[trissim]

Defect

BusProjectionTables.create_schema rebuilds every bus projection whenever the Core commit changes. Two parts of that replay did needless work:

1. Per-row reads. A header stream, then kind.read_from(db, seq) issued one point SELECT per capability table for every record, each through the strict decode boundary.
2. Claim rereads. Every replayed claim called HeldResources.claims_for. That reselected and fully decoded the current holders' Claimed rows, which the same in-order pass had already decoded.

Fix

Commit 1: one joined cursor.

• BusCapability.history(db) reads every record in sequence order from one cursor: the history header LEFT JOINed to each capability table on seq. The table set is derived from TypedTable.members_with(CapabilityRow). Each stored row is decoded once, at its column offset.
• read_from and history share one assembly step, BusCapability.assembled.
• TypedTable.qualified() and decode_at() let one select join tables that share column names and decode each table's slice.

Commit 2: carried claim holders.

• HeldResources.project and apply take a ClaimHolders member.
• StoredHolders reads the holders of one live append from their table. That read is required: each append is its own transaction.
• ReplayedHolders, owned by the create_schema pass, carries the owners it already projected and never rereads them.
• apply stays the one owner of which resources leave, and hands that set to the holders.

Deleted

• The per-row BusRow.stream + header.kind.read_from(db, header.seq) replay in create_schema.
• The duplicated record assembly in read_from.
• The claims_for reread during the rebuild.
• Claimed.replay_claim, folded into project_into.

Not changed

• ProjectionProducer.records_for: the commit stamp is kept. Stamp derived projections and index answer eligibility at append #2636 replaced per-class producer identities with it so that derived rows can't survive a change to the code that produces them while the SQL stays equal. Those producers are selected at runtime by role, type and reply scope (messages.py:413-414, bus_publication.py:57, read_basis.py:866), so a per-declaration stamp would need a hand-written producer list.
• No cache and no second reader.

Measurements

All runs were niced, on backup-API copies of the live bus.sqlite3. Each deleted every projection table, replayed all records and rolled back.

Path	Records	Seconds
Per-row read_from (main)	19,448	46.5
Joined cursor (commit 1)	19,448	37.7
Joined cursor + stored holders (commit 1 shape)	19,457	41.5
Joined cursor + carried holders (commit 2)	19,457	34.5

• The two 19,457-record runs come from one process on one copy, and every projection table's full contents are identical.
• The 19,448-record runs compared row counts and decoded records, and both were equal.
• Where the time goes (cProfile; the run took 109 s under the profiler):
  • All projection inserts together take about 6 s. received_source, at 96,786 rows, is 2.7 s of that.
  • The rest is decoding records (FieldCodec) and replaying claim transitions.

No pytest was run here; the integrator tests this once on integration. tests/test_bus_resource_claims.py:188 now replays through ReplayedHolders and compares against the live projection.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[github-actions]

debt-src at 5d5290c, every finding:

grew GodClassExcess:src/agent_comms/typed_table.py::TypedTable +15
grew LintFindings:src/agent_comms/bus_tables.py +5
grew LintFindings:src/agent_comms/typed_table.py +2
grew TypeErrors:src/agent_comms/bus_tables.py +2