REG-019 — DOCUMENTED: ledger compaction + cold tiering does not exist;
recorded as roadmap rather than fabricated or left silently absent.

=== Search for an existing mechanism ===

$ grep -rli "compaction\|cold.tier\|cold storage" aegis/
aegis/proxy/app.py               (unrelated: WAF pattern name)
aegis/config.py                  (unrelated: config field name substring)
aegis/connectors/lakehouse/parquet_exporter.py   (unrelated: PyArrow engine compaction, not WAL)
aegis/storage/s3_worm.py         (unrelated match)
aegis/core/mifid_record_keeper.py (unrelated match)
aegis/core/archival_bundle.py    (unrelated: this is algorithm-agile long-term
                                   hash/signature migration for sealed bundles,
                                   not WAL segment compaction or storage tiering)

No module implements compacting the JSONL WAL (removing/merging old segments
while preserving hash-chain and MMR-proof reachability) or moving older
records to a cheaper storage tier.

=== Confirmed the gap is not already claimed elsewhere ===

$ grep -n -i "compaction\|cold.tier\|tiering" docs/CLAIMS_MATRIX.md
139: CLM-071 mentions "compaction" only for PyArrow/Parquet engine-side
     encoding in the lakehouse *connector* (a derivative copy, explicitly
     stated as never the evidence authority) — unrelated to the primary WAL.

$ grep -n -i "compaction\|cold.tier\|tiering" ROADMAP.md docs/ROADMAP.md
(no matches before this session's edit)

docs/architecture/DEEP_DIVE.md §5 "Evidence Export & Long-Term Retention"
lists archival_bundle.py's algorithm-agility mechanism but nothing about
segment compaction or tiering.

docs/operations/STORAGE_REQUIREMENTS.md's only relevant guidance: "Provision
so the segment cannot reach capacity during a retention window" — i.e. the
operational answer today is "give it enough disk," which is honest but
means the absence of compaction is a real operational cost, not a defect
that silently misleads (the storage doc already tells operators to plan for
unbounded growth).

=== Disposition ===

This is a genuine absent capability, not a bug and not a false claim to
retract. Building it (compacting a hash-chained, MMR-proof-backed WAL
without breaking inclusion-proof reachability for archived leaves, plus a
cold-tier storage adapter) is substantial design work — not a bounded fix
AGENTS.md rule 2's "smallest authorized change" would license unilaterally.

Recorded as `[ ]` open work in `docs/ROADMAP.md`'s "P2 performance and
operations" section (matching REG-019's own P2 severity) and as a new
paragraph in the summary `ROADMAP.md`'s "Evidence and storage" section,
alongside the existing "Durable WAL backend options" entry it's adjacent to
in scope. Both entries state the MMR-proof-reachability constraint any
future implementation must satisfy, so this isn't a bare "not done" note.

No code change. Closed DOCUMENTED rather than WONT-FIX: this is plausible
future work with no stated reason to reject it, not a deliberately declined
feature.
