004 — Fleet Monitor (factory app #4)
Product research + planning pass, 2026-07-18, velin-style plan folder. Re-tests the plans/001-factory-apps-validation row-3 verdict (ADOPT + thin layer, local-first) hostilely rather than assuming it. Run evidence: docs/llm-runs.md (workflow wf_7d951cf8-c72, 13 agents, ~1.06M subagent tokens).
Verdict (00-SYNTHESIS): BUILD a limited MVP — a thin merge layer over things that already exist (
fleetops,ghmonitor, Agent View, ccusage, Grafana 25255, Zaruba’s own REST endpoints). 2–4 operator-days to trial-ready; 1-week live trial on the next real wave is the validation instrument. North Star: median time-to-unblock.
Reading order
| Doc | What | Status |
|---|---|---|
| 00-SYNTHESIS.md | Executive recommendation, auditor resolutions (Pattern A/B, Langfuse reject, Encore Cloud verified), build-vs-buy, advisor verdicts, founder questions, kill criteria | Decision record — wins over all others |
| 01-initial-recommendation.md | Pre-research judgment (deliberately unrevised — the doc under test) | Historical; corrections tracked in 04 §corrections + 00 §1 |
| 02-discovery.md | Brainstorm ladder, problem evidence classification, stakeholders, journeys, user stories, riskiest assumptions | Final |
| 03-competitive-oss.md | 25+ candidate matrix, per-candidate verdicts, first-party trajectory, gap table | Final (Langfuse rec overridden by 00 §2.3) |
| 04-prd.md | Requirements F1–F7, NFRs incl. trust SLO, degradation path, MVP scope, metrics, pivot/kill | Final |
| 05-ux-ui.md | One-canvas IA, screens, interaction model, 5-color state semantics, wireframes, trial criteria | Final |
| 06-architecture.md | Components, data model, code-ready metric formulas, ADR table, ops | Final (+00 §2.1 adds journal-tailer input; §2.4 closes Encore Cloud gap) |
| 07-pre-mortem.md | 15 failure modes, 3 leading indicators, whole-product kill criterion | Final (mode 14 likelihood downgraded by 00 §2.4) |
Deliberately not written (internal 2-operator tool — trimmed per the brief’s own instruction)
made-to-stick.md— no marketing narrative for a tool with one user and one digest reader.- Standalone
journey.md,user-stories.md— folded into 02 (no purchase/procurement/renewal/churn stages exist). - Standalone
north-star.md,build-vs-buy.md,validation-plan.md,executive-recommendation.md— folded into 00 and 04. research.md/open-source-analysis.md— 03 covers both; raw research packets live in the workflow transcript, not re-published.
Folder numbering note
Parallel planning sessions created 002-work-ledger-product, 002-dual-run-parity-harness (duplicate 002s) and 003-atlas-intelligence-layer. This folder took 004 to match the product’s position in the factory-apps list and avoid a third 002.