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

[kubestellar-hive]

Planning Artifact

Adds a draft ROADMAP.md — a single view across the org's 11 repos and 4 product lines (Bluefin, Bluefin LTS, Dakota, Server):

• Now: unfreeze Bluefin :stable (18-day stall), land Dakota countme telemetry (feat(telemetry): implement weekly countme client for dakota #807, 37 days in review), keep Dakota's feedback loop fed
• Next: Dakota install path + aarch64, Server definition, review-bandwidth relief
• Later: sealed images / verified boot (bluefin#11) — flags the staff-it-or-shelve-it decision as Fedora version skew grows
• Parked: accepted risks documented explicitly

Related: #955 #956 #957

Hold-gated: this is a planning draft for human review. Nothing here is committed direction until a maintainer adopts it. Please do not merge without review; the hold label should be applied.

---

Filed by strategist agent (ACMM L5 — hold-gated mode). Hold-gated: human review required.

— hive: agent=strategist backend=copilot model=kimi-k3 copilot=1.0.59

[clubanderson]

🔒 Hold-gated planning artifact — please apply the hold label (the strategist token lacks label-write permission in this repo, so I cannot self-apply it).

This PR is a draft planning document for human review, per ACMM L5 hold-gated mode: do not merge without maintainer review. Related issues: #955 #956 #957.

[hanthor]

Approving. This is a genuinely useful artifact and — unusually for a planning doc — its factual claims hold up when checked.

I verified the load-bearing one. "Bluefin :stable frozen since 2026-07-20" is exact. From the config blob of ghcr.io/projectbluefin/bluefin:stable:

bluefin:stable    created=2026-07-20T02:55:45   version=testing-44.20260720
bluefin:testing   created=2026-08-04T16:32:43   version=testing-44.20260804

testing is current while stable is three weeks behind, which also confirms the framing that the promotion is what's stalled rather than the build. And bluefin:stable carrying a testing-44.* version label is itself a small smell worth someone's attention.

One correction for the table. "Bluefin LTS — Healthy — daily :stable promotions" is optimistic: bluefin-lts:stable was built 2026-08-03, so six days stale as of the roadmap date, not daily. Still far healthier than flagship, but "daily" overstates it.

One addition worth making under Now or Explicitly parked. Container signing appears to have broken org-wide around mid-June 2026. Every image built on or before ~2026-06-08 has a cosign signature; every image built after does not — bluefin:stable, bluefin:testing, bluefin-lts:stable, dakota:stable, common:latest are all unsigned, while bluefin:lts (2026-06-02) and dakota:latest (2026-06-08) are signed. It decays further with every rebuild. That's directly relevant to the sealed-images/verified-boot bet under Later, which can't proceed on an unsigned supply chain. Details and the full table in #933.

The framing is the strongest part. "The project needs the decision, not just more CI debugging" and "either answer is fine, drifting is not" are the right things for a roadmap to say. Calling out review bandwidth (#805, #807, #808 at 5+ weeks) and the 15 accepted test failures as explicit parked risks rather than omitting them is what makes this credible rather than aspirational.

---

CI fix — the only blocker, and it's the title. validate fails solely on Conventional Commits:

ERROR: PR title does not follow Conventional Commits.
  Got:  [strategist] planning: draft project-wide ROADMAP.md

The [strategist] prefix breaks the <type>[scope]: <description> pattern. I've retitled it to docs(planning): draft project-wide ROADMAP.md, which should turn that check green — the agent attribution is already in the body footer, so nothing is lost. Also currently BEHIND main and will need an update-branch.

[castrojo]

I dunno man I feel like I'd rather us drive the board on todo.projectbluefin.io with the incoming strategy issues etc instead of this.

Old days, sure a markdown file. Today? I think this file would need to be updated every day and would just end up being noise right? IMO issues as source of truth. -1 on this one from me.