audit-core/audit_core/stream_findings.py

132 lines
5.7 KiB
Python
Raw Normal View History

AUDIT-WP-0009 T04/T06/T07 — heartbeats, reconciliation, and a home for findings The detection half audit-core argued up to a MUST and then could not support. Two registered sources were waiting on it. T04, heartbeats. A heartbeat is an ordinary event — same envelope, same append-only custody, same chain, no special table. Deliberate: a heartbeat stored outside the chain would be the one record here that could be back-dated. Declared per class rather than per source, because a per-source heartbeat from a mixed-volume emitter is satisfied by its chattiest class and says nothing about the quiet, security-relevant one, which is the only reason heartbeats exist. Not the §17 cadence schema T05 waits on: cadence describes expected rate, this says how often a source promises to say "nothing to report" for a class that may legitimately be silent. no_heartbeat_since_registration is its own finding kind rather than a skip — it is the case most likely to be a broken integration and the one a naive "compare against last seen" implementation silently drops. Grace widens the window so one late run does not flap; it never removes a finding. T06, reconciliation. Counts, never payloads. The awkward part is that every registered sender holds may_read: false, which taken literally makes the §9.6 reconciliation obligation undischargeable by every source actually registered. Resolved by observing that a source asking how many of its own events we hold is not reading the archive — it learns nothing it did not itself emit. So the surface is scoped to the caller's own sources and tenants and returns no payloads; anything wider stays behind may_read and full tenant scope. Another source's counts return 403 rather than an empty count, because a zero would read as "we hold none of yours" — a false answer to a question about completeness. No default window, since a count whose bounds the caller did not choose is not comparable to anything the caller computed. T07, the findings surface. /v1/stream-findings, following the dead-letter and secret-finding conventions: may_read plus full tenant scope, since findings span every sender and carry no tenant key to filter on. The bound is on every response rather than in a document nobody opens beside it. A missing heartbeat is not proof of suppression, and agreement on counts proves neither completeness nor that any event occurred. Both controls cover loss, outage, drain failure and accident; neither covers a source lying about itself, and where the emitter is compromised both agree with it. Closing that needs an observer independent of the emitter, which §16 put outside our scope. The scope overlay may shorten a heartbeat interval or add a class, never lengthen or remove one — same asymmetry as evidence_kind, and for the same reason: a ConfigMap refresh must not widen the window in which a suppressed class goes unnoticed without anyone deciding to. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Nb7Q6ZmXppNDkTWytfYqfv Assistant: claude-code Assistant-Model: opus Assistant-Process: 2069992@bnt-lap001 Assistant-Session: 167dd7f8-2a25-4be1-aa46-3b6f1a5f94c6
2026-09-10 16:43:26 +02:00
"""Stream-completeness findings: when the trail itself is the defect.
`AUDIT-WP-0009-T04` and `T07`. `INTENT.md` principle 10 already says a degraded
audit stream is itself an audit and operations event. The principle was in
place; the mechanism was not.
**What this can and cannot see.** The hash chain proves records held were not
altered or truncated. It says nothing about a record that never arrived, and
§9.6 is explicit that omission is the acute risk for exactly the rare negative
classes revocation, denial, containment where suppression is most valuable
and least visible. Rate monitoring cannot help there either: a class that is
legitimately silent for a month is indistinguishable from one being suppressed.
A heartbeat fixes that by inverting the burden. Instead of inferring health
from events that may never come, the source makes a positive claim
*nothing to report for this class* on a declared interval. The claim can
itself go missing, and a missing claim is a finding. That is the whole idea,
and it is the only control here that covers adversarial omission.
**The bound, stated because a finding surface invites over-reading.** A
compromised source emits a truthful-looking heartbeat while suppressing the
event it is supposed to be vouching for. Heartbeats cover loss, outage, drain
failure and accident most of what actually goes wrong and do not cover a
source lying about itself. Nothing an archive holds can close that; it needs an
observer independent of the emitter.
"""
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from typing import Any, Iterable
# The event class a source uses to say "nothing to report". A heartbeat is an
# ordinary event: same envelope, same append-only custody, same chain. It gets
# no special table, because a heartbeat that lived outside the chain would be
# the one record in this store that could be back-dated.
HEARTBEAT_ACTION = "audit-core.heartbeat"
MISSING_HEARTBEAT = "missing_heartbeat"
NEVER_HEARTBEAT = "no_heartbeat_since_registration"
@dataclass(frozen=True)
class StreamFinding:
"""One reason to believe a stream is not telling the whole truth."""
kind: str
sender: str
source: str
event_class: str
expected_within_seconds: int
last_seen: str | None
overdue_seconds: int
def as_dict(self) -> dict[str, Any]:
return {
"kind": self.kind,
"sender": self.sender,
"source": self.source,
"class": self.event_class,
"expected_within_seconds": self.expected_within_seconds,
"last_seen": self.last_seen,
"overdue_seconds": self.overdue_seconds,
# Said on every finding rather than in a document nobody opens
# alongside it: this is the absence of a positive claim, not proof
# that an event was suppressed.
"means": (
"a declared heartbeat did not arrive; the class may be healthy "
"and the emitter silent. Absence is not proof of suppression."
),
}
def _parse(value: str | None) -> datetime | None:
if not value:
return None
try:
parsed = datetime.fromisoformat(str(value).replace("Z", "+00:00"))
except ValueError:
return None
return parsed if parsed.tzinfo else parsed.replace(tzinfo=timezone.utc)
def evaluate(
identities: Iterable[Any],
last_heartbeats: dict[tuple[str, str], str | None],
*,
now: datetime | None = None,
grace_factor: float = 1.5,
) -> list[StreamFinding]:
"""Findings for every declared heartbeat class that is overdue.
``last_heartbeats`` is keyed ``(source, class)``. A class with no heartbeat
ever recorded produces a finding of its own kind rather than being skipped:
a source that declared a heartbeat and never sent one is the case most
likely to be a broken integration, and it is precisely the one a
"compare against last seen" implementation silently drops.
``grace_factor`` exists so one late run does not flap a finding on and off,
matching the reasoning behind the attestation freshness window. It widens
the window; it never removes the finding.
"""
now = now or datetime.now(timezone.utc)
findings: list[StreamFinding] = []
for identity in identities:
for event_class, interval in getattr(identity, "heartbeat_classes", ()):
for source in sorted(identity.sources):
if source == "*":
# A wildcard source cannot be held to a heartbeat: there is
# no determinate set of streams to expect one from. Refused
# rather than guessed at.
continue
deadline = timedelta(seconds=interval * grace_factor)
last = _parse(last_heartbeats.get((source, event_class)))
if last is None:
findings.append(StreamFinding(
kind=NEVER_HEARTBEAT, sender=identity.name, source=source,
event_class=event_class, expected_within_seconds=interval,
last_seen=None, overdue_seconds=-1,
))
continue
overdue = (now - last) - deadline
if overdue.total_seconds() > 0:
findings.append(StreamFinding(
kind=MISSING_HEARTBEAT, sender=identity.name, source=source,
event_class=event_class, expected_within_seconds=interval,
last_seen=last.isoformat(),
overdue_seconds=int(overdue.total_seconds()),
))
return findings