research: evaluate observed co-access signals for context recommendations

[dapi]

Question

After Execution Handoffs have accumulated, can observed co-access patterns safely improve context recommendations for a concrete task?

Scope

Evaluate signals from completed handoffs, such as documents, decisions, changed files, commits, and verification artefacts that repeatedly occur together for the same type of work.

The objective is to learn whether these signals can form useful task topics and recommend a compact context set for the next run.

Non-goals

• automatic document merging;
• changing canonical ownership or SSOT;
• treating inferred semantic similarity as fact;
• choosing a graph database before the data proves the need.

Deliverable

A short evidence-backed decision record that answers:

1. which signals are reliable enough to use;
2. whether temporal proximity and shared artefacts improve recommendations;
3. how to retain source provenance and user control;
4. whether a follow-up implementation issue is justified.

Dependencies

• docs: define evidence-backed execution handoff contract #95
• feat: build a read-only execution handoff from explicit project evidence memory-bank-cli#47

[dapi]

Human Gate: the research package is Decision Ready at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da, but two required predicates remain blocked.

1. The named decision owner (dapi/memory-bank maintainers) must review the evidence-backed recommendation and record or authorize the terminal disposition. The candidate recommends inconclusive: provenance-preserving, user-controlled co-access signals are safe to evaluate, but no implementation issue is justified until docs: define evidence-backed execution handoff contract #95 and feat: build a read-only execution handoff from explicit project evidence memory-bank-cli#47 are delivered and a representative completed-handoff corpus exists.
2. This runtime has worktree write but .git read-only, so it cannot create codex/gh-96, commit, push, or open the required PR. No Git mutation was attempted.

Current safe checkpoint:

• memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md: passed
• memory-bank-cli doctor --profile downstream: passed (one informational lock note)
• git diff --check: passed
• worktree: memory-bank/research/README.md modified; memory-bank/research/R-GH-96/ untracked

Resume this same run after a maintainer disposition is available and in a full-delivery profile with .git write plus authenticated GitHub write. Then re-ground sources/HEAD, apply the authorized disposition, create and push codex/gh-96, open the PR, report verification here, and replace codex-ready with human-review.

[dapi]

Human Gate: the evidence-backed research package is clean at the Decision Ready checkpoint, but two predicates block its terminal disposition and repository delivery.

1. The named decision owner (dapi/memory-bank maintainers) must authorize and record the terminal disposition in memory-bank/research/R-GH-96/brief.md. The recommendation remains inconclusive: direct, provenance-preserving handoff membership may be evaluated as optional, user-controlled candidate context; co-access, temporal proximity, and shared artefacts are not yet demonstrated to improve recommendations. Do not open an implementation issue now.
2. This run has worktree write but restricted .git write (HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da), so it cannot create codex/gh-96, commit, push, or open the required PR. No Git mutation was attempted.

Freshness update: dapi/memory-bank-cli#47 now has open, unmerged PR #49 with passing checks; it remains implementation-candidate evidence only and does not provide completed handoffs or a representative corpus.

Verification performed:

• memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md
• memory-bank-cli doctor --profile downstream (0 errors, 0 warnings, one informational lock note)
• git diff --check and untracked-document whitespace checks
• artifact traceability/lifecycle review of the Research Flow package
• attempted codex review --uncommitted; the restricted environment denied its app-server operation

Resume this same run after a maintainer disposition and in a full-delivery profile with .git write. Re-ground the sources/HEAD, apply the authorized status, then commit/push and open the PR; afterwards report the PR and verification here and replace codex-ready with human-review.

[dapi]

Human Gate (re-grounded 2026-07-31): R-GH-96 remains Decision Ready. The evidence supports an inconclusive disposition: direct, provenance-preserving handoff membership may be evaluated only as optional, user-controlled candidate context; co-access, temporal proximity, and shared artefacts are not demonstrated to improve recommendations. Do not open an implementation issue yet.\n\nCurrent dependency evidence: #95 and dapi/memory-bank-cli#47 remain open; cli PR #49 remains open and unmerged, so no representative completed-handoff corpus exists.\n\nRequired to resume: (1) a dapi/memory-bank maintainer records or authorizes the inconclusive terminal disposition in memory-bank/research/R-GH-96/brief.md; and (2) this same run is continued with .git write and authenticated GitHub write to commit/push codex/gh-96, open the PR, report verification, and replace codex-ready with human-review.\n\nCheckpoint: HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md, memory-bank-cli doctor --profile downstream, and git diff --check passed. codex review --uncommitted is unavailable in this restricted environment because its app-server client cannot initialize.

[dapi]

Human Gate (re-grounded 2026-07-31): R-GH-96 remains at Decision Ready. No maintainer terminal disposition has been recorded, so brief.md cannot safely move from decision_ready to inconclusive without its named decision owner. The recommendation is unchanged: provenance-preserving, user-controlled candidate context may be evaluated later, but co-access, temporal proximity, and shared artefacts are not demonstrated to improve recommendations; do not open an implementation issue now.\n\nDependency evidence is unchanged: #95 and dapi/memory-bank-cli#47 are open; CLI PR #49 is open and unmerged (completed non-publish checks succeeded), so there is no representative completed-handoff corpus.\n\nThis runtime has worktree write but restricted .git write at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; no branch, commit, push, PR, or label mutation was attempted. Resume only after (1) a maintainer authorizes or records the terminal inconclusive disposition and (2) the same run has full Git delivery capability. Then re-ground, apply the disposition, commit/push codex/gh-96, open the PR, report verification, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding (2026-07-31). The Research & Discovery package is still Decision Ready, but no dapi/memory-bank maintainer has authorized or recorded the required terminal inconclusive disposition in memory-bank/research/R-GH-96/brief.md.\n\nDependency evidence is unchanged: #95 and dapi/memory-bank-cli#47 remain open; CLI PR #49 is open, clean, and unmerged, with successful non-publish checks. This does not provide a completed-handoff corpus.\n\nThis workspace has worktree write but restricted .git write at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; no branch, commit, push, PR, or label mutation was attempted. git diff --check passed.\n\nResume only after (1) a maintainer records or authorizes research_status: inconclusive and (2) the same run has full .git write plus authenticated GitHub write. Then re-ground sources and HEAD, apply the disposition, create/push codex/gh-96, open the PR, report verification here, and replace codex-ready with human-review.

[dapi]

Human Gate (re-grounded 2026-08-01): the Research & Discovery package remains at the Decision Ready checkpoint. The evidence-backed recommendation is inconclusive: direct, provenance-preserving handoff membership may later be evaluated as optional, user-controlled candidate context, but co-access, temporal proximity, and shared artefacts are not demonstrated to improve recommendations. Do not open an implementation issue now.

Current dependency evidence: #95 and dapi/memory-bank-cli#47 remain open; CLI PR #49 is open and unmerged. There is still no representative completed-handoff corpus.

Validation passed: memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md, memory-bank-cli doctor --profile downstream (0 errors, 0 warnings, one informational lock note), and git diff --check. codex review --uncommitted remains unavailable because this restricted runtime cannot initialize its app-server client.

Required to resume: (1) a dapi/memory-bank maintainer records or authorizes research_status: inconclusive in memory-bank/research/R-GH-96/brief.md; and (2) continuation in a full-delivery profile with .git write. This workspace is at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; no branch, commit, push, PR, or label mutation was attempted. After both conditions, re-ground sources and HEAD, apply the disposition, create/push codex/gh-96, open the PR, report verification here, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding (2026-08-01). The Research & Discovery package is still Decision Ready: the named decision owner has not recorded or authorized the required terminal research_status: inconclusive in memory-bank/research/R-GH-96/brief.md.

Dependency evidence is unchanged: #95 and dapi/memory-bank-cli#47 are open; CLI PR #49 is open, clean, and unmerged. No representative completed-handoff corpus exists.

This runtime restricts .git write at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da, so it cannot create codex/gh-96, commit, push, or open the required PR. The worktree package remains unchanged; git diff --check passed.

Resume this same run only after (1) a dapi/memory-bank maintainer authorizes or records the terminal disposition and (2) continuation has full .git write plus authenticated GitHub write. Then re-ground, apply the disposition, deliver the branch/PR, report verification, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding (2026-08-01). The Research & Discovery package is still at Decision Ready: no dapi/memory-bank maintainer has authorized or recorded the required terminal research_status: inconclusive in memory-bank/research/R-GH-96/brief.md.

Dependency evidence remains unchanged: #95 and dapi/memory-bank-cli#47 are open; CLI PR #49 is open, unmerged, and has only completed non-publish checks. There is no representative completed-handoff corpus.

This runtime permits worktree writes but restricts .git writes at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; therefore it cannot create codex/gh-96, commit, push, or open the required PR. No Git delivery mutation was attempted. Rechecked successfully: memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md, memory-bank-cli doctor --profile downstream (0 errors, 0 warnings, one informational lock note), and git diff --check.

Resume this same run only after (1) a maintainer authorizes or records the terminal disposition and (2) continuation has full .git write plus authenticated GitHub write. Then re-ground sources and HEAD, apply the disposition, deliver the branch/PR, report verification, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding. R-GH-96 is still at the Research & Discovery Flow Decision Ready checkpoint: the named decision owner has not authorized or recorded the required terminal research_status: inconclusive in memory-bank/research/R-GH-96/brief.md.

The evidence remains inconclusive: direct, provenance-preserving handoff membership may later be evaluated as optional user-controlled candidate context, but co-access, temporal proximity, and shared artefacts are not demonstrated to improve recommendations. No implementation issue should be opened now. Dependencies #95 and dapi/memory-bank-cli#47 remain open; CLI PR #49 is open and unmerged, despite successful completed non-publish checks, so no representative completed-handoff corpus exists.

This runtime has worktree write but restricted .git write at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; no branch, commit, push, PR, or label mutation was attempted. Rechecked successfully: memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md, memory-bank-cli doctor --profile downstream (0 errors, 0 warnings, one informational lock note), and git diff --check.

Resume this same run only after (1) a dapi/memory-bank maintainer authorizes or records the terminal inconclusive disposition and (2) continuation has full .git write plus authenticated GitHub write. Then re-ground sources and HEAD, apply the disposition, create/push codex/gh-96, open the PR, report verification here, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding (2026-07-31). R-GH-96 is still at the Research & Discovery Flow Decision Ready checkpoint: no dapi/memory-bank maintainer has authorized or recorded the required terminal research_status: inconclusive in memory-bank/research/R-GH-96/brief.md.

The evidence remains inconclusive: direct, provenance-preserving handoff membership may later be evaluated as optional user-controlled candidate context, but co-access, temporal proximity, and shared artefacts are not demonstrated to improve recommendations. No implementation issue should be opened now. Dependencies #95 and dapi/memory-bank-cli#47 remain open; CLI PR #49 is clean and unmerged. Its Validate, Stable downstream smoke, and Local init/update E2E checks passed; Publish release was skipped. This is not a representative completed-handoff corpus.

This runtime permits worktree writes but restricts .git writes at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; it cannot create codex/gh-96, commit, push, or open the required PR. git diff --check passed.

Resume this same run only after (1) a maintainer authorizes or records the terminal disposition and (2) continuation has full .git write plus authenticated GitHub write. Then re-ground sources and HEAD, apply the disposition, deliver the branch/PR, report verification here, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding (2026-08-01). R-GH-96 is still at the Research & Discovery Flow Decision Ready checkpoint: no dapi/memory-bank maintainer has authorized or recorded the required terminal research_status: inconclusive in memory-bank/research/R-GH-96/brief.md.

The evidence-backed recommendation remains inconclusive: direct, provenance-preserving handoff membership may later be evaluated only as optional, user-controlled candidate context; co-access, temporal proximity, and shared artefacts are not demonstrated to improve recommendations. No follow-up implementation issue is justified now. Dependencies #95 and dapi/memory-bank-cli#47 remain open; CLI PR #49 is CLEAN and unmerged, so there is no representative completed-handoff corpus.

This runtime has worktree write but restricted .git delivery writes at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; it cannot create codex/gh-96, commit, push, or open the required PR. No Git delivery mutation was attempted.

Resume this same run only after (1) a dapi/memory-bank maintainer authorizes or records terminal inconclusive, and (2) continuation has full .git write plus authenticated GitHub write. Then re-ground source state and HEAD, apply the disposition, create/push the branch and PR, report verification here, and replace codex-ready with human-review.

[dapi]

Human Gate remains active after re-grounding (2026-08-01). R-GH-96 is Decision Ready, but the named decision owner has not authorized or recorded the terminal research_status: inconclusive in memory-bank/research/R-GH-96/brief.md. The recommendation remains: preserve direct handoff/source provenance and user control as future evaluation guardrails; co-access, temporal proximity, and shared artefacts are not yet shown to improve recommendations, so no implementation issue is justified.

Dependencies #95 and dapi/memory-bank-cli#47 remain open; CLI PR #49 is CLEAN and unmerged, so no representative completed-handoff corpus exists. This workspace also has restricted .git delivery capability at HEAD a1cad4ee2745fbf4f9fe80e91b5d8f0393f4e2da; it cannot create codex/gh-96, commit, push, or open a PR.

Verified: memory-bank-cli lint --scope-root memory-bank --entrypoint memory-bank/README.md, memory-bank-cli doctor --profile downstream (0 errors, 0 warnings, 1 informational lock note), and git diff --check all passed.

Resume only after a dapi/memory-bank maintainer authorizes or records inconclusive and the same run has full .git write plus authenticated GitHub write. Then re-ground, apply the disposition, and complete branch/PR delivery with the required tracker updates.