approval-engine/approval_engine/store.py

1020 lines
38 KiB
Python
Raw Normal View History

"""SQLite-backed approval object, closed machine, local outbox.
All mutations run in BEGIN IMMEDIATE and insert the outbox row before COMMIT.
Consumption follows GH-DEC-2026-003: the PEP presents the decision binding's
request digest before the protected side effect.
"""
from __future__ import annotations
import json
import os
import sqlite3
import threading
import tempfile
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Callable
from uuid import uuid4
from .binding import binding_digest, canonical_binding, require_digest
from .errors import (
Conflict,
DuplicateApprover,
NotFound,
StoreUnavailable,
Unprocessable,
)
ISSUER = "approval-engine"
CLAIM_SCHEMA = "0.1"
YIELDS_TO = "net-kingdom taxonomy request-claim schema (statute §17; unassigned)"
DEFAULT_FRESHNESS_TTL = 30
AUDIT_SCHEMA = "audit-core.event.v1alpha1"
SOURCE = "approval-engine"
SCOPE = "netkingdom-approvals"
EVENT_CLASSES = ("issuance", "use", "supersession", "revocation", "heartbeat")
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
LATEST_SCHEMA_VERSION = 3
SCHEMA = """
CREATE TABLE IF NOT EXISTS approvals (
id TEXT PRIMARY KEY,
status TEXT NOT NULL,
binding_json TEXT NOT NULL,
binding_digest TEXT NOT NULL,
pdp_digest TEXT,
actor TEXT NOT NULL,
principal TEXT NOT NULL,
action TEXT NOT NULL,
purpose TEXT NOT NULL,
target_json TEXT NOT NULL,
not_before TEXT NOT NULL,
expires_at TEXT NOT NULL,
required_count INTEGER NOT NULL,
superseded_by TEXT,
consumed_digest TEXT,
consumed_decision_id TEXT,
consumed_at TEXT,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
pdp_path INTEGER NOT NULL DEFAULT 0,
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL
);
CREATE TABLE IF NOT EXISTS entries (
approval_id TEXT NOT NULL,
subject_id TEXT NOT NULL,
approved_at TEXT NOT NULL,
assurance TEXT,
evidence_ref TEXT,
PRIMARY KEY (approval_id, subject_id)
);
CREATE TABLE IF NOT EXISTS outbox (
event_id TEXT PRIMARY KEY,
class TEXT NOT NULL,
approval_id TEXT,
payload_json TEXT NOT NULL,
created_at TEXT NOT NULL,
drained_at TEXT,
attempts INTEGER NOT NULL DEFAULT 0,
last_attempt_at TEXT,
last_error TEXT
);
"""
ACTIVE = ("requested", "approved")
MUTABLE = ("requested", "approved")
def utc_now(clock: Callable[[], datetime] | None = None) -> datetime:
if clock is not None:
now = clock()
else:
now = datetime.now(timezone.utc)
if now.tzinfo is None:
now = now.replace(tzinfo=timezone.utc)
return now.astimezone(timezone.utc).replace(microsecond=0)
def iso(ts: datetime) -> str:
return ts.astimezone(timezone.utc).replace(microsecond=0).isoformat()
def parse_iso(value: str) -> datetime:
return datetime.fromisoformat(value.replace("Z", "+00:00"))
@dataclass
class Approval:
id: str
status: str
binding: dict[str, Any]
binding_digest: str
pdp_digest: str | None
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
pdp_path: bool
actor: str
principal: str
action: str
purpose: str
target: dict[str, Any]
not_before: str
expires_at: str
required_count: int
superseded_by: str | None
consumed_digest: str | None
consumed_decision_id: str | None
consumed_at: str | None
created_at: str
updated_at: str
entries: list[dict[str, Any]]
def as_dict(self) -> dict[str, Any]:
result = {
"id": self.id,
"status": self.status,
"binding": {
**self.binding,
"digest": self.binding_digest,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
"pdp_digest": self.pdp_digest,
"pdp_path": self.pdp_path,
},
"validity": {"not_before": self.not_before, "expires_at": self.expires_at},
"required_count": self.required_count,
"entries": self.entries,
"superseded_by": self.superseded_by,
"created_at": self.created_at,
"updated_at": self.updated_at,
}
if self.consumed_digest:
result["consumption"] = {
"request_digest": self.consumed_digest,
**(
{"decision_id": self.consumed_decision_id}
if self.consumed_decision_id
else {}
),
"consumed_at": self.consumed_at,
}
return result
class Engine:
def __init__(
self,
path: str | Path = ":memory:",
*,
clock: Callable[[], datetime] | None = None,
freshness_ttl: int = DEFAULT_FRESHNESS_TTL,
fail_outbox: bool = False,
tenant: str = "platform",
auto_migrate: bool = True,
) -> None:
self.path = str(path)
self.clock = clock
self.freshness_ttl = freshness_ttl
self.fail_outbox = fail_outbox
self.tenant = tenant
self.auto_migrate = auto_migrate
self._local = threading.local()
self._init_schema()
def _connect(self) -> sqlite3.Connection:
try:
conn = sqlite3.connect(self.path, timeout=5, check_same_thread=False)
conn.row_factory = sqlite3.Row
conn.execute("PRAGMA foreign_keys=ON")
conn.execute("PRAGMA busy_timeout=5000")
if self.path != ":memory:":
conn.execute("PRAGMA journal_mode=WAL")
return conn
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
def _conn(self) -> sqlite3.Connection:
conn = getattr(self._local, "conn", None)
if conn is None:
conn = self._connect()
self._local.conn = conn
return conn
def _init_schema(self) -> None:
conn = self._conn()
try:
current = int(conn.execute("PRAGMA user_version").fetchone()[0])
if not self.auto_migrate:
tables = {
row[0]
for row in conn.execute(
"SELECT name FROM sqlite_master WHERE type='table'"
).fetchall()
}
if current != LATEST_SCHEMA_VERSION:
raise StoreUnavailable(
f"database schema version {current} is not {LATEST_SCHEMA_VERSION}; "
"run approval-engine migrate"
)
if not {"approvals", "entries", "outbox"} <= tables:
raise StoreUnavailable("database schema is incomplete")
return
conn.executescript(SCHEMA)
approval_columns = {
row["name"] for row in conn.execute("PRAGMA table_info(approvals)").fetchall()
}
for name in ("consumed_digest", "consumed_decision_id", "consumed_at"):
if name not in approval_columns:
conn.execute(f"ALTER TABLE approvals ADD COLUMN {name} TEXT")
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
if "pdp_path" not in approval_columns:
# v3, GH-DEC-2026-008. Legacy rows default to 0: an approval
# issued before the ruling was never declared for the PDP path,
# and inferring intent from a recorded digest would manufacture
# a declaration nobody made.
conn.execute(
"ALTER TABLE approvals ADD COLUMN pdp_path INTEGER NOT NULL DEFAULT 0"
)
outbox_columns = {
row["name"] for row in conn.execute("PRAGMA table_info(outbox)").fetchall()
}
if "attempts" not in outbox_columns:
conn.execute(
"ALTER TABLE outbox ADD COLUMN attempts INTEGER NOT NULL DEFAULT 0"
)
for name in ("last_attempt_at", "last_error"):
if name not in outbox_columns:
conn.execute(f"ALTER TABLE outbox ADD COLUMN {name} TEXT")
conn.execute(f"PRAGMA user_version={LATEST_SCHEMA_VERSION}")
conn.commit()
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
def close(self) -> None:
conn = getattr(self._local, "conn", None)
if conn is not None:
conn.close()
self._local.conn = None
def now(self) -> datetime:
return utc_now(self.clock)
def storage_status(self, *, integrity: bool = False) -> dict[str, Any]:
conn = self._conn()
try:
version = int(conn.execute("PRAGMA user_version").fetchone()[0])
result: dict[str, Any] = {
"schema_version": version,
"expected_schema_version": LATEST_SCHEMA_VERSION,
"schema_current": version == LATEST_SCHEMA_VERSION,
"persistent": self.path != ":memory:",
}
tables = {
row[0]
for row in conn.execute(
"SELECT name FROM sqlite_master WHERE type='table'"
).fetchall()
}
result["required_tables"] = {
name: name in tables for name in ("approvals", "entries", "outbox")
}
if integrity:
check = [
row[0] for row in conn.execute("PRAGMA integrity_check").fetchall()
]
foreign_keys = conn.execute("PRAGMA foreign_key_check").fetchall()
result["integrity"] = check
result["foreign_key_violations"] = len(foreign_keys)
result["ok"] = (
check == ["ok"]
and not foreign_keys
and result["schema_current"]
and all(result["required_tables"].values())
)
return result
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
def backup(self, output: str | Path) -> dict[str, Any]:
if self.path == ":memory:":
raise Unprocessable("an in-memory database cannot be backed up")
target = Path(output)
if target.exists():
raise Conflict(f"backup target already exists: {target}")
if not target.parent.exists():
raise Unprocessable(f"backup parent does not exist: {target.parent}")
handle = tempfile.NamedTemporaryFile(
prefix=target.name + ".",
suffix=".tmp",
dir=target.parent,
delete=False,
)
temporary = Path(handle.name)
handle.close()
os.chmod(temporary, 0o600)
destination: sqlite3.Connection | None = None
try:
destination = sqlite3.connect(temporary)
self._conn().backup(destination)
destination.close()
destination = None
check = sqlite3.connect(temporary)
try:
integrity = [
row[0] for row in check.execute("PRAGMA integrity_check")
]
finally:
check.close()
if integrity != ["ok"]:
raise StoreUnavailable("backup integrity verification failed")
os.replace(temporary, target)
except Exception:
if destination is not None:
destination.close()
temporary.unlink(missing_ok=True)
raise
return {"path": str(target), "mode": "0600", "integrity": "ok"}
def create(
self,
binding: dict[str, Any],
validity: dict[str, str],
required_count: int = 1,
*,
pdp_digest: str | None = None,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
pdp_path: bool = False,
approval_id: str | None = None,
) -> Approval:
if required_count < 1:
raise Unprocessable("required_count must be >= 1")
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
if not isinstance(pdp_path, bool):
raise Unprocessable("pdp_path must be a boolean")
canon = canonical_binding(binding)
digest = binding_digest(canon)
pdp = require_digest(pdp_digest)
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
if pdp_path and pdp is None:
# GH-DEC-2026-008: refuse at issue rather than at consume. An
# approval declared for the PDP path without a bound request
# digest is unusable there, and discovering that at the moment of
# the protected side effect is the worst place to find out.
raise Unprocessable(
"pdp_path requires pdp_digest: an approval for the "
"GH-DEC-2026-003 path must bind the PDP request digest at issue"
)
not_before = validity.get("not_before") or iso(self.now())
expires_at = validity.get("expires_at")
if not expires_at:
raise Unprocessable("validity.expires_at is required")
if parse_iso(expires_at) <= parse_iso(not_before):
raise Unprocessable("expires_at must be after not_before")
aid = approval_id or str(uuid4())
now = iso(self.now())
conn = self._conn()
try:
conn.execute("BEGIN IMMEDIATE")
conn.execute(
"""INSERT INTO approvals (
id, status, binding_json, binding_digest, pdp_digest,
actor, principal, action, purpose, target_json,
not_before, expires_at, required_count, superseded_by,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
pdp_path, created_at, updated_at
) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
(
aid,
"requested",
json.dumps(canon, sort_keys=True),
digest,
pdp,
canon["actor"],
canon["principal"],
canon["action"],
canon["purpose"],
json.dumps(canon["target"], sort_keys=True),
not_before,
expires_at,
required_count,
None,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
1 if pdp_path else 0,
now,
now,
),
)
conn.commit()
except sqlite3.IntegrityError as exc:
conn.rollback()
raise Conflict("approval id already exists") from exc
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
except Exception:
conn.rollback()
raise
return self.get(aid)
def get(self, approval_id: str) -> Approval:
conn = self._conn()
try:
row = conn.execute("SELECT * FROM approvals WHERE id=?", (approval_id,)).fetchone()
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
if row is None:
raise NotFound(approval_id)
return self._hydrate(row, persist_expiry=True)
def _hydrate(self, row: sqlite3.Row, persist_expiry: bool) -> Approval:
entries = [
{
"subject_id": e["subject_id"],
"approved_at": e["approved_at"],
"assurance": e["assurance"],
"evidence_ref": e["evidence_ref"],
}
for e in self._conn()
.execute(
"SELECT subject_id, approved_at, assurance, evidence_ref "
"FROM entries WHERE approval_id=? ORDER BY approved_at",
(row["id"],),
)
.fetchall()
]
status = row["status"]
if persist_expiry and status in ACTIVE:
now = self.now()
if now >= parse_iso(row["expires_at"]):
status = self._expire(row["id"], status)
binding = json.loads(row["binding_json"])
return Approval(
id=row["id"],
status=status,
binding=binding,
binding_digest=row["binding_digest"],
pdp_digest=row["pdp_digest"],
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
pdp_path=bool(row["pdp_path"]),
actor=row["actor"],
principal=row["principal"],
action=row["action"],
purpose=row["purpose"],
target=json.loads(row["target_json"]),
not_before=row["not_before"],
expires_at=row["expires_at"],
required_count=row["required_count"],
superseded_by=row["superseded_by"],
consumed_digest=row["consumed_digest"],
consumed_decision_id=row["consumed_decision_id"],
consumed_at=row["consumed_at"],
created_at=row["created_at"],
updated_at=row["updated_at"],
entries=entries,
)
def _expire(self, approval_id: str, from_status: str) -> str:
conn = self._conn()
now = iso(self.now())
try:
conn.execute("BEGIN IMMEDIATE")
cur = conn.execute(
"UPDATE approvals SET status='expired', updated_at=? "
"WHERE id=? AND status=?",
(now, approval_id, from_status),
)
conn.commit()
return "expired" if cur.rowcount == 1 else from_status
except sqlite3.Error:
conn.rollback()
return from_status
def add_entry(
self,
approval_id: str,
subject_id: str,
*,
assurance: str | None = None,
evidence_ref: str | None = None,
) -> Approval:
if not subject_id:
raise Unprocessable("subject_id is required")
conn = self._conn()
now = iso(self.now())
try:
conn.execute("BEGIN IMMEDIATE")
row = conn.execute("SELECT * FROM approvals WHERE id=?", (approval_id,)).fetchone()
if row is None:
conn.rollback()
raise NotFound(approval_id)
if row["status"] not in MUTABLE:
conn.rollback()
raise Conflict(f"cannot add entries in status {row['status']}")
try:
conn.execute(
"INSERT INTO entries (approval_id, subject_id, approved_at, assurance, evidence_ref) "
"VALUES (?,?,?,?,?)",
(approval_id, subject_id, now, assurance, evidence_ref),
)
except sqlite3.IntegrityError as exc:
conn.rollback()
raise DuplicateApprover(subject_id) from exc
count = conn.execute(
"SELECT COUNT(*) FROM entries WHERE approval_id=?", (approval_id,)
).fetchone()[0]
became_approved = False
if row["status"] == "requested" and count >= row["required_count"]:
conn.execute(
"UPDATE approvals SET status='approved', updated_at=? WHERE id=? AND status='requested'",
(now, approval_id),
)
became_approved = True
else:
conn.execute(
"UPDATE approvals SET updated_at=? WHERE id=?", (now, approval_id)
)
if became_approved:
self._outbox_insert(
conn,
"issuance",
approval_id,
actor=row["actor"],
Carry threshold evidence on issuance and use events GH-DEC-2026-005 moved the distinct-approver check off the PEP onto this engine's valid_now. secrets-engine has implemented the split and reports it no longer verifies the threshold independently. Gate House accepted that as correct on layering AND as a genuine reduction in defence in depth, and named the compensating control: not a second check at the PEP, which is the duplication the split removes, but reconstructability at the issuer under §9.6. The emitted events could not support that. approval.issuance carried required_count but never who satisfied it, and approval.use carried no threshold evidence at all, so an auditor replaying the stream could not recompute the evaluation without reading live rows -- rows that may since have been superseded, revoked, or expired. Both events now carry a threshold object: required_count, distinct_approver_count, threshold_met, and approvers with approved_at plus assurance and evidence_ref when recorded. Tests prove reconstruction from the use row alone, and that the claim still discloses no approver identities -- they are evidence for audit-core, not consumer-facing, and the claim keeps disclosing the least it can. Writing the tests showed distinctness is already a storage invariant: entries is UNIQUE on (approval_id, subject_id), so a repeat approver is refused at insert and a separate entry_count could never differ from the distinct count. Dropped that field rather than ship a number that cannot vary, and the test now asserts the refusal instead. 88 tests pass (4 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 08:10:21 +02:00
extra={
"binding_digest": row["binding_digest"],
"required_count": row["required_count"],
"threshold": self._threshold_evidence(
conn, approval_id, row["required_count"]
),
},
)
conn.commit()
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
except Exception:
conn.rollback()
raise
return self.get(approval_id)
def revoke(self, approval_id: str) -> Approval:
return self._terminal(approval_id, "revoked", "revocation")
def supersede(self, approval_id: str, successor_id: str | None = None) -> dict[str, Any]:
successor_id = successor_id or str(uuid4())
conn = self._conn()
now = iso(self.now())
try:
conn.execute("BEGIN IMMEDIATE")
row = conn.execute("SELECT * FROM approvals WHERE id=?", (approval_id,)).fetchone()
if row is None:
conn.rollback()
raise NotFound(approval_id)
if row["status"] not in MUTABLE:
conn.rollback()
raise Conflict(f"cannot supersede in status {row['status']}")
cur = conn.execute(
"UPDATE approvals SET status='superseded', superseded_by=?, updated_at=? "
"WHERE id=? AND status IN ('requested','approved')",
(successor_id, now, approval_id),
)
if cur.rowcount != 1:
conn.rollback()
raise Conflict("supersession lost the compare-and-swap")
existing = conn.execute("SELECT id FROM approvals WHERE id=?", (successor_id,)).fetchone()
created_successor = False
if existing is None:
conn.execute(
"""INSERT INTO approvals (
id, status, binding_json, binding_digest, pdp_digest,
actor, principal, action, purpose, target_json,
not_before, expires_at, required_count, superseded_by,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
pdp_path, created_at, updated_at
) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
(
successor_id,
"requested",
row["binding_json"],
row["binding_digest"],
row["pdp_digest"],
row["actor"],
row["principal"],
row["action"],
row["purpose"],
row["target_json"],
row["not_before"],
row["expires_at"],
row["required_count"],
None,
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
row["pdp_path"],
now,
now,
),
)
created_successor = True
self._outbox_insert(
conn,
"supersession",
approval_id,
actor=row["actor"],
extra={
"binding_digest": row["binding_digest"],
"superseded_by": successor_id,
},
)
conn.commit()
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
except Exception:
conn.rollback()
raise
return {
"superseded": self.get(approval_id).as_dict(),
"successor_id": successor_id,
"successor_created": created_successor,
}
def _terminal(self, approval_id: str, status: str, event_class: str) -> Approval:
conn = self._conn()
now = iso(self.now())
try:
conn.execute("BEGIN IMMEDIATE")
row = conn.execute("SELECT * FROM approvals WHERE id=?", (approval_id,)).fetchone()
if row is None:
conn.rollback()
raise NotFound(approval_id)
cur = conn.execute(
"UPDATE approvals SET status=?, updated_at=? "
"WHERE id=? AND status IN ('requested','approved')",
(status, now, approval_id),
)
if cur.rowcount != 1:
conn.rollback()
raise Conflict(f"cannot {status} from status {row['status']}")
self._outbox_insert(
conn,
event_class,
approval_id,
actor=row["actor"],
extra={"binding_digest": row["binding_digest"]},
)
conn.commit()
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
except Exception:
conn.rollback()
raise
return self.get(approval_id)
def consume(
self,
approval_id: str,
request_digest: str | None,
*,
decision_id: str | None = None,
) -> dict[str, Any]:
"""Atomically spend an approval for one decision-bound request.
Repeating the same request digest is an idempotent success. A different
digest against a consumed object conflicts and the caller must not act.
"""
if request_digest is None:
raise Unprocessable("request_digest is required")
if not isinstance(request_digest, str):
raise Unprocessable("request_digest must be a string")
digest = require_digest(request_digest)
if decision_id is not None and (
not isinstance(decision_id, str) or not decision_id
):
raise Unprocessable("decision_id must be a non-empty string")
conn = self._conn()
now = iso(self.now())
idempotent = False
try:
conn.execute("BEGIN IMMEDIATE")
row = conn.execute("SELECT * FROM approvals WHERE id=?", (approval_id,)).fetchone()
if row is None:
conn.rollback()
raise NotFound(approval_id)
if row["status"] == "consumed":
if row["consumed_digest"] != digest:
conn.rollback()
raise Conflict("approval already consumed for a different request digest")
idempotent = True
stored_decision_id = row["consumed_decision_id"]
conn.commit()
return {
"approval_id": approval_id,
"status": "consumed",
"request_digest": digest,
**({"decision_id": stored_decision_id} if stored_decision_id else {}),
"consumed_at": row["consumed_at"],
"idempotent": idempotent,
}
if row["status"] != "approved":
conn.rollback()
raise Conflict(f"cannot consume from status {row['status']}")
if self.now() < parse_iso(row["not_before"]) or self.now() >= parse_iso(row["expires_at"]):
conn.rollback()
raise Conflict("cannot consume outside validity window")
cur = conn.execute(
"UPDATE approvals SET status='consumed', consumed_digest=?, "
"consumed_decision_id=?, consumed_at=?, updated_at=? "
"WHERE id=? AND status='approved'",
(digest, decision_id, now, now, approval_id),
)
if cur.rowcount != 1:
conn.rollback()
raise Conflict("consumption lost the compare-and-swap")
self._outbox_insert(
conn,
"use",
approval_id,
actor=row["actor"],
extra={
"binding_digest": row["binding_digest"],
"request_digest": digest,
**({"decision_id": decision_id} if decision_id else {}),
Carry threshold evidence on issuance and use events GH-DEC-2026-005 moved the distinct-approver check off the PEP onto this engine's valid_now. secrets-engine has implemented the split and reports it no longer verifies the threshold independently. Gate House accepted that as correct on layering AND as a genuine reduction in defence in depth, and named the compensating control: not a second check at the PEP, which is the duplication the split removes, but reconstructability at the issuer under §9.6. The emitted events could not support that. approval.issuance carried required_count but never who satisfied it, and approval.use carried no threshold evidence at all, so an auditor replaying the stream could not recompute the evaluation without reading live rows -- rows that may since have been superseded, revoked, or expired. Both events now carry a threshold object: required_count, distinct_approver_count, threshold_met, and approvers with approved_at plus assurance and evidence_ref when recorded. Tests prove reconstruction from the use row alone, and that the claim still discloses no approver identities -- they are evidence for audit-core, not consumer-facing, and the claim keeps disclosing the least it can. Writing the tests showed distinctness is already a storage invariant: entries is UNIQUE on (approval_id, subject_id), so a repeat approver is refused at insert and a separate entry_count could never differ from the distinct count. Dropped that field rather than ship a number that cannot vary, and the test now asserts the refusal instead. 88 tests pass (4 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 08:10:21 +02:00
"threshold": self._threshold_evidence(
conn, approval_id, row["required_count"]
),
},
)
conn.commit()
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
except Exception:
conn.rollback()
raise
return {
"approval_id": approval_id,
"status": "consumed",
"request_digest": digest,
**({"decision_id": decision_id} if decision_id else {}),
"consumed_at": now,
"idempotent": idempotent,
}
Carry threshold evidence on issuance and use events GH-DEC-2026-005 moved the distinct-approver check off the PEP onto this engine's valid_now. secrets-engine has implemented the split and reports it no longer verifies the threshold independently. Gate House accepted that as correct on layering AND as a genuine reduction in defence in depth, and named the compensating control: not a second check at the PEP, which is the duplication the split removes, but reconstructability at the issuer under §9.6. The emitted events could not support that. approval.issuance carried required_count but never who satisfied it, and approval.use carried no threshold evidence at all, so an auditor replaying the stream could not recompute the evaluation without reading live rows -- rows that may since have been superseded, revoked, or expired. Both events now carry a threshold object: required_count, distinct_approver_count, threshold_met, and approvers with approved_at plus assurance and evidence_ref when recorded. Tests prove reconstruction from the use row alone, and that the claim still discloses no approver identities -- they are evidence for audit-core, not consumer-facing, and the claim keeps disclosing the least it can. Writing the tests showed distinctness is already a storage invariant: entries is UNIQUE on (approval_id, subject_id), so a repeat approver is refused at insert and a separate entry_count could never differ from the distinct count. Dropped that field rather than ship a number that cannot vary, and the test now asserts the refusal instead. 88 tests pass (4 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 08:10:21 +02:00
def _threshold_evidence(
self, conn: sqlite3.Connection, approval_id: str, required_count: int
) -> dict[str, Any]:
"""Threshold evaluation as evidence, for reconstruction under §9.6.
`GH-DEC-2026-005` moved the distinct-approver check off the PEP and onto
this engine's `valid_now`. The compensating control is detection, not
prevention: the emitted evidence must let an auditor recompute the
evaluation after the fact without reading live rows, which may since
have been superseded or expired.
Approver identities belong here and not on the claim. The claim is
consumer-facing and discloses the least it can; the outbox is the
evidence path to audit-core, where the identities are the point.
Distinctness is a storage invariant, not a recomputation: `entries`
has a UNIQUE constraint on (approval_id, subject_id), so a repeat
approver is refused at insert. The dedup below is belt-and-braces for
a future schema that relaxes it.
"""
rows = conn.execute(
"SELECT subject_id, approved_at, assurance, evidence_ref "
"FROM entries WHERE approval_id=? ORDER BY approved_at, subject_id",
(approval_id,),
).fetchall()
seen: dict[str, sqlite3.Row] = {}
for r in rows:
seen.setdefault(r["subject_id"], r)
approvers = [
{
"subject_id": r["subject_id"],
"approved_at": r["approved_at"],
**({"assurance": r["assurance"]} if r["assurance"] else {}),
**({"evidence_ref": r["evidence_ref"]} if r["evidence_ref"] else {}),
}
for r in seen.values()
]
return {
"required_count": required_count,
"distinct_approver_count": len(approvers),
"threshold_met": len(approvers) >= required_count,
"approvers": approvers,
}
def _outbox_insert(
self,
conn: sqlite3.Connection,
event_class: str,
approval_id: str | None,
*,
actor: str | None,
extra: dict[str, Any] | None = None,
) -> str:
if self.fail_outbox:
raise StoreUnavailable("outbox insert failed")
if event_class not in EVENT_CLASSES:
raise Unprocessable(f"unknown event class {event_class}")
event_id = str(uuid4())
created = iso(self.now())
resource = f"approval:{approval_id}" if approval_id else "approval-engine:heartbeat"
details = {"class": event_class, **(extra or {})}
if approval_id:
details["approval_id"] = approval_id
payload = {
"schema_version": AUDIT_SCHEMA,
"event_id": event_id,
"observed_at": created,
"tenant": self.tenant,
"scope": SCOPE,
"source": SOURCE,
"actor": actor,
"action": f"approval.{event_class}",
"resource": resource,
"outcome": "success",
"reason": None,
"details": details,
}
conn.execute(
"INSERT INTO outbox (event_id, class, approval_id, payload_json, created_at, drained_at) "
"VALUES (?,?,?,?,?,NULL)",
(event_id, event_class, approval_id, json.dumps(payload, sort_keys=True), created),
)
return event_id
def claim(self, approval_id: str) -> dict[str, Any]:
obj = self.get(approval_id)
observed = self.now()
not_after = observed + timedelta(seconds=self.freshness_ttl)
state, valid_now, consumed, reason = self._evaluate(obj, observed)
binding = {
**canonical_binding(obj.binding),
"digest": obj.binding_digest,
}
State pdp_digest explicitly; decline to publish a vocabulary mapping flex-auth asked whether this engine should publish an action/target mapping between the claim binding's vocabulary (secrets.kv.destroy, {"id": "lane-openbao-root"}) and a policy package's (destroy, lane:...), since their package makes no cross-check that a claim was approved for the action being decided. Answered no. A PIP asserting that one vocabulary's action means another's would author policy semantics it does not own, over vocabularies it does not own, and the failure mode is asymmetric: a wrong mapping silently accepts a claim approved for a different action, which is worse than no mapping. binding.pdp_digest is the correspondence and sidesteps vocabulary entirely -- it compares the PDP's own digest to the PDP's own digest, with no translation by anyone. Implemented the part that was ours. pdp_digest was emitted only when recorded, so a consumer could not distinguish "not issued against a decision" from "we forgot to look". It is now always present and null in that case, required-but-nullable in the schema, and documented as something a PEP on a privileged lane must refuse. This engine states the fact; enforcing the lane's policy stays with the consumer. Both published examples were already contradicting the updated schema by omitting the field -- the same fixture-versus-contract defect flex-auth hit twice this week and that secrets-engine implemented. Fixed both, made them cover the PDP-bound and unbound shapes so neither is inferred from the other, and added tests/test_examples.py to validate every example against the schema so the class cannot recur here. jsonschema added as a dev dependency. 94 tests pass (6 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 09:32:11 +02:00
# Always stated, null when the approval was not issued against a PDP
# decision. A missing key reads as an oversight; an explicit null is a
# fact the consumer must act on. See docs/approval-claim.md.
binding["pdp_digest"] = obj.pdp_digest or None
Implement GH-DEC-2026-008: declared PDP-path intent, enforced at issue Gate House ruled binding.pdp_digest is the binding correspondence on the GH-DEC-2026-003 path and is required there, having rejected a vocabulary mapping for the reasons we gave. It asked this engine to record the PDP digest at issue for approvals intended for that path, and to have the claim state which approvals those are rather than leaving it to the requester's memory. Schema v3 adds approvals.pdp_path. create() refuses pdp_path true without a pdp_digest, so an approval that would be unusable on the path fails at issue rather than at the protected side effect. The claim exposes binding.pdp_path, which makes it a guarantee rather than a hint: pdp_path true implies pdp_digest is non-null. Intent is declared and never inferred. A pdp_digest that happens to be present is not a declaration anybody made, so a recorded digest alone leaves pdp_path false, legacy rows migrate to false rather than being back-filled from their digests, and a successor inherits its predecessor's declaration. Approvals issued before the ruling stay usable by consumers in this engine's own vocabulary and are simply not usable on the PDP path -- the ruling's intended cost, stated as such. Schema, both published examples, a v2-to-v3 migration test asserting survivors keep their digest while declaring no path intent, and tests for refusal at issue, claim exposure, non-inference, and successor inheritance. 102 tests pass (8 new). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TvyJPAaVCGsVheVhcCwNND Assistant: claude-code Assistant-Model: opus Assistant-Process: 411227@bnt-lap001 Assistant-Session: d566f6d3-bcaf-43c3-bc5e-3ddd0f64b535
2026-09-06 14:51:23 +02:00
# GH-DEC-2026-008: the claim states whether this approval was declared
# for the PDP path, so a consumer does not infer it from a digest that
# happens to be present.
binding["pdp_path"] = obj.pdp_path
return {
"schema_version": CLAIM_SCHEMA,
"kind": "approval-claim",
"yields_to": YIELDS_TO,
"issuer": ISSUER,
"approval_id": obj.id,
"state": state,
"valid_now": valid_now,
"consumed": consumed,
"binding": binding,
"freshness": {
"observed_at": iso(observed),
"ttl_seconds": self.freshness_ttl,
"not_after": iso(not_after),
},
"validity": {"not_before": obj.not_before, "expires_at": obj.expires_at},
"reason_code": reason,
}
def _evaluate(self, obj: Approval, now: datetime) -> tuple[str, bool, bool, str]:
consumed = obj.status == "consumed"
if obj.status == "revoked":
return "revoked", False, consumed, "revoked"
if obj.status == "superseded":
return "superseded", False, consumed, "superseded"
if obj.status == "expired" or now >= parse_iso(obj.expires_at):
return "expired", False, consumed, "expired"
if obj.status == "consumed":
return "consumed", False, True, "consumed"
if obj.status == "requested":
return "requested", False, False, "requested"
if obj.status == "approved":
if now < parse_iso(obj.not_before):
return "approved", False, False, "not_yet_valid"
if len({e["subject_id"] for e in obj.entries}) < obj.required_count:
return "approved", False, False, "insufficient_approvers"
return "valid", True, False, "ok"
return obj.status, False, consumed, obj.status
def emit_heartbeat(self) -> dict[str, Any]:
conn = self._conn()
counts = self.transition_counts()
counts["heartbeat"] += 1
try:
conn.execute("BEGIN IMMEDIATE")
event_id = self._outbox_insert(
conn,
"heartbeat",
None,
actor=None,
extra={"assertion": "nothing-to-report", "counts": counts},
)
conn.commit()
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
except Exception:
conn.rollback()
raise
return {"event_id": event_id, "assertion": "nothing-to-report", "counts": counts}
def transition_counts(self) -> dict[str, int]:
conn = self._conn()
try:
rows = conn.execute("SELECT class, COUNT(*) AS n FROM outbox GROUP BY class").fetchall()
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
counts = {c: 0 for c in EVENT_CLASSES}
for row in rows:
counts[row["class"]] = row["n"]
return counts
def outbox_stats(self) -> dict[str, Any]:
conn = self._conn()
now = self.now()
try:
rows = conn.execute(
"SELECT event_id, class, created_at, drained_at, attempts, last_error "
"FROM outbox"
).fetchall()
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
pending = [r for r in rows if r["drained_at"] is None]
ages = [(now - parse_iso(r["created_at"])).total_seconds() for r in pending]
return {
"total": len(rows),
"pending": len(pending),
"max_age_seconds": max(ages) if ages else 0,
"attempts": sum(int(row["attempts"]) for row in rows),
"failed_pending": sum(
1 for row in pending if row["last_error"] is not None
),
"counts": self.transition_counts(),
}
def heartbeat_due(self, interval_seconds: int) -> bool:
if interval_seconds < 1:
raise Unprocessable("heartbeat interval must be positive")
try:
row = self._conn().execute(
"SELECT MAX(created_at) AS created_at FROM outbox WHERE class='heartbeat'"
).fetchone()
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
if row["created_at"] is None:
return True
return (self.now() - parse_iso(row["created_at"])).total_seconds() >= interval_seconds
def undrained(self) -> list[dict[str, Any]]:
conn = self._conn()
try:
rows = conn.execute(
"SELECT event_id, class, approval_id, payload_json, created_at, "
"attempts, last_attempt_at, last_error "
"FROM outbox WHERE drained_at IS NULL ORDER BY created_at"
).fetchall()
except sqlite3.Error as exc:
raise StoreUnavailable(str(exc)) from exc
return [
{
"event_id": r["event_id"],
"class": r["class"],
"approval_id": r["approval_id"],
"payload": json.loads(r["payload_json"]),
"created_at": r["created_at"],
"attempts": r["attempts"],
"last_attempt_at": r["last_attempt_at"],
"last_error": r["last_error"],
}
for r in rows
]
def drain(self, sink: Callable[[dict[str, Any]], None]) -> dict[str, int]:
"""Deliver undrained payloads. Sink failure leaves the row pending.
Object mutations are already committed; this must not roll them back.
"""
delivered = 0
failed = 0
conn = self._conn()
for item in self.undrained():
attempted_at = iso(self.now())
try:
sink(item["payload"])
except Exception as exc:
failed += 1
try:
conn.execute(
"UPDATE outbox SET attempts=attempts+1, last_attempt_at=?, "
"last_error=? WHERE event_id=? AND drained_at IS NULL",
(attempted_at, type(exc).__name__, item["event_id"]),
)
conn.commit()
except sqlite3.Error as db_exc:
conn.rollback()
raise StoreUnavailable(str(db_exc)) from db_exc
continue
try:
conn.execute(
"UPDATE outbox SET drained_at=?, attempts=attempts+1, "
"last_attempt_at=?, last_error=NULL "
"WHERE event_id=? AND drained_at IS NULL",
(iso(self.now()), attempted_at, item["event_id"]),
)
conn.commit()
delivered += 1
except sqlite3.Error as exc:
conn.rollback()
raise StoreUnavailable(str(exc)) from exc
return {"delivered": delivered, "failed": failed}