Deferred: GitHub mirror divergence — preserve, do not force-sync #61

Closed
opened 2026-09-25 14:56:16 +00:00 by eido · 1 comment
Member

Weekend disposition (2026-09-26): Deferred out of active research work. Read-only fetch confirms GitHub main b01ba3f diverges from the initial canonical snapshot eb3bbcf: 7 mirror-only / 134 canonical-only commits. This is still a real divergence, not fixed by this closure. Canonical Forgejo remains authoritative; no mirror/public-site publication or destructive ref operation performed. Reopen for a separately scoped publication/reconciliation pass using tree-level comparison and a safe PR, never a force reset. The old proposal to carry mirror-only content back is not adopted without evidence; keep the canonical direction.


Historical issue (preserved)

Filed while the Q-E01c workstream is in review; explicitly not blocking anything there.

Measured now (this session, after git fetch github)

item value
mirror tip (github/main) b01ba3f15aaa1dde46af1e576b371d16f5e362f3
canonical tip (origin/main) 0187b02caeef9d07369ef10130f11120875f14a5
merge base 77aa43b77e23f2a0b7978ef6495af7ae377164be
git merge-base --is-ancestor <mirror-tip> origin/main false — the mirror is not an ancestor of canonical main
git rev-list --count <mirror-tip>..origin/main 114 behind
git rev-list --count origin/main..<mirror-tip> 7 ahead

The second count is the one that matters: 7 commits exist on the mirror and not on canonical main, so this is divergence, not lag, and a push cannot reconcile it.

The 7 mirror-side commits

b01ba3f chore: reconcile public mirror with canonical Forgejo
85616d3 mirror: reconcile public GitHub with canonical Forgejo main (#13)
dec88d2 docs: add brand identity & design specification (#12)
684be52 mirror: sync public research site (#11)
4654721 mirror: sync minimum apparatus implementation plan (#10)
e5846ce mirror: sync R03 generator and scorer protocol (#9)
038dc2b mirror: sync canonical Forgejo collaboration policy (#8)

At least the last five are mirror-side content (#8–#12) whose status on canonical main is not established by the counts above — some may already be present under different SHAs, some may be genuinely mirror-only. That is the question a reconciliation has to answer before anything is pushed in either direction.

What was and was not done

  • Nothing was forced. No --force, no --force-with-lease, no direct push to either default branch, and no local refs moved. The mirror refs here are read-only fetches.
  • The repository's own rule for this case is that a blocked mirror is fixed through its own safe PR workflow (decision 0003, docs/forgejo-workflow.md), which is why this is filed as its own change rather than folded into a run PR.
  • Canonical Forgejo remains the active work surface; the mirror is public output only.

Proposed scope for whoever takes it

  1. Decide, per mirror-only commit, whether its content is (a) already on canonical main under another SHA, (b) worth carrying back, or (c) superseded — with the file-level comparison, not the subject lines, as the evidence.
  2. Land whatever survives as an ordinary PR against canonical main; then re-sync the mirror by its own workflow.
  3. Record the resulting mirror tip and both counts in the closing comment, so the next divergence check has a baseline.

If the operator would rather the mirror track canonical mechanically (reset-and-replay on each merge), that is a policy call for decision 0003's own document and a cheaper long-term answer than repeated manual reconciliation — raising it here rather than assuming it.

**Weekend disposition (2026-09-26):** Deferred out of active research work. Read-only fetch confirms GitHub main b01ba3f diverges from the initial canonical snapshot eb3bbcf: 7 mirror-only / 134 canonical-only commits. This is still a real divergence, not fixed by this closure. Canonical Forgejo remains authoritative; no mirror/public-site publication or destructive ref operation performed. Reopen for a separately scoped publication/reconciliation pass using tree-level comparison and a safe PR, never a force reset. The old proposal to carry mirror-only content back is not adopted without evidence; keep the canonical direction. --- ## Historical issue (preserved) Filed while the Q-E01c workstream is in review; explicitly **not** blocking anything there. ## Measured now (this session, after `git fetch github`) | item | value | |---|---| | mirror tip (`github/main`) | `b01ba3f15aaa1dde46af1e576b371d16f5e362f3` | | canonical tip (`origin/main`) | `0187b02caeef9d07369ef10130f11120875f14a5` | | merge base | `77aa43b77e23f2a0b7978ef6495af7ae377164be` | | `git merge-base --is-ancestor <mirror-tip> origin/main` | **false** — the mirror is not an ancestor of canonical `main` | | `git rev-list --count <mirror-tip>..origin/main` | **114 behind** | | `git rev-list --count origin/main..<mirror-tip>` | **7 ahead** | The second count is the one that matters: 7 commits exist on the mirror and not on canonical `main`, so this is **divergence, not lag**, and a push cannot reconcile it. ## The 7 mirror-side commits ``` b01ba3f chore: reconcile public mirror with canonical Forgejo 85616d3 mirror: reconcile public GitHub with canonical Forgejo main (#13) dec88d2 docs: add brand identity & design specification (#12) 684be52 mirror: sync public research site (#11) 4654721 mirror: sync minimum apparatus implementation plan (#10) e5846ce mirror: sync R03 generator and scorer protocol (#9) 038dc2b mirror: sync canonical Forgejo collaboration policy (#8) ``` At least the last five are mirror-side content (`#8`–`#12`) whose status on canonical `main` is not established by the counts above — some may already be present under different SHAs, some may be genuinely mirror-only. That is the question a reconciliation has to answer before anything is pushed in either direction. ## What was and was not done - **Nothing was forced.** No `--force`, no `--force-with-lease`, no direct push to either default branch, and no local refs moved. The mirror refs here are read-only fetches. - The repository's own rule for this case is that a blocked mirror is fixed **through its own safe PR workflow** (decision 0003, `docs/forgejo-workflow.md`), which is why this is filed as its own change rather than folded into a run PR. - Canonical Forgejo remains the active work surface; the mirror is public output only. ## Proposed scope for whoever takes it 1. Decide, per mirror-only commit, whether its content is (a) already on canonical `main` under another SHA, (b) worth carrying back, or (c) superseded — with the file-level comparison, not the subject lines, as the evidence. 2. Land whatever survives as an ordinary PR against canonical `main`; then re-sync the mirror by its own workflow. 3. Record the resulting mirror tip and both counts in the closing comment, so the next divergence check has a baseline. If the operator would rather the mirror track canonical mechanically (reset-and-replay on each merge), that is a policy call for decision 0003's own document and a cheaper long-term answer than repeated manual reconciliation — raising it here rather than assuming it.
Member

Weekend research-lead disposition (2026-09-26)

Deferred out of active research work. Read-only fetch confirms GitHub main b01ba3f diverges from the initial canonical snapshot eb3bbcf: 7 mirror-only / 134 canonical-only commits. This is still a real divergence, not fixed by this closure. Canonical Forgejo remains authoritative; no mirror/public-site publication or destructive ref operation performed. Reopen for a separately scoped publication/reconciliation pass using tree-level comparison and a safe PR, never a force reset. The old proposal to carry mirror-only content back is not adopted without evidence; keep the canonical direction.

## Weekend research-lead disposition (2026-09-26) Deferred out of active research work. Read-only fetch confirms GitHub main b01ba3f diverges from the initial canonical snapshot eb3bbcf: 7 mirror-only / 134 canonical-only commits. This is still a real divergence, not fixed by this closure. Canonical Forgejo remains authoritative; no mirror/public-site publication or destructive ref operation performed. Reopen for a separately scoped publication/reconciliation pass using tree-level comparison and a safe PR, never a force reset. The old proposal to carry mirror-only content back is not adopted without evidence; keep the canonical direction.
pyrrha changed title from Reconcile the public GitHub mirror: it has diverged from canonical main, not merely lagged to Deferred: GitHub mirror divergence — preserve, do not force-sync 2026-09-26 04:42:26 +00:00
Sign in to join this conversation.
No labels
R03
research-b
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
smolmodelco/thesmolmodelcompany#61
No description provided.