docs(triage): quarterly upstream sweep, REBUILD vocabulary, cadence change - #152
Merged
Merged
Conversation
…hange Backlog 72 -> 92 open: 31 opened, 11 closed (5 opened and closed inside the window), leaving 26 new PRs to triage. Of those, 21 were already covered or superseded, 1 is a verified defect in our tree, 3 prompted native work without being adoptable themselves, and 1 is hostile. - Renames the CHERRY-PICK disposition to REBUILD throughout. We never apply upstream diffs; every adopted idea is implemented against our own tree, so the old label misdescribed the work. Re-letters the snapshot legend and the two live row tokens that used it. - Moves the review cadence to quarterly and adds a title-level pre-filter as an explicit step. Yield per cycle is falling while upstream is unmaintained and this fork is 0 commits behind it, so waiting accrues no merge risk and most of the backlog can be dispositioned without reading a diff. - Records the sweep with a ledger that reconciles, the disposition of all 26 new PRs, three new action items with their our-code anchors, and the declines with reasons so future cycles do not re-litigate them. - Extends the settled clusters, which are the cache the new pre-filter reads. - Adds a hostile submission to the do-not-merge list.
Review of this entry surfaced one factual error and several arithmetic inconsistencies. Fixes, in order of consequence: - Correct the characterization of the full-feature upstream PR. It was described as shelling out to an "undisclosed" AGPL tool; it is not undisclosed. That PR carries a README section on the AGPL boundary, module docstrings explaining the licensing chain, and a test asserting nothing under evolution/ imports the AGPL package. The SKIP disposition stands on its own merits — the canary is premature here and the entanglement is real — but the characterization was wrong and unfair, so it is restated accurately. - Drop two PRs from this cycle's remainder list. They were dispositioned in the previous cycle, so listing them here inflated the entry's own enumeration past the 26 it claims to cover. - Remove a test count that does not reproduce. The figure came from a verification run that cannot be reconstructed at any plausible suite scope; the substantive claim (the importer suites pass with a real populated state.db present, so the autouse fixtures do isolate them) is kept without it. - Reconcile the disposition split with the snapshot's own S/R/X tags: 24 skip, 1 action item, 1 do-not-merge. Three of those skips still prompted native work, and are now named explicitly rather than folded into a separate count that agreed with nothing. - Move the prompt-section and MIPROv2 reconstructions out of the phase-mega-PR bucket into the cluster this same change defines for them. The title pre-filter matches against these buckets, so mislabeling works against it. - Fix an identifier: the per-importer concatenation lives in build_dataset_from_external. - Add action-item rows for the importer/dataset-builder hardening and the MIPROv2 valset items, so every tracked item has a row to flip and the reference to "the hardening item" resolves. - Restore the author attribution the two prior sweep entries use, add a group count that reconciles, describe the verifier PR's oracle as a 17-paper fact table rather than 17 facts, and cross-reference the parked statistics item to the action item that now absorbs half of it.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Records this cycle's review of the upstream backlog (
NousResearch/hermes-agent-self-evolution) and changes how the review runs.Ledger. 72 -> 92 open: 31 opened, 11 closed (5 opened and closed inside the window), 6 of the previously-tracked 72 closed. That leaves 26 new PRs to disposition: 21 already covered or superseded, 1 a verified defect in our tree, 3 not adoptable but worth rebuilding from, 1 hostile.
Changes
CHERRY-PICK->REBUILD. We never apply upstream diffs — every adopted idea is implemented against our own tree, so the old label misdescribed the work. Re-letters the snapshot legend and the two live row tokens that used it.Notes
Docs-only; no code paths touched. The three action items land as their own PRs, each flipping its own row in the same PR as the fix.
One correction worth flagging: the draft ledger did not reconcile because the previously-tracked set had been derived by grepping issue and cluster references out of the doc rather than from the API. The figures here come from
ghdirectly and close exactly.