docs(planning): draft project-wide ROADMAP.md

ROADMAP.md

@@ -0,0 +1,78 @@
+# Project Bluefin Roadmap
+
+> **Status: DRAFT — hold-gated planning artifact.** Filed by the strategist agent for
+> human review. Nothing here is committed direction until a maintainer adopts it.
+> Tracking issue: #957.
+
+This document is a single view across the Project Bluefin org
+(11 repos, 4 product lines) so contributors, users, and agents can see what
+matters now, what comes next, and what is explicitly parked.
+
+## Product lines at a glance
+
+| Product | Repo | State | Update health (2026-08-07) |
+|---|---|---|---|
+| Bluefin (flagship, rolling) | `bluefin` | Production | **`:stable` frozen since 2026-07-20** — GNOME 50 smoke gate (#929, #989, #995, #1025) |
+| Bluefin LTS | `bluefin-lts` | Production | Healthy — daily `:stable` promotions |
+| Dakota (GNOME OS, from source) | `dakota`, `dakota-iso` | **Alpha** | Active; user feedback loop producing issues daily |
+| Bluefin Server | `server` | Early | Architecture roadmap lives in `server/docs/skills/architecture-roadmap.md` |
+| Shared layer | `common` (this repo) | Production | OCI layer for every image; very active |
+| Tooling | `bluefinctl`, `finpilot`, `actions`, `testsuite`, `fsdk-containers` | Various | — |
+
+## Now (next ~30 days)
+
+1. **Unfreeze Bluefin `:stable`.** The GNOME 50 smoke gate has failed since
+   2026-06-25. Either the gate gets fixed or a risk-accepted manual promotion
+   happens — but the project needs the *decision*, not just more CI debugging.
+   See #955 (release-gate stall policy) and the CI issues above.
+2. **Land Dakota adoption telemetry.** The privacy-preserving weekly countme
+   client (PR #807) has been in review since 2026-07-01. Alpha prioritization
+   is currently blind to install-base size and growth. See #956.
+3. **Keep Dakota's feedback loop fed.** The `ujust report/confirm/verify` loop
+   is the product differentiator; user-filed alpha issues (audio, LUKS
+   keyboard layout, WiFi regdb, GVFS playback) need triage velocity more than
+   new features.
+
+## Next (1–2 quarters)
+
+1. **Dakota install path + upgrade hardening.** The README lists installation
+   and upgrades/rollbacks as the known gaps; both block a beta label.
+2. **Dakota aarch64 bring-up.** Qualcomm x13s service failures on
+   non-Qualcomm hardware (dakota#1193) and aarch64 boot-test gaps
+   (dakota#1306) are the visible edge of the ARM story.
+3. **Bluefin Server definition.** Grow the server architecture roadmap into a
+   user-facing "what is Bluefin Server" story.
+4. **Review-bandwidth relief.** Multiple PRs in `common` have waited 5+ weeks
+   in `4-review` (#805, #807, #808). Consider widening the reviewer pool or
+   explicit review SLAs for agent-authored PRs.
+
+## Later (long-term bets)
+
+1. **Sealed images / verified boot** (bluefin#11). systemd-boot + signed UKI +
+   composefs, targeting TPM-backed passwordless disk unlock. **Decision
+   needed:** the epic has had no checklist progress since 2026-05-30 and the
+   Fedora version skew is growing (Bluefin defaults to F42; sealed upstream
+   publishes F44/45). Staff the `next`-branch spike or explicitly shelve it —
+   either answer is fine, drifting is not.
+2. **TPM-backed disk unlock** — the user-facing payoff of the sealed-images
+   bet; blocked on it.
+
+## Explicitly parked / accepted risks
+
+- Bluefin `:stable` gate policy — until #955 lands, promotions during gate
+  stalls are ad hoc (see #995 for what that looks like in practice).
+- `common` carries 15 pre-existing test failures on main (#946) — accepted
+  noise that reduces signal for new failures.
+
+## How to use this document
+
+- **Contributors:** pick from *Now* first; Dakota's queued issue queue is the
+  agent-ready entry point.
+- **Maintainers:** this doc should be revisited monthly; the strategist agent
+  will propose updates as hold-gated PRs.
+- **Users:** "Which image should I run?" — Bluefin LTS for stability today,
+  Bluefin once `:stable` unfreezes, Dakota if you want to help shape the alpha.
+
+---
+*Drafted by the strategist agent (ACMM L5 — hold-gated mode). Human review
+required before this becomes canonical.*