QEP-2 adoption: standard label set rollout

[mmcky]

Goal: every in-scope QuantEcon repository carries the QEP-2 label set and nothing else, with issue and pull-request history intact and a filed migration report per repository.

Where we stand (verified 2026-08-27)

Next: #370 — decide the estate scope and populate config.yml for both profiles. It is the one item that can proceed today: it is an offline decision, it waits on nothing upstream, and every later phase reads its output.

QEP-2 — Standard GitHub Label Set and Labelling Policy — was Accepted on 2026-08-08 (QuantEcon/qeps#2, recorded in QuantEcon/qeps#10), and nothing has been migrated since. The reason is upstream: the qe engine converges repositories toward a bundled schema that is stale against the accepted standard, so a sync today would produce a non-conformant repository and --prune against it would delete standard labels. That work is QuantEcon/cli#20, a separate project — see Related work — and the pilot carries native blocked by edges on it rather than restating its state here.

This tracker was reshaped on 2026-08-27. It had been carrying its work as body checkboxes above a prose roster of the CLI's issues. The roster had drifted in both directions within three weeks of being written — it named four issues that are no longer part of that project and missed six that are, including both live data-loss paths — and the checkboxes were never measured. The work is now native sub-issues. The CLI engine project is related by dependency rather than held as a child: the projects collector reads one level of sub-issues and never grandchildren, so parenting it published 0 of 1 while hiding thirteen items, and double-counted a project that is separately registered.

What the standard fixes: 19 core labels, a lecture extension (new-lecture, editor), and a one-label software extension (refactor) — names, colours and descriptions defined normatively in qep-0002-labels.yml. Adoption is a migration, not an addition: variants are renamed in place so issue and pull-request history survives, and every sync emits a migration report.

Plan

Work items and their order: the sub-issue list (top = next; completed items keep their place). Constraints between items are native dependency edges. This repository's phases have no milestones, so the table below is the phase authority.

Phase	Intent	Exit criterion
Ground state	Fix what a new repository starts from, and which existing repositories are in scope	config.yml names every in-scope repository with no undecided candidates; the org default label set is the core 19
Pilot	Prove the migration on one lecture repository before touching any other	lecture-python-programming carries the 21 labels and nothing else, and its migration report has been read and accepted
Widen	Every in-scope repository migrated, one at a time, each with its own report	One filed report per repository; no label carrying tags deleted unattended; unmapped labels collected across the batch so recurrence is visible
Conform & close	The estate stays conformant without supervision, and the historical record is closed	Org default in place; all three emitters conformant under qe gh bots check; superseded trackers closed

Gates:

• Nothing in Pilot or later runs against a production repository before QuantEcon/cli#20 closes. A sync against the stale bundled schema silently produces a non-conformant repository, and --prune against it deletes standard labels. Ground state is unaffected — it is a decision and a settings change, both made offline, and its output is what the pilot reads.
• The pilot also waits on the tool being installable. QuantEcon/cli#25 and QuantEcon/cli#35: the v0.2.0 tag exists on origin but no release is published, and the README pins a tag that has never existed.
• Bot conformance waits on the software widening batch. An emitter cannot be made to emit build-failure before the repositories it opens issues in carry that label; it would create the label ad hoc and reintroduce the drift.
• The superseded trackers close after the estate is migrated, not before. Until the migration has worked they are the fallback record.

Sequencing rationale: ground state first, because every later phase reads the scope decision and because it is the only phase that can proceed while the engine is still being fixed — and a default set now stops new repositories being born drifted while the migration runs. Pilot before widening because QEP-2 requires it, and because the first real report is the only evidence that tag survival works outside a test fixture. Conformance last because both halves of it, the org default aside, depend on the estate already carrying the standard set.

Related work

• QuantEcon/cli#20 — qe CLI: deliver the QEP-2 label migration — builds the engine this project runs. It gates this project's Pilot phase onward; the gate is stated above and carried as a blocked by edge on Pilot: migrate lecture-python-programming and review its migration report #372. Not a child: it is a separately registered project, and the collector reads one level of children only.
• QuantEcon/cli#24 — qe CLI: test infrastructure for production readiness — its integration tier is what proves --prune preserves tags against a real repository rather than only against offline fixtures. It gates cli#20's own Safe migration phase, which is where that gate is stated.

Feeding amendments back

A label recorded as unmapped across several repositories is a gap in the set, and closing it is a normal in-place amendment under QEP-1 that bumps QEP-2's version. The bar is recurrence: one repository's oddity is a local exception; the same label unmapped in four repositories is a missing label. Migration reports produce the evidence; proposing the amendment stays a human judgement.

QEP-4 (QuantEcon/qeps#8, draft) will re-apply its parked registry hooks to QEP-2 in its own acceptance pull request, which moves QEP-2 to version: 1.

Out of scope / what does not change

• QEP-2's text and qep-0002-labels.yml. Evidence flows to QuantEcon/qeps as amendment proposals; nothing here edits the standard.
• The qe engine. Every tool fix — the schema pin, the case-insensitive matching, the prune guard, the retired-label decision path — belongs to QuantEcon/cli#20.
• Test infrastructure. QuantEcon/cli#24.
• Dependabot. Settled 2026-08-08: its labels belong to the automation, not to QEP-2's contract. QEP-2 defines the label dependencies — name, colour, description — and what a bot applies to its own output is automation behaviour, which is QEP-4's subject. labels: [dependencies] in .github/dependabot.yml is correct as it stands and needs no change.
• Translation forks and *.notebooks build repositories. Excluded by QEP-2 §Scope.
• The programme layer. This project sits under the governance programme in the QuantEcon/status-projects registry, which is where programme grouping lives. It is not itself a programme and carries no programme tracker; whether the programme layer ever becomes structural on GitHub is QuantEcon/status-projects#10.

This issue carries the native Project issue type and no Type label — per QEP-6 the type marks the unit, and per QEP-2 a tracker is structure, not work.

[mmcky]

Reshaped 2026-08-27 — checkboxes to sub-issues, roster deleted, cli#20 detached, Project type applied. This tracker was written on 2026-08-08 as a sequenced execution checklist and had not been touched since. Reviewed against QEP-2's accepted text, the QEP-6 draft (QuantEcon/qeps#18) and the QuantEcon/status-projects tracker contract; what follows is what changed and why.

The prose roster is gone. §1 enumerated eleven QuantEcon/cli issues. That is a mirror of another tracker's sub-issue list, which QEP-6 §6 bans by name, and it had drifted in both directions within three weeks: it still named #2, #3, #4 and #11, which cli#20's membership review removed on 2026-08-27, and it missed #13, #15, #16, #18, #36 and #37, which include both live data-loss paths and the stale-schema risk. Its "Blocking for the pilot: #8, #7, #9" line named one issue that is now closed and missed four that are not. Deleting it is not a loss of information: the list it mirrored is one click away and always live.

The checkboxes are now eleven sub-issues (#370–#380), in plan order across four phases. §§2–5 held roughly a dozen real deliverables as unmeasured body text, which is why this tracker reported 0% while carrying the entire estate rollout. Two things fell out of writing them properly:

• The widening checkbox concealed an unmade decision. "Remaining lecture-* repos" and "software / tooling repos" are not enumerable from QEP-2, which fixes the three profiles but never their membership. That membership exists only in src/qe_cli/data/config.yml — six lecture repos, five commented-out candidates, an empty software: profile, and a header comment still saying "core 16 … all 18" against QEP-2's 19/21/20. Measured against the org's 255 non-fork, non-archived repositories, QEP-2's two named exclusions remove 26, six lecture repos are decided, and 16 lecture-family repos, 6 .theme repos and 102 non-lecture repos are undecided. That is now Decide the QEP-2 estate scope and populate config.yml for both profiles #370, and it is the only item that can proceed today.
• The bot-conformance items existed nowhere. All three emitter fixes were checkboxes here and had no issue in their host repositories. They are now action-check-warnings: emit automated + build-failure, not execution / python-warnings #377, action-link-checker: add the missing automated label #378 and action-activity-report: drop dead label aliases and settle discuss / low-priority handling #379, each blocked by the software widening batch and by QuantEcon/cli#5.

The org default moved early. It was §4, late in the sequence; it is now #371 in Ground state. It depends on nothing, and docs/rollout.md already has it at Step 0 for the right reason — every repository created while the migration runs is otherwise born drifted.

cli#20 is no longer a child of this tracker, deliberately. It was added on 2026-08-26 to repair the parent theft that emptied this tracker (QEP-6 Adoption §2's failure mode, live). The repair was right; the shape was not. The projects collector reads exactly one level of sub-issues and never grandchildren, so with cli#20 as the only child this tracker published 0 of 1 with thirteen items invisible — and, because QuantEcon/cli is private, an empty children array with children_private_count: 1. cli#20 is separately registered as project qep2-cli-engine, so the parentage also double-counted it. The relationship is now carried the way QuantEcon/status-projects#10 leans: a Related work entry each way, plus native blocked by edges from #372 onto cli#20, cli#25 and cli#35.

Two blockers nobody was carrying. The pilot cannot run because the tool is not installable as documented: the v0.2.0 tag exists on origin with no published release (cli#35), and the README pins @v0.1.0, a tag that has never existed (cli#25). Both were classified as general CLI work and parented nowhere. They are now blocked by edges on #372.

Project issue type applied, matching this tracker's registry row (qep2-label-rollout, programme governance) and QEP-6 §1. The programme layer is the registry's, not an issue's — recorded on QuantEcon/status-projects#10 as evidence for ruling R7.

Unchanged: the goal, the settled scope decisions (Dependabot out — its labels are QEP-4's subject; translation forks and *.notebooks excluded; meta's local labels resolved by migration rather than held as an exception), the amendment feedback loop and its recurrence bar, and the rule that meta#178 and meta#290 close only after the migration has worked. #323 joins them in #380: it is superseded in substance and needs a paired registry change, not a plain close.

---

Generated by Claude Code