Finalize native memory architecture and qualification evidence by seathatflowsinourveins · Pull Request #603 · native-agent-stack/native-agent-stack

seathatflowsinourveins · 2026-10-02T06:26:13Z

Scope

Close the native memory decision with an evidenced architecture: retain native Codex/Claude, one ai-memory owner and separate source-grounded document/code retrieval. Hindsight and other replacement backends are not selected for this deployment; no further candidate test or canary is queued by this unit. This is not a claim of universal retrieval superiority or full-stack readiness.

The package supplies pinned upstream Inspect execution, corpus/provenance/decision checks, a preserved prospective protocol and actual isolated restore evidence. Register its two dependency files in the existing OSV inventory and update only that inventory's existing evidence hash/byte count to repair the initial CI failure.

Base: b9dbe3c5a09cdefca435cd78c7f3dad46ca883a4
Lane: lane:foundation
Owned paths: blueprints/memory-stack/native-qualification/; the two additions to .github/osv-scanner-lockfiles.json; that file's existing hash/byte-count entry in manifests/evidence.json.
Global services, native authentication, hooks and deployed selection are unchanged.

SOTA sources

Inspect AI, official wheel 0.3.275: task/solver/scorer/log contracts and tests/solver/test_solver.py::test_solvers_termination. This is a scorer/importer on the upstream runner.
ai-memory, official 2.5.2 ARM64 container pinned by digest: lifecycle, backup/restore and messaging commands. Release metadata refreshed at closure.
Hindsight, official 0.10.2 ARM64 container; coding-agents, official 0.8.0 archive integrity. Source review does not establish native capture acceptance.
Official Claude ACP adapter, 0.23.1, src/acp-agent.ts: source-qualified native executable/login option, not adopted or queued by this closure. The separately completed OmniRoute SDK route is distinct.
LongMemEval-V2 informs the authored fixture categories; memory-systems characterization informs accounting boundaries. These synthetic scores are not benchmark-comparable.
The canonical docs/convergence-architecture.md and docs/decisions/2026-10-01-new-wsl-definitive-defaults.md at base b9dbe3c5 supply the separation of instruction/retrieval/state and adverse historical memory-retrieval qualifications. The existing OSV inventory format and tests.test_osv_lockfile_coverage define the dependency registration repair.

Evidence-class table

Claim	Evidence class	Receipt
Contract regressions pass	local_integration	20 tests, exit 0; `VALIDATION.md`
Inspect executes authored development data	synthetic	12 samples, no errors, all scores 1.0, zero model events
One unchanged upstream solver test passes	local_integration	Pinned upstream test, 1 pass; unavailable moto/xdist setup excluded explicitly
Wiki and DB-only handoff survive isolated restore/restart	native_proven	`runtime/observed-20261002.json`; external config restored separately
Native accounts authenticated; legacy adapter remains unqualified	source_review	`runtime/native-gates-20261002.json`, `runtime/passwordless.md`
Separately owned SDK route completed	native_proven	`closure.json`: 27/27 protocol checks and exact receipt hash; not blind memory convergence
Deployment verdict retains core without promoting challenger	source_review	`VERDICT.md`, `closure.json`; adverse retrieval evidence preserved

Local commands run

uv sync --frozen --no-build --group dev
uv run --frozen python -m pytest
uv run --frozen inspect eval qualification.py@synthetic_pipeline --model mockllm/model -T cases=corpus/dev/cases.jsonl -T sources=corpus/dev/sources.json --log-dir "$TASK_SCRATCH/inspect-synthetic" --log-format json --display plain
python3 scripts/validate.py
python3 -m unittest -v tests.test_osv_lockfile_coverage
python3 -m unittest -v tests.test_pre_push_gate
git diff --check origin/main

Initial failures remain preserved. Original full CI reported one macOS inventory failure and eight Linux failures (the same inventory failure plus seven pre-push cascades). The narrow repair passed 38 inventory tests, 16 pre-push tests at committed HEAD and repository validation, all exit 0. Current full CI is available on this PR rather than inferred from those focused results.

Decision record

blueprints/memory-stack/native-qualification/VERDICT.md and closure.json are the final decision. PREREGISTRATION.md is archived, not an active experiment. The 60-case holdout, native comparison and genuine-session canary are closed as not executed by this unit; they cannot support a replacement verdict. Future replacement requires a new explicitly scoped unit.

The owned Colima profile is stopped, rollback volumes retained, and no model/provider experiment is scheduled here. Historical failures, full-readiness limitations and unknown whole-task usage/net savings remain explicit. Other owners' work and host-specific decisions are not closed by this PR. Source integration remains with the existing owner via issue #384; a handoff is not recipient acknowledgement.

Checklist

No workflow files or global settings changed.
No secrets, personal histories or native authentication stores committed.
No new paid hosting or subscription introduced.
Peer-owned worktrees preserved; shared inventory evidence change is narrowly scoped.
The prior open-testing handoff was withdrawn; closure is recorded explicitly.

…vidence

…alification

socket-security · 2026-10-02T06:27:22Z

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	pypi/inspect-ai@0.3.275

View full report

…alification

… through #622 Each new gate row cites an in-repository evidence file that its PR added and quotes that file's own facts: - wsl-retrieval-retirement-20261003 (#622) - new-wsl-definitive-defaults-20261001 (#589, #591, #602) - new-wsl-local-model-server-20261002 (#598) - new-wsl-distro-recipe-20261002 (#593) - new-wsl-install-plan-20261002 (#606, #607) - new-wsl-client-configuration-20261002 (#608) - mac-memory-qualification-closure-20261002 (#603) - sdk-useful-task-preparation-20261002 (#609, #612) - two-host-architecture-20261002 (#610) The three lanes, the six workers and the 48 existing gates are unchanged; recorded_at_utc comes from date -u and meaning is rewritten for this checkpoint. The state.json row of manifests/evidence.json is re-registered with host_receipts.register_file (docs/lanes.md hot-file protocol). The generation holds 117 entities with the workflow adapter unconfigured, 127 with ten Dagu runs (cap 128). Co-Authored-By: Claude Opus 5.5 <[email protected]>

… through #622 (#640) Each new gate row cites an in-repository evidence file that its PR added and quotes that file's own facts: - wsl-retrieval-retirement-20261003 (#622) - new-wsl-definitive-defaults-20261001 (#589, #591, #602) - new-wsl-local-model-server-20261002 (#598) - new-wsl-distro-recipe-20261002 (#593) - new-wsl-install-plan-20261002 (#606, #607) - new-wsl-client-configuration-20261002 (#608) - mac-memory-qualification-closure-20261002 (#603) - sdk-useful-task-preparation-20261002 (#609, #612) - two-host-architecture-20261002 (#610) The three lanes, the six workers and the 48 existing gates are unchanged; recorded_at_utc comes from date -u and meaning is rewritten for this checkpoint. The state.json row of manifests/evidence.json is re-registered with host_receipts.register_file (docs/lanes.md hot-file protocol). The generation holds 117 entities with the workflow adapter unconfigured, 127 with ten Dagu runs (cap 128). Co-authored-by: Scout <scout@local> Co-authored-by: Claude Opus 5.5 <[email protected]>

seathatflowsinourveins added 9 commits October 2, 2026 01:55

Add isolated pinned memory runtime qualification recipe and restore e…

4820694

…vidence

Add synthetic native memory qualification corpus and preregistration

63fbef9

Document Inspect workflow and preserve native promotion prerequisites

29a2d30

Add pinned Inspect native-memory qualification contracts

8c3720e

Clarify frozen extension inputs and passwordless native auth boundary

347422f

Merge remote-tracking branch 'origin/main' into feat/native-memory-qu…

39ded05

…alification

Verify frozen native log snapshots and retain extension observations

15fa129

Bind extension parent provenance to frozen manifest bytes

61c7e87

Record bounded qualification acceptance and preserved failures

1eba68f

seathatflowsinourveins added the lane:foundation Foundation lane: Claude/Codex setup, hosts, memory, RAG, research, workers label Oct 2, 2026

Separate native passwordless authentication from gateway compatibility

da16256

seathatflowsinourveins mentioned this pull request Oct 2, 2026

[mac-coordinator] memory E2E: clean agentmemory 0.9.29 beside ai-memory, C4/D2h arms and token-efficiency measurement #384

Open

2 tasks done

seathatflowsinourveins added 3 commits October 2, 2026 10:33

Merge remote-tracking branch 'origin/main' into feat/native-memory-qu…

1c6105b

…alification

Register native qualification dependencies in OSV inventory

9a7d7f1

Close memory architecture decision with evidence and no queued trials

9ac024a

seathatflowsinourveins changed the title ~~Add native memory qualification with Inspect and isolated restore evidence~~ Finalize native memory architecture and qualification evidence Oct 2, 2026

seathatflowsinourveins added 2 commits October 2, 2026 11:23

Use classified retention status in closed verdict

eca7903

Merge current main and refresh qualification evidence registration

1d35742

seathatflowsinourveins merged commit b507989 into main Oct 2, 2026
25 checks passed

seathatflowsinourveins deleted the feat/native-memory-qualification branch October 2, 2026 19:29

seathatflowsinourveins mentioned this pull request Oct 3, 2026

Grand dashboard checkpoint 2026-10-03: nine gate rows for work merged through #622 #640

Merged

5 tasks done

Navigation Menu

Repository navigation

Finalize native memory architecture and qualification evidence - #603

Finalize native memory architecture and qualification evidence#603
seathatflowsinourveins merged 15 commits into
mainfrom
feat/native-memory-qualification

seathatflowsinourveins commented Oct 2, 2026 •

edited

Loading

Uh oh!

socket-security Bot commented Oct 2, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

seathatflowsinourveins commented Oct 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Scope

SOTA sources

Evidence-class table

Local commands run

Decision record

Checklist

Uh oh!

socket-security Bot commented Oct 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

seathatflowsinourveins commented Oct 2, 2026 •

edited

Loading

socket-security Bot commented Oct 2, 2026 •

edited

Loading