The Cold Hand

VERDICT: FAIL

STRONGEST: Timing and the second disagreement are solid. All six onsets measure within 0.07 s of the score. The beat at film 22.5 to 23.0 s (scene 19.5 to 20.0) is a frame a viewer can point to: the red dashed thread runs from the official's banner to a moon hidden behind a cloud while new dew beads keep appearing on the mirror. The final scene frame (film 35.0) shows all three remainders, covered mirror, tucked hand and the lit hall with 明水 tablet and filled vessel, and the moon is separate from the mirror throughout.

VIOLATION: The teller changes the world without touching it. At scene 20.6 to 21.6 (film 23.6 to 24.6) the mirror tilts by itself while the official stands about 150 px away. The dew pours onto the ground, and his vessel fills anyway. This tilt and collection exist only when the teller is present (cf.noTeller gates it), so the page's own invariant fails: "re-rendered without the teller, every world trace is identical." The Proof test passes only because it compares bead birth times and ignores the beads shown on the mirror. Three more problems: the "square receptacle open to the sky" is an upright panel facing the camera, so it reads as a window or TV screen rather than a dish facing upward. The first disagreement (film 14 to 18 s: moon in a gap, dew stops) is visible only as old beads sitting still, which a viewer without the meter strip will not register. And the cold person's goal is unclear: why someone cold seeks dew, and why they cover the mirror, is never shown.

EVIDENCE: Frames at film 24.0 and 25.0 show the mirror rotated about 30 degrees with beads falling below it and no hand on it, while the official sits at x≈640 with his vessel. Code: drawMirror tilt = cf.noTeller?0:0.5*sm(COLLECT...), and the collected beads are removed only when the teller exists. In frames at film 15 to 18 the bead count stays at about 8 and nothing on the mirror changes. The only change from the dew-forming moments is a ring-shaped hole around the moon in a flat grey curtain whose lower edge sits well above the mirror, so the mirror's own sky does not look covered. In the Ledger, H4's claim that Li Shizhen "rejected" the shell reading needs a citation: Gao You glosses fangzhu as a large clam (大蛤), and Li Shizhen's position is not sourced. P2 admits that polished speculum is a poor radiator, which undercuts the film's central mechanism. The rest checks out: H1/H2 (Huainanzi chs. 6 and 3), H3 (Zhouli 司烜氏 with 鑒) and Monteith 1957 QJRMS 83 are accurate. The phone viewport (390x844) keeps the film and meter strip on top and the panel below, and it works.

REPAIR: Have the official walk to the mirror and tip it with his own hand into his vessel, or drop the collection beat, so the teller's only effect on the world is physical contact or none. Then extend the Proof test to compare the displayed bead state with and without the teller. Turn the mirror face up toward the sky, or angle it clearly, and in the clouded-sky stretch make the stop visible: beads shrinking or evaporating, or the meter's breath and shiver cue changing. Do not rely on an unchanging count. Before the reach, add one gesture that establishes the person's motive (thirst, or gathering water for the hall), and either source or delete the Li Shizhen clause in H4.

VERDICT: FAIL

REPAIRS MET: (1) Met. At film 23.6–24.0 s the official's vessel rim touches the plate's lower corner, the plate tips and the beads run into the vessel. The Proof test reports identical bead state with and without the teller until contact at scene 20.55 s, and the tilt starts at 20.61 s. (2) Not met. On its stand the plate leans back only about 18° from vertical (quad 48 px tall, 16 px lean). In every frame from film 12 s to 27 s it still reads as a monitor or easel facing the cold person, not a plate open to the sky. (3) Met. Under the cloud bank the beads visibly shrink and vanish between film 16 and 17.7 s, and the test confirms 7 beads go to 0. However, the moon's reflection is drawn as a white oval the same colour as dew and stays on the plate, so the last thing visible on the plate looks like a surviving bead. (4) Partly met. Nothing new happens before the reach at film 4.0 s: the figure sits still with breath puffs, and the empty bowl lies unused. The open-hand "ask" at scene 22.3–22.9 s (film 25.3–25.9) exists, but the official's descending banner crosses in front of it, so the reason for the cover is only half legible. (5) Met. H4 no longer has the Li Shizhen clause. It now cites Needham 1962, pp. 87–91, and Ware 1966 and keeps the Gao You clam-shell caveat. The Ware/Baopuzi citation for "square, bronze" should still be checked.

STRONGEST: The told passage works. At film 22–24 s the moon is behind a small cloud and dew is visibly forming, while the red banner 方諸取露於月 and the subtitle credit the moon. The official then collects the dew only by contact with his vessel rim. The ending holds all three remainders: the plate is under cloth, the cold hand holds up an empty bowl, and the hall's doorway is lit with the vessel and the 明水 tablet.

VIOLATION: The teller's marks touch the world. When the banner comes down at film 25.9–26.2 s (scene 22.9–23.2), the red paper with 方/諸/取 passes over the plate face and the cold person's open hand. For about 0.3 s, at 26.1 s, a glyph sits on the plate. The dashed red claim-thread also ends on the moon's disk. Separately, the plate still reads as a screen, which was round-1 condition 2.

EVIDENCE: Frames were rendered with __renderAt every 1 s from 0 to 38, plus 23.6, 23.8, 24.0, 24.4, 25.5–26.3 in 0.1 s steps, 26.8, 27.0 and 34.9. Counterfactual renders were made for no teller, no moon and no clouds. The page's own score timing passes, with the worst onset off by 0.07 s, and all seven Proof tests pass. The 390x844 view keeps the film and the meter in view. In the no-moon run, an empty dark disc stays where the moon was, while cloud and dew are unchanged.

REPAIR: Lay the plate nearly flat, or tilt it at least 45° back, so its face plainly opens to the sky. Draw the moon's reflection in a different shape or tone from the dew. Retract the banner upward or away from the plate so no mark crosses the plate or the hand, and stop the thread short of the moon. Add a gesture before the reach at scene 1.0 s, such as rubbing the hands or tipping the empty bowl, and keep the open-hand ask at scene 22.3 s clear of anything in front of it.

VERDICT: FAIL

CONDITIONS: (a) PARTIAL. The plate is now raised on its pedestal with the face tilted back and up, so it does open to the sky. But it has a luminous blue gradient face, a black bezel with pale rivet dots, and a single post. At every size from about film 12–27 s it still reads as a lit tablet or kiosk screen, not as dark bronze. (b) MET. The moon's reflection is a soft amber ellipse with no edge (film 13–18 s). The dew is hard, round, cold-white pinpricks with a dark offset. Shape and tone are clearly different. (c) MET. The banner hangs above the teller's cap and never crosses the plate or the cold person's hand, including during the ask at 25.2–25.6. The red dashed thread ends at the cloud edge, short of the moon disc (film 23–25 s). The subtitle sits below the ground line. (d) MET. From film 3.0 to 3.9 s the hand is at the mouth with a breath puff, which sets up cold as the motive before the reach. The open-hand ask at film 25.2–25.6 is held palm-up toward the teller with nothing in front of it. The banner retracts above it, and the staff comes down only after the arm has dropped.

STRONGEST: The disagreement is clear on screen. In the sim, under the scene-12–15 cloud bank, the full moon shines through a gap while all of the plate's dew dries off (shown only as bead counts on the meter). Then the moon is behind a cloud while beads multiply on the plate and the red banner 方諸取露於月 rises, so the claim and the light visibly disagree. The counterfactuals behave correctly: no moon gives identical dew, no clouds gives 71 beads against 40, and no teller gives identical beads up to the rim contact. The last frame (film 34.9) shows all three remainders: the raised empty bowl, the covered plate, and the hall with its vessel and 明水 tag.

VIOLATION: The body does not answer to the physical variable. The cold person's coldness is a time script: 0.35 + 0.012·s, plus a fixed spike at scene 14.8 (coldness() at line 233). It never reads the plate temperature. The recoil from "cold metal" at scene 14.7–15.35 comes just after the cloud bank has warmed the plate back near air temperature, warmer than at any point since the lift. The film itself shows the plate warmed, because every dew bead has just evaporated, yet the hand flinches as if from the coldest metal in the scene.

EVIDENCE: I rendered every 1 s from 0 to 38 and every 0.1 s from 22 to 27, plus spot frames, at a 1000x700 viewport with 2x scale; the 390x844 check had no horizontal scroll and the film fit. Note for the harness: __renderAt does not pause autoplay, so early screenshots drifted. Final frames were taken after pressing pause, and they reproduce exactly. The Proof tab passes all 7 tests, with a worst onset error of 0.04 s. Code at lines 190–215 computes the plate temperature: under clear sky it heads toward −6 °C by scene 11; c1 covers the sky from scene 11–12; with a 1.2 s time constant the plate is about −0.9 °C by 14.7. The bead count falls from 7 to 0 by 14.6, and the touch is measured at 14.69, followed by the CH recoil pose and the large breath puff at 15.3.

REPAIR: Drive the touch spike from the simulated plate temperature at the moment of contact, e.g. spike ∝ max(0, −2.5 − Tm). If the plate is warm under cloud, the hand finds it dry and only mildly cool and gets nothing, so it stays open rather than recoiling. Or move the plate under clear sky at 14.7 so a strong recoil is earned. For (a), replace the glowing gradient face with matte dark bronze that only faintly tints with the sky, drop the bezel dots, and seat the plate low on the ground or on a squat tripod so it cannot be read as a screen.

VERDICT: FAIL

CONDITIONS: (i) Met. Coldness comes from the simulation: body += dt*0.022*skyClear, so it accumulates with sky exposure. buildKeys() chooses the touch keys from contactCold(), which reads the plate's simulated temperature SIM.T at TOUCH=14.8. With the cold plate (no clouds) you get the TM to CH recoil and a breath enlarged by 1+0.8*contactCold. With the plate above the dew point (base run) you get TM held 14.8 to 15.6, then a slow return to H by 16.4. This is in the code and visible in frames. (ii) Not met. The stand is low (pivot 22 px above the ground, three short legs). The plate face, however, is a flat mid-tan fill (RGB about 92–108, 78–96, 56–82) with a soft moon highlight, set in a black bezel at a laptop angle in front of a seated figure. In a scene where everything else is black or night-blue, it is the brightest object on the ground and reads as a lit screen or open laptop, not as dark bronze.

STRONGEST: The world model holds up. The proof tests pass and I confirmed them in renders. Removing the moon gives identical dew. Removing the teller leaves the dew identical until his vessel visibly touches the plate's corner (20.57 s, tilt at 20.61 s). Clearing the clouds gives 71 beads against 40. The disagreement is easy to see: from 16.4 to 22.5 s the moon sits behind its own small cloud while the red banner 方諸取露於月 names it and beads keep forming. The final frame holds all three remainders: the empty bowl raised in the hand, the plate under the cloth, and the hall lit with 明水 and the water vessel. All six onsets land within 0.04 s.

VIOLATION: The case's object fails the round-3 material condition. At every exposed moment from 9 to 24 s (for example scene 13.0 and 20.0), the fangzhu reads as a glowing beige panel, not cold dark bronze. That undercuts the case's physical point, which is that a dark metal surface open to the sky goes cold.

EVIDENCE: The base run at scene 13.0 (film 16.0) shows the plate as a light tan quad with a radial glow inside a black bezel, tilted toward the viewer on a short stand. It is the brightest surface below the horizon. In the counterfactual touch frames (film 17.5–19.5), "Clear every cloud" gives a quick recoil and a large breath plume. The base run gives a lingering open hand and a small breath. That correctly tracks contactCold (base: dry plate, no recoil). One minor issue: in the no-clouds case the hand lands on the beads, but no bead is disturbed and the hand shows no wetness. The 390x844 viewport keeps the film on top and the meter and controls below, with no horizontal scroll (scrollWidth 390). There were no page errors.

REPAIR: Darken the plate face toward near-black bronze, for example about #2a241c to #3a3024 with a cool sky-tinted sheen only along the rim edge. Keep the moon reflection as a small hard point, not a soft glow, so the dew beads become the brightest marks on it. Show the plate's thickness or its cast bronze edge rather than a uniform bezel. Lower the tilt so it lies closer to flat, open to the zenith rather than angled toward the viewer like a screen. Optionally, have the hand disturb two or three beads at the cold-plate touch.

VERDICT: FAIL

CONDITION: Met. From the first reach to the cover, the fangzhu is a dark brown-bronze square tray on the ground, tilted up to the sky, with white dew beads on it. It looks like metal, not a screen. It has no emitted glow, no stand and no bezel. Under the cloud it is darker than the sky around it, and the only light on it is the beads.

STRONGEST: The disagreement is easy to see and the physics is right. From film 20 to 23.5 s the moon is fully behind the cloud and the beads keep growing (tests report moon visibility 0.05 and at least 2.03 beads/s). Earlier the moon shines while the plate dries under cloud. After the cloth goes on, the moon is back and the dew rate is exactly 0. The three counterfactuals hold: removing the moon leaves the 40 beads at identical times, removing the clouds gives 71 beads, and removing the teller changes nothing until his vessel visibly touches the plate corner (contact at scene 20.57 s, tilt at 20.61 s). All six onsets land within 0.04 s.

VIOLATION: The teller's marks touch the world. The red dashed arc that links his banner 方諸取露於月 to the moon ends inside the cloud's silhouette, overlapping its right lobe from about film 22.4 s to 24.9 s. That breaks "the marks never touch the world" as drawn, even though the cloud's behaviour does not change. A smaller problem: in the last frame the plate is a black cloth block with only a sliver of bronze corner showing. The object barely reads as itself, and the cold person's bowl raised toward the moon can be read as thanking the moon, which goes against the film's own end card.

EVIDENCE: The arc meets the cloud in 0.1 s renders of film 22.4–24.9 s; frames 23.5 and 25.0, zoomed, show the dashes ending on the cloud's edge. The plate's appearance was checked at film 4–27 s; the last frame was checked at 34.9 s. The three counterfactuals were rendered at film 12, 18, 21, 24, 27 and 34.9 s. At 390x844 the film stays at the top, the trace sits under it and there is no horizontal scroll. Frames and contact sheets are in /tmp/claude-0/-home-claude/21362850-8176-5f25-9feb-97722a7915b9/scratchpad (r5arc.png, r5cover.png, r5last.png, cf_noMoon.png, cf_noClouds.png, cf_noTeller.png).

REPAIR: Stop the dashed arc at least 15–20 px short of the cloud's outer edge, or route it to point at the hidden moon through open sky, so no mark pixel overlaps a world shape at any frame; then add a pixel-overlap test for the arc against the cloud and moon masks. In the last frame, let the cloth fold back or leave one full bronze edge showing so the plate reads as the plate. Have the raised bowl angle toward the hall rather than the moon.