Pipeline status

---

title: "Pipeline Status" description: "Honest snapshot of what is running, queued, blocked, recently completed, and on the three-month horizon for the DevPlane research program." status: v0.1 publishedAt: 2026-04-30 authors: ["Mike West"]

v0.1 — manually maintained. The program is young; this snapshot reflects actual current state, not aspirational state. "Forthcoming" is acceptable for many slots — the seven-slot baseline (overview, program, proposal, literature review, methodology, pipeline status, instrumentation spec) is designed so that visible gaps stay visible. Last updated 2026-04-30.

Running

Nothing is producing research data yet. The program is in pre-instrumentation setup.

The operational substrate is running: DevPlane in production, single operator, heterogeneous AI agents (Cursor, Claude Code) on a multi-project portfolio. Assignment registry, handoff protocol, two-phase actor handoff, and the shared-index-aware commit protocol are live and have been for several months. Git history, session reports under docs/handoff/, and devplane_log_completion records exist as a partial pre-instrumentation corpus and are recoverable for dispatch, merge_outcome, and rework event types — but not for dwell_ms or operator_inspect, which require the coordination-event log itself.

The coordination-event log (instrumentation-spec.md, DP-50) is specified at v0.1 but not yet implemented. This is the critical-path unblocker.

Queued

The next three concrete pieces, in order:

1. DP-50 — coordination-event log implementation. New SQLite table in the existing DevPlane database, capture points wired in src/agent-workflow.ts, the kanban UI, the auto-resolver/reconciler, and the session manager. Acceptance includes one week of live capture against current DevPlane usage with spot-check verification. Estimated ~2 weeks of implementation work; nothing else moves until this lands.
2. C1 pre-registration formalization. PROPOSAL.md is the working pre-registration in spirit; OSF deposit is queued for filing once the analysis pipeline is built and the run-in period has accumulated enough data to confirm the instrumentation works. The deposit must precede unblinding to post-intervention rate data.
3. Reading-queue completion. The four-paper queue at reading-queue.md — Bainbridge (1983), Lee & See (2004), Parasuraman & Manzey (2010), one CSCW anchor — is the minimum to put the program in conversation with its primary literature. Mike's evening reading.

Blocked

• C1 run-in period. Cannot start until DP-50 ships. The pre-registered design requires ≥30 days of pre-intervention data captured by the live event log; partial backfill from git history covers some event types but not the dwell-time and inspection events that anchor H2.
• C1 intervention deployment. Candidate intervention is the auto-resolve heuristic for assignment-status reconciliation (lineage ASN-953/979). Deployment timestamp is the intervention boundary; it cannot be set until the run-in baseline is captured.
• External-operator replication. Single-operator N=1 is the largest threat to external validity. Replication requires recruiting at least one external operator with comparable instrumentation, plus a consent protocol and likely IRB oversight depending on institutional partnership. Not committed in the C1 proposal; a successor assignment.
• Faculty outreach. faculty-map.md is being built up over weeks. Explicit hold: do not email until the four-paper queue is read and PROGRAM.md has been sharpened against that reading.
• Deep-research literature gaps. deep-research-prompts.md names five low-confidence areas in the literature review (post-2020 Ironies of Automation applications to AI/LLM contexts, recent CMU HCII/S3D/LTI work on human-AI dev collaboration, risk-compensation outside automotive, empirical multi-agent LLM coordination failures, contemporary stigmergy treatment). Outputs land in deep-research-outputs/ and reconcile back into the literature review before public posting.

Recently completed

• 2026-04-29 — Research scaffold landed (PROGRAM.md, OVERVIEW.md, PROPOSAL.md, LITERATURE-REVIEW.md, instrumentation-spec.md, reading-queue.md, deep-research-prompts.md, faculty-map.md). The three-arm program is named, the lead study is specified, the literature is reviewed with confidence flags on uncertain references, and the apparatus is specified at v0.1.
• 2026-04-30 — Methodology surface published (methodology.md). Operational discipline (assignment registry, handoff protocol, two-phase actor handoff, shared-index-aware commit protocol, coordination-event log) and research discipline (pre-registered yes/no worlds, falsifiable constructs, pre-committed coding rules for attribution, acknowledged researcher position, threats-to-validity register) are documented.

Coming soon — three-month horizon

Day 0–30 (current). Build the corpus by shipping DP-50. Read the four-paper queue. Sharpen PROGRAM.md against that reading. Map 3–5 CMU faculty whose work touches this. Reconcile deep-research outputs back into the literature review.

Day 30–60. Run-in period begins (≥30 days of pre-intervention corpus). First pre-registration filed at OSF before the post-intervention period opens. Outreach to faculty using the sharpened materials. Begin recruiting one external operator for the replication arm.

Day 60–90. Mid-run-in checkpoint: spot-check that the event log is capturing what the analysis plan needs. Begin descriptive A-arm work in parallel using accumulated corpus (false-complete base rates, stigmergic-coordination patterns visible in the queue). The C1 intervention deployment is still ahead.

This is a slow program. It is designed to be slow. Coordination cost is the kind of construct that punishes rushed measurement.

---

Master spec: PROGRAM.md Lead study: PROPOSAL.md Methodology: methodology.md Apparatus: instrumentation-spec.md