rdmsm4x-changelog-20260902-2329-replicantdb-memory-bound-checkpoint

Checkpointed a measured ReplicantDB GUI-memory fix in an isolated worktree; passive status refresh no longer hydrates more than one million duplicate-member records, while fleet build/deployment remains held behind the serialized portfolio queue and a SwiftPM test-harness self-lock.

Scope

• Host changed: rdmsm4x only.
• Source worktree: /Users/richh/dev/_handoff/codex-out/replicantdb-memory-bounds-20260902.
• Branch/commit: codex/replicantdb-memory-bounds-20260902 at 259c7917539ff2be678505789a4ed16a63f30678.
• Ticket: ISSUE-20260901-26, lease retained by codex@rdmsm4x/replicantdb-memory-bounds.
• Release-gate defect filed as ISSUE-20260902-20.
• Canonical checkout, installed applications, live databases, daemon registrations, TCC settings, and the other five hosts were not changed.

Files changed

• Sources/ReplicantDBCore/Database.swift
• Sources/replicantDB/ReplicantDBApp.swift
• Tests/ReplicantDBTests/DuplicateStatisticsMemoryBoundTests.swift
• SESSION-STATE.md
• ISSUES.md
• /Users/richh/.agent-coordination/checkins/codex-rdmsm4x-replicantdb-memory-bounds-20260902.json
• Fleet coordination message 20260902-232834-F59094AF

What changed

• Replaced getStats() duplicate object hydration with one SQLite grouped aggregate.
• Removed duplicate-member loading from startup and four-second Status telemetry.
• Added lazy browsing limits of 250 groups and 100 hydrated members per group, while retaining complete group counters.
• Added exact-hash complete-group resolution immediately before an explicit per-group deduplication action.
• Moved GUI-triggered duplicate reads/actions off the MainActor.
• Added four isolated regression tests for aggregate semantics, liveness/zero-byte exclusions, preview limits, and full-group action resolution.
• Recorded Rich's next-rollout cleanup requirement: after canary acceptance, retain one active revision plus one rollback package per host, archive every other old app/CLI/MCP/staging copy outside launch/search paths, and prove that with a second six-host inventory. Preserve the shared code-identity TCC grants because they are not version-specific.

Commands and verification

• swift test --filter DuplicateStatisticsMemoryBoundTests — 4 executed, 0 failures; app target compiled under Swift 6.
• swift test — 1,095 executed, 1 failure, 1,094 passed, 447.712 seconds.
• sample 88618 1 1, sample 89716 1 10, and lsof on the worktree build lock proved the sole failure was a nested SwiftPM self-lock: the outer test process held .build.lock while LocalIntelligenceTests launched swift package describe --type json in the same worktree.
• kill -TERM 89716 terminated only the deadlocked child so the suite could record the failure and complete.
• git diff --check passed before the explicit-path commit.
• No ./build.sh, signing, installation, process restart, live-data command, scan, OCR, classification, hashing, tagging, deduplication, or file move was run.

Preservation and undo

• The complete source checkpoint is commit 259c7917539ff2be678505789a4ed16a63f30678 in the isolated worktree; the deployed 1.19.0 (34) fleet remains unchanged with its existing per-host rollback archives.
• To discard only this candidate before integration, delete the linked worktree/branch after first preserving the commit reference. Do not reset or clean the dirty canonical checkout.
• If a later deployed build fails, use the per-host rollback workflow recorded in deployment-evidence/REPLICANTDB-1.19.0-BUILD34-FLEET-ROLLOUT-20260902.md; no new rollback payload exists yet because no new artifact was built.

Outstanding owner/queue actions

• Wait for the serialized portfolio coordinator to release the ReplicantDB heavyweight lane.
• Repair the nested SwiftPM source-audit test without weakening its dependency coverage, then rerun the full suite.
• If and only if that gate is green: bump 1.19.1 (35), build/sign universal2, verify the artifact, canary, deploy all six hosts, perform the requested old-copy archive sweep, and collect a post-install memory curve.
• Do not resolve ISSUE-20260901-26 until comparable long-duration runtime evidence exists; the separate Vision/IOSurface worker-retention risk also remains open for follow-up.