docs(atlas): publish the decision roster, lessons and portable delivery handoff #3168

Proposals are welcome before implementation. For substantive changes, wait
for a responsible human maintainer to approve the scope and identify a review
contact. An automated recommendation, label alone, or silence is not approval.
See Contributing.

Is your feature request related to a problem? Please describe.

The product-decision session has accumulated an 86-issue roster and source-linked strategy decisions, but most Atlas knowledge is still committed only locally. A new session must be able to recover the human decisions, derived lessons and the actual blocked delivery state without private chat history or a false assumption that accepted work has shipped.

This tracks the maintainer's October 6 request to preserve that knowledge locally and remotely. It is administrative knowledge publication, not a ninth implementation-intake item or an expansion of the fixed 86-issue decision roster.

Describe the solution you'd like

Publish the existing Atlas history and a bounded knowledge update through a review branch and a pull request targeting the protected atlas branch:

• A complete 86-row register that separates 32 accepted, 14 deferred, 8 rejected and 32 pending human outcomes from older agent proposals.
• Source-linked, explicitly agent-derived product and delivery lessons.
• A portable next-session handoff with exact current PR heads, qualified components, unresolved gates and consumed capacity.
• A dated publication decision superseding the old local-only storage hold, without rewriting historical product decisions.

Acceptance: Atlas compilation succeeds; all 86 unique rows reconcile and every decided row retains its public decision link; local and remote commit contents can be read back; the publication PR preserves repository review requirements. The four implementation PRs must remain accurately marked blocked unless their real gates change.

Exclusions: production code, Starlight documentation, normative specification/schema changes, implementation-PR repairs, new product decisions, raw private transcripts, credentials, machine-specific paths, budget resets, review bypasses, enqueueing and merges. Existing parent-worktree Atlas mount changes must not be discarded or silently committed as application changes.

Describe alternatives you've considered

Keeping knowledge solely in session files or unpushed commits does not satisfy cross-session remote durability. Copying raw transcripts would add unnecessary private context and obscure the current decision. Pushing directly to protected atlas, or merging main into the knowledge-only branch, is not the selected route.

Additional context

Maintainer request, transcribed by Copilot from Daniel Meppiel's October 6 conversation (the quotation is the human request; this issue is an agent-authored record):

we are trying here to capture the learnings, ensure we have stored our knowledge in atlas, and that we can move to another session while not loosing that knowledge, so that we can reuse it with atlas, so it should be stored locally and on the remote repo durably

Responsible maintainer and existing confirmed default review contact: Daniel Meppiel (@danielmeppiel). This does not assign another reviewer or claim that review has occurred.

Related delivery records: #3145, #3147, #3149 and #3150 remain protected BLOCKED; #3144 is the earlier merged documentation slice. Contributor-owned #3148/#3153 and the held #3128 redesign remain out of scope. The remote atlas branch requires a PR; no bypass or merge is requested.

AI disclosure: GitHub Copilot prepared this issue as a transcription and bounded execution record of the maintainer's explicit knowledge-publication request. It is not an independent agent approval of product work.