Repository navigation
Conversation
…al form, guide and harness follow-ups The rc.21 repin did not rerun anything. Regrading what the repository holds on rc.21 (eval/results/rc21-regrade-2026-11-05): - S-latest-reask records DV-19: `latest` of a pending guess answers the re-asked truth once a re-ask lands under it (solidjs/solid#4689; ref/oracle/repros/dv19-latest-reask.mjs). S-guess-reask-refresh records the D-35 hold under an outside re-ask, with no divergence; the effect-only shape that does revert is solidjs/solid#4690 (repros/4690-guess-reask-refresh.mjs), which no .rx program reaches. - The chain2, chain3, chain4 and count references read the open count outside the loading on(filter) region, which CN-3e has refused since 2026-10-30; they now read `latest issues`, the program H-7 wants. Each passes its task's hidden and overlap tests; chain4's one count hole is DV-19. - guide/async.md §4: refresh every question you guessed on; a write whose value is not on screen yet guesses nothing. - eval: Playwright is a dependency; CHROMIUM=… drives a system browser in the harness and the server-render runner; a round is regradable only with its runs committed. Suites on rc.21: oracle and compiled 114 pass with the one recorded divergence; server render 107 pass; types 154; the crate agrees with ref/ on 319/319 programs. S-store-8 fails on the interpreter before and after this change. Co-authored-by: tembo[bot] <208362400+tembo-io[bot]@users.noreply.github.com>
…asses, same-key variant added The interpreter kept one guess per node with one owner (`guessBy`), so when a second action guessed on the same row, the first action's end found nothing of its own to revert and its declined guess stayed up (S-store-8 instant 4, `Z+B-` for `Z-B-`, failing since edb76be). Guesses are now entries per (key, hold); the node's guess is the truth with the live entries applied in order, an action's end removes its entries and leaves the others, the body-end correction (D-35) corrects only that action's keys, and a new answer at a guessed key spends that entry (D-34). The oracle and the compiled runtime agree on all 114 scenarios, S-store-8's new same-key variant included. Co-authored-by: tembo[bot] <208362400+tembo-io[bot]@users.noreply.github.com>
… DV-20; DV-19 caveat - runtime: a correction marks the action's guess entry spent instead of dropping it, so the action's re-ask still holds the corrected guess (a closed row came back between the write landing and the re-ask). S-latest-reask variant covers it; Solid rc.21 agrees. - #4690 is reachable: an effect over a source nothing renders. S-guess-reask-refresh gets an effect-only variant with DV-20; divergence entries may now name effects. - DV-19: note that it rests on reading H-11's correction narrowly, and what changes if Solid declines #4689. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0117BoWMbSPnht6GSU1vp8Pe
…DV-20 fixed on next - H-11: a landing answer is compared with the guess, not with the truth the guess was made over, so an outside re-ask's answer corrects it (`latest x` the truth); a corrected guess's effects wait for the commit as its views do. S-latest-reask instant 4 is `2 / 3`, no divergence. - Interpreter: corrects on any landing answer that differs from the guess; effects over a frozen (corrected) guess wait. - chain2/3/4 and count references read the open count in a small component under the list's region instead of `latest issues`: chain4 77/77, 38/38, fuzz holes 3 -> 2 (both the app's); chain2/3/count pass their hidden and overlap tests. - CN-3e message (rxc and ref/) and spec/08 say when each fix fits; guide/reference.md one line. - DV-20 (#4690) noted fixed on next, recorded until the repin. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0117BoWMbSPnht6GSU1vp8Pe
ryansolid
changed the title
…he commit; DV-19 is Solid's - H-11: an answer is compared with the guess and corrects the graph now, but the screen, the guess's effects and `latest x` of a guess the screen shows keep it until the hold commits (Solid D-22, D-34, decision 3). H-7's "after a correction, the truth" narrowed to a guess the screen never showed. - Interpreter: `latest` of a shown, answer-corrected guess answers the guess until the commit. - S-latest-reask back to `2 / 2`, with the changed-truth variant; DV-19 recorded at four instants (#4689 ruled, not built). - CN-3e message, spec/08 and guide/reference.md back to what holds: both fixes show the guess. - References keep the count under the list's region (right by the spec either way, and not dependent on #4689). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0117BoWMbSPnht6GSU1vp8Pe
ryansolid
changed the title
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
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.
Requested by Ryan · project thread
The rc.21 repin (#30) carried the fixes for every Solid issue the eval had filed but did not rerun anything. This regrades what the repository holds on rc.21, records what it found, and lines the spec up with Solid's rulings in solidjs/solid#4695.
DV-19:
latestof a shown guess (solidjs/solid#4689, ruled in #4695)The fuzzer found a tab count read with
latest issuesshowing 3 beside 2 rows while a close was in flight. A re-ask from outside the action had landed under the guess. #4695 rules this (decision 3; D-22 "a slot that is displaying a guess has the guess as its newest value", D-34 "the screen, untracked reads andlatestof the displayed slot keep the guess until the commit"). It is not built yet.latest xof a guess the screen shows keep the guess until the hold commits. H-7's "after a correction, the truth" now applies only to a guess the screen never showed.2 / 2throughout, plus a variant where the outside answer changes the truth (the closed row stays out when the write lands under the action's refresh). DV-19 is recorded at four instants on the oracle and the compiled runner.ref/src/runtime.ts): corrects on any landing answer that differs from the guess.latestof a shown, answer-corrected guess answers the guess until the commit. Effects over a corrected guess wait.References
chain2, chain3, chain4 and count read the open count outside the
loading on(filter)region, which CN-3e has refused since 2026-10-30. They now read it where the list is read, in a smallOpenCountcomponent under the list's region. By the spec this is as right aslatest issues, but it doesn't depend on #4689, and it is what 10 of 12 agent desks in rounds 10 and 11 wrote. chain4 on the desk9 brief: 77/77 hidden, 38/38 overlap (including "retry after not found"), and 2 fuzz seeds with a hole, both from a rename sent before its target was on screen. chain2 64/64 + 29/29, chain3 70/70 + 32/32, count 56/56 + 24/24.DV-20 (solidjs/solid#4690)
A guess read by effects alone reverts under the action's refresh once an outside re-ask has landed under it. An
effectover a source nothing renders reaches it, so S-guess-reask-refresh has an effect-only variant recording it at instants 5 and 6 (divergence entries may now name effects). Fixed onnext(4fe7532); recorded until the repin.Interpreter: a guess is its own action's, per key
S-store-8 had failed on the interpreter since edb76be (one guess per node with one owner). Guesses are now entries per (key, hold): an action's end removes its own entries and leaves the others, and the body-end correction (D-35) corrects only that action's keys. A correction (D-34) marks the entry spent rather than dropping it, so the action's re-ask still holds the corrected guess. A same-key variant was added to S-store-8.
Guide
guide/async.md§4, two bullets stated as semantics: refresh every question you guessed on (round 4's one surviving defect), and a write whose value is not on screen yet guesses nothing.Harness
Playwright is a dependency of
eval/, andCHROMIUM=/path/to/chromiumdrives a system browser inharness/driver.mjsandref/ssr/run.ts. The eval README says a round can be regraded only if its runs are committed (rounds 5, 7, 10 and 11 are not).Results note
eval/results/rc21-regrade-2026-11-05/has the tables above, the graded outputs, and the 18 round 4 .rx apps regraded on rc.21: 14 pass everything, 3 are refused bySTALE_WRITE_BACKas the round 4 write-up predicted, and 1 keeps the missing-refresh defect.Suites on rc.21
Interpreter 114 pass. Oracle and compiled 114 pass, with 6 recorded divergences (DV-19 at four instants, DV-20 at two). Types 156 pass. The crate agrees with ref/ on 321/321 programs. Server render was 107 pass, 7 skip before the review fixes and has not been rerun since.
🤖 Generated with Claude Code
https://claude.ai/code/session_0117BoWMbSPnht6GSU1vp8Pe