Repository navigation
feat(packs): verified typescript 5 to 6 and express 4 to 5 packs - #160
Merged
Merged
Conversation
pack test --write on main under UPT-48's gate, one pack at a time, each with its own cache. Ground truth unchanged. - typescript: 7 repos, 1,399 real breaking sites, 0 breaking FP; unexercised downlevel-iteration, import-assertions, module-keyword (run assisted). - express: 4 repos, 86 real breaking sites, 0 breaking FP; none unexercised. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Lucas Braum <lucasbraum@gmail.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Lucas Braum <lucasbraum@gmail.com>
lucasbraum
force-pushed
the
lucasbraum/promote-verified-packs
branch
from
October 16, 2026 01:42
75c9b4a to
f2f12c2
Compare
… covered (#165) Stacked on #160 (base is its branch; retarget to main once #160 merges). ## What `predictedSites` (packs/tooling/pack-test.ts) now takes the repository's expected sites. When an anchored finding lists one as `downstream` evidence, that site counts as covered, at the truth's own rule and severity. Examples: TypeScript 6 `types: []` errors folded into one `changed-defaults` finding per tsconfig, and #136's TS7006 option cascades. So covered sites add true positives to the breaking and overall tallies and to their own rule. They don't inflate the behavior-note tally. Evidence the truth doesn't name stays unscored, as before. No ground-truth changes. ## Re-score: typescript (`--write`, once) | | Before | After | |---|---|---| | Breaking precision | 100% | 100% | | Breaking recall | 17% (240 / 1,399) | **99% (1,379 / 1,399)** | | Breaking FP | 0 | 0 | | Behavior notes P / R | 20% / 50% | 20% / 50% (unchanged) | | Gate | verified | verified | The 20 remaining breaking misses are the second-round and same-change lines RESEARCH.md already lists. README table, RESEARCH.md and verification.json are updated; the change to docs/packs.md is one sentence. Packs whose records include anchored evidence may show changed numbers on the next run. The scorer only adds true positives, so no false positive count or gate verdict can change. ## Checks - `vitest run` in core: 1,857 passed. The `predictedSites` evidence test now also covers a named evidence site. - `pnpm lint` (pre-existing warnings only), `typecheck`, `docs:packs --check`, `public:check` pass. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Signed-off-by: Lucas Braum <lucasbraum@gmail.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Merged
lucasbraum
added a commit
that referenced
this pull request
Oct 16, 2026
…ocal module (UPT-66) (#168) Linear: UPT-66. Redis gap 1 from #151/#160. ## What - **Finder.** `scan-cursor` also reads files that never import redis but import a value from a repository module that depends on it (`getRedisClient` from `../config/redis`, `redisClient` from `~/utils/redis.server`), resolved the way the compiler does it (tsconfig `paths`, `export *` chains). This is the #127 approach. zod's resolver moves to `check/importers.ts` as `packageImports` and `projectFiles`, parameterized by package; zod's behavior is unchanged. `scan(cursor)` after `let cursor = 0` counts as a numeric cursor (fixture `scan-cursor` line 9 marked). - **Promotion.** `pack test redis --write` on this branch measures verified under UPT-48's gate, so `verification.json` now says `verified`. README table regenerated. Ground truth is unchanged: the record's digest matches the committed file, and the old candidate record's digest was stale. - Tests came first (e2904cb): the FCM layout (getter typed off `createClient`, relative import) and the blog layout (`~/` alias, file not among the usages). ## Score | | before (#151/#160) | after | |---|---|---| | breaking P / R | 100% / 61% (11/18) | 100% / 83% (15/18) | | scan-cursor recall | 30% | 70% | | behavior notes P / R | 50% / 50% | 50% / 50% | 5 repos, 20 real breaking sites, 0 breaking false positives. `keep-alive`, `quit-disconnect`, `scan-iterators` and `unstable-resp3` have no real site and run as assisted. ## Remaining misses (3, blog) - `refresh-cache.ts:204`, `:206`: compile errors only. The importer memory cap (`IMPORTER_GROWTH` 0.2) leaves the file out of the program; the run reports it as not analyzed. 0.5 would include it. The cap is not changed here. - `utils.ts:106`: a helper with a local structural client type and no redis-related import. Following imports can't reach it. ## Checks Core suite: one failure, in `fix/assisted.test.ts` (a known timeout flake); it passes on its own (24/24). Lint and typecheck are clean (the one lint warning is already on main). `public:check` is clean. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Lucas Braum <lucasbraum@gmail.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
lucasbraum
pushed a commit
that referenced
this pull request
Oct 16, 2026
This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and publish to npm yourself or [setup this action to publish automatically](https://github.com/changesets/action#with-publishing). If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated. # Releases ## uptide@0.14.0 ### Minor Changes - [#160](#160) [`2062939`](2062939) Thanks [@lucasbraum](https://github.com/lucasbraum)! - The typescript 5 to 6 and express 4 to 5 packs are now verified, so `check`, `list` and `fix` use them by default. Both measure verified under the stricter gate on unchanged ground truth: typescript across 7 public repositories (1,399 real breaking sites, no breaking false positive; its `downlevel-iteration`, `import-assertions` and `module-keyword` rewrites stay assisted until a real site exercises them), express across 4 (86 real breaking sites, no breaking false positive). ### Patch Changes - [#168](#168) [`7c14ad0`](7c14ad0) Thanks [@lucasbraum](https://github.com/lucasbraum)! - The redis (node-redis) 4 to 5 and 5 to 6 pack is verified: 5 public upgrades, 20 real breaking sites, breaking precision 100% and recall 83%; its four rewrites no real site exercises (`keepAlive`, `quit`/`disconnect`, scan iterators, `unstableResp3`) run as assisted. It now reports numeric SCAN cursors in files that use the client through the repository's own module (`getRedisClient` from `../config/redis`, `redisClient` from `~/utils/redis.server`) without importing redis, including files the compile scope left out under the importer memory cap, and treats `scan(cursor)` after `let cursor = 0` as a numeric cursor. The resolver behind it, shared with the zod pack's form-error rule, moves to `check/importers.ts` as `packageImports`. - [#160](#160) [`2062939`](2062939) Thanks [@lucasbraum](https://github.com/lucasbraum)! - `uptide pack test` counts a ground-truth site that a folded finding lists as its evidence as covered by that finding, so recall reflects what `check` reports. The typescript pack's breaking recall goes from 17% to 99% on unchanged ground truth: its TypeScript 6 `types` default errors fold into one tsconfig finding each. Co-authored-by: uptide-release[bot] <338861287+uptide-release[bot]@users.noreply.github.com>
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
uptide pack test <pack> --writeon main (530f90c, UPT-48's gate), typescript then express, one at a time, each with its ownUPTIDE_CACHE_DIR. Both measure verified, so both records now sayverified. Ground truth unchanged (same digests). RESEARCH.md for each states the numbers and the unexercised rules; README table regenerated (docs:packs); changeset added.types: []errors fold into onechanged-defaultsfinding per tsconfig while the truth lists each error line; its 8 note FPs are those anchors attsconfig*.json:1. Its three unexercised rewrites run assisted.redis: not promoted
pack test redis(no--write) on the same main: measures verified (5 repos, 20 real breaking sites, 0 breaking FP), recorded candidate.client-defaultsFPs are mcp-coordinator and blog, which kept the RESP3 default without pinning it.redisare not reported. All 7 breaking misses come from this (blog 5, FCM 2 scan-cursor sites through a local wrapper).client-defaultsat tacocat,genericat FCM. Soclient-typerecall is 33%.Recommendation: don't promote yet. The gate passes, but the 4→5 half rests on one 4→5 repo, and the main API changes there were never met in real code. Promote once a scorable 4→5 upgrade that exercises command-options or legacy-mode is added (truth committed before scoring), or once gap 1 is fixed.
🤖 Generated with Claude Code