_session={
+                "complete": True,
+                "clean_before": True,
+                "clean_after": True,
+                "command": "pytest",
+                "result": "passed",
+            },
+        )
+        self.assertTrue(result["proven"], result["reasons"])
+
+    def test_inventory_page_size_assumption_blocks(self):
+        report = (
+            "Inventory is complete because 13 results are less than page size 50."
+        )
+        result = assess_proof_backed_handoff_report(report)
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("page-size" in r.lower() for r in result["reasons"]))
+
+    def test_skip_without_conflict_proof_blocks(self):
+        report = "\n".join([
+            "Earlier PRs skipped: #372, #374",
+            "## Controller Handoff",
+            "- Earlier PRs skipped: #372, #374",
+        ])
+        result = assess_proof_backed_handoff_report(report)
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("skip" in r.lower() for r in result["reasons"]))
+
+    def test_baseline_without_command_blocks(self):
+        report = "Baseline validation passed; same as master failures."
+        result = assess_proof_backed_handoff_report(report)
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("baseline" in r.lower() for r in result["reasons"]))
+
+    def test_cleanup_without_worktree_list_blocks(self):
+        report = "Worktrees were cleaned after merge."
+        result = assess_proof_backed_handoff_report(report)
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("cleanup" in r.lower() for r in result["reasons"]))
+
+    def test_cleanup_with_worktree_list_passes(self):
+        report = "Cleanup: git worktree list confirmed removal."
+        result = assess_proof_backed_handoff_report(
+            report,
+            action_log=[{"command": "git worktree list"}],
+        )
+        self.assertTrue(result["proven"], result["reasons"])
+
+    def test_master_integration_requires_command(self):
+        report = "Merged master into the review worktree before validation."
+        result = assess_proof_backed_handoff_report(report)
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("master integration" in r.lower() for r in result["reasons"]))
+
+    def test_master_integration_with_action_log_passes(self):
+        report = "Merged master into review worktree; merge_exit=0"
+        result = assess_proof_backed_handoff_report(
+            report,
+            action_log=[{"command": "git merge --no-commit prgs/master"}],
+        )
+        self.assertTrue(result["proven"], result["reasons"])
+
+    def test_live_proof_requires_source_classification(self):
+        report = "Live proof shows inventory complete."
+        result = assess_proof_backed_handoff_report(report)
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("proof source" in r.lower() for r in result["reasons"]))
+
+
+class TestExport(unittest.TestCase):
+    def test_review_proofs_reexport(self):
+        from review_proofs import assess_proof_backed_handoff_report as exported
+
+        self.assertTrue(callable(exported))
+        result = exported(_good_report(), inventory_session={"inventory_complete": True})
+        self.assertTrue(result["proven"], result["reasons"])
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_reviewer_validation_failure_history.py

@@ -0,0 +1,147 @@
+"""Tests for transient validation failure history verifier (#396)."""
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from final_report_validator import assess_final_report_validator  # noqa: E402
+from reviewer_validation_failure_history import (  # noqa: E402
+    assess_validation_failure_history_report,
+)
+
+
+def _full_history_report() -> str:
+    return "\n".join([
+        "Validation failure history",
+        "Failure #1:",
+        "Command: venv/bin/python -m pytest tests/test_worktrees.py -q",
+        "Failing test: tests/test_worktrees.py::TestWorktreeStart::test_rejects_untraceable_branches",
+        "Suspected cause: stale /tmp/gitea_issue_lock.json from prior session",
+        "Reproduced: no",
+        "On baseline master: no",
+        "PR-caused: no",
+        "What changed between runs: removed /tmp/gitea_issue_lock.json",
+        "Environmental cleanup: lock file removed before rerun",
+        "Validation: passed after transient failure investigation",
+        "Final validation command: venv/bin/python -m pytest tests/ -q",
+        "Final result: 1497 passed, 6 skipped",
+    ])
+
+
+class TestValidationFailureHistory(unittest.TestCase):
+    def test_no_failures_observed_passes(self):
+        result = assess_validation_failure_history_report(
+            "Validation: passed",
+            validation_session={"observed_failures": []},
+        )
+        self.assertTrue(result["proven"])
+
+    def test_omitted_failure_blocks(self):
+        result = assess_validation_failure_history_report(
+            "Validation: passed\n1497 passed, 6 skipped",
+            validation_session={
+                "observed_failures": [{
+                    "command": "pytest tests/test_worktrees.py -q",
+                    "failing_test": (
+                        "tests/test_worktrees.py::TestWorktreeStart::"
+                        "test_rejects_untraceable_branches"
+                    ),
+                }],
+                "final_validation_status": "passed",
+            },
+        )
+        self.assertFalse(result["proven"])
+        self.assertTrue(result["block"])
+
+    def test_full_history_with_transient_investigation_passes(self):
+        result = assess_validation_failure_history_report(
+            _full_history_report(),
+            validation_session={
+                "observed_failures": [{
+                    "command": (
+                        "venv/bin/python -m pytest tests/test_worktrees.py -q"
+                    ),
+                    "failing_test": (
+                        "tests/test_worktrees.py::TestWorktreeStart::"
+                        "test_rejects_untraceable_branches"
+                    ),
+                    "suspected_cause": "stale /tmp/gitea_issue_lock.json",
+                    "reproduced": "no",
+                    "on_baseline_master": "no",
+                    "pr_caused": "no",
+                }],
+                "final_validation_status": "passed_after_transient_failure_investigation",
+                "environmental_contamination": True,
+            },
+        )
+        self.assertTrue(result["proven"], result["reasons"])
+
+    def test_baseline_equivalent_failure_documented(self):
+        report = "\n".join([
+            "Validation failure history",
+            "Command: pytest tests/test_example.py -q",
+            "Failing test: tests/test_example.py::test_flaky",
+            "Suspected cause: pre-existing master failure",
+            "On baseline master: yes",
+            "PR-caused: no",
+            "What changed between runs: none; failure matches baseline",
+            "Validation: passed after transient failure investigation",
+        ])
+        result = assess_validation_failure_history_report(
+            report,
+            validation_session={
+                "observed_failures": [{
+                    "command": "pytest tests/test_example.py -q",
+                    "failing_test": "tests/test_example.py::test_flaky",
+                    "on_baseline_master": "yes",
+                    "pr_caused": "no",
+                }],
+                "final_validation_status": "passed_after_transient_failure_investigation",
+            },
+        )
+        self.assertTrue(result["proven"], result["reasons"])
+
+    def test_unknown_cause_plain_pass_blocks(self):
+        result = assess_validation_failure_history_report(
+            "Validation: passed",
+            validation_session={
+                "observed_failures": [{
+                    "command": "pytest tests/ -q",
+                    "failing_test": "tests/test_foo.py::test_bar",
+                    "suspected_cause": "unknown",
+                }],
+                "final_validation_status": "passed",
+                "cause_unknown": True,
+            },
+        )
+        self.assertFalse(result["proven"])
+        self.assertTrue(any("transient" in r.lower() for r in result["reasons"]))
+
+    def test_final_report_validator_rejects_omitted_failure(self):
+        result = assess_final_report_validator(
+            "Validation: passed",
+            "review_pr",
+            validation_session={
+                "observed_failures": [{
+                    "command": "pytest tests/ -q",
+                    "failing_test": "tests/test_foo.py::test_bar",
+                }],
+            },
+        )
+        self.assertTrue(result["blocked"] or result["downgraded"])
+        self.assertTrue(
+            any(
+                f.get("rule_id") == "reviewer.validation_failure_history"
+                for f in result.get("findings") or []
+            )
+        )
+
+    def test_exported_from_review_proofs(self):
+        from review_proofs import assess_validation_failure_history_report as exported
+
+        self.assertTrue(callable(exported))
+
+
+if __name__ == "__main__":
+    unittest.main()