Skip to content

fix(sources): widen 52l2 guard + reunify retired siblings live #3236

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Sinity merged 2 commits into from
Jul 21, 2026
Merged

fix(sources): widen 52l2 guard + reunify retired siblings live #3236

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7d1c2b2
fix(sources): widen 52l2 guard to recognize legacy retirement detail
Sinity Jul 21, 2026
4c516ec
fix(sources): reunify retired revision siblings in the live watcher
Sinity Jul 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 22 additions & 0 deletions polylogue/archive/revision_authority.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,28 @@ class RawRevisionAuthority(StrEnum):
#: later-discovered raw as an unconditional singleton byte-proven baseline.
HISTORICAL_NON_PREFIX_GOVERNANCE_DETAIL = "historical non-prefix full revision governance"

#: All ``raw_membership_census.detail`` marker literals that mean "this raw
#: was retired from full-revision byte governance to membership governance"
#: (polylogue-hm2f, residual of polylogue-52l2). Durable ``raw_membership_
#: census`` rows written before the #3234 fix used the literal
#: ``"cross-route full revision governance"`` at the live-watcher call site
#: (``sources/live/batch.py``, pre-fix) -- a DIFFERENT string from
#: ``HISTORICAL_NON_PREFIX_GOVERNANCE_DETAIL``, which only the offline
#: backfill call site used at the time. Those pre-fix rows are durable
#: (``source.db``) and were never rewritten in place (durable-tier changes
#: need an explicit additive migration, not a silent detail-string rewrite),
#: so identities retired under the legacy literal remain invisible to a guard
#: query keyed on the new marker alone. ``ArchiveStore.raw_membership_
#: retired_full_revision_siblings`` matches against every literal in this
#: tuple so old and new retirements are both recognized. The legacy literal
#: is frozen here for read-compatibility only -- it must never be written by
#: new code (both current call sites use ``HISTORICAL_NON_PREFIX_GOVERNANCE_
#: DETAIL`` exclusively).
RETIRED_FULL_REVISION_GOVERNANCE_DETAILS: tuple[str, ...] = (
HISTORICAL_NON_PREFIX_GOVERNANCE_DETAIL,
"cross-route full revision governance", # legacy pre-#3234 literal -- read-only, never write.
)


@dataclass(frozen=True)
class RawRevisionEnvelope:
Expand Down
128 changes: 101 additions & 27 deletions polylogue/sources/live/batch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1823,36 +1823,92 @@ def _ingest_full_records_archive(
),
)
plan = archive.classify_raw_revision_cohort(logical_source_key)
if not plan.accepted_raw_ids:
# classify_raw_revision_cohort is a synchronous,
# purely byte-level classification (no async
# conveyor revisits it) -- an empty cohort here
# is a genuine, terminal rewrite/divergence
# conflict, not a pending hand-off. Leave this
# raw out of both raw_ids and deferred_raw_ids so
# the caller's aggregation counts it as a real
# failure (fail-closed), with a log line so the
# cause is diagnosable.
logger.warning(
"live.watcher: no unique byte-revision candidate accepted for %s "
"(logical_source_key=%s) -- surfacing as failed",
if plan.accepted_raw_ids:
parsed_by_raw_id = self._parse_raw_revision_chain(archive, plan)
session_id, applied_raw_ids = archive.apply_raw_revision_replay(
plan,
parsed_by_raw_id,
acquired_at_ms=acquired_at_ms,
stage_timings_s=record_timings,
stage_timing_prefix="full",
)
record_session_ids.append(session_id)
record_session_count = 1
record_message_count = sum(
len(parsed_by_raw_id[raw_id].messages) for raw_id in applied_raw_ids
)
else:
# polylogue-hm2f (live-path half of the
# polylogue-52l2 guard): an empty cohort here
# can mean two different things.
#
# (1) This identity has NO retired sibling
# evidence -- a genuine, terminal
# rewrite/divergence conflict between raws
# that are all still live 'full' rows. There
# is nothing this tick can safely resolve;
# leave this raw out of both raw_ids and
# deferred_raw_ids so the caller's aggregation
# counts it as a real failure (fail-closed),
# with a log line so the cause is diagnosable.
#
# (2) This identity DOES have retired sibling
# evidence (raw_membership_retired_full_
# revision_siblings is non-empty): the 52l2
# guard is exactly what emptied this cohort.
# Offline backfill reunites retired siblings
# with a newly discovered raw via its own
# connected-component/membership_candidates
# bookkeeping across the whole census pass;
# the live incremental path processes one
# record per tick with no such durable
# bookkeeping, so it re-derives the sibling
# set from the same durable marker the guard
# itself reads. Fold this raw into membership
# governance -- the same call sequence used
# above for pre-existing-membership and
# multi-session identities -- and pass the
# retired siblings in explicitly so the real
# content-prefix classifier
# (classify_membership_revisions) weighs this
# raw against every known sibling instead of
# evaluating it alone. If the siblings turn
# out to genuinely disagree, that classifier
# still refuses to pick a winner (ambiguous
# quarantine), so this can only ever recover a
# real resolution, never fabricate one.
retired_siblings = archive.raw_membership_retired_full_revision_siblings(
logical_source_key
)
if not retired_siblings:
logger.warning(
"live.watcher: no unique byte-revision candidate accepted for %s "
"(logical_source_key=%s) -- surfacing as failed",
record.source_path,
logical_source_key,
)
continue
logger.info(
"live.watcher: reunifying %s with %d retired sibling(s) under membership "
"governance (logical_source_key=%s)",
record.source_path,
len(retired_siblings),
logical_source_key,
)
continue
parsed_by_raw_id = self._parse_raw_revision_chain(archive, plan)
session_id, applied_raw_ids = archive.apply_raw_revision_replay(
plan,
parsed_by_raw_id,
acquired_at_ms=acquired_at_ms,
stage_timings_s=record_timings,
stage_timing_prefix="full",
)
record_session_ids.append(session_id)
record_session_count = 1
record_message_count = sum(
len(parsed_by_raw_id[raw_id].messages) for raw_id in applied_raw_ids
)
(
record_session_ids,
record_session_count,
record_message_count,
raw_authority_complete,
) = self._apply_membership_sessions(
archive,
source_raw_id,
sessions,
acquired_at_ms=acquired_at_ms,
stage_timings_s=record_timings,
allow_current_complete_raw=True,
extra_member_raw_ids=retired_siblings,
)
else:
archive.replace_raw_membership_census(
source_raw_id,
Expand Down Expand Up @@ -1958,7 +2014,22 @@ def _apply_membership_sessions(
acquired_at_ms: int,
stage_timings_s: dict[str, float] | None = None,
allow_current_complete_raw: bool = False,
extra_member_raw_ids: tuple[str, ...] = (),
) -> tuple[list[str], int, int, bool]:
"""Apply membership-governed classification for one logical identity.

``extra_member_raw_ids`` (polylogue-hm2f) names raws that carry known
membership evidence for this identity but are not currently
discoverable through ``archive.raw_membership_raw_ids`` -- concretely,
raws previously retired from full-revision byte governance
(``raw_membership_retired_full_revision_siblings``): their
``raw_session_memberships`` row survives retirement with
``revision_authority`` left at the default ``'quarantined'``, so the
ordinary byte-proven-or-caller-owned query never surfaces them. Forced
inclusion here is what lets a later-discovered raw be weighed by the
real content-prefix classifier against siblings the 52l2 guard is
aware of instead of being evaluated alone.
"""
session_ids: list[str] = []
session_count = 0
message_count = 0
Expand Down Expand Up @@ -1996,6 +2067,9 @@ def _apply_membership_sessions(
include_complete_raw_id=source_raw_id if allow_current_complete_raw else None,
)
)
for extra_raw_id in extra_member_raw_ids:
if extra_raw_id not in member_raw_ids:
member_raw_ids.append(extra_raw_id)
accepted_head_raw_id = archive.raw_revision_head_raw_id(logical_source_key)
if accepted_head_raw_id is not None and accepted_head_raw_id not in member_raw_ids:
member_raw_ids.append(accepted_head_raw_id)
Expand Down
17 changes: 13 additions & 4 deletions polylogue/storage/sqlite/archive_tiers/archive.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,7 @@
QueryTextPredicate,
)
from polylogue.archive.revision_authority import (
HISTORICAL_NON_PREFIX_GOVERNANCE_DETAIL,
RETIRED_FULL_REVISION_GOVERNANCE_DETAILS,
HistoricalRawRevisionStream,
RawRevisionAuthority,
RawRevisionEnvelope,
Expand Down Expand Up @@ -2003,18 +2003,27 @@ def raw_membership_retired_full_revision_siblings(self, logical_source_key: str)
transition, so a later-arriving sibling for the same identity can
still be told this identity has known, unresolved ambiguous
evidence (polylogue-52l2) instead of being evaluated alone.

Matches every literal in ``RETIRED_FULL_REVISION_GOVERNANCE_DETAILS``
(polylogue-hm2f), not only the current
``HISTORICAL_NON_PREFIX_GOVERNANCE_DETAIL`` marker: durable
``raw_membership_census`` rows written before #3234 used a different,
now-legacy literal at the live-watcher call site, and durable-tier
detail strings are never silently rewritten in place.
"""
detail_placeholders = ", ".join("?" for _ in RETIRED_FULL_REVISION_GOVERNANCE_DETAILS)
where_clause = f"c.detail IN ({detail_placeholders})"
rows = (
self._ensure_source_conn()
.execute(
"""
f"""
SELECT m.raw_id
FROM raw_session_memberships AS m
JOIN raw_membership_census AS c ON c.raw_id = m.raw_id
WHERE m.logical_source_key = ? AND c.detail = ?
WHERE m.logical_source_key = ? AND {where_clause}
ORDER BY m.raw_id
""",
(logical_source_key, HISTORICAL_NON_PREFIX_GOVERNANCE_DETAIL),
(logical_source_key, *RETIRED_FULL_REVISION_GOVERNANCE_DETAILS),
)
.fetchall()
)
Expand Down
Loading