Operating Model — Decisions Log

Planning Decisions Log

Running log of decisions from the ops/roadmap reorg. Newest at top of each section. Companion to the operating model.

Decided

#	Date	Decision	Rationale
14	2026-07-13	Support/community/OSS = an operational function, not a pillar: incident on-call (Rootly, Platform-anchored) + support/community on-call (DevRel-anchored, incl. OSS PR/issue triage), one Triage funnel; enabling tooling = Platform deliverables	Continuous, relationship-driven work; give it a rotation + intake, not a roadmap slot
13	2026-07-13	Repo tagging via an optional repo: label group; actually set priority; projects stay outcome-scoped and may span repos	One issue shows every repo it touches; linked PRs already reveal the repo, so tag only when non-obvious/cross-repo
12	2026-07-13	Draft the operating-model + this decisions log at repo root, not the wiki (move later)	Keep it a fast-iterating working doc while it churns
11	2026-07-13	Proactive process/automation work = scheduled deliverables owned by Platform (Dev-Experience / Eng-Productivity); the check they run afterward = maintenance	”Build the automation” has a done+date; “run the check forever” doesn’t — different shapes
10	2026-07-13	Maintenance handling: work-type labels, ~20% sprint capacity reservation, one sanctioned evergreen “maintenance lane” per pillar, detectors over reminders, DRI per integration	Roadmaps can’t hold recurring work; give it capacity + a home, not a deadline. Closes the reactive-integration-breakage gap
9	2026-07-13	Backlog: systematically carve near-term into completable deliverables, keep a real Backlog holding-pen for genuine someday work	Kill evergreen buckets without pretending everything has a deadline
8	2026-07-13	Intake: GitHub issues on OSS repos → synced to Linear Triage; internal repos file directly in Linear; one Platform backlog for loose infra repos. One source of truth (Linear) for priority	Don’t run two backlogs
7	2026-07-13	4 technical pillars: Honcho/Product, Platform, Research (ML), Side-Projects. Growth/PLG untouched (existing Honcho-in-Space + Wed sync)	Matches team clusters; side-projects isolated so they don’t clutter the core roadmap
6	2026-07-13	Vineeth = Product pillar lead; Rajat = Platform; Matt/Yuya = Research	—
5	2026-07-13	Platform ↔ Research split on infrastructure vs. instrument: Platform owns serving (Metagross) + eval-harness runtime; Research owns bench logic + experiments. ML-335 stays Research with a Platform dependency	Lets Platform own the “bench too expensive” (ML-330) problem; keeps the science with ML
4	2026-07-13	Meetings: weekly Mon all-hands + biweekly Research/Platform/Product syncs at sprint boundaries + existing Wed Growth sync	Separate the genuinely independent work (ML, Platform); keep Honcho+FE+DevRel together where they blur
3	2026-07-13	Cadence: 3-month plan → 2-week sprints → weekly standup. Change DEV cycles weekly→2wk	User’s call; aligns all pillars to shared sprint boundaries
2	2026-07-13	Model: pillars (initiatives) → deliverables (completable projects w/ lead+date) → sprints (cycles); roadmap = derived Linear view, not a hand-maintained doc	Evergreen projects are the root anti-pattern; the old wiki roadmap rotted because it was maintained in parallel to execution
1	2026-07-13	Archive the stale Q2-2026 roadmap; don’t patch it	Only 1 of 14 specs shipped; disconnected from execution — a system problem, not a doc problem

Resolved this turn (2026-07-13)

#	Decision	Rationale
18	SDK/CLI under Product: CLI lead = Abigail; SDKs = Eugene/Aakash/Vineeth cadence; integration DRIs openclaw+Hermes → Eri, Claude Code TBD	Fold tooling under Product; match existing review cadence
17	Capacity: DRIs + ~20% reservation for integration maintenance; rotating on-call for support + OSS contributions	Integrations have named owners; support/OSS load is bursty + shared
16	Side-Projects membership: Kyogre, YouSim, Dedenne, Mimikyu, Honcho Demo UI	Keep experimental/demo work isolated from the core roadmap
15	Excadrill/Eevee live in both pillars (Platform + Research)	Matches the shared runtime-vs-logic operational reality

Open

None blocking Phase 1. Minor TBD: Claude Code integration DRI.

Product Q3 scope + working agreements (2026-07-13)

• Q3 = cadence-establishment quarter for Product: finish in-flight, support conclusion/harness work, get OSS-contribution triage + review cadence in order. Modest feature scope by design.
• Flagship epic: Composable Peers (DEV-1629). Multi-modal + SQLite defer to Q4.
• Reliability/foundation work may be dual-listed across Product + Platform (projects can span initiatives); Platform takes point but it need not live solely under Platform.
• Working agreements (WIP + review-first): (1) clear PRs you’re responsible for before starting new work; (2) finish/park in-progress code before opening new (WIP ~1–2, enforced on Linear In Progress column). Responsible for = requested reviewer / area DRI / blocking a teammate. Distribute review authority (fill review-chains) so reviews don’t funnel to Vineeth.
• Future workload skill computes per-person review-debt, WIP, policy violations, reviewer-bottleneck, per-project progress/stalls — reuse the pr-triage skill.

Product prioritization principles (2026-07-13)

Adopted a six-axis lens for Product (detail in operating-model.md): (1) Visibility, (2) Memory quality [Product×Research: Research measures, Product improves], (3) DX / topologies, (4) Expanding the substrate — sharpened; headline bet = native for CLI/coding agents (codebase-as-peer, filesystem, “company brain”), currently unstaffed, (5) Groudon UX, (6) Foundation / reliability [NEW] — reliability/perf/cost (2/M) + the data-model migration & primitive renames; **Platform takes point**. Ownership is a current lean, not fixed DRIs. Tie-breaker: design-partner → unblocks-others → cost-to-2/M → net-new.

Executed

• 2026-07-13 — Phase 3 (Platform) carve. Broke the Groudon evergreen bucket into 11 Platform deliverable-projects (KloudStax/K8s [Rajat], GCP Consolidation, Xatu→Pub/Sub [Adavya], PII Redaction [DEV-1971, Adavya+Yuya], Stripe Hardening, Gateway & Data-Layer, OAuth & Connected Apps [Aakash+Eri], Enterprise Features [backlog], + 3 Dashboard-UX [Eri]); moved 75 issues in (owners untouched). Tentacruel → Platform backlog (no project). Capacity: 45 backend issues were unassigned → for Ulysse (starts Wed) / Phil (next Mon); Eri owns ~all frontend (bus-factor flag). Groudon retains ~26 loose/stale/misc as de-facto backlog.

• 2026-07-13 — Phase 3 (Product) started, Option A. Created 9 deliverable-projects in Honcho/Product with target dates (Scopes v1, CloudEvents, Connection Hardening [dual w/ Platform], Dialectic Structured Outputs & Evidence, Deriver Conclusion Quality & Dedup, Renameable Primitives, Composable Peers [flagship, Oct 5], Honcho CLI V2, SDK Quality); moved 44 issues in (owners/state untouched). Deferred: cycle/capacity balancing, milestones, area: labels + area-project archive, cleanup pass, most leads.

• 2026-07-13 — Phase 1 done in Linear. Created initiatives Platform (owner Rajat) + Side-Projects; renamed Honcho → Honcho / Product (owner Vineeth) and Evaluation Benchmarks → Research (ML) (owner Yuya); reassigned 16 active projects (Groudon/Metagross/Watchog→Platform; Excadrill/Eevee→Platform+Research; Deriver-Conclusion/Dreaming-Investigations/Neuromancer-MR/Machamp/Top-Down/MemoryArena→Research; Kyogre/YouSim/YouSim-Platform/Mimikyu/Honcho-Demo-UI→Side-Projects); archived the emptied Demos initiative. Legacy/completed/canceled projects left for a later archive sweep.

Prior context (this session, before the reorg)

• Linear + wiki audit completed; conclusion-quality eval work filed as epic ML-335 (see [operating-model.md] and memory). Specs index de-drifted; Q2 roadmap archived.