ummaya 0.2.4 → 0.2.6

package/src/ummaya/tools/documents/authoring.py

@@ -0,0 +1,283 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+import re
+from enum import StrEnum
+from hashlib import sha256
+from typing import Final, assert_never
+
+from pydantic import BaseModel, ConfigDict, Field, field_serializer, model_validator
+
+from ummaya.tools.documents.models import DocumentToolResult, ToolResultStatus
+
+_SHA256_PATTERN: Final[str] = r"^[a-f0-9]{64}$"
+_REDACTED_DOCUMENT_CONTENT: Final[str] = "[redacted-document-content]"
+_RAW_DOCUMENT_MARKERS: Final[tuple[str, ...]] = (
+    "%pdf",
+    "raw document bytes",
+    "/users/",
+    "\\users\\",
+)
+_COMMON_PII_PATTERNS: Final[tuple[tuple[re.Pattern[str], str], ...]] = (
+    (re.compile(r"\b\d{6}-\d{7}\b"), "[redacted-rrn]"),
+    (re.compile(r"\b01[016789]-\d{3,4}-\d{4}\b"), "[redacted-phone]"),
+    (re.compile(r"[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}"), "[redacted-email]"),
+)
+
+
+class StrictAuthoringModel(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid", strict=True)
+
+
+class EvidenceSourceKind(StrEnum):
+    user_provided = "user_provided"
+    document_derived = "document_derived"
+    policy_derived = "policy_derived"
+
+
+class ApprovalDecisionKind(StrEnum):
+    approved = "approved"
+    edited = "edited"
+    leave_blank = "leave_blank"
+    cancel = "cancel"
+
+
+class AuthoringState(StrEnum):
+    needs_input = "needs_input"
+    draft_ready_for_approval = "draft_ready_for_approval"
+    approved_for_mutation = "approved_for_mutation"
+    blocked_missing_evidence = "blocked_missing_evidence"
+
+
+class AmbiguityClassification(StrEnum):
+    blocking = "blocking"
+    non_blocking_engineering = "non_blocking_engineering"
+
+
+class AuthoringEvidenceItem(StrictAuthoringModel):
+    evidence_id: str = Field(min_length=1)
+    source_kind: EvidenceSourceKind
+    summary: str = Field(min_length=1, max_length=1200)
+    source_ref: str | None = Field(default=None, min_length=1, max_length=300)
+    redacted_excerpt: str | None = Field(default=None, min_length=1, max_length=2000)
+
+    @field_serializer("summary", "source_ref", "redacted_excerpt", when_used="json")
+    def _serialize_redacted_text(self, value: str | None) -> str | None:
+        if value is None:
+            return None
+        return redact_authoring_text(value)
+
+
+class SocraticQuestion(StrictAuthoringModel):
+    question_id: str = Field(min_length=1)
+    target_id: str = Field(min_length=1)
+    prompt: str = Field(min_length=1, max_length=800)
+    required: bool
+
+
+class UserAnswer(StrictAuthoringModel):
+    answer_id: str = Field(min_length=1)
+    question_id: str = Field(min_length=1)
+    response_summary: str = Field(min_length=1, max_length=1200)
+    evidence_refs: tuple[str, ...] = Field(min_length=1)
+
+    @field_serializer("response_summary", when_used="json")
+    def _serialize_response_summary(self, value: str) -> str:
+        return redact_authoring_text(value)
+
+
+class DraftClaim(StrictAuthoringModel):
+    claim_id: str = Field(min_length=1)
+    text: str = Field(min_length=1, max_length=2000)
+    evidence_refs: tuple[str, ...] = Field(min_length=1)
+
+    @field_serializer("text", when_used="json")
+    def _serialize_text(self, value: str) -> str:
+        return redact_authoring_text(value)
+
+
+class DraftCandidate(StrictAuthoringModel):
+    draft_id: str = Field(min_length=1)
+    target_id: str = Field(min_length=1)
+    draft_text: str = Field(min_length=1, max_length=8000)
+    draft_sha256: str = Field(pattern=_SHA256_PATTERN)
+    claims: tuple[DraftClaim, ...] = Field(min_length=1)
+    evidence_items: tuple[AuthoringEvidenceItem, ...] = ()
+
+    @field_serializer("draft_text", when_used="json")
+    def _serialize_draft_text(self, value: str) -> str:
+        return redact_authoring_text(value)
+
+    @model_validator(mode="after")
+    def _validate_draft_hash_and_claim_refs(self) -> DraftCandidate:
+        if self.draft_sha256 != hash_authoring_text(self.draft_text):
+            raise ValueError("draft_sha256 must match draft_text")
+        if self.evidence_items:
+            _require_known_evidence_refs(self.claims, self.evidence_items)
+        return self
+
+
+class ApprovalDecision(StrictAuthoringModel):
+    approval_id: str = Field(min_length=1)
+    draft_id: str = Field(min_length=1)
+    decision: ApprovalDecisionKind
+    draft_sha256: str = Field(pattern=_SHA256_PATTERN)
+    approved_text_sha256: str | None = Field(default=None, pattern=_SHA256_PATTERN)
+
+    @model_validator(mode="after")
+    def _validate_approved_text_hash(self) -> ApprovalDecision:
+        match self.decision:
+            case ApprovalDecisionKind.approved | ApprovalDecisionKind.edited:
+                if self.approved_text_sha256 is None:
+                    raise ValueError("approved decisions require approved_text_sha256")
+            case ApprovalDecisionKind.leave_blank | ApprovalDecisionKind.cancel:
+                if self.approved_text_sha256 is not None:
+                    raise ValueError("blank or cancel decisions cannot carry approved_text_sha256")
+            case unreachable:
+                assert_never(unreachable)
+        return self
+
+
+class AmbiguityRecord(StrictAuthoringModel):
+    ambiguity_id: str = Field(min_length=1)
+    classification: AmbiguityClassification
+    evidence_path: str | None = Field(default=None, min_length=1)
+    reviewer_verdict_id: str | None = Field(default=None, min_length=1)
+    required_next_question: str | None = Field(default=None, min_length=1, max_length=800)
+
+    @field_serializer("evidence_path", "required_next_question", when_used="json")
+    def _serialize_redacted_metadata(self, value: str | None) -> str | None:
+        if value is None:
+            return None
+        return redact_authoring_text(value)
+
+
+class AuthoringClosureVerdict(StrictAuthoringModel):
+    state: AuthoringState
+    target_id: str = Field(min_length=1)
+    evidence_items: tuple[AuthoringEvidenceItem, ...] = ()
+    questions: tuple[SocraticQuestion, ...] = ()
+    answers: tuple[UserAnswer, ...] = ()
+    draft: DraftCandidate | None = None
+    approval: ApprovalDecision | None = None
+    ambiguities: tuple[AmbiguityRecord, ...] = ()
+
+    @model_validator(mode="after")
+    def _validate_state_contract(self) -> AuthoringClosureVerdict:
+        _validate_answer_refs(self.answers, self.questions, self.evidence_items)
+        if self.draft is not None:
+            _require_known_evidence_refs(self.draft.claims, self.evidence_items)
+        _validate_closure_state(self)
+        return self
+
+
+class DocumentAuthoringResult(DocumentToolResult):
+    authoring: AuthoringClosureVerdict
+
+    @model_validator(mode="after")
+    def _validate_authoring_status(self) -> DocumentAuthoringResult:
+        match self.authoring.state:
+            case AuthoringState.needs_input | AuthoringState.draft_ready_for_approval:
+                if self.status is not ToolResultStatus.needs_input:
+                    raise ValueError(
+                        f"{self.authoring.state.value} authoring results require status needs_input"
+                    )
+            case AuthoringState.approved_for_mutation:
+                if self.status is not ToolResultStatus.ok:
+                    raise ValueError("approved_for_mutation authoring results require status ok")
+            case AuthoringState.blocked_missing_evidence:
+                if self.status is not ToolResultStatus.blocked:
+                    raise ValueError(
+                        "blocked_missing_evidence authoring results require status blocked"
+                    )
+            case unreachable:
+                assert_never(unreachable)
+        return self
+
+
+def hash_authoring_text(value: str) -> str:
+    return sha256(value.encode("utf-8")).hexdigest()
+
+
+def redact_authoring_text(value: str) -> str:
+    lowered = value.lower()
+    if any(marker in lowered for marker in _RAW_DOCUMENT_MARKERS):
+        return _REDACTED_DOCUMENT_CONTENT
+    redacted = value
+    for pattern, replacement in _COMMON_PII_PATTERNS:
+        redacted = pattern.sub(replacement, redacted)
+    return redacted
+
+
+def _require_known_evidence_refs(
+    claims: tuple[DraftClaim, ...],
+    evidence_items: tuple[AuthoringEvidenceItem, ...],
+) -> None:
+    known_ids = {item.evidence_id for item in evidence_items}
+    for claim in claims:
+        unknown_refs = tuple(ref for ref in claim.evidence_refs if ref not in known_ids)
+        if unknown_refs:
+            raise ValueError(f"unknown evidence reference: {unknown_refs[0]}")
+
+
+def _validate_answer_refs(
+    answers: tuple[UserAnswer, ...],
+    questions: tuple[SocraticQuestion, ...],
+    evidence_items: tuple[AuthoringEvidenceItem, ...],
+) -> None:
+    known_questions = {question.question_id for question in questions}
+    known_evidence = {item.evidence_id for item in evidence_items}
+    for answer in answers:
+        if answer.question_id not in known_questions:
+            raise ValueError("answers must reference known questions")
+        if known_evidence:
+            unknown_refs = tuple(ref for ref in answer.evidence_refs if ref not in known_evidence)
+            if unknown_refs:
+                raise ValueError(f"unknown answer evidence reference: {unknown_refs[0]}")
+
+
+def _validate_closure_state(verdict: AuthoringClosureVerdict) -> None:
+    match verdict.state:
+        case AuthoringState.needs_input | AuthoringState.blocked_missing_evidence:
+            _validate_question_waiting_state(verdict)
+        case AuthoringState.draft_ready_for_approval:
+            _validate_draft_ready_state(verdict)
+        case AuthoringState.approved_for_mutation:
+            _validate_approved_state(verdict)
+        case unreachable:
+            assert_never(unreachable)
+
+
+def _validate_question_waiting_state(verdict: AuthoringClosureVerdict) -> None:
+    if not verdict.questions:
+        raise ValueError(f"{verdict.state.value} requires questions")
+    if verdict.draft is not None:
+        raise ValueError(f"{verdict.state.value} cannot carry draft")
+    if verdict.approval is not None:
+        raise ValueError(f"{verdict.state.value} cannot carry approval")
+
+
+def _validate_draft_ready_state(verdict: AuthoringClosureVerdict) -> None:
+    if verdict.draft is None:
+        raise ValueError("draft_ready_for_approval requires draft")
+    if verdict.approval is not None:
+        raise ValueError("draft_ready_for_approval cannot carry approval")
+
+
+def _validate_approved_state(verdict: AuthoringClosureVerdict) -> None:
+    if verdict.draft is None:
+        raise ValueError("approved_for_mutation requires draft")
+    if verdict.approval is None:
+        raise ValueError("approved_for_mutation requires approval")
+    match verdict.approval.decision:
+        case ApprovalDecisionKind.approved | ApprovalDecisionKind.edited:
+            pass
+        case ApprovalDecisionKind.leave_blank | ApprovalDecisionKind.cancel:
+            raise ValueError("approved_for_mutation requires an approval decision")
+        case unreachable:
+            assert_never(unreachable)
+    if verdict.approval.draft_id != verdict.draft.draft_id:
+        raise ValueError("approval draft_id must match draft")
+    if verdict.approval.draft_sha256 != verdict.draft.draft_sha256:
+        raise ValueError("approval draft_sha256 must match draft")

package/src/ummaya/tools/documents/baselines.py

@@ -0,0 +1,132 @@
+# SPDX-License-Identifier: Apache-2.0
+"""Public-form conformance baseline models and fixture loader."""
+
+from __future__ import annotations
+
+from importlib import resources
+from pathlib import Path
+from typing import Literal, cast
+
+import yaml
+from pydantic import BaseModel, ConfigDict, Field, model_validator
+
+from ummaya.tools.documents.models import DocumentFormat, MetadataValue
+
+CONFORMANCE_BASELINE_FIXTURE_PATH = (
+    Path(__file__).resolve().parents[4]
+    / "tests"
+    / "fixtures"
+    / "documents"
+    / "public_forms"
+    / "baselines.yaml"
+)
+
+
+class BaselineField(BaseModel):
+    """Required field declared by a public-form conformance baseline."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    field_id: str = Field(min_length=1)
+    label: str = Field(min_length=1)
+    path: str = Field(min_length=1)
+
+
+class BaselineTextAnchor(BaseModel):
+    """Protected text or label that must remain present at a known anchor."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    text: str = Field(min_length=1)
+    anchor: str = Field(min_length=1)
+
+
+class BaselineTableGeometry(BaseModel):
+    """Expected row and column geometry for a protected public-form table."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    table_id: str = Field(min_length=1)
+    anchor: str = Field(min_length=1)
+    rows: int = Field(ge=1)
+    columns: int = Field(ge=1)
+
+
+class ConformanceBaseline(BaseModel):
+    """Format-specific public-form conformance rule set."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    template_id: str = Field(min_length=1)
+    schema_id: str = Field(min_length=1)
+    format: DocumentFormat
+    authoritative_standard: str = Field(min_length=1)
+    authority_refs: tuple[str, ...] = Field(min_length=1)
+    supports_conformance: bool
+    unsupported_reason: str | None = None
+    expected_page_count: int | None = Field(default=None, ge=1)
+    required_fields: tuple[BaselineField, ...] = ()
+    protected_text: tuple[BaselineTextAnchor, ...] = ()
+    required_labels: tuple[BaselineTextAnchor, ...] = ()
+    table_geometries: tuple[BaselineTableGeometry, ...] = ()
+    signature_regions: tuple[BaselineTextAnchor, ...] = ()
+    metadata_exact_matches: dict[str, MetadataValue] = Field(default_factory=dict)
+
+    @model_validator(mode="after")
+    def _enforce_unsupported_reason(self) -> ConformanceBaseline:
+        if not self.supports_conformance and not self.unsupported_reason:
+            raise ValueError("unsupported conformance baselines require unsupported_reason")
+        return self
+
+
+class ConformanceBaselineCatalog(BaseModel):
+    """Offline catalog of public-form conformance baselines."""
+
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    version: Literal[1]
+    catalog_id: str = Field(min_length=1)
+    source_policy: Literal["offline_fixtures_only"]
+    live_network_allowed: Literal[False]
+    baselines: tuple[ConformanceBaseline, ...] = Field(min_length=1)
+
+    @model_validator(mode="after")
+    def _require_unique_template_ids(self) -> ConformanceBaselineCatalog:
+        template_ids = [baseline.template_id for baseline in self.baselines]
+        if len(set(template_ids)) != len(template_ids):
+            raise ValueError("template_id values must be unique")
+        return self
+
+    def by_template_id(self, template_id: str) -> ConformanceBaseline:
+        """Return one baseline by stable template ID."""
+        for baseline in self.baselines:
+            if baseline.template_id == template_id:
+                return baseline
+        raise KeyError(f"Unknown public-form conformance baseline: {template_id}")
+
+
+def load_conformance_baselines(
+    path: Path = CONFORMANCE_BASELINE_FIXTURE_PATH,
+) -> ConformanceBaselineCatalog:
+    """Load checked-in public-form conformance baselines from YAML."""
+    raw = yaml.safe_load(_read_conformance_baseline_text(path))
+    if not isinstance(raw, dict):
+        raise ValueError(f"Conformance baseline manifest must be a YAML object: {path}")
+    return ConformanceBaselineCatalog.model_validate(cast(dict[str, object], raw))
+
+
+def _read_conformance_baseline_text(path: Path) -> str:
+    if path.is_file():
+        return path.read_text(encoding="utf-8")
+
+    try:
+        bundled = resources.files("ummaya._canonical").joinpath("baselines.yaml")
+        with resources.as_file(bundled) as resource_path:
+            if Path(resource_path).is_file():
+                return Path(resource_path).read_text(encoding="utf-8")
+    except (FileNotFoundError, ModuleNotFoundError, AttributeError):
+        pass
+
+    raise FileNotFoundError(
+        f"Conformance baseline manifest not found in source tree or bundled wheel resource: {path}"
+    )