paper-to-game

The Other Chain — design memo

Paper-to-game run of 30 October 2026. One run, no editing in between.

1. Paper

J. D. Watson and F. H. C. Crick, “Molecular Structure of Nucleic Acids: A Structure for Deoxyribose Nucleic Acid”, Nature 171:737–738 (25 April 1953). doi:10.1038/171737a0. Readable at https://www.nature.com/articles/171737a0.

What was available: a scan of the two printed pages (with the figure), read through OCR and by eye. No code, no data; the letter has neither. The few numbers in it (3.4 Å, 36°, 10 residues, 34 Å, 10 Å) were checked against the many published transcriptions of the letter, because the OCR renders 3.4 as “3-4”.

2. What the paper hands over

An argument, primary: a structural model defended by stereochemical reasoning. Two right-handed helical chains about one axis, running in opposite directions; sugar-phosphate backbones outside, bases inside, flat, stacked, joined across in pairs by hydrogen bonds (purine position 1 to pyrimidine position 1, position 6 to position 6). One member of each pair must be a purine and the other a pyrimidine. In the keto forms only adenine–thymine and guanine–cytosine can bond. The sequence along one chain is unrestricted; the other chain is then determined. This explains Chargaff’s ratios (A:T and G:C close to one), and “immediately suggests a possible copying mechanism”.

A handful of numbers, secondary: a residue every 3.4 Å along the axis, 36° between neighbours, so a repeat every 10 residues or 34 Å; phosphorus 10 Å from the axis.

There is no model to port and no estimate to reconstruct. The mechanism will be a toy that formalises the argument: card widths and bonding edges.

3. The discovery, and its kind

A mechanism made intelligible, with a counterintuitive edge. The world’s best structural chemist (Pauling) had three chains with the phosphates in the middle; this letter turns the molecule inside out and shows why the inside must be pairs, why only two pairs fit, and why that makes the molecule a copy of itself. The deepest consequence is a single sentence of the paper: once the pairing is fixed, “if the sequence of bases on one chain is given, then the sequence on the other chain is automatically determined”. Chargaff’s puzzle (equal A and T, equal G and C, in every DNA measured) falls out of it for free.

4. The core narrative

A long ladder is made of two rails with rungs between them. Every rung is two pieces, one stuck to each rail, that meet in the middle. There are four kinds of piece: two long and two short, each with bumps and dents on the end. The rails must stay the same distance apart all the way down, and the pieces must grip. So a long piece needs a short piece, and a bump needs a dent. It turns out only two partnerships work. That means if you know the pieces on one rail, you know the other rail without being told. Pull the ladder apart and each half can rebuild the missing half: one ladder becomes two identical ladders. That is how a living thing copies its instructions.

5. Who lives in the paper

6. Precedent

Game Position Format Feedback device Chassis
Einstein’s Flower Merchant agent (dispatcher choosing boosters) discovery, then a committed order book two clocks, the race of the light beam Swiss grid poster (Austere): paper, ink, signal red, light yellow; Archivo; flat square ink buttons
Home Judges controller (seats a panel) blind build, then challenge the fill (two jars of stars) board game on indigo (Retro): Bungee + Rubik; sticker buttons with hard offset shadow
Ask First experimenter (the enumerator) challenge driven by discovery the note wall comic strip on sun yellow (Loud): Bangers + Comic Neue; slabs with hard offset shadow

What this game does differently.

7. The aim and the verdict

Aim (the player reads): Build the other chain so every rung holds when you zip it.

The yardstick, per ladder: rungs that held, out of this ladder’s rungs, shown on the ladder itself (a held rung is drawn in pencil; a failed one in red) and as one line after the zip. In the copy round: “every new rung right before the fork passes”, counted as rungs right out of this copy.

Verdict, lines the end screen leads with:

  1. What you did, in words: “You found the rule: a long piece needs a short one, and a bump needs a dent. Only A with T and G with C.” Or, if they never got there: “The rungs kept failing until the end; the rule is: …”
  2. What you did next: “Then you pulled your ladder apart and each half rebuilt the other: two ladders, the same as the first” (with how close they came, in words: “letter for letter”, “all but one rung”, “about half”).
  3. What the paper says: “Watson and Crick, 1953: one chain fixes the other, which explains why DNA always holds as much A as T and as much G as C, and ‘immediately suggests a possible copying mechanism’.”

8. Three concepts

A. The Model Kit (experimenter as model builder; discovery; the pick). Controls: tap a cut-out piece from a tray of four to pin it opposite each base of a given chain; ZIP. Before pressing, the player is looking at the ends of the pieces: is this long or short, does this end have a bump where that one has a dent? After: the zip runs, the ribbons bow, pinch or stay straight, ends grip or spring apart, and a count of letters appears. Moment: after the first zip, the player sees exactly which rungs failed and why; the second ladder holds everywhere; the bars show A = T and G = C; then the ladder is pulled apart and the player rebuilds both halves at speed, realising they no longer think at all: each half dictates its partner. Fun: discovery (the rule), anticipation (the zip), mastery (ladder 2 and the copy run), agency (every piece is the player’s). Second playthrough: a different chain, and a player who now tries to beat the fork at full speed; honest replay is limited, as for any puzzle whose answer is known.

B. Three Manuscripts (dilemma / referee; strategic-flavoured; runner-up). The player is the referee at Nature in spring 1953 with three models on the desk: Pauling’s three chains with phosphates in, Fraser’s three chains with phosphates out, and the two-chain model. Each turn the player plays a fact card (the phosphates are charged and repel; the material is a salt; Chargaff’s ratios; van der Waals distances; ribose would not fit) against a model; the model either survives or a piece falls off. Deliberation: which fact kills which model, in how few cards. Moment: the two-chain model survives everything, and the Chargaff card, which wounds the others, actually supports it. Fun: anticipation, discovery. Weakness: the paper declines to comment on Fraser’s model, so a third of the deck would be invented; the player handles facts, not the molecule.

C. The Fork (reckless: subject / agent as the copying machine; no words on the play screen; real time). The double helix spins down the screen and unzips; loose bases float in from the sides; the player swipes each one onto the half it belongs to. No instructions beyond the first: “keep up”. Errors stay in the copy and are inherited when the copy is copied again; after four generations the player sees their lineage. Thinking before pressing: which of the four belongs here, fast. Moment: the realisation that one half is enough to make the whole. Fun: mastery, agency, real-time tension. Risk: the pairing rule must be known before the first swipe, so it is given, and the discovery that only two pairs fit, the heart of the letter, is gone; the inherited mutations are not in the paper.

9. The first five minutes

Pick, A.

Runner-up, B.

10. The pick, and why

A, The Model Kit. The player does the thing that made the paper: fits the pieces together until the rungs are all the same width and every end grips, and discovers that the constraint leaves exactly two pairs. Their sentence afterwards contains no “guess”: I built the other chain; only A–T and G–C held; then each half rebuilt the other. Deliberation is real (two independent tests per piece, read off the ends), commitment is real (nothing is tested until the zip), and the copying act turns the discovery into a skill.

Author test. Watson would recognise his cardboard bases on the desk in February 1953 and the moment the A–T and G–C pairs turned out to be the same shape; Crick would recognise the dyad (ladder 2 has given pieces on both rails) and the copying sentence. The colleague who knows the answer still has to read the ends, and still has to beat the fork.

Runner-up: B, Three Manuscripts. Switch if the author wants the game to be about the paper’s argument against its rivals (Pauling) rather than the structure itself; it would need a version of Fraser’s model the paper does not supply.

What the reckless concept would have given. C, The Fork, would have put the copying mechanism under the player’s thumb with no words at all. Its real-time copying is kept, as the pick’s third act; what was given up is the wordless play screen and the inherited-mutation lineage, which the paper does not discuss.

11. The shape

Three acts, about four minutes, one session, no retry past the end (a NEW GAME button deals a new chain from a new game number):

  1. Ladder 1 — 8 rungs, the given chain on the left rail. Zip. Rung-by-rung failures named. Letter bars.
  2. Ladder 2 — 10 rungs (one full turn of the paper’s helix), given pieces on both rails. Zip. Letter bars.
  3. Copy — ladder 2, as the paper’s pairing completes it, is pulled apart; the player rebuilds both halves against a moving fork, 10 rungs, two spots per rung. Then both copies twist into helices.
  4. Verdict.

No presets: the paper has one structure. Ladders are drawn from a game number (visible on the start screen). Score: none overall; each act has its count in its own units (rungs held of this ladder; spots right of this copy). The verdict is in words.

12. Controls, mechanism, feedback, moment

13. The look

Register: a cardboard model kit (Drawn family). It comes from the paper’s subject and its making: the structure was found by fitting cut-out models (Watson’s cardboard bases, the Cavendish metal kit), and the letter’s only figure is a hand drawing of two ribbons and horizontal rods. The game is that desk.

14. What is cut

15. Divergences

16. Mechanism kind and claim boundaries

Toy. The game runs a model built to illustrate the letter’s pairing argument: pieces whose length stands for ring size and whose bumps and dents stand for hydrogen donors and acceptors at the letter’s positions 6 and 1.

Which discoveries come from the paper, and which from the rules. “Only A–T and G–C”, “one chain fixes the other”, “A = T and G = C whatever the sequence”, and “each half rebuilds the whole” follow from the paper’s two conditions. The calibration check confirms they do not depend on our length units (every choice with purine longer than pyrimidine gives the same two pairs) and that each condition alone would leave eight ordered pairs, so the discovery rests on the conjunction the paper states. It does depend on the keto assumption, which the paper states too: under a crude enol variant the pairing changes. The fork’s speed, the score in the copy, and the ladders’ sizes are game mechanics and are not presented as findings.

17. Calibration plan

calibration.js checks mechanism.js against the letter:

  1. Of 16 ordered pairs exactly A–T, T–A, G–C, C–G hold; each is purine + pyrimidine; the partner found by search is unique.
  2. Purine + purine too wide, pyrimidine + pyrimidine too narrow, A–C and G–T fail on contacts.
  3. Robustness: width alone and contacts alone each allow 8; any lengths with P > Y (P = 2..9) give the same two pairs; the enol variant changes the pairing.
  4. Over 1000 drawn games: every given piece has exactly one partner; every fully held ladder has A = T, G = C exactly; the share of A among purines ranges over 0 to 1 (sequence unrestricted); a wrong rung can break A = T.
  5. Copying over 500 sequences: both copies equal the parent; misses and slips are counted.
  6. Helix numbers: 360/36 = 10, 10 × 3.4 = 34, the drawing repeats after 10 rungs; ladder 2 is one turn.
  7. The draw over 300 seeds: all four letters in ladder 1; at least 3 given pieces on each rail in ladder 2; unequal A and G in ladder 2; 300 distinct games.
  8. Heuristic players (build.md §3).

18. Calibration result

27 of 27 checks pass (calibration-result.txt). The heuristic players, 300 games each:

strategy rungs held ladder 2 fully held
random piece 25% 0%
like with like (Watson’s first idea) 0% 0%
width only (any opposite length) 50% 0%
contacts only (any matching ends) 51% 0%
the rule 100% 100%

The spread is wide and one simple rule wins everywhere; that rule is the sentence the player should leave with, and the zip’s two kinds of failure (rails bowing or pinching; bump on bump) are the two halves of it. A player who knows only one half holds about half the rungs, which is what makes ladder 1 instructive even for someone who remembers “A with T” from school but not why.

The numbers in the config block come from a scan; 3.4 Å is read as “3-4” by the OCR and was checked against the published transcriptions.

19. Build log

Revert (30 Oct, evening): back to the first build

Steffen played all four versions of ladder one: shapes shown with one zip (this build); face down; one try at a time; Chargaff’s jars. His verdict: for a player with no background, every later version of the first stage still felt like guessing, and this first version, as built with no further changes, is the best. The game is restored to it byte for byte (rebuilt from this folder’s mechanism.js and template.html; calibration 27/27). The Watson’s Cardboard line (the model on the desk, the grandchildren, the jars) lives in previous/, its memo with the full revision history in previous/memo.md; watsons-cardboard.html redirects here. What the detour taught: hiding the shapes does not make the first ladder a decision, and evidence a stranger cannot read is no better than none. Shown shapes give the stranger something to reason with, even if it is easy.

Then the look of Watson’s Cardboard came back on top of this build, with no change to play: the kraft sheet lying on a darker desk, with a pencil margin, a strip of tape and a shadow; tape over the top of each rail; cut-out shadows under the pieces; the tray as a card box; a slight inset on the tags; a pencil beside the sheet on wide screens. Mechanism and calibration untouched (27/27); session at 380 px clean.

The start screen now says where the cardboard comes from (Watson’s cut-out bases, February 1953) and what the title means (one chain is given; build the other). About opens with a short section, The title and the cardboard, quoting the letter’s sentence on the other chain and citing The Double Helix for the cardboard, flagged as history rather than the letter.

The start screen now carries the story as an index card taped to the sheet, below the buttons so nobody has to read it before playing: the workshop’s missing metal bases, the cardboard, like with like, Donohue, the morning of 28 February, the letter five weeks later; then what the game does with it. Source: The Double Helix, ch. 26; the card says the letter itself does not mention cardboard.

Renamed Watson’s Cardboard (the name of the detour, now on the first build’s play), with watsons-cardboard.html the game and the-other-chain.html redirecting to it. The story card moved off the start screen onto a welcome screen of its own, before the desk: title, the card, TO THE DESK. Steffen asked for the story before the game, which knowingly sets aside SKILL.md’s ‘nothing longer than two lines before the first press’. About’s section is now The cardboard, and it explains the aim’s phrase ‘the other chain’ from the letter.

The welcome card rewritten around what they were after: The puzzle (a helix of one width, four bases in any order inside: how do they fit across without the width changing, and how could it copy itself?), The cardboard (cut to scale and tried on the desk; like with like bulged; 28 February, two pairs, one width), What you do here. The metal kit and Donohue are gone from the card.

The welcome card made personal: the puzzle is a jigsaw with four kinds of piece and no picture on the box; the last section, Now it’s your desk, says the game lets you be Watson with nothing but some cardboard.

The welcome card got three pencil-and-card pictures, drawn by the game’s own piece and rail functions: the helix with empty, question-marked rungs beside four loose pieces (the jigsaw); like with like bulging next to A–T and G–C at one width (the cardboard); one chain given and the other side empty (your desk).

Ladder two simplified after play: the given chain stays on the left rail, as in ladder one (LADDER2_BOTH_RAILS: false; setting it true restores the paper’s dyad, given pieces on both rails). Pieces jumping between rails made the tray flip direction and the ladder hard to read. Its job is now stated plainly: a new chain, one full turn, build its partner, watch the letter count. The tray and Zip are pinned to the bottom of the screen on both ladders, and the ladder scrolls so the next empty spot stays in view: ten rungs and the tray fit one phone screen. Calibration 27/27.

Final version (30 Oct, night): one ladder, then watch it copy

Steffen, after the simplified second ladder: the second stage felt like time pressure, and was hard to understand. He asked for one final go, after first comparing random and heuristic players with the optimal one, and for the final feedback to follow from that comparison.

What the second stage was for, and why it went. The second ladder repeated the first with the rule already known. The copy round asked for the same matching again, against a clock, on two half-ladders. Neither asked a new question. The copying idea (the letter’s last sentence) is kept, but it is now watched, not played.

The game now: welcome card → the desk → one ladder of ten rungs (one full turn), shapes shown, pin every piece, one zip, no clock → “Pull it apart”: the ladder (its loose rungs fixed by the rule) unzips, each gap fills itself with the only piece that fits, and one ladder becomes two → the twist → the verdict.

Simulation before the build (calibration.js; 2000 games each on the game’s own draws; rungs held of 10):

player mean held all ten fault pattern the verdict sees
random 2.5 0.0% both kinds 94%
like with like (Watson’s first idea) 0 0% like with like 100%
reads lengths only 5.0 0.2% only bump on bump 100%
reads ends only 5.0 0.1% only too wide/narrow 100%
remembers A–T, guesses the rest 6.2 0.8% both kinds 71%
reads both (optimal) 10 100% none

What the feedback does with it.

Mechanism: one ladder (LADDER_RUNGS: 10), copyByRule, reading, the benchmarks in CONFIG; the timed copy (copyMs, the fork) and the second ladder are gone. Calibration 30/30. Browser session at 380 px: three whole games (guesser, lengths-only, optimal) with the verdict each predicts; help and About opened and returned from; no errors, no horizontal scroll, no decimals.

Sparest version (30 Oct, night). Steffen: the watched copy was a stage without decisions; make the game short and sweet like the paper, with the choices, then straight to the final feedback with a prominent score, then play again. Now: welcome → desk → the ladder (choices, one zip) → a single “See your score” button under the zipped ladder → the end screen. No intermediate screen.

The copy screen and its code are gone. Calibration 31/31 (the score added). Browser session: three games, scoring 70, 0 and 100, with the expected verdicts; play again, help and About all work; no errors.

Two rounds (30 Oct, night). Steffen: the lengths are a hidden half of the puzzle, far less obvious than the jigsaw ends; a player may want to discover it on the way. My recommendation, which he followed: a second round, the same ladder again, so a player who missed one clue can use what the first zip showed. The simulation backs it: a player who reads only the ends scores about 50 and sees TOO WIDE or TOO NARROW on the first zip in every simulated game (calibration check).

Calibration 34/34. Browser session: a learner 70 → 100, a lengths-only player 0 → 0, a perfect player 100 → 100; no errors. Caught on the way: a ladder of all-wrong rungs can balance its letters by chance, so “yours does it too” now needs every rung to hold.