GEWU/src/gewu/domain/validation.py

"""Deterministic proposal and evidence validation rules."""

from __future__ import annotations

import hashlib
import json
import re
import unicodedata
from collections.abc import Iterable
from copy import deepcopy
from dataclasses import dataclass
from typing import Any

from gewu.domain.contracts import (
    AutoAcceptPolicy,
    ErrorCode,
    EvidenceReference,
    ExtractionBlock,
    ExtractionRecord,
    Finding,
    KnowledgeNode,
    Locator,
    ProjectRecord,
    ProjectStatus,
    ProposalRevision,
    SourceRevisionRecord,
    SupportStatus,
    TypedEdge,
)


def canonical_json(value: object) -> bytes:
    """Serialize contract data deterministically for digests and replay checks."""

    return json.dumps(
        value,
        ensure_ascii=False,
        allow_nan=False,
        sort_keys=True,
        separators=(",", ":"),
    ).encode("utf-8")


def assertion_digest(text: str) -> str:
    normalized = unicodedata.normalize("NFC", text.replace("\r\n", "\n").replace("\r", "\n"))
    return hashlib.sha256(canonical_json({"assertion": normalized})).hexdigest()


def evidence_digest(evidence: EvidenceReference) -> str:
    """Hash a typed evidence reference using its public JSON representation."""

    return hashlib.sha256(canonical_json(evidence.model_dump(mode="json"))).hexdigest()


def _numeric_tokens(text: str) -> set[str]:
    """Return normalized numeric literals for a conservative evidence check."""
    matches = re.findall(r"(?<!\w)\d+(?:,\d{3})*(?:\.\d+)?", text)
    return {match.replace(",", "") for match in matches}


def extraction_digest(extraction: ExtractionRecord) -> str:
    value = extraction.model_dump(mode="json", exclude={"extraction_id"})
    return hashlib.sha256(canonical_json(value)).hexdigest()


def proposal_payload_digest(proposal: ProposalRevision) -> str:
    operations = [item.model_dump(mode="json") for item in proposal.operations]

    def omit_empty_online_evidence(value: Any) -> None:
        if isinstance(value, dict):
            if value.get("online_evidence") == []:
                value.pop("online_evidence")
            for child in value.values():
                omit_empty_online_evidence(child)
        elif isinstance(value, list):
            for child in value:
                omit_empty_online_evidence(child)

    omit_empty_online_evidence(operations)
    return hashlib.sha256(canonical_json(operations)).hexdigest()


def prepare_proposal_digests(draft: dict[str, Any]) -> ProposalRevision:
    """Validate a proposal draft, fill its dependent evidence digests, and pin its payload."""

    value = deepcopy(draft)
    operations = value.get("operations")
    if not isinstance(operations, list) or not operations:
        raise ValueError("Proposal must contain a non-empty operations array")

    for operation in operations:
        if not isinstance(operation, dict):
            raise ValueError("Proposal operations must be objects")
        node = operation.get("node")
        if isinstance(node, dict):
            for group in ("claims", "assumptions", "limitations"):
                claims = node.get(group, [])
                if not isinstance(claims, list):
                    raise ValueError(f"Node {group} must be an array")
                for index, claim in enumerate(claims):
                    if not isinstance(claim, dict) or not isinstance(claim.get("text"), str):
                        raise ValueError(f"Node {group}[{index}] must contain claim text")
                    digest = assertion_digest(claim["text"])
                    evidence = claim.get("evidence")
                    if not isinstance(evidence, list) or not evidence:
                        raise ValueError(f"Node {group}[{index}] must contain evidence")
                    path = f"{group}[{index}].text"
                    evidence_digests = []
                    for item in evidence:
                        if not isinstance(item, dict):
                            raise ValueError("Evidence references must be objects")
                        item["assertion_path"] = path
                        item["assertion_digest"] = digest
                        parsed_evidence = EvidenceReference.model_validate(item, strict=True)
                        item.clear()
                        item.update(parsed_evidence.model_dump(mode="json"))
                        evidence_digests.append(evidence_digest(parsed_evidence))
                    assessment = claim.get("assessment")
                    if not isinstance(assessment, dict):
                        raise ValueError(f"Node {group}[{index}] must contain an assessment")
                    assessment["assertion_digest"] = digest
                    assessment["evidence_digests"] = evidence_digests

        edge = operation.get("edge")
        if isinstance(edge, dict):
            edge_evidence = edge.get("evidence", [])
            if not isinstance(edge_evidence, list):
                raise ValueError("Edge evidence must be an array")
            for item in edge_evidence:
                if not isinstance(item, dict):
                    raise ValueError("Evidence references must be objects")
                item["assertion_path"] = "edge"
                item["assertion_digest"] = "0" * 64
            # The draft originated as JSON, where arrays represent tuple fields.
            # Revalidate through the JSON path so strict mode accepts the typed
            # edge evidence array without weakening Python-object validation.
            parsed_edge = TypedEdge.model_validate_json(
                json.dumps(edge, ensure_ascii=False, allow_nan=False), strict=True
            )
            digest = edge_assertion_digest(parsed_edge)
            for item in edge_evidence:
                item["assertion_digest"] = digest
                parsed_evidence = EvidenceReference.model_validate(item, strict=True)
                item.clear()
                item.update(parsed_evidence.model_dump(mode="json"))
            edge["evidence"] = edge_evidence

    value["payload_digest"] = "0" * 64
    provisional = ProposalRevision.model_validate_json(
        json.dumps(value, ensure_ascii=False, allow_nan=False), strict=True
    )
    return provisional.model_copy(update={"payload_digest": proposal_payload_digest(provisional)})


def edge_assertion_digest(edge: TypedEdge) -> str:
    """Bind edge evidence to the relationship, conditions and rationale."""

    value = edge.model_dump(mode="json", exclude={"edge_id", "evidence"})
    return hashlib.sha256(canonical_json(value)).hexdigest()


def _locator_matches_block(locator: Locator, block: ExtractionBlock) -> bool:
    if locator.kind == "pdf_page_span":
        return (
            locator.page == block.page and block.time_start_ms is None and block.time_end_ms is None
        )
    if locator.kind == "transcript_timestamp_span":
        return (
            locator.page is None
            and locator.time_start_ms == block.time_start_ms
            and locator.time_end_ms == block.time_end_ms
        )
    return (
        locator.page is None
        and locator.time_start_ms is None
        and locator.time_end_ms is None
        and block.page is None
        and block.time_start_ms is None
        and block.time_end_ms is None
    )


@dataclass(frozen=True, slots=True)
class ValidationReport:
    status: str
    eligible_for_auto_accept: bool
    findings: tuple[Finding, ...]
    affected_policy_ids: tuple[str, ...]


def _finding(code: str, message: str, path: str | None = None) -> Finding:
    return Finding(code=code, message=message, path=path)


def _entity_names(node: KnowledgeNode) -> set[str]:
    return {
        unicodedata.normalize("NFC", " ".join(value.split())).casefold()
        for value in (node.title, *node.aliases)
        if value.strip()
    }


def _validate_node_enrichment(
    previous: KnowledgeNode,
    updated: KnowledgeNode,
    proposer_id: str,
    path: str,
) -> tuple[Finding, ...]:
    """Require an update to preserve accepted knowledge and add only assessed material."""
    findings: list[Finding] = []
    for field in ("node_id", "project_id", "node_type", "created", "lifecycle"):
        if getattr(previous, field) != getattr(updated, field):
            findings.append(
                _finding("NODE_ENRICHMENT_IDENTITY_CHANGED", f"Enrichment cannot change {field}", path)
            )
    if updated.updated <= previous.updated:
        findings.append(
            _finding("NODE_ENRICHMENT_TIMESTAMP_INVALID", "Enrichment must advance updated", path)
        )
    if updated.summary != previous.summary:
        findings.append(
            _finding(
                "NODE_ENRICHMENT_SUMMARY_CHANGED",
                "Enrichment cannot replace the summary; add supported claims instead",
                f"{path}.summary",
            )
        )
    for field in ("claims", "assumptions", "limitations"):
        old_items = {claim.claim_id: claim for claim in getattr(previous, field)}
        new_items = {claim.claim_id: claim for claim in getattr(updated, field)}
        if not old_items.keys() <= new_items.keys():
            findings.append(
                _finding("NODE_ENRICHMENT_REMOVED_CLAIM", "Enrichment cannot remove accepted claims", f"{path}.{field}")
            )
            continue
        for claim_id, old_claim in old_items.items():
            new_claim = new_items[claim_id]
            claim_path = f"{path}.{field}.{claim_id}"
            if (
                old_claim.text != new_claim.text
                or old_claim.evidence_type != new_claim.evidence_type
            ):
                findings.append(
                    _finding("NODE_ENRICHMENT_CHANGED_CLAIM", "Enrichment cannot alter an accepted assertion", claim_path)
                )
                continue
            if not set(old_claim.online_evidence) <= set(new_claim.online_evidence):
                findings.append(
                    _finding("NODE_ENRICHMENT_REMOVED_ONLINE_EVIDENCE", "Enrichment cannot remove prior online evidence", claim_path)
                )
            old_evidence = set(old_claim.evidence)
            if not old_evidence <= set(new_claim.evidence):
                findings.append(
                    _finding("NODE_ENRICHMENT_REMOVED_EVIDENCE", "Enrichment cannot remove accepted evidence", claim_path)
                )
            if new_claim.evidence != old_claim.evidence and new_claim.assessment.assessor_id != proposer_id:
                findings.append(
                    _finding("ASSESSOR_ATTRIBUTION_MISMATCH", "New evidence requires an assessment by its proposer", claim_path)
                )
    if not set(previous.tags) <= set(updated.tags):
        findings.append(_finding("NODE_ENRICHMENT_REMOVED_FACET", "Enrichment cannot remove existing tags", f"{path}.tags"))
    for field in ("audience", "domain", "use_for"):
        if not set(getattr(previous, field)) <= set(getattr(updated, field)):
            findings.append(_finding("NODE_ENRICHMENT_REMOVED_FACET", "Enrichment cannot remove existing facets", f"{path}.{field}"))
    return tuple(findings)


def _entity_visible(
    candidate: KnowledgeNode,
    existing: KnowledgeNode,
    readable_project_ids: frozenset[str],
) -> bool:
    return (
        existing.project_id == candidate.project_id or existing.project_id in readable_project_ids
    )


def validate_proposal(
    proposal: ProposalRevision,
    *,
    projects: dict[str, ProjectRecord],
    policies: dict[str, AutoAcceptPolicy],
    sources: dict[str, SourceRevisionRecord],
    extractions: dict[str, ExtractionRecord],
    existing_nodes: dict[str, KnowledgeNode],
    current_base_commit: str,
    readable_project_ids: frozenset[str] = frozenset(),
) -> ValidationReport:
    """Validate evidence, scope, relationships, policy and replay bindings."""

    findings: list[Finding] = []
    affected_policy_ids: set[str] = set()
    eligibility_checks: list[tuple[tuple[str, ...], str, str]] = []
    scoped_projects = set(projects) if proposal.scope.all_projects else set(proposal.scope.projects)
    if proposal.scope.include_shared:
        scoped_projects.add("shared")
    comparison_project_ids = readable_project_ids & frozenset(scoped_projects)
    if (
        not proposal.scope.all_projects
        and proposal.initiating_project_id != "shared"
        and proposal.initiating_project_id not in proposal.scope.projects
    ):
        findings.append(
            _finding(
                "OUT_OF_SCOPE", "Initiating project is outside explicit scope", "scope.projects"
            )
        )
    known_sources = dict(sources)
    proposed_nodes = {
        operation.node.node_id: operation.node
        for operation in proposal.operations
        if operation.node is not None and operation.operation == "upsert_node"
    }
    known_nodes = {**existing_nodes, **proposed_nodes}
    entity_nodes_seen: list[KnowledgeNode] = []

    if proposal.base_commit != current_base_commit:
        findings.append(_finding("BASE_CONFLICT", "Canonical base revision has changed"))
    if proposal_payload_digest(proposal) != proposal.payload_digest:
        findings.append(_finding("PAYLOAD_DIGEST_MISMATCH", "Proposal payload digest is incorrect"))

    for index, operation in enumerate(proposal.operations):
        path = f"operations[{index}]"
        if operation.source is not None:
            source = operation.source
            previous = known_sources.get(source.revision_id)
            if previous and previous != source:
                findings.append(
                    _finding(
                        "SOURCE_REVISION_CONFLICT", "Source revision ID is already bound", path
                    )
                )
            known_sources[source.revision_id] = source
            # Source relevance is explicit: a source ingested for one project
            # does not become shared knowledge merely because its record is
            # stored in the canonical source registry.
            source_domains = set(source.ingested_for)
            for project_id in source.ingested_for:
                if project_id != "shared" and project_id not in projects:
                    findings.append(
                        _finding(
                            "PROJECT_NOT_REGISTERED",
                            "Source relevance names an unregistered project",
                            f"{path}.source.ingested_for",
                        )
                    )
                if (
                    project_id != "shared"
                    and project_id not in proposal.scope.projects
                    and not proposal.scope.all_projects
                ):
                    findings.append(
                        _finding(
                            "OUT_OF_SCOPE",
                            "Source relevance names a project outside explicit scope",
                            f"{path}.source.ingested_for",
                        )
                    )
            affected_policy_ids.update(source_domains)
            eligibility_checks.append((tuple(sorted(source_domains)), "source", "register_source"))

        if operation.node is not None:
            node = operation.node
            previous_node = (
                existing_nodes.get(operation.target_id or "")
                if operation.operation == "update_node"
                else None
            )
            eligibility_checks.append(((node.project_id,), "node", node.node_type.value))
            if operation.operation == "upsert_node" and node.node_id in existing_nodes:
                findings.append(
                    _finding(
                        "NODE_ID_EXISTS",
                        "Node IDs are stable and cannot be reused",
                        f"{path}.node.node_id",
                    )
                )
            elif operation.operation == "update_node":
                if previous_node is None:
                    findings.append(
                        _finding("NODE_NOT_FOUND", "Update target is not an accepted node", f"{path}.target_id")
                    )
                else:
                    findings.extend(
                        _validate_node_enrichment(previous_node, node, proposal.proposer_id, f"{path}.node")
                    )
            if node.node_type.value == "entity":
                possible_duplicates = [
                    other
                    for other in (*existing_nodes.values(), *entity_nodes_seen)
                    if other.node_id != node.node_id
                    and other.node_type.value == "entity"
                    and _entity_visible(node, other, comparison_project_ids)
                    and _entity_names(node) & _entity_names(other)
                ]
                if possible_duplicates:
                    findings.append(
                        _finding(
                            "POSSIBLE_DUPLICATE_ENTITY",
                            "Entity name or alias matches an in-scope entity; review provenance "
                            "before linking or merging",
                            f"{path}.node",
                        )
                    )
                entity_nodes_seen.append(node)
            if node.project_id != "shared" and node.project_id not in projects:
                findings.append(
                    _finding(
                        "PROJECT_NOT_REGISTERED",
                        "Node project is not registered",
                        f"{path}.node.project_id",
                    )
                )
            affected_policy_ids.add(node.project_id)
            if node.agent_id != proposal.proposer_id:
                findings.append(
                    _finding(
                        "AGENT_ATTRIBUTION_MISMATCH",
                        "Node attribution must match the authenticated proposal author",
                        f"{path}.node.agent_id",
                    )
                )
            claim_items = (
                (group_name, claim_index, claim)
                for group_name, claims in (
                    ("claims", node.claims),
                    ("assumptions", node.assumptions),
                    ("limitations", node.limitations),
                )
                for claim_index, claim in enumerate(claims)
            )
            for group_name, claim_index, claim in claim_items:
                claim_path = f"{path}.node.{group_name}[{claim_index}]"
                previous_claims = (
                    {item.claim_id: item for item in previous_node.evidence_claims}
                    if previous_node is not None
                    else {}
                )
                old_claim = previous_claims.get(claim.claim_id)
                unchanged_prior_assessment = bool(
                    old_claim
                    and old_claim.text == claim.text
                    and old_claim.evidence_type == claim.evidence_type
                    and old_claim.evidence == claim.evidence
                    and old_claim.assessment == claim.assessment
                )
                if claim.assessment.assessor_id != proposal.proposer_id and not unchanged_prior_assessment:
                    findings.append(
                        _finding(
                            "ASSESSOR_ATTRIBUTION_MISMATCH",
                            "Support assessment must be attributed to the proposal author",
                            f"{claim_path}.assessment.assessor_id",
                        )
                    )
                expected_assertion_digest = assertion_digest(claim.text)
                assessments = {claim.assessment.assertion_digest}
                if assessments != {expected_assertion_digest}:
                    findings.append(
                        _finding(
                            "ASSESSMENT_STALE",
                            "Support assessment does not bind the current claim text",
                            claim_path,
                        )
                    )
                expected_evidence_digests = {
                    evidence_digest(evidence) for evidence in claim.evidence
                }
                if set(claim.assessment.evidence_digests) != expected_evidence_digests:
                    findings.append(
                        _finding(
                            "ASSESSMENT_EVIDENCE_MISMATCH",
                            "Support assessment does not bind the current evidence",
                            claim_path,
                        )
                    )
                if claim.assessment.status != SupportStatus.SUPPORTED:
                    findings.append(
                        _finding(
                            "SUPPORT_UNAVAILABLE",
                            "Claim lacks a supported semantic assessment",
                            claim_path,
                        )
                    )
                quoted_text = "\n".join(evidence.quote for evidence in claim.evidence)
                unquoted_numbers = _numeric_tokens(claim.text) - _numeric_tokens(quoted_text)
                if unquoted_numbers:
                    findings.append(
                        _finding(
                            "NUMERIC_CLAIM_UNGROUNDED",
                            "Claim contains numeric literals absent from attached evidence; "
                            "semantic review is required",
                            claim_path,
                        )
                    )
                for evidence_index, evidence in enumerate(claim.evidence):
                    evidence_path = f"{claim_path}.evidence[{evidence_index}]"
                    expected_assertion_path = f"{group_name}[{claim_index}].text"
                    if evidence.assertion_path != expected_assertion_path:
                        findings.append(
                            _finding(
                                "EVIDENCE_ASSERTION_PATH_MISMATCH",
                                "Evidence assertion path does not match its typed claim group",
                                evidence_path,
                            )
                        )
                    if evidence.assertion_digest != expected_assertion_digest:
                        findings.append(
                            _finding(
                                "EVIDENCE_ASSERTION_MISMATCH",
                                "Evidence is not bound to the current claim text",
                                evidence_path,
                            )
                        )
                    source_record = known_sources.get(evidence.source_revision_id)
                    if source_record is None:
                        findings.append(
                            _finding(
                                "SOURCE_REVISION_MISSING",
                                "Evidence source revision is unavailable",
                                evidence_path,
                            )
                        )
                    extraction = extractions.get(evidence.extraction_id)
                    if (
                        extraction is None
                        or extraction.source_revision_id != evidence.source_revision_id
                    ):
                        findings.append(
                            _finding(
                                "EXTRACTION_MISMATCH",
                                "Evidence extraction does not belong to its source revision",
                                evidence_path,
                            )
                        )
                        continue
                    if extraction_digest(extraction) != extraction.extraction_id:
                        findings.append(
                            _finding(
                                "EXTRACTION_HASH_MISMATCH",
                                "Stored extraction does not match its immutable identity",
                                evidence_path,
                            )
                        )
                        continue
                    if source_record and source_record.archive_status != "archived":
                        findings.append(
                            _finding(
                                "SOURCE_NOT_ARCHIVED",
                                "Reference-only sources cannot support accepted factual claims",
                                evidence_path,
                            )
                        )
                    block = next(
                        (
                            item
                            for item in extraction.blocks
                            if item.block_id == evidence.locator.block_id
                        ),
                        None,
                    )
                    if block is None:
                        findings.append(
                            _finding(
                                "EVIDENCE_LOCATION_MISMATCH",
                                "Evidence block does not exist",
                                evidence_path,
                            )
                        )
                        continue
                    if not _locator_matches_block(evidence.locator, block):
                        findings.append(
                            _finding(
                                "EVIDENCE_LOCATION_MISMATCH",
                                "Evidence locator metadata does not match its source block",
                                evidence_path,
                            )
                        )
                        continue
                    if evidence.locator.end > len(block.normalized_text):
                        findings.append(
                            _finding(
                                "EVIDENCE_LOCATION_MISMATCH",
                                "Evidence span extends beyond its source block",
                                evidence_path,
                            )
                        )
                        continue
                    located_text = block.normalized_text[
                        evidence.locator.start : evidence.locator.end
                    ]
                    normalized_quote = unicodedata.normalize(
                        "NFC", evidence.quote.replace("\r\n", "\n").replace("\r", "\n")
                    )
                    location_map = block.normalization_map[
                        evidence.locator.start : evidence.locator.end
                    ]
                    original_start = min((span[0] for span in location_map), default=-1)
                    original_end = max((span[1] for span in location_map), default=-1)
                    original_quote = (
                        block.original_text[original_start:original_end]
                        if original_start >= 0
                        else ""
                    )
                    normalized_original_quote = unicodedata.normalize(
                        "NFC", original_quote.replace("\r\n", "\n").replace("\r", "\n")
                    )
                    if (
                        located_text != normalized_quote
                        or normalized_original_quote != normalized_quote
                    ):
                        findings.append(
                            _finding(
                                ErrorCode.EVIDENCE_MISMATCH.value,
                                "Quote does not match text at the stated location",
                                evidence_path,
                            )
                        )

        if operation.edge is not None:
            edge = operation.edge
            expected_edge_digest = edge_assertion_digest(edge)
            for evidence_index, evidence in enumerate(edge.evidence):
                evidence_path = f"{path}.edge.evidence[{evidence_index}]"
                if evidence.assertion_digest != expected_edge_digest:
                    findings.append(
                        _finding(
                            "EVIDENCE_ASSERTION_MISMATCH",
                            "Evidence is not bound to the current relationship",
                            evidence_path,
                        )
                    )
                source_record = known_sources.get(evidence.source_revision_id)
                extraction = extractions.get(evidence.extraction_id)
                if source_record is None:
                    findings.append(
                        _finding(
                            "SOURCE_REVISION_MISSING",
                            "Evidence source revision is unavailable",
                            evidence_path,
                        )
                    )
                if (
                    extraction is None
                    or extraction.source_revision_id != evidence.source_revision_id
                ):
                    findings.append(
                        _finding(
                            "EXTRACTION_MISMATCH",
                            "Evidence extraction does not belong to its source revision",
                            evidence_path,
                        )
                    )
                    continue
                if extraction_digest(extraction) != extraction.extraction_id:
                    findings.append(
                        _finding(
                            "EXTRACTION_HASH_MISMATCH",
                            "Stored extraction does not match its immutable identity",
                            evidence_path,
                        )
                    )
                    continue
                if source_record and source_record.archive_status != "archived":
                    findings.append(
                        _finding(
                            "SOURCE_NOT_ARCHIVED",
                            "Reference-only sources cannot support accepted relationships",
                            evidence_path,
                        )
                    )
                block = next(
                    (
                        item
                        for item in extraction.blocks
                        if item.block_id == evidence.locator.block_id
                    ),
                    None,
                )
                if block is None or not _locator_matches_block(evidence.locator, block):
                    findings.append(
                        _finding(
                            "EVIDENCE_LOCATION_MISMATCH",
                            "Evidence block or locator metadata does not match",
                            evidence_path,
                        )
                    )
                    continue
                if evidence.locator.end > len(block.normalized_text):
                    findings.append(
                        _finding(
                            "EVIDENCE_LOCATION_MISMATCH",
                            "Evidence span extends beyond its source block",
                            evidence_path,
                        )
                    )
                    continue
                actual_quote = block.normalized_text[evidence.locator.start : evidence.locator.end]
                if actual_quote != unicodedata.normalize(
                    "NFC", evidence.quote.replace("\\r\\n", "\\n").replace("\\r", "\\n")
                ):
                    findings.append(
                        _finding(
                            ErrorCode.EVIDENCE_MISMATCH.value,
                            "Quote does not match text at the stated location",
                            evidence_path,
                        )
                    )
            source_node = known_nodes.get(edge.source.id) if edge.source.kind == "node" else None
            target_node = known_nodes.get(edge.target.id) if edge.target.kind == "node" else None
            edge_domains: set[str] = set()
            for endpoint, node_endpoint, label in (
                (edge.source, source_node, "source"),
                (edge.target, target_node, "target"),
            ):
                if endpoint.kind == "node":
                    if node_endpoint is None:
                        findings.append(
                            _finding(
                                "EDGE_TARGET_MISSING",
                                f"Edge {label} node {endpoint.id} is unavailable",
                                f"{path}.edge.{label}.id",
                            )
                        )
                    else:
                        affected_policy_ids.add(node_endpoint.project_id)
                        edge_domains.add(node_endpoint.project_id)
                elif endpoint.kind == "source":
                    if endpoint.id not in known_sources and endpoint.id not in {
                        source.revision_id
                        for source in (
                            item.source for item in proposal.operations if item.source is not None
                        )
                    }:
                        findings.append(
                            _finding(
                                "EDGE_SOURCE_MISSING",
                                f"Edge source revision {endpoint.id} is unavailable",
                                f"{path}.edge.{label}.id",
                            )
                        )
                    affected_policy_ids.add("shared")
                    edge_domains.add("shared")
                elif endpoint.id not in projects:
                    findings.append(
                        _finding(
                            "EDGE_PROJECT_MISSING",
                            f"Edge project {endpoint.id} is not registered",
                            f"{path}.edge.{label}.id",
                        )
                    )
                else:
                    affected_policy_ids.add(endpoint.id)
                    edge_domains.add(endpoint.id)
            if edge_domains:
                eligibility_checks.append((tuple(sorted(edge_domains)), "edge", edge.edge_type))
            if edge.edge_type in {"solves", "addresses"} and source_node and target_node:
                if (
                    source_node.node_type.value != "technique"
                    or target_node.node_type.value != "problem"
                ):
                    findings.append(
                        _finding(
                            "EDGE_TYPE_MISMATCH",
                            f"{edge.edge_type} must connect a technique to a problem",
                            f"{path}.edge",
                        )
                    )

        if operation.operation == "flag_stale" and operation.target_id is not None:
            target = existing_nodes.get(operation.target_id)
            if target is None:
                findings.append(
                    _finding(
                        "NODE_NOT_FOUND", "Staleness target does not exist", f"{path}.target_id"
                    )
                )
            else:
                affected_policy_ids.add(target.project_id)
                eligibility_checks.append(((target.project_id,), "stale", "flag_stale"))

    policy_by_domain: dict[str, AutoAcceptPolicy] = {}
    for project_id in sorted(affected_policy_ids):
        project = projects.get(project_id)
        if project and project.status != ProjectStatus.ACTIVE:
            findings.append(
                _finding(
                    "PROJECT_NOT_ACTIVE",
                    f"Project {project_id} is {project.status.value} and cannot accept changes",
                )
            )
        policy_id = project.autonomy_policy_id if project else project_id
        policy = policies.get(policy_id)
        if policy is None or policy.project_id != project_id:
            findings.append(
                _finding("POLICY_MISSING", f"No applicable policy is registered for {project_id}")
            )
            continue
        if not policy.require_supported_assessment:
            findings.append(
                _finding(
                    "POLICY_WEAKENING",
                    "Policy cannot waive required claim support assessments",
                    f"policies.{policy_id}.require_supported_assessment",
                )
            )
        policy_by_domain[project_id] = policy
        if project_id == "shared" and not proposal.scope.include_shared:
            findings.append(
                _finding(
                    "OUT_OF_SCOPE", "Proposal changes shared outside its explicit scope", "scope"
                )
            )
        elif project_id != "shared" and project_id not in proposal.scope.projects:
            findings.append(
                _finding(
                    "OUT_OF_SCOPE",
                    "Proposal changes a project outside its explicit scope",
                    "scope.projects",
                )
            )

    automatic = bool(policy_by_domain) and len(policy_by_domain) == len(affected_policy_ids)
    for domains, change_kind, change_type in eligibility_checks:
        applicable = [policy_by_domain.get(project_id) for project_id in domains]
        if any(policy is None for policy in applicable):
            automatic = False
        elif change_kind == "source":
            automatic = automatic and all(
                bool(policy and policy.automatic_source_registration) for policy in applicable
            )
        elif change_kind == "node":
            automatic = automatic and all(
                bool(policy and change_type in policy.automatic_node_types) for policy in applicable
            )
        elif change_kind == "edge":
            automatic = automatic and all(
                bool(policy and change_type in policy.automatic_edge_types) for policy in applicable
            )
        else:
            automatic = False
    if not automatic and not findings:
        findings.append(
            _finding(
                "AUTO_ACCEPT_UNSPECIFIED", "Applicable policy requires an exceptional decision"
            )
        )

    blocking_codes = {"BASE_CONFLICT", "PAYLOAD_DIGEST_MISMATCH"}
    invalid_codes = {
        "SOURCE_REVISION_CONFLICT",
        "PROJECT_NOT_REGISTERED",
        "NODE_ID_EXISTS",
        "NODE_NOT_FOUND",
        "NODE_ENRICHMENT_IDENTITY_CHANGED",
        "NODE_ENRICHMENT_TIMESTAMP_INVALID",
        "NODE_ENRICHMENT_SUMMARY_CHANGED",
        "NODE_ENRICHMENT_REMOVED_CLAIM",
        "NODE_ENRICHMENT_CHANGED_CLAIM",
        "NODE_ENRICHMENT_REMOVED_ONLINE_EVIDENCE",
        "NODE_ENRICHMENT_REMOVED_EVIDENCE",
        "NODE_ENRICHMENT_REMOVED_FACET",
        "ASSESSMENT_STALE",
        "ASSESSMENT_EVIDENCE_MISMATCH",
        "SUPPORT_UNAVAILABLE",
        "EVIDENCE_ASSERTION_MISMATCH",
        "SOURCE_REVISION_MISSING",
        "EXTRACTION_MISMATCH",
        "EXTRACTION_HASH_MISMATCH",
        "SOURCE_NOT_ARCHIVED",
        "EVIDENCE_LOCATION_MISMATCH",
        ErrorCode.EVIDENCE_MISMATCH.value,
        "EDGE_TARGET_MISSING",
        "EDGE_SOURCE_MISSING",
        "EDGE_PROJECT_MISSING",
        "EDGE_TYPE_MISMATCH",
        "NODE_NOT_FOUND",
        "POLICY_WEAKENING",
        "OUT_OF_SCOPE",
    }
    codes = {item.code for item in findings}
    if codes & blocking_codes:
        status = "conflict"
    elif codes & invalid_codes:
        status = "invalid"
    elif findings:
        status = "needs-review"
    else:
        status = "eligible"
    return ValidationReport(
        status=status,
        eligible_for_auto_accept=status == "eligible" and automatic,
        findings=tuple(findings),
        affected_policy_ids=tuple(sorted(policy.policy_id for policy in policy_by_domain.values())),
    )


def unique_findings(findings: Iterable[Finding]) -> tuple[Finding, ...]:
    """Stable deduplication for findings shared across bundle members."""

    result: dict[tuple[str, str, str | None], Finding] = {}
    for item in findings:
        result[(item.code, item.message, item.path)] = item
    return tuple(result.values())