Skip to content

fix(transcription): restore Arabic punctuation onto unpunctuated words - #552

Merged
chhoumann merged 9 commits into
masterfrom
fix/punctuation-restore-gaps
Nov 21, 2026
Merged

chhoumann merged 9 commits into
masterfrom
fix/punctuation-restore-gaps

Conversation

@chhoumann

@chhoumann chhoumann commented Nov 20, 2026 •

Copy link
Copy Markdown
Owner

Summary

Follow-up to #544 for 2.25.1. Some answers from Qwen3 ASR, and from OpenAI-compatible servers, list bare words and keep the punctuation in text. PodNotes puts that punctuation back on the words. The verifier on #544 accepted four gaps. This PR fixes one and explains why the other three stay open.

  1. Not changed: a curly contraction can take a nearby comma. With text Well, we’ll see. and words we’ll see, the note reads we’ll, see.. I tried three rules, and an independent verifier failed each one with a case master gets right:

    • Keys that keep ’ lost the marks of words split at an apostrophe (Rock n roll against Rock’n’roll!).
    • Rules for split words and digits lost the mark of 90s against 90’s..
    • Matching without apostrophes, with a tie-break for exact spellings, moved the mark of 90’s. 90s! onto the wrong 90s.

    Which token a word belongs to when the two differ only by an apostrophe cannot be told without context. The real Qwen3 answers never hit it either: their words and text agree on every apostrophe (87 of 87 and 97 of 97). Matching stays as on master.

  2. Fixed: Arabic ، ؛ ؟ were dropped. They are now in the closing marks (TRAILING), which also restores the Urdu ؟. No word can already hold one, because a word with a mark keeps the whole answer as given (PLAIN_WORD), so none can be added twice.

  3. Not changed: more than 200 tokens of filler stop the restore for the rest of the answer. I replayed the two real Qwen3 answers from feat(transcription): offer Voxtral Mini Transcribe 3 and Qwen3 ASR on OpenRouter #544 (Up First, 2,262 words, and Planet Money, 3,785 words) with the resync window at 3, 4, 6, 10, 20, 50, 200 and unbounded. Every run restored the same marks, 301 of 311 and 612 of 616. A wider window brings back the quadratic worst case feat(transcription): offer Voxtral Mini Transcribe 3 and Qwen3 ASR on OpenRouter #544 measured (2.2 s for 17,500 unmatched words).

  4. Not changed: one non-plain word turns the restore off for the whole answer. The real answers have 0 non-plain words out of 6,047. Restoring around such words is the boundary behind most of feat(transcription): offer Voxtral Mini Transcribe 3 and Qwen3 ASR on OpenRouter #544's doubled-punctuation rounds. It would also change notes from Grok, Scribe v2 and MAI, which share this reader. The existing test keeps every Qwen word as given when one has a symbol, an emoji, a quote or an end apostrophe pins the current behavior.

The verify skill's fixture now answers Qwen3 ASR with bare words, as the real model does. Before this, the fixture never exercised the restore end to end.

Verification

  • gives Qwen's words Arabic punctuation from its text fails on 9ef1834 with مرحبا كيف حالك أنا بخير شكرا.. npm run plant with ،؛؟ removed from TRAILING makes it fail, and it passes again once the file is restored.

  • A replay of the reader on master and this head covers every case the three verifiers raised, plus 420 generated quote and apostrophe shapes. Every output is identical to master, except Arabic and Urdu, which gain their marks. Both real Qwen3 answers are byte-identical to master.

  • Obsidian 1.14.4 on Linux, isolated instance, fake OpenRouter key, Qwen3 ASR 1.7B. I played Fixtures/Local Fixture.wav with "Play with PodNotes" and ran "Transcribe current episode from audio" from the palette, with net.mjs watch --turns text Well, we’ll see. ‘Okay.’ Bye. and مرحبا، كيف حالك؟ أنا بخير؛ شكرا.:

    • master: مرحبا كيف حالك أنا بخير شكرا.
    • this head: مرحبا، كيف حالك؟ أنا بخير؛ شكرا.
    • dev:errors showed no errors.
  • npm run lint, format:check, typecheck, build, size, test (2,993) and test:tooling (56) pass, run with fnm exec --using 22.23.2. I skipped docs:build because no docs changed.

  • Size: on top of master, the budgets rise by the measured 6 raw and 12 gzip bytes, to 334,171 and 109,282.

  • An independent verifier (GPT 6 Astra) gave PASS+NOTES at 3907092. Merging master after that changed only the size budgets and one line of the verify recipe. The reader, test and fixture patch is unchanged (same git patch-id).

Release impact

None. No settings, storage, API or URI changes.


Note

Low Risk
Localized change to punctuation matching for unpunctuated ASR words; no auth, storage, or API surface changes.

Overview
Extends the verbose JSON punctuation restore in verboseJson.ts so closing Arabic marks (، ؛ ؟, including Urdu ؟) are applied when Qwen3 ASR and OpenAI-compatible servers return bare words with punctuation only in text.

Adds a unit test for Arabic text and updates the verify fixture (net.mjs) so Qwen3 ASR responses omit word-level punctuation like the live model, with matching notes in transcripts.md.

Bundle budgets rise by 6 raw and 12 gzip bytes (334,171 / 109,282) in size.mjs, tests, AGENTS.md, and size-baseline.md.

Reviewed by Cursor Bugbot for commit bbeccc4. Bugbot is set up for automated code reviews on this repo. Configure here.

Note

Fix trailing punctuation restoration for unpunctuated Qwen transcription words

  • Adds a trailing-punctuation pattern next to the existing leading-punctuation pattern in verboseJson.ts, so a token's closing punctuation is assigned to its last word and opening punctuation to its first word. This restores Arabic punctuation that Qwen omits from word tokens but keeps in the response text.
  • Adds an Arabic punctuation test case in TranscriptionService.test.ts and updates the Qwen fixture in net.mjs to return words without surrounding punctuation.
  • Raises the raw and gzip bundle size budgets in size.mjs, with matching docs and test updates.
  • Behavioral Change: transcripts from Qwen now keep trailing punctuation on the final word of a token; bundle size checks now pass against higher raw and gzip thresholds.

Macroscope summarized bbeccc4.

…punctuation

Both cases fail on 9ef1834: we’ll takes the comma of the Well before it, and
Arabic ، ؛ ؟ are dropped.
…rom Arabic marks

Keys read a curly apostrophe as a straight one, so we’ll is not well, and the
closing marks include ، ؛ ؟. Budgets rise to the measured 334,117 raw and
109,265 gzip bytes.
@chhoumann
chhoumann marked this pull request as draft November 20, 2026 21:33
@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 20, 2026 •

Copy link
Copy Markdown

Deploying podnotes with  Cloudflare Pages  Cloudflare Pages

Latest commit: bbeccc4
Status:🚫  Build failed.

View logs

@chhoumann

Copy link
Copy Markdown
Owner Author

FAIL
Verified head 66ea40b by an independent verifier (GPT 6 Astra).

  • P2 regression at verboseJson.ts:93. With text: "Rock’n’roll! Next." and timed words ["Rock", "n", "roll", "Next"], parent 9ef18342 saves Rock n roll! Next.; head saves Rock n roll Next. Starts are 1, 1.3, 1.6 and 1.9 seconds, each lasting 0.2 seconds. Keeping the normalized apostrophe in the token key prevents the remaining split words from matching. The subsequent prefix match can consume another token. Reproduced from the palette and direct audio command in real Obsidian on both bundles.
  • Two further synthetic replays reproduce sentence-mark loss through the palette. text: "Yes, we’ll. Next!", words Yes we ll Next: parent Yes, we ll. Next!, head Yes, we ll Next!. text: "It costs 1’000. Next!", words It costs 1000 Next: parent It costs 1000. Next!, head It costs 1000 Next!. These are adversarial fixture answers, not claims that the cached provider responses contain these shapes.
  • Both intended fixes pass. The omitted-Well case changes we’ll, see. to we’ll see.. Arabic ، ؛ ؟ survive in the saved note. Compared 16 palette runs and 2 direct-command runs on Obsidian 1.14.4, Linux, in this worktree's isolated vault. Played the 60-second Fixtures/Local Fixture.wav with “Play with PodNotes,” chose OpenRouter/Qwen3 ASR in Settings, and used net.mjs watch --turns plus targeted local replies. Fake key only. Ordinary punctuated Grok, Scribe v2 and MAI note pairs are byte-identical.
  • Replayed 30 targeted cases and 420 generated quote/apostrophe shapes against both readers. Original word characters survive; no doubled marks found. The unchanged real Up First answer keeps 2,262/2,262 words and restores 301/311 ASCII punctuation marks on both revisions. Planet Money keeps 3,785/3,785 and restores 612/616 on both. Complete outputs are byte-identical.
  • The existing 210-token filler and non-plain-word limits are reasonable to defer in this focused PR. The window bounds the current search; the whole-answer guard protects against the duplication demonstrated in feat(transcription): offer Voxtral Mini Transcribe 3 and Qwen3 ASR on OpenRouter #544. Neither cached answer needs those rules relaxed. Urdu ۔, Persian joiners and RTL-mark limitations also reproduce on parent and do not cause this verdict.
  • All requested checks pass after npm ci: npm run lint, format:check, typecheck, build, size, test (125 files, 2,993 tests), and test:tooling (9 files, 56 tests). Used fnm exec --using 22.23.2; its installed Node reports v22.23.4. Head is exactly 334,117 raw and 109,265 gzip bytes.
  • Read the saved notes and settings, inspected all 13 screenshots, and checked runtime errors on both bundles. Every dev:errors result is clear. No tracked file changed, committed or pushed. Fixture cleanup ran; the isolated instance is stopped and the dry run reports pids: none.
  • Evidence and rerunnable drivers: /home/christian/.t3/worktrees/PodNotes/28318b32-80de94373c0f2100/.obsidian-e2e-artifacts/review552/. Start with verification.md, replay-results.json, runtime-results.json, direct-runtime-results.json, and the paired parent-split-compound-note.md / head-split-compound-note.md and screenshots.

… with one

Both fail on 66ea40b: Rock n roll loses the ! of Rock’n’roll!, and 1000 the
period of 1’000. The independent verifier on #552 found them.
…ouped with one

A word that ends at an apostrophe inside its token moves past it, so we and ll
both fall in we'll, and an apostrophe between digits is dropped from the key.
Budgets rise to the measured 334,180 raw and 109,299 gzip bytes.
@chhoumann

Copy link
Copy Markdown
Owner Author

On the 66ea40b verdict: fixed in 3e44ed3. The verifier was right that this PR caused it. Once ’ counted in the keys, a word that ended at an apostrophe inside its token left the next word facing that apostrophe, so n and roll never matched rock'n'roll. Straight ' already failed the same way on master. Two changes fix it:

  • A word that ends at an apostrophe inside its token moves past it, so we and ll both fall in we'll. well still does not match we'll.
  • An apostrophe between digits is dropped from the key, so 1000 matches 1’000 and 2'000.

Your three examples now save as Rock n roll! Yes, we ll. It costs 1000. Next!.

  • New tests, each red on 66ea40b:
    • "gives the last of the Qwen words split at an apostrophe in its text that token's punctuation" (curly and straight)
    • "restores punctuation after a Qwen number its text groups with an apostrophe"
  • Plants caught by the named test:
    • the apostrophe skip reverted
    • the digit rule reverted
    • every apostrophe stripped from the keys, which the curly contraction test catches
    • a digit rule for ’ only, which the straight 2'000 case catches
  • The real Qwen3 answers are byte-identical to 66ea40b: 301 of 311 and 612 of 616 marks, every word kept.
  • Checks: test (2,995) and test:tooling (56) pass. Obsidian 1.14.4 on Linux saved the note above, and the earlier curly and Arabic notes are unchanged. Budgets rise to the measured 334,180 raw and 109,299 gzip bytes.

@chhoumann

Copy link
Copy Markdown
Owner Author

FAIL
Verified head 3e44ed3 by an independent verifier (GPT 6 Astra).

  • P2 regression at verboseJson.ts:93. With text: "The 90’s. Next!" and timed words ["The", "90s", "Next"], parent 9ef18342 saves The 90s. Next!; head saves The 90s Next!. Starts are 1, 1.3 and 1.6 seconds, each lasting 0.2 seconds. The normalized apostrophe remains before the plural s, so 90s no longer matches 90’s and loses its period. Reproduced through both the palette and direct audio command in real Obsidian. 1990’s versus 1990s reproduces in the reader replay too. These are synthetic adversarial answers, not cases claimed to occur in the cached provider responses. Preserve this numeric-plural equivalence without collapsing well/we'll or its/it's.
  • The three previous failures are fixed: Rock n roll! Next., Yes, we ll. Next!, and It costs 1000. Next! now match parent. The intended curly-contraction and Arabic fixes pass. 90's when present in the timed word, and 1’000s against 1000s, also pass.
  • All requested checks pass after npm ci: npm run lint, format:check, typecheck, build, size, test (125 files, 2,995 tests), and test:tooling (9 files, 56 tests). Used fnm exec --using 22.23.2; its installed Node reports v22.23.4. The head bundle is 334,180 raw and 109,299 gzip bytes at level 9.
  • Replayed 53 targeted inputs and 420 generated quote/apostrophe shapes. Word characters and timing survive, and already-punctuated generated answers remain unchanged. The real Up First answer keeps 2,262/2,262 words and restores 301/311 ASCII marks on both revisions. Planet Money keeps 3,785/3,785 and restores 612/616. Both complete outputs are byte-identical.
  • Obsidian 1.14.4 on Linux, this worktree's isolated vault, 60-second Fixtures/Local Fixture.wav, “Play with PodNotes,” Settings selection of OpenRouter/Qwen3 ASR 1.7B, fake key only. Compared 22 palette runs and 4 direct-command runs across parent/head. Ordinary punctuated Grok, Scribe v2 and MAI notes are byte-identical. Read persisted notes/settings and inspected all 19 screenshots. Every dev:errors report is clear.
  • Existing filler-window, non-plain-word, RTL-mark and Urdu-period limits reproduce on parent and do not cause this verdict. No tracked files changed, committed or pushed. Fixture and instance cleanup completed; the stop dry run reports pids: none.
  • Evidence and rerunnable drivers: /home/christian/.t3/worktrees/PodNotes/0bee0a2b-40c68a2c65aab78a/.obsidian-e2e-artifacts/review552/. Start with verification.md, replay-results.json, extra-runtime-results.json, direct-decade-runtime-results.json, and the paired parent-decade-normalized-note.md / head-decade-normalized-note.md and PNGs.

The 90s loses the ! of 90’s! on 3e44ed3, which the second independent verifier
on #552 found. Two more cases guard against a word spelled like an apostrophe
word, its against It's, and a later token pulling a split word out of its own.
…exact spelling

Words match the text with apostrophes dropped, so 90s finds 90’s, 1000 finds
1’000 and rock n roll finds rock'n'roll, which replaces the apostrophe skip and
the digit rule. A word that matches a whole token only that way, as its does
It's, moves to a later token in reach that spells it exactly, which keeps we’ll
off a nearby Well,. Budgets rise to the measured 334,318 raw and 109,351 gzip
bytes.
@chhoumann

Copy link
Copy Markdown
Owner Author

On the 3e44ed3 verdict: fixed in 412f4c1 by changing the rule instead of adding another shape. 90s against 90’s was the third failure in a row from the same premise: matching on keys that keep apostrophes. Every time the words leave out an apostrophe the text has (90s, 1000, split rock n roll), matching breaks.

  • Words now match the text with apostrophes dropped, as master did for curly ones. That covers 90s, 1000, Rock n roll and we ll. It replaces the apostrophe skip and the digit rule from 3e44ed3.
  • Apostrophes only break a tie. A word that matches a whole token only once they are dropped, as its matches It's,, moves to a later token in reach that it spells exactly. That keeps we’ll off a nearby Well,, which was the original gap.

Your case now saves The 90s. Next!.

  • I replayed your replay.cjs plus seven more cases on master, 3e44ed3 and this head. No case restores fewer marks than master, except the four where master gave a contraction the comma of a nearby Well, or Its,. That comma was the bug. The 420 generated shapes pass. Up First and Planet Money are byte-identical to master: 301 of 311 and 612 of 616 marks, every word kept.
  • New tests:
    • "restores punctuation to Qwen's words that leave out an apostrophe their text has" (1’000, 2'000, 90’s) was red on 3e44ed3.
    • "does not give a Qwen word the punctuation of one that is spelled like it with an apostrophe" (its against It’s,)
    • "keeps the first of the Qwen words split from one token of its text in that token when a later one spells it" (Short stays in Short-term,)
  • Plants, each caught by the named test:
    • keys that keep apostrophes, the 3e44ed3 rule: caught by the leave-out test
    • no exact-spelling preference: caught by the its test and both contraction tests
    • the preference without its whole-token check: caught by the Short-term test
  • Checks: test (2,997) and test:tooling (56) pass. Obsidian 1.14.4 on Linux saved Back in the 90s! its own. Rock n roll! Yes, we ll. It costs 1000. we’ll see., and the fixture's curly and Arabic note is unchanged. Budgets are the measured 334,318 raw and 109,351 gzip bytes.

@chhoumann

Copy link
Copy Markdown
Owner Author

FAIL
Verified head 412f4c1 by an independent verifier (GPT 6 Astra).

  • P2 regression at verboseJson.ts:135. With text: "We’ll go. Well, yes." and words ["Well", "go", "Well", "yes"], parent 9ef18342 saves Well go. Well, yes.; head saves Well, go Well yes. The exact-spelling preference jumps over the first occurrence, moves the later comma onto it, and loses the intervening period. Both occurrences are present, so this is not the intended omitted-filler fix. Starts are 1 + index * 0.3 seconds, ends 1.2 + index * 0.3.
  • Minimal reproduction: text: "90’s. 90s!", words ["90s", "90s"]. Parent saves 90s. 90s!; head saves 90s! 90s. Both this case and the previous case reproduce through the palette and direct audio command. Also reproduced through the palette: It’s done. Its value. / Its done Its value loses the period after done; its quoted variant loses both quotes too. These are synthetic adversarial answers, not cases observed in the cached real answers.
  • The previous decade, split-compound, split-contraction and grouped-number failures now match parent. The intended omitted-Well, omitted-It’s and Arabic fixes pass. Ordinary punctuated Grok, Scribe v2 and MAI note pairs remain byte-identical.
  • After npm ci, all requested checks pass through fnm exec --using 22.23.2: npm run lint, format:check, typecheck, build, size, test (125 files, 2,997 tests), test:tooling (9 files, 56 tests). Svelte reports zero errors and warnings. Installed Node reports v22.23.4. Size is 334,318 raw / 109,351 gzip bytes at level 9.
  • Replayed 66 targeted inputs and 420 generated quote/apostrophe shapes against parent/head. Word characters and timing survive. Real Up First keeps 2,262/2,262 words and restores 301/311 ASCII marks on both; Planet Money keeps 3,785/3,785 and restores 612/616. Both complete outputs are byte-identical. Existing filler-window, non-plain-word and RTL limitations reproduce on parent and do not cause this verdict.
  • Obsidian 1.14.4 on Linux, isolated worktree vault, 60-second Fixtures/Local Fixture.wav, Play with PodNotes, Settings selection of OpenRouter/Qwen3 ASR 1.7B, fake key only. Compared 28 palette and 4 direct-command runs across both bundles. Read all saved notes and settings, inspected all 25 screenshots. All 36 dev:errors checks are clear. No tracked files changed, committed or pushed. Fixture and instance cleanup completed; stop dry run reports pids: none.
  • Evidence and rerunnable drivers: /home/christian/.t3/worktrees/PodNotes/d0e86a20-98624f62ebb5ddfa/.obsidian-e2e-artifacts/review552/. Start with verification.md, replay-results.json, runtime-results.json, direct-runtime-results.json, and paired parent-well-retained-note.md / head-well-retained-note.md, parent-decade-repeated-note.md / head-decade-repeated-note.md, and their PNGs.

…arks

Three independent verifier rounds each found a case where a new apostrophe rule
lost or misplaced a mark that master restores: words split at an apostrophe,
90s against 90’s, and 90s twice against 90’s. 90s!. Which token a word with or
without an apostrophe belongs to cannot be told without context, and the real
Qwen3 answers never hit it. Matching goes back to master's, so a curly we’ll
can still take a nearby omitted Well,'s comma, as before. Budgets are the
measured 334,097 raw and 109,248 gzip bytes.
@chhoumann chhoumann changed the title fix(transcription): restore punctuation onto curly contractions and from Arabic marks fix(transcription): restore Arabic punctuation onto unpunctuated words Nov 21, 2026
@chhoumann

Copy link
Copy Markdown
Owner Author

On the 412f4c1 verdict: you were right, and this is the third apostrophe rule an independent verifier has failed. No matching rule can tell, without context, whether a word belongs to a token that differs from it only by an apostrophe. So 3907092 reverts the matching to master's, and gap 1 (a curly we’ll taking the comma of a nearby omitted Well,) stays open, as the PR body explains. The PR now ships only the Arabic marks and the bare-word Qwen fixture. A replay of every case the three verifiers raised is identical to master, except Arabic and Urdu, which gain their marks. Both real Qwen3 answers are byte-identical to master. test (2,992) and test:tooling (56) pass.

@chhoumann

Copy link
Copy Markdown
Owner Author

PASS+NOTES
Verified head 3907092 by an independent verifier (GPT 6 Astra).

  • No regression found against parent 9ef1834. All cases from the three earlier FAIL comments now match parent. Replayed 91 targeted cases and 420 generated quote/apostrophe cases; inputs without the three added marks produce identical words and timing.
  • No newly doubled Arabic marks found. Already-punctuated words, standalone punctuation entries, emoji entries and RTL-marked words retain the whole answer unchanged. Marks next to emoji or RTL controls in text restore once when they end the token; trailing RTL controls still block restoration. Two marks already present in source text remain two.
  • Real cached Qwen answers are byte-identical between parent and head. Up First keeps 2,262/2,262 words and restores 301/311 ASCII marks; Planet Money keeps 3,785/3,785 and restores 612/616.
  • Obsidian 1.14.4 on Linux, isolated worktree vault, fake OpenRouter key, Qwen3 ASR 1.7B selected in Settings. Played the 60-second Fixtures/Local Fixture.wav with "Play with PodNotes" and compared 14 palette runs across both bundles. For text: "مرحبا، كيف حالك؟ أنا بخير؛ شكرا.", bare words مرحبا كيف حالك أنا بخير شكرا, the saved parent note says مرحبا كيف حالك أنا بخير شكرا.; head says مرحبا، كيف حالك؟ أنا بخير؛ شكرا.. Default Qwen turns, ordinary Scribe v2, the earlier split-compound/decade/retained-Well cases and already-marked Arabic produce identical paired notes. Read every saved note and inspected all 9 screenshots. All 16 runtime error reports are clear.
  • After npm ci, all requested checks pass through fnm exec --using 22.23.2 (installed Node reports v22.23.4): npm run lint, format:check, typecheck, build, size, test (125 files, 2,992 tests), test:tooling (9 files, 56 tests). Svelte reports zero errors/warnings. Size is 334,097 raw / 109,248 gzip bytes at level 9.
  • The verify fixture's default note still matches its documented recipe. tests/e2e/journey/fixtures.ts has its own unchanged stub and does not import net.mjs. npm run test:e2e -- -t 'J-TR-11|J-TR-12' passes both targeted tests; 258 are filtered out. The full E2E suite was not run.
  • The PR's deferral is justified for this diff. Curly-apostrophe ambiguity, the filler window, and the whole-answer plain-word guard retain parent behavior; the cached answers do not require relaxing them. Unspaced Arabic internal commas, Urdu full stops and RTL/joiner limits remain pre-existing limitations. No tracked files changed, committed or pushed. Fixture and instance cleanup completed; stop dry run reports pids: none.
  • Evidence and rerunnable drivers: /home/christian/.t3/worktrees/PodNotes/d34f7ef5-bec811383ff34929/.obsidian-e2e-artifacts/review552/. Start with verification.md, replay-results.json, runtime-results.json, paired parent-turns-note.md / head-turns-note.md, the PNGs, and the check logs. This verdict applies to the specified head and parent; GitHub currently reports the PR behind master with conflicts.

Budgets are the measured 334,171 raw and 109,282 gzip bytes: master's 334,165
and 109,270 plus the Arabic marks.
@chhoumann
chhoumann marked this pull request as ready for review November 21, 2026 00:12
@chhoumann
chhoumann merged commit 1813e47 into master Nov 21, 2026
20 of 21 checks passed
@chhoumann
chhoumann deleted the fix/punctuation-restore-gaps branch November 21, 2026 00:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant