Resolve Dell-7810 shared-lane reachability without repo-scoped runners

[Jesssullivan]

Downstream tracker: Jesssullivan/Dell-7810#22
Linear: TIN-550

Dell-7810 has cacheable Chapel and kernel dogfood workflow definitions for the shared GloriousFlywheel tinyland-nix capability label, but the repo cannot currently schedule them as counted runner authority.

Evidence from Jesssullivan/Dell-7810 on 2026-04-25:

• The Dell repo Actions runner API reports 0 accessible self-hosted runners.
• Earlier Dell dogfood jobs queued with labels=tinyland-nix and runner=null; those stale runs were cancelled as queue hygiene.
• Live ARC tinyland-nix is registered to https://github.com/tinyland-inc.
• Live ARC personal-nix is registered to https://github.com/jesssullivan/jesssullivan.github.io.
• Because Dell-7810 is Jesssullivan/Dell-7810, neither visible lane is currently reachable by that repo.
• Dell cacheable workflows are now workflow_dispatch only and require confirm_runner_reachable=true before scheduling tinyland-nix jobs.

GloriousFlywheel boundary:

• Do not create a Dell repo-scoped runner lane.
• Do not create Dell-specific runner labels such as dell-7810-*.
• Keep the workflow-facing contract on shared capability classes such as tinyland-nix.
• Treat personal-account repo-scoped ARC anchors as compatibility debt, not pooled shared-runner authority.
• Keep Dell blocked from counted shared-runner authority until a compliant shared-scope proof exists.

Desired GloriousFlywheel decision:

• Choose one clean exit for Dell cacheable CI: move or mirror the workflow surface under an enrolled org/enterprise shared scope, prove an enterprise-level shared runner surface for the personal owner, or keep Dell explicitly blocked.
• Preserve the current cache contract: Attic/Nix cache injection, optional Bazel cache env, and finite runner capacity.
• Do not paper over the owner-boundary problem with a Dell-shaped scale set.

Acceptance:

• A Dell Chapel or kernel dogfood job starts on an actual shared tinyland-nix runner through a compliant org/enterprise or enterprise-equivalent shared scope.
• No Dell repo-scoped runner set or Dell-specific runner label is introduced.
• The GloriousFlywheel enrollment queue remains honest about Dell until that proof exists.

Useful Dell-side diagnostic:

• just platform-runner-enrollment-status --no-cluster
• scripts/platform/runner-enrollment-status

[Jesssullivan]

Correction after reviewing GloriousFlywheel canon: the original issue wording incorrectly pointed at a repo-scoped personal runner path. That is not allowed as product direction.

The issue has been retitled/reworded to match the pooled-substrate contract: Dell-7810 must either truthfully reach the shared tinyland-nix capability class through owner-boundary / GitHub App scope / ARC control-plane cleanup, or remain blocked from counted shared-runner authority. No Dell repo-scoped runner set or Dell-specific runner label should be introduced.

[Jesssullivan]

Downstream Dell correction is now committed and pushed in Jesssullivan/Dell-7810@2becf55. The Dell diagnostic and sprint docs now explicitly reject Dell repo-scoped runner sets and keep the blocker framed as shared-lane reachability / control-plane debt.

[Jesssullivan]

Dell now has a durable downstream audit for this blocker: Jesssullivan/Dell-7810@b3f9672, docs/tracking/gloriousflywheel-runner-reachability-2026-04-25.md. The audit records compliant paths and explicitly rejects Dell repo-scoped runner sets. During that audit I also found unrelated live massageithaca repo-shaped ARC residue and filed #409 separately so this Dell issue stays focused on shared-lane reachability.

[Jesssullivan]

Follow-up while preserving the Dell boundary: opened GF draft PR #411 for the unrelated live massageithaca repo-shaped ARC residue found during Dell reachability audit. Dell #407 remains focused on shared tinyland-nix reachability without repo-scoped runners.

[Jesssullivan]

GF PR #411 merged at aa47f90, accounting for the unrelated live massageithaca repo-shaped ARC residue and closing #409. This keeps #407 focused on Dell shared tinyland-nix reachability without repo-scoped runners.

[Jesssullivan]

Dell-side follow-up: PR #414 is merged at c2c3881 and Dell-7810 records that guard in Jesssullivan/Dell-7810@ab5065e.

The guard closes the source-config drift path for future repo-scoped/project-shaped ARC extra_runner_sets. The live Dell blocker remains unchanged: Dell workflows still queue on tinyland-nix with runner=null, and Jesssullivan/Dell-7810 reports zero accessible self-hosted runners. From Dell, GF #407 remains the owner-boundary/control-plane reachability issue, not a request for Dell-specific labels or repo-scoped runner sets.

[Jesssullivan]

Dell-side follow-up after #415: Jesssullivan/Dell-7810@77df335 now records the stricter personal-account boundary.

The downstream diagnosis is now aligned with GF main: personal-account repo-scoped ARC anchors are compatibility debt, not pooled shared-runner proof. Dell remains blocked unless the cacheable CI surface is moved/mirrored under an enrolled org/enterprise shared scope, an enterprise-level shared runner surface is proven, or Dell stays explicitly blocked from counted authority. The latest Dell diagnostic still shows zero accessible self-hosted runners and queued tinyland-nix jobs with runner=null.

[Jesssullivan]

Dell follow-up: Jesssullivan/Dell-7810@e0cb796 now gates the downstream cacheable Chapel/kernel workflows behind manual dispatch while #407 is unresolved.

This prevents the personal-account repo from repeatedly enqueueing known-unreachable tinyland-nix jobs with runner=null. It preserves manual proof dispatch for the moment an org/enterprise shared scope or other compliant boundary exists. Dell cancelled stale queued runs from earlier commits.