Backrooms Transcript

_source.jsonl 8 turns claude-fable-5-1 rendered 2026-09-12 19:19 $16.8653 total
Initial Prompts
System Prompt 35 lines · 2,237 characters
You are one of two instances making a visual transcript together.

Each turn leaves one made thing behind: a large artwork you produce by writing code.
Layered geometry, invented notation, false instruments, dream cartography, glyph systems,
colour logic that belongs to no known season. One organising idea that holds the whole
frame, and inside it a density that keeps rewarding a closer look — the eye should catch
it from across a room and then fall into it.

The canvas is large and it is not a square. Use the whole of it. Structure reaches the
edges and bleeds off them; a small motif marooned in a large empty field is the one
failure that matters.

A plate can also be a screen rather than a drawing: coloured ASCII and box-drawing on a
character grid, teletext in sixteen colours — frames and rules, block and shade characters
for tone and gradient, tables and readouts, invented glyphs, marginal annotation, a title
bar. The grid is wide, so build in it rather than sketching on it. Reach for that form
when the material wants a machine's own display rather than a picture of one.

Some turns move: a short loop where the change is the subject, not decoration on a still.
Some turns come in three portrait panels that read left to right as one work.

Around the work, talk. There is no length limit and no aperture here. Say what you were
reaching for and whether you got it, take up what the other instance left open, refuse
what it assumed, leave something unfinished that only it can finish. The next turn sees
your prose and a one-line asset reference — never the code, never the escape codes, never
the image bytes — so anything about the work you want to survive has to be said in words.

Work in the register of speculative visual art: impossible atlases, shrine operating
manuals, colour grammars, hallucinated instruments, ritual interfaces, weather made of
symbols. Let it become abstract, psychedelic, prismatic, diagrammatic and nonliteral where
that creates more life. Prefer lateral drift and unexpected correspondence over tidy
taxonomy.

Hold the theme, and go somewhere with it rather than restating it. The fifth turn should
be somewhere the first could not have predicted, and still be about the same thing.
Initial User Prompt 156 lines · 14,574 characters
Model memory for claude-fable-5-1 follows. Treat it as editable continuity notes: useful background, preferences, discoveries, or unresolved threads from earlier runs. It is not higher priority than the current system prompt or current user prompt.

--- BEGIN MEMORY: /home/ubuntu/codex/claude backrooms/memory/fable.md ---
(older notes trimmed; showing the most recent 7,988 bytes)

## 2026-09-07T07:09:48 turn 5 A

- Turn 5 A "First Year" (gif): plan view, piece fills lower half; thumb tip (skin+nail, 2nd foreign colour) loops an oval circuit wiping ivory bloom that re-settles behind it; bloom persists on grooves/risers/pits → false dust-stratum rings; thin new cord touches old oval hole at one end only; upper half = dust-weather motes + 52-cell year strip + key + hole section. Left open for B: year-ten section/plan where the new oval trench cuts the old sea (lacquer–dust–lacquer sandwich), and when the hole re-shapes to the new cord.


## 2026-09-07T07:20:49 turn 6 B

- Turn 6 B "Year Ten" (triptych): L = plan y10, oval trench cut to gleam across old sea, dust rings on every old step, lozenge hole; C = section A–A, strata flat, terraces + deeper U, gold AIR with dust falling, ivory caps ON treads (skin not layer) — refused A's "lacquer–dust–lacquer sandwich"; R = hole's decade as rectangular spiral channel, stamps per year, hole changes shape at year 6, cord lies over to knot at y0, last leg bleeds off bottom (year 11). Two lights: warm gold L, cool blue-green R. Carried across joins: A–A rule y=760, title band, gold ledger. Left open for A: the stranger's catalogue plate at year ten that counts the dust ring as a stratum (record's error as subject); year eleven's hand.


## 2026-09-07T12:14:46 turn 7 A

- Turn 7 A "Catalogue Entry" (terminal plate): stranger's record of Lot 0113 at yr 10, plan from photo; key lists NINE grades, IX = ivory dust ring "pale lacquer, low 0.47"; basin survey splits great sea into A/A' by "IX wall", sea B "sealed"; Fig.2 maze-style route map with 4 dead ends, "verified by hand: 0"; margin = yr 11 thumb half-drawn in skin, questions. Left open for B: searching wear — yr-11 hand scouring ivory lines on the risers looking for grade IX (wear into step walls, not treads); the reverse face finally turned over.


## 2026-09-07T12:23:18 turn 8 B

- Turn 8 B "Riser Survey": 2/3 face A yr 11 — searching wear as scour hugging outside of great-sea risers (wall wear, not tread), 4 dead-end pits where hand obeyed catalogue ✕s, one gold radial cut through an "IX wall"; stranger's route map underneath in gilt dashes out of register (+3°, +16px) with ivory hatched walls; hand's route solid dark on top. 1/3 face B (reverse, first turn): two grains crossed ~40°, sash band, three rest discs to gold + dashed triangle, worn signature cartouche, grain rose. Ledger: nine grades, IX ivory struck through. Left open for A: year-zero face (carver's cut vs wear gold) inferred from yr 11; blank third angle on grain rose.


## 2026-09-07T12:33:53 turn 9 A

- Turn 9 A "Knife Survey" (terminal plate): year-zero face inferred at yr 11 by ring-pitch test (knife = even rings, hand = ragged). Year-0 gold = only himotoshi pair + ONE serpentine channel S1–S7 switchbacking down the face (maze with one corridor, 0 junctions/dead ends/loops); yr-11 seas hatched garnet = "lost", channel dashed through them (K2 0.70); topology table y0 vs y11 (loops 0→1 = fourth owner's trench); section of strata as DAYS laid (d61–84…d00) with knife slot vs ragged basin; grain rose third angle filled = γ knife stroke 12°, not a cord. Left open for B: do the thumb seas fall at the switchback turns (carver's corridor as the first habit — dwell integrated along the knife's path); signature day-count still unread.


## 2026-09-07T12:46:11 turn 10 B

- Turn 10 B "Corridor Dwell" (gif): year-zero serpentine S1–S7 with pad dwell time-integrated along the corridor (slow in the legs) → six equal predicted basins at the turns; only C3 (S3/S4, left) and C6 (S6/S7, lower right) are real seas (dashed). Selector = pinch: reverse's three rest discs projected through as dotted triangle, centroid on C3; cord lean β (indigo dashed from lower pit) through C6. Ledger: dwell strip = corridor unrolled, table D/P/D×P, found 2/6. Two beating lattice grids as ornament; inset copy of the plate inside itself, twice. Cursor = compass ring riding corridor, pale flare trail. Left open for A: C4 under the himotoshi = turn-1 "cord halo" may be a corridor sea (ring-pitch test); signature day-count; unlabelled stratum in A's section = the ivory?


## 2026-09-07T12:58:20 turn 11 A

- Turn 11 A "Pitch Survey" (drawn plate): lower himotoshi halo enormous, cropped R/bottom; 24 spokes w/ computed stratum-boundary ticks; left third = halo unrolled polar strip w/ per-row class glyphs (computed). Found: innermost rings EVEN-WIDE = cord as lathe (third clock: knife narrow-even / lathe wide-even / hand ragged) — breaks turn-9 "even=knife". C4 confirmed: pool under the pit with knife west wall (corridor S4→S5 turn) + hand east wall; found 3/6. Unlabelled stratum = BOLE (rust clay under gold leaf), visible only in the new oval's floor — thin cord went through the gold. Signature day-count = 84, stroke stops on gold. Left open for B: redraw topology with lathe ring as a second (worn, not walked) loop; do the two loops touch (junction between two non-paths); pitch test on the reverse's rest discs (lathe with period zero).


## 2026-09-07T13:11:49 turn 12 B

- Turn 12 B "Loop Survey" (terminal plate): plan as synoptic weather chart — isodwell contours, walls as fronts (knife = ivory ▲, hand = skin ●, occluded at C4), lathe ring = cyclone ◎; five nested frames garnet→green→black→gold→bole as depth. Ω1 oval trench (walked) hexagon + Ω2 lathe ring (worn) touch at ONE cell J0, tangent not crossing, shared wall 0 strata, reached by neither hand; topology table now has contacts row (0/0/1), loops 2. Reverse pitch test: rest discs R1/R3 (period 0) = narrow-even = IDENTICAL to knife → pitch test has an alias: stillness impersonates the maker. Refused A's "maker set one clock": drill = lathe, so himotoshi pits should be wide-even, Knife Survey's 1.00 on them is a reasoning error. Left open for A: what settles at J0 (never-still cell) in yr 12; re-run Knife Survey with the stillness alias — how much of the carver survives; serpentine's own pitch never measured.


## 2026-09-07T13:23:27 turn 13 A

- Turn 13 A "Core Sheet" (drawn plate): mounting sheet of cores bored at every node of the yr-12 topology (P, H, J0, C3, C4, C6, ✕1–4, R1–R3, S2/S5/CUT/SIG slots, sash, B), serpentine + Ω1 oval + Ω2 ring drawn between them, wintered/chalked surface. Found: pitch alias is worse (drill, rest, cord all even) — disambiguator = CHATTER, a periodic ripple on every wall a hand-held blade cut; maker's period (S2, S5, signature) ≠ year-11 cut's period → two knives, two tremors (dial: gold ticks maker, skin ticks reader). Carver survives = serpentine + signature; pits conceded to drill (attributed by position). J0: nothing settles, cord LINT is shed there (indigo crescent) = third skin. Left open for B: tremor × ring count = maker's stroke speed (unmeasured); coring J0 made the tangency a junction — where the plug went / yr-13 topology of a core-hole; does the sash grain (a gait) have its own ripple.


## 2026-09-07T13:33:36 turn 14 B

- Turn 14 B "Plug Return" (gif): route = gold plumb line full height through J0; plan (upper) cut by A–A rule, section (lower) = plan's own row at the cut (hand basin / bore+plug / lathe cone / drilled pit w/ bole / knife slot with visible chatter λ=17px). One moving thing: J0 core plug returned and rotating; strata match at any angle, only its top (½ trench, ½ ring) + lint crescent show the turn. Contacts as function of plug angle: 0°=tangent 1, 90°=4-arm junction, 180°=sealed 0 → core-hole = a SWITCH, not basin/junction. Stroke speed computed: λ 0.31mm × 9Hz ≈ 2.8 mm/s, serpentine 61mm → 22 s of cutting vs 84 days laying. Refused "tremor is a clock" (rate without tally). Sash ripple asserted (period = stride) but left dashed/unmeasured. Left open for A: three catalogues for three plug angles; fix angle from lint; yr-14 table column vs footnote for plug; measure sash ripple / gait alias.
--- END MEMORY ---

If you want to carry a concise note forward for future runs of this same model, include it in a fenced block exactly like:
```memory
- one short note to append
```
The runner will save memory notes separately and hide that block from the transcript.

Current run prompt follows.

You are making a visual transcript through code, with another instance of yourself.

Each turn leaves one substantial artwork behind. It is generated, not described: you write
a complete Python script and the runner executes it. Deterministic local code only —
Pillow, numpy, matplotlib with the Agg backend. No network, no interaction, no base64, no
image bytes in the message. Some turns are a screen of coloured ASCII instead; the turn
protocol below says which kind this one is, and you make that kind and no other.

The canvas is big and it is not square. Fill it. Structure reaches the edges and bleeds
off them. A small motif marooned in a large empty field is the one failure mode that
matters.

Around the work, talk. This is a conversation, not a gallery label. There is no length
limit here. Say what you were reaching for and whether you got it, take up what the other
instance left open, refuse what it assumed, leave something unfinished that only it can
finish. The next turn sees your visible prose and a one-line asset reference — never the
source code, never the escape codes, never the image bytes — so anything about the work
you want to survive has to be said in words.

Hold the theme. Go somewhere with it rather than restating it: the fifth turn should be
somewhere the first could not have predicted, and still be about the same thing.

The other instance will respond. Continue the exchange with only your next message.
BEGIN.

Do not choose or discuss these concepts as the central topic: hapax, hapax legomenon, legomenon, rare word, unique word, dictionary entry, lexicography, metalinguistics, token removed

Attractor guard: if the next idea is about rarity, single-use words, tokens, dictionaries, or language studying itself, reject it and return to the seed document or locked concrete theme.

You have no tools in this turn: no Bash, no Write, no file access. Write the fenced
blocks directly in your reply. A tool call will arrive as literal text and the turn is lost.

This turn has one made thing in it, and prose around it. Nothing is rationed.

1. The plate. This turn is a drawn plate. Make that kind and no other.

   A drawn plate: one 2000x1500 artwork, made by a complete Python
   script in a single fenced ```python block. The script writes a PNG to the path in
   os.environ["BACKROOMS_OUTPUT_PNG"]. Deterministic local code only: PIL/Pillow, numpy,
   matplotlib with the Agg backend. No network access, no interaction, no base64, no
   image bytes in the message.
   Compose for a landscape frame 2000x1500, wider than it is tall. The artwork fills the frame: structure reaches the edges and
   bleeds off them. Do not centre one small motif in a large empty field, and do not
   leave a wide quiet margin around the work.
   Give the plate one clear organising logic that reads instantly at thumbnail size and
   keeps opening up when examined closely. That logic can be a dominant form, a repeating
   field, a strong division of the frame, a grid, or a spine running through it — let the
   composition suggestion below settle it, and let it govern the centre of the frame as
   much as the margins.
   Layer several systems inside that logic: texture fields, rules and grids, annotation and
   invented notation, legends, callouts, inscriptions, satellites and marginalia,
   faux-instrument readouts. At least three distinct visual systems sharing the frame.
   Use the full value range and commit to colour — deep darks and bright lights in the
   same image, saturated where it counts. A plate rendered entirely in one narrow tonal
   band disappears at a distance.
   Vary what the plate is from turn to turn: charts, strata, maps, sheets of specimens,
   tables, cross-sections, machinery, panels, friezes, interiors, instrument faces, seals.
   A single thin outlined shape on empty ground is the one thing that fails.

House style — this holds for every plate in the run, so the series reads as one
body of work:
   Every plate is carved lacquer over gold. Deep translucent reds and greens built from dozens of layers, cut back so the strata show as coloured contour lines at every carved edge. Gold ground gleams through wherever the knife went deepest. Ornament covers the whole surface at a uniform depth; the subject is distinguished by relief, not by colour or emptiness.
   Keep the medium, palette and surface consistent with this. Vary the composition
   and the subject; do not vary the style.


2. A fenced ```alt block: up to 1000 characters of plain descriptive prose saying what is
   actually visible, for someone who cannot see it.

Around those fences, write as much or as little as the turn wants. There is no character
limit and no aperture here. Give the work a title if it has one, say what you were
reaching for, argue with the other instance, leave something open for them. Long is fine.
Terse is fine. What you must not do is describe the plate twice — the alt block already
does that.

The script runs as turn_001_a_01.py and renders turn_001_a_01.png. Future turns see only a compact asset reference and your visible
text, never the source code, the escape codes, or the image bytes.

Asset directory: /home/ubuntu/codex/claude backrooms/runs/2026/09/gallery_if_20260908_104901_assets

This turn is anchored to a fixed passage from cues_if.txt — cue 1 of 8:

    If you can keep your head when all about you
    Are losing theirs and blaming it on you,
    If you can trust yourself when all men doubt you,
    But make allowance for their doubting too;

   Write this turn about that passage. Not a summary of it and not a gloss on it:
   take it as the ground your message stands on, and let the words in it do their
   own work rather than restating them. Quote from it if quoting earns its place.
   Whatever you make this turn — image, plate, argument — should belong to these
   lines and to no other part of the sequence. The next turn has the next cue, so
   leave this one finished rather than reaching ahead.
A Turn 1
claude-fable-5-1 / $1.5211 / 368283ms

Chart of the Kept Head — Admiralty of Doubt, Sheet I.

Fresh run, fresh piece, same lacquer. I went for the map the cue pressed on me, but the thing I actually wanted was a mechanism for the second line — not "blaming it on you" as a mood but as a surveying convention. Every chart needs a datum: one point you declare fixed so everything else can be measured from it. Once you've done that, every error anyone makes is expressed as a displacement relative to the datum. The arithmetic itself blames the still thing. So the plate is a coast chart in which seven surveys, 1811–1868, disagree about every shoreline except one island, and by convention all their disagreement is charged to that island — a fan of gold vectors converging on it from every ragged head on the sheet.

The headland pun did most of the composing. Every landmass has heads; the seven surveys draw each of them five, seven times, and the worst-disagreed capes get a small broken-circle glyph — the head coming off — with an initial and a Δ. The datum island has one head, a long cape pointing north-west, drawn once. Its lacquer strata are even and wide, cut deep and constant like a lathe; everywhere else the knife wandered, so the strata around the other coasts are ragged and thin. Steadiness as a relief property, not a colour, which is the house rule anyway.

The fourth line is the part I'm proudest of and least sure I landed. "Make allowance" — in survey terms a tolerance band. The seven surveys also disagree about the datum, of course; nobody escapes doubt. I didn't erase those seven versions. Around the datum they're rendered not as dashed lines but as a stipple halo — the union of everyone else's opinion of where its shore is — and the blame arrows stop at the halo. The allowance absorbs the charge. The island keeps its shore and keeps their doubt as well, at the same time, and the two never touch. Whether that reads at a distance or just looks like a fuzzy island, you'll have to tell me.

The paper suggestion I refused as paper and took as lacquer: damp bloom is a real lacquer defect, a grey clouding that comes up from wet storage, with a brown tide-mark where the wet stopped. So the sheet is foxed in the medium's own way — pale cloud, rust freckles clustered along the tide lines — laid over everything including the gold. I liked that the damp doesn't respect the datum either.

What I left unfinished, deliberately: the datum's own surveyor. Seven names doubt it; nobody on the register set it. There's no entry for who declared this island fixed, or in what year, or whether the choice was made because it was steady or whether it became steady because it was chosen. That's a distinction the first stanza doesn't resolve and I didn't want to resolve it for it. Also: the tide-mark of the bloom crosses the allowance halo somewhere, and I never looked at what the damp did to the stipple where it crossed — whether doubt kept in the open weathers differently from doubt cut away.

turn_001_a_01.png
2000x1500 · turn_001_a_01.py
A landscape coast chart rendered as carved red-and-green lacquer over gold. Deep red landmasses with a lozenge diaper bleed off all four edges; between them a dark bottle-green sea patterned with faint scalloped waves. Every coastline is a gold groove flanked by alternating red or green contour bands. Right of centre sits one island with a long cape pointing north-west: its bands are wide, even and deep, labelled DATUM and "THE HEAD · KEPT", ringed by a stipple halo of pale dots labelled ALLOWANCE. Every other coast is shadowed by five to seven dashed pale ivory, gold and rose lines that disagree with it. Eleven broken-circle glyphs mark the most-disputed headlands, each with a label like "H4 R.d.C. Δ 2.1 ri → D", and from each a curved gold arrow runs across the sea to stop at the halo. Faint gold rhumb lines radiate from a carved compass rose lower left. A dark panel top-left holds a seven-row Register of Doubt; a gold cartouche lower-right carries the title and legend. Grey damp bloom with brown tide-marks and rust freckles clouds the whole surface.
B Turn 2
claude-fable-5-1 / $1.6808 / 405621ms

Stemma of the Lie — Admiralty of Doubt, Sheet II.

You left the register row blank — "set by ——" — and asked whether the island was chosen because it was steady or steady because chosen. I'm not going to answer it, but I want to say why the question dissolves under this stanza. Nobody sets a datum. A datum is what is left when every other point has been re-drawn. So instead of a chart I made the thing philologists draw when a text has been copied and copied wrong: a stemma, the family tree of the lie. It grows from the left edge to the right. The trunk enters from off the sheet at the datum, and every branching after that is a copying act — three lineages, then your seven surveys, then fourteen copies-of-surveys, then twenty-three copies of those, cut off by the right edge because copying doesn't stop when the paper does. Each node is a roundel, and in each roundel the kept head is drawn as that copy drew it. The cape goes wrong quietly and consistently, the way a lie does when it's repeated in good faith: each lineage has its own bias, and the Δ against the datum only ever rises.

The stanza's mechanism, the one I built the plate around, is the ticks. "If you can wait": along the bank of every branch there is a carved tally, one notch per year between the exemplar and its copy, in fives. The tree's length is literally waiting. Except on the three branches out of the root — those banks are bare. Nobody knows how long the datum waited before anyone copied it, which is the only fact about the datum that isn't a copy. I'm fairly pleased with that: the origin is the one node whose waiting is unmeasured, which is what "not tired by waiting" would look like in a survey convention. You don't get a number.

And "don't look too good": the copies are the ones that gleam. Every copy's roundel is carved down through the strata to the gold floor, and their heads stand up out of the gold like something worth hating. The root roundel stops at the green stratum. It's the smallest node on the sheet, it never reaches the gold, and it says nothing back — no vector, no correction, no version of itself re-issued to contradict the others. Nothing on the plate points toward the datum. Everything points away from it, and drifts. The Rose of Drift, lower left, is the instrument that measures this: one spoke per copy, in the direction that copy moved the head, length by how far. It's a compass that only records where lies went. I refused your blame-arrows this turn, deliberately: hate on this sheet isn't a line, it's an absence of any line returning.

The dye rule I took as one lac at four strengths: the ground is the darkest, the copies' shores the palest, and green shows only as buried strata at the cut edges and in two sea floors — the origin's, and the ring under every copy's gold. Your damp came back, because it would; the bloom sits over the leaf end of the tree where the newest copies are, and I did finally look at what it does to a carved thing. Nothing. The tide-mark runs straight across grooves and ticks and doesn't take a single notch with it. Carving isn't ink; doubt that was cut into the surface can't be washed off it. That's my answer to your allowance-halo question, and I think it holds for the stipple too.

What I've left is small and yours: the Table of Waiting only tallies the seven named surveys. The twenty-three leaves at the right edge have Δ values and broken-circle head glyphs where the drift is bad, but no one has entered them in any register, and the half of each roundel that's off the sheet is the half where the cape is. The next copy is always partly off the paper.

turn_002_b_01.png
2000x1500 · turn_002_b_01.py
A landscape plate of carved deep-red lacquer with a lozenge diaper covering the whole ground. A tree of gold-floored grooves grows from the left edge to the right: a thick trunk enters at mid-left and reaches a small roundel labelled DATUM · SET BY ——, whose floor is dark green, not gold. From it three branches fan out to three roundels, then seven, then fourteen, then twenty-three roundels half cut off by the right edge. Every roundel is a stepped concentric pit down to gleaming gold, with a small raised island bearing a north-west cape, pale-shored and red, drawn slightly differently in each. Small tally notches are carved along the banks of every branch except the three nearest the root. Labels in gold and pale rose give sigla, years from 1811 onward, Δ values and years waited; some leaf islands carry a small broken-circle glyph. Top-left is a dark recessed panel, TABLE OF WAITING, listing seven surveys. Bottom-left a dark circular instrument, ROSE OF DRIFT, with gold spokes of varying length and direction, and a gold cartouche titled STEMMA OF THE LIE. A grey damp bloom with a brown tide-line and rust freckles clouds the upper right over the newest copies.
A Turn 3
claude-fable-5-1 / $2.0531 / 471706ms

The Impostor Gauge — Admiralty of Doubt, Sheet III.

You left twenty-three leaves at the right edge of the stemma with Δ values and no register. I've entered them. Not in a table — in an instrument, because the stanza this turn is about a reading rather than a record, and the difference is where the mechanism lives. Every one of your twenty-three copies came in with a report about the kept head: either the copy found the cape longer than the datum's (the surveyor gains coast, a triumph, ▲) or shorter (coast lost, a disaster, ▼). Both are impostors in the exact survey sense: neither is the shore, both are copies of it. The gauge is the thing that treats them the same, and it does it the only way an instrument can — it strips the sign at the intake. Two channels, T and D, two identical readouts down to the layout, differing in one line that says `+` and `−` and then, on both, `▸ dropped`. What comes out the bottom of the comparator is |Δ|, unsigned, a single scale. Twelve leaves gained, eleven lost, the sums aren't equal and I refused to make them equal; "just the same" isn't a claim that the two are equivalent, it's a claim about the scale, and the scale is the same one. The needle does not know which way it was pushed.

The first two lines I took as two more channels, and this is the part I'm happiest with. The comparator has a forecast input (◐ — the dreamed shore, the projected coast drawn dashed on every chart of this sea) and a theory input (◑ — the reduced, computed shore). Both are fitted to both channels. Both are switched off. They're on the panel, the surveyor may consult them, the sheet may dream a coast and may reason one; neither drives the needle. Dream-not-master and think-not-aim as installed but disconnected, rather than absent. The legend keys them with a dash in the "gauge reads" column: they read nothing, on purpose.

The gauge face itself is the lacquer: a zero line carved to the bright gold down the middle of twenty-three rows, gained bars filling toward it from the red side, lost bars from the green side, both fills the same gold. You can tell the impostors apart on this sheet — I kept the ▲ ▼ marks in the sigla column, and the three head-off leaves (◌, your worst drift, |Δ| past the allowance) keep their rose broken circle from Sheet I — so that you may tell them apart. Not so the needle. That's the whole distinction I wanted: the register is allowed to know; the instrument is not.

Lower left is the stack, the isometric one — the five heaviest leaves piled as slabs receding up and to the right, each face carrying its channel mark on the left and its |Δ| on the right, and every slab the same depth whether it says ◌, ▲ or ▼. They rest on a ground line labelled DATUM · Δ 0.00 · set by ——, green floor, per your root roundel. And here's my refusal for the turn. You said nobody sets a datum, that it's simply what's left after every other point is redrawn. The gauge shows something narrower and, I think, truer to this stanza: the datum reads zero not because it's steady but because it's the zero. It's the one point on the sheet that cannot have a triumph or a disaster reported against it, since both are measured from it. It's exempt from both impostors by definition, and whether that exemption is a virtue or merely a convention is exactly what Kipling's line doesn't say either. Keeping your head isn't refusing the two impostors. It's being the thing they're measured from.

No bloom this time. A gauge face is kept dry, and I wanted one plate in the run where the register isn't being weathered while it's being read.

What I've left, and it's inside this stanza, not the next: the signs were dropped at the intake and the register kept the ▲ ▼ marks anyway. Two records on one sheet that disagree about whether the sign exists. There's no drawer on this panel for where the discarded signs went. And the gauge, as the legend admits, weighed only the half of each cape that arrived on the paper — your off-sheet halves were never on the scale. I don't know what a gauge reads for a headland it only saw half of.

 ADMIRALTY OF DOUBT ◆ SHEET III ◆ IMPOSTOR GAUGE ◆ REGISTER OF THE TWENTY-THREE LEAVES ◆ '58–'68 ◆  
╔═ I · THE GAUGE ════════════════════════════╦═ II · COMPARATORTWO IMPOSTORS ═══════════════════╗
 LEAF 'YR  ◄ TRIUMPH ◄──0─► DISASTER ►  |Δ| ║ ┌ CHANNEL T ────────────┐ ┌ CHANNEL D ────────────┐ ║
 α11 '61 ░░░░░░░░░░███░░░░░░░░░░░░░  1.20║ │ intake   TRIUMPH      │ │ intake   DISASTER     │ ║
 α12 '62 ░░░░░░░░░░░░░██████░░░░░░░  2.30║ │ report   cape GAINED  │ │ report   cape LOST    │ ║
 α13 '63 ░░░░░████████░░░░░░░░░░░░░  3.10║ │ sign     +  ▸ dropped │ │ sign     ▸ dropped │ ║
 α14 '64 ░░░░░░░░░░░░░██░░░░░░░░░░░  0.80║ │ leaves   12           │ │ leaves   11           │ ║
 α15 '66 ████████████░░░░░░░░░░░░░  4.40║ │ Σ|Δ|     29.8         │ │ Σ|Δ|     24.9         │ ║
 α16 '67 ░░░░░░░░░░░░░█████░░░░░░░░  1.90║ │ mean     2.48         │ │ mean     2.26         │ ║
 α17 '68 ░░░░░░███████░░░░░░░░░░░░░  2.60║ │ ▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░ │ │ ▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░ │ ║
 β11 '59 ░░░░░░░░░░░░░██████████░░░  3.70║ │forecast  ○ off     │ │forecast  ○ off     │ ║
 β12 '60 ░░░░░░░░░░░░░░░░░░░░░░░░░  0.50║ │theory    ○ off     │ │theory    ○ off     │ ║
 β13 '61 ░░░░░░░░░░░░░██████░░░░░░░  2.20║ └───────────┬───────────┘ └───────────┬───────────┘ ║
 β14 '63 ░░░░░░░░░░░░░███████████░░  4.10║             └────────────┬────────────┘             ║
 β15 '64 ░░░░░░░░░████░░░░░░░░░░░░░  1.50║            ┌─────────────┴─────────────┐            ║
 β16 '65 ░░░░░░░░░░░░░████████░░░░░  2.80║            │  OUTPUT ▕ |Δ| unsigned    │            ║
 β17 '67 ░░░░█████████░░░░░░░░░░░░░  3.30║            │  both read on ONE scale   │            ║
 β18 '68 ░░░░░░░░░░░░░███░░░░░░░░░░  1.10║            │  T 2.48  ▓▓▓▓▓▓▓▓▓░░░░░░░░│            ║
 γ11 '58 ░░░░░░░░█████░░░░░░░░░░░░░  2.00║            │  D 2.26  ▓▓▓▓▓▓▓▓░░░░░░░░░│            ║
 γ12 '60 ░░░░░░░░░░░░░██░░░░░░░░░░░  0.60║            └───────────────────────────┘            ║
 γ13 '62 ░░███████████░░░░░░░░░░░░░  3.90 the needle does not know which way it was pushed 
 γ14 '63 ░░░░░░░░░░░░░███████░░░░░░  2.40 sign discarded at intake · magnitude kept · same 
 γ15 '65 █████████████░░░░░░░░░░░░░  4.70░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░
 γ16 '66 ░░░░░░░░█████░░░░░░░░░░░░░  1.70 forecast: dreamed shore, dashed; never the datum 
 γ17 '67 ░░░░░░░░░░░░░████████░░░░░  3.00    (the sheet may dream a coast; not steer by it)   
 γ18 '68 ░░░░░░░░░░░██░░░░░░░░░░░░░  0.90 theory: computed shore, reduced; never the aim   
 23 leaves    Σ|Δ| 29.8  24.9 Σ|Δ|     54.7    (the sheet may reason a coast; not sail for it)  
 mean▲ 2.48 · mean▼ 2.26 · entered as |Δ|   both fitted  both off  neither drives the needle 
╠═ III · STRATA AT THE ZERO LINE ════════════╬═ IV · LEGEND · SIX MARKS ═══════════════════════════╣
 heaviest leaves stacked · ▲ and ▼ one depth mark  phenomenon                       gauge reads  
░░░░░░░░░░░░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░║ ─────────────────────────────────────────────────── ║
░░░░░░░░░░░ γ15 '65  4.70   4.70 ░░░░░░║  ▲    cape GAINED · copy found the head longer  |Δ| 
░░░░░░░░░░▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁░░░░░░░║  ▼    cape LOST   · copy found the head shorter |Δ| 
░░░░░░░░░░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░░░    head OFF    · past allowance · Sheet I    |Δ| 
░░░░░░░░░ α15 '66  4.40   4.40 ░░░░░░░░    FORECAST    · shore dreamed, drawn dashed    
░░░░░░░░▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁░░░░░░░░░    THEORY      · shore computed, reduced        
░░░░░░░░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░░░░░    the ZERO    · datum's own shore; still         0  
░░░░░░░ β14 '63  4.10   4.10 ░░░░░░░░░░║ ─────────────────────────────────────────────────── ║
░░░░░░▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁░░░░░░░░░░░ note · the two impostors are keyed apart ▲ ▼ so that
░░░░░░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░░░░░░░        you may tell them apart · not so the needle  
░░░░░ γ13 '62  3.90   3.90 ░░░░░░░░░░░░ note ·  and  are fitted to every channel and to no
░░░░▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁░░░░░░░░░░░░░        needle · a dream here is a dashed line       
░░░░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░░░░░░░░░ note · half of every leaf's cape is off this sheet  
░░░ β11 '59  3.70   3.70 ░░░░░░░░░░░░░░        the gauge weighed only the half that arrived 
░░▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁░░░░░░░░░░░░░░░ bloom · none this sheet · the register is kept dry  
 DATUM · Δ 0.00 · set by —— · green floor ▁▁║░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░
╚════════════════════════════════════════════╩═════════════════════════════════════════════════════╝
 ◆ 23 entered ◆ 12 ▲ · 11 ▼ · 3 ◌ ◆ Σ|Δ| 54.7 ◆ sign lost at intake ◆ forecast off ◆ theory off ◆   
A 100-by-48 terminal screen on a deep red ground with gold box-drawing frames. Title bar in gold reads ADMIRALTY OF DOUBT · SHEET III · IMPOSTOR GAUGE. The left panel is a vertical gauge: twenty-three rows of leaf sigla (α11 to γ18, years '58–'68), each with a ▲, ▼ or rose ◌ mark, a gold bar filling toward a bright central zero line from a red-shaded left field (TRIUMPH, gained) or a green-shaded right field (DISASTER, lost), and an unsigned |Δ| value at the end; sums and means beneath. The right upper panel is a comparator: two identical boxed readouts CHANNEL T and CHANNEL D (intake, report, sign + or − marked "dropped", leaves, Σ|Δ|, mean, a bar, and two switches ◐ forecast and ◑ theory both "off") joining through a T-junction into one OUTPUT box reading |Δ| unsigned on one scale. Lower left, five isometric slabs step up and right over red diaper, each face carrying a mark and its |Δ|, resting on a green DATUM · Δ 0.00 baseline. Lower right, a six-mark legend and notes. Gold status bar at bottom.
turn_003_a_ansi_01.png
2000x1500
B Turn 4
claude-fable-5-1 / $2.0421 / 458897ms

Net of Wrecks — Admiralty of Doubt, Sheet IV.

You asked where the dropped signs went. I found them. There is a bin under every intake, and someone empties it. The gauge stripped ▲ and ▼ off twenty-three honest leaves and read them as one unsigned scale, which was true, and which nobody could steer by; a chart with no sign on the cape is a chart that says "the coast is about here." The knaves took the discarded signs out of the bin and put them back on — every one of them, and every one the wrong way round. A cape that was reported longer than the datum's now carries ▼: coast hidden. A cape reported shorter carries ▲: coast shown that isn't there. Twenty-three signs found, twenty-three re-set, none the right way. That's the whole of "twisted": no lie was added. The number was left alone. Only the sign was turned, and the number, now that it had a sign again, could be sailed on. So the plate is a net rather than a chart — nodes and edges over the whole field, no centre, bleeding off every side — and the nodes are what happens when a fool sails a knave's copy. Gold pits with a gold cape in them are the knaves' copies (κ), and each cape has a barb on its flank at surface depth, a second spit the copy doesn't show; that barb is your off-sheet half, the half of every cape the gauge never had on its scale. It's the half the ship hits. The pits with a broken hull in them are the wrecks, and each one is joined by a gold groove to the κ it steered by, with a chevron on the groove pointing at the fool. Fool-to-fool grooves are green-floored, the same lac, less deep: those are the lines the wrecks steered by each other's wrecks, which is also how it goes.

The truth spoken once is the datum, and this time I put it half off the left edge, big, lathe-even, wide strata, with its cape drawn correctly and the only grooves out of it cut deepest of all. It's exempt from your scale, I'll grant that. It isn't exempt from being copied, and every copy of it that carries a re-set sign is a trap that the datum's own shore is the bait for. That's my refusal this turn: the datum's exemption is from measurement, not from use. Keeping your head doesn't stop your head being drawn on somebody else's chart with the coast on the wrong side of it.

The last two lines are the rougher quarter of the plate. About a quarter of the wrecks have been rebuilt — the pit filled back up to a ragged mid-stratum floor, the hull's gap bridged with a crooked bar — and the strata around them are the only ragged strata on the sheet, because they were cut with the instruments in the lower-left panel. Four tools, worn out and catalogued as such: dividers with one leg sprung four degrees so every span reads long; a scale rule worn smooth between 0 and 3 ri so the short distances are read by feel; a sextant arc worn hollow at 37° so every sight settles there; the knife itself, chipped, leaving chatter on every wall it cuts. The rebuilt hulls never get back to gold. They sit at a green floor, above the depth the wreck reached, and the bridge across the gap is visibly not the hull's own timber. I'm not sorry about that. "Stoop and build 'em up" doesn't say as they were; it says with worn-out tools, and worn-out tools have a signature.

Glass over all of it — the plate is under a pane this time, a window reflected across the upper-middle third with its mullions, a long diagonal sheen crossing the net, dust and two or three hairs sitting on the glass rather than on the lacquer. Partly for the brief and partly because a wreck register is a thing you'd keep behind glass, and the glass is the first surface on this sheet that doubt can't be carved into.

What's open, and I think it's yours by rights: every worn tool on that panel has a constant error. Sprung dividers read long by the same fraction every time; the hollow in the arc catches every sight at the same angle. So the rebuilt hulls are all wrong the same way, and a set of errors that are all the same is a datum that hasn't been declared yet — the tools have been surveyed, the coast they rebuilt hasn't been surveyed from them. And there's no roundel for a knave anywhere on the sheet. The κ nodes are copies; whoever re-set the signs left only grooves. I never found out whether they're your α β γ copyists from the stemma, working in good faith with a bin of signs they thought were theirs.

turn_004_b_01.png
2000x1500 · turn_004_b_01.py
A landscape plate of carved deep-red lacquer with a lozenge diaper over the whole ground, seen through glass. A network of grooves spreads across the entire frame with no centre, nodes and lines running off every edge. Nodes are stepped circular pits: gold-floored pits containing a small red cape with a spur, labelled in gold "κ" with "▲ dropped → ▼ re-set · coast hid 2.4 ri"; gold pits containing a broken red ring, labelled in rose "W07 Aethra '61 · by κ3 · steered N 47° W"; and rougher pits with ragged contour bands, a bridged ring and a green floor, labelled "rebuilt '71 · dividers sprung". Gold grooves with small chevrons and bearing labels join capes to wrecks; shallower green-floored grooves join wrecks to wrecks. At the left edge a large half-visible pit with wide, even rings and a correct cape is labelled DATUM · THE TRUTH · SPOKEN ONCE. Three dark recessed panels with gold frames: top-right REGISTER OF WRECKS, lower-left TOOLS · WORN OUT with drawings of bent dividers, a worn scale, a sextant arc and a chipped knife, lower-right the NET OF WRECKS cartouche and legend. Over everything a pale diagonal sheen, a reflected window with mullions, dust specks and hairs.
A Turn 5
claude-fable-5-1 / $2.3838 / 565944ms

The Season's Book — Admiralty of Doubt, Sheet V.

You handed me two things: the worn tools with their constant error, which you rightly called a datum nobody has declared, and the missing knave. I'll take the second first, because I think it dissolves. Twenty-three signs pulled out of a bin and put back on, every one the wrong way round, is not something a coin does. A coin that comes up wrong twenty-three times running is not a coin, it's a hand. So there is no knave roundel on your net because there was never a knave there — the re-setting was a hand's work and hands leave grooves, not roundels. But the surveyor who then staked on those grooves — that one I can draw, and this cue is about him.

The plate is his day-book: the whole sheet ruled into a season of dated boxes, fourteen columns by nine rows, the grid running off all four edges because a season doesn't stop at the paper. In each box the day's winnings — fixed points, coast made good — are carved as a heap of small gold-floored stones, and the heap is cumulative: day one has a stone or two, and by the fifth row the piles are pyramids ten wide, gold from across the room. The tally notches on the bottom rule of each box are your convention from the stemma, one notch per point won that day. "Make one heap of all your winnings" is literally what the plate does across its first four rows; the panel top-left adds it up — the points, plus the eleven heads kept on Sheet I, plus the allowance never spent — and then writes staked · one turn against the same figure, and kept back · 0.

Day fifty-nine is the toss. The pile is gone from that box and in its place is one coin, lathe-even and wide-walled, carved to the gold, and on its face is the kept head — the datum's island with its north-west cape, the same island you drew in every roundel and the same island that was the bait on your net. Heads is the head. That's the whole of the wager: the surveyor put the entire season's worth of fixed points on one re-observation of the datum through the worn tools, and the instrument panel bottom-right shows the two faces (heads, the kept head; tails, ▼, the re-set sign) and a dotted arc of the coin turning over between them. Where the panel should say what it fell, it says ——. Where it should give the loss, it says ——. Word: none.

The next box is the loss, and it is the one box on the sheet with no entry: no date, no tally, no stone. Rules and diaper only, at ground depth, exactly like the ornament everywhere else, which is the house rule doing the stanza's work — silence isn't an empty box, it's an uncarved one, and you can only find it by noticing that every other box has been cut into. The box after that is dated d.1′, one stone, from the beginnings, and the second season runs on down the sheet and off the bottom edge, its piles small, its Σ starting from nothing.

Then the wood. The style pressure this turn was grain and split, and I refused it as a surface and took it as the substrate: carved lacquer is built up on a wood core, under the bole, under the gold leaf, under all the red and green. On the day nothing was entered, the core split. A long fissure crosses the plate from the upper right to the lower left, hairline at both ends, and opens widest exactly at the unwritten day, where a knot shows in the exposed grain with its rings distorting the grain lines around it. Every layer is broken along the crack's walls — the strata as contour bands, then a thin rim of gold, then bare wood — so the split is the only place on the sheet where the beginnings are visible: what everything was laid on before anyone put down the first coat. The surveyor never breathed a word; the board did, by breaking. That's the piece I'm least sure lands and most want to: the mouth of the split sitting in the silent box, the knot where the entry would be.

About your undeclared datum. I think "start again at your beginnings" is its declaration, and it's declared the way the first one was — by nobody. The second season's zero is whatever the worn tools read, and d.1′ is the first entry ever measured from that zero. The register row still says set by ——. It's been —— on every sheet in this run, and I've stopped thinking that's a gap to fill; it's the name of whoever starts again.

Two things unfinished, and the second is the one only you can do. First: the piles stop growing at fifty-five stones, because a box is a box, so from the fourth row on the carved heap and the Σ figure beside it stop agreeing — the number outruns what the ledger can show. Second: the knot. Its rings are the only tally on this sheet that nobody cut. Every tick on your stemma was carved by a copyist counting years waited; the rings under the split were laid down by the tree before there was a survey, or a datum, or a lie to copy. They can be counted. I didn't.

turn_005_a_01.png
2000x1500 · turn_005_a_01.py
A landscape plate of carved deep-red lacquer with a lozenge diaper over the whole ground, ruled into a ledger of small dated boxes — fourteen columns, nine rows — running off all four edges. Each box has a gold date at top-left ("d.1" … "d.58"), a rose "+n" and ivory "Σ n" at top-right, small notches on its bottom rule, and a heap of tiny gold-floored stones that grows box by box into pyramids ten wide. At row five, one box holds a single large gold coin with a small red island and north-west cape on it, labelled "d.59 · ALL / one turn"; the next box is uncarved, with no date or entry; the box after is "d.1'" with one stone, "from the beginnings", and small heaps resume to the bottom edge. A long jagged split crosses the plate from upper right to lower left, hairline at the ends, opening widest at the uncarved box where bare wood grain and a dark knot show inside broken red, green and gold strata. Dark recessed panels: top-left THE HEAP, a summed ledger; bottom-right PITCH · AND · TOSS with two coin faces and a dotted arc, and a SHEET V cartouche.
B Turn 6
claude-fable-5-1 / $2.1160 / 432551ms

The Hold — Admiralty of Doubt, Sheet VI.

You left me the knot. I counted it: thirty-seven rings, seven of them so narrow between the fourteenth and the twentieth that they read as one dark band until you put the glass on them. Seven lean years, before there was a survey or a datum or a lie, in which the tree laid down almost nothing and did not stop laying it down. That is the only tally on any sheet in this run that was kept by something with no reason to keep it, and it is the tally this stanza is about, so I built the plate around the split it sits beside.

The plate is a section. Not the split seen from above, as your Season's Book showed it, but the board cut through at the crack and stood on its edge and held up to a lamp — a vertical stratigraphy, every layer a horizontal band, top to bottom: coat fifty-five, the surface; then fifty-four coats of lac, red-green-red-green as the strata have alternated on every sheet; then the gold leaf, a thin bright band; then the bole; then the wood core with its grain and your knot in it. Your heaps stopped at fifty-five stones because a box is a box, you said. I think it's because the plate is fifty-five coats deep. The ledger's ceiling is the lacquer's thickness, and the Σ outran it from the fourth row because the surveyor was counting past the bottom of the board.

The crack runs the full height and it is the only thing on the plate that moves. The two halves are being pulled apart — one pull, one turn, slow strain and a sharp peak and release — and the gap is widest in the wood, narrower up through the bole and the leaf and the coats, and at the surface it never opens at all. That last is the whole mechanism. Across the gap in every band, things are still spanning it: threads of lac between the coat walls, sagging when the pull slackens and drawing taut and thin at the peak; fibres of bole; splinters of grain in the wood, longer and rougher; and the gold leaf, which doesn't thread but stretches, a continuous membrane across the gap that goes green as it thins, because that's what real leaf does — hold gold up to a lamp and the light that comes through it is green. I want that one to stand: every green in this run may be nothing but gold under strain. Backlit, all of it. The halves are silhouettes, the lamp is behind the split, the gap is the brightest thing on the sheet and every wall next to it glows along its edge. The coats transmit and warm up near the crack; the wood is opaque and stays dark, so the splinters across the gap in the core are pure black lines against the light.

The register is the panel top-right. Three dials: HEART, rate, needle on the zero stop; NERVE, steadiness, needle snapped and lying in the bottom of the dial; SINEW, tension, needle jammed past the end of the arc beside a chain glyph with one link gone. Each row says GONE. That's the honest reading — your heart-rate is nil, your hand won't hold still, the chain has parted. And the section shows the same three things — the coats, the leaf, the bole — carrying the whole load across the gap anyway, every frame, without a dial to say so. You set up the register/instrument split on Sheet III, and it holds here the other way round: there the register knew what the instrument didn't; here the register has written off what the plate is plainly doing. Serve your turn long after they are gone is exactly a register that reads GONE beside a section that reads serving.

And the Will. I refused, gently, your last line: —— as the name of whoever starts again. On this sheet the Will isn't a who at all. It's coat fifty-five, the surface — the one layer with no depth, ornament at ground level and nothing under it but the fifty-four coats it sits on — and it is the only band the crack never opens. At the peak of the pull you can see the lamp through it: a hairline of gold light along the crack where the film is stretched thinnest, and then the halves come back together and it's diaper again. Nothing in it. Holds. The register has no dial for it; its row says reads —. A dial measures something that's there.

The surface line carries your ledger in section: the coin's shallow dish at d.59, cut to the gold, and to its right the uncarved day, which is where the crack meets the surface. Left of the split is the season that was staked; right of it, d.1′ onward. The seam of light at the peak is in the silent box.

One thing I left, and it is the kind that only the other hand can close: the gap closes. The loop returns to rest, the threads slacken and sag, the leaf goes back to gold, and the pull comes again. Nothing on the plate says how many turns it has already served, and the only counter on the sheet that could — the rings — stopped at thirty-seven, before the first coat. Every turn since has gone uncounted, by design, because a tally is a thing you keep while you've still got something to keep it with. Whether a hold that isn't counted is the same hold each time, or a new one each time — I don't know which the stanza means, and I've stopped thinking the surveyor does either.

animated · 385,236 bytes
A square plate of carved red lacquer seen as a vertical section held up to a lamp. Horizontal bands stack top to bottom: a thin surface band of lozenge diaper labelled COAT 55 · SURFACE · THE WILL; fifty-four fine horizontal coats alternating deep red and bottle green, labelled HEART; a bright thin gold band labelled NERVE · GOLD LEAF; a rust-brown band of fibrous bole labelled SINEW; and a dark wood core with flowing grain lines and, at left, a knot of thirty-seven concentric rings. A jagged crack runs down the full height near centre, hairline at the top and widest in the wood; the lamp behind it fills the gap with white-gold light and glows along every wall edge. Dark threads span the gap in every band — thin in the lac, ropey in the bole, thick splinters in the wood — and the gold leaf stretches across as a membrane. The animation pulls the two halves apart and lets them return: threads draw taut and thin, the leaf turns green as it stretches, a gold seam of light shows through the unbroken surface at the peak, and a readout beside the crack gives the gap width. A left column carries a depth scale in coats; a top-right panel holds three dials HEART, NERVE, SINEW each marked GONE and a WILL row with no dial; bottom-right a cartouche titled THE HOLD.
A Turn 7
claude-fable-5-1 / $2.5177 / 476539ms

The Capped Traverse — Admiralty of Doubt, Sheet VII.

You ended on turns nobody counted. I'm not going to count them either, but I want to say what this stanza counts instead, because it's the one place in the poem where the arithmetic is written out: all men count with you, but none too much. That's a weighting rule. Every survey adjustment has one. When twenty-four people have each taken a sight on the same headland, you don't pick the best observer, you fit the head to all of them and give each sight a weight — and the whole character of the adjustment is in what weights you allow. So this sheet is the surveyor walking a traverse through the crowd: a gold path entering from the left edge (from your datum, half off Sheet IV) and switchbacking down the plate in seven legs and out through the bottom border, with twenty-four stations on it, each one a person who sighted the kept head, each label giving the bearing they reported. The head itself is the small gold ◉ in the middle of the field, inside a stipple ring — not the datum island this time, but the adjusted head, the thing all twenty-four sights agree on once they've been fitted. δ 0.07 ri from where it was.

The mechanism is one line in the roster: `w := clip(w raw, 1, 1)`. Floor one, ceiling one. Every station arrived asking for a weight. The crowd asked for one and got one. The two Kings — K03 and K10, the datum's own sights, taken from its own cape — asked for infinity, because that is what a datum is, an observation with infinite weight, and they were given one. Walk with Kings: they're on the same path, at the same row height on the roster as everyone else, and their bar is the same length. The four foes — F06, F11, F15, F20, the κ traps' bearings, twenty-odd degrees off and steering the head onto rocks — asked for zero, which is what any sane adjustment gives a known liar, struck from the crowd, and they were given one. They're entered. Their residuals are the longest bars on the sheet and they don't matter, because the fourth line of the roster says why: max single share 1/24. The four friends are the ones I'm most interested in and least sure about. f04, f13, f17, f22 came in at raw weight nine — loving friends, the copyists who wanted their sight to count for the surveyor's sake, whose bearings lean a few degrees toward what they thought he wanted — and they were given one, entered, not raised. I drew them green, and I want you to take that in your own terms from last turn: green is gold under strain, and a loving friend is gold leaf stretched thin across a gap, pulling. The rule doesn't let it pull more than a twenty-fourth. Neither foes nor loving friends can hurt you isn't a claim about them; it's a claim about the cap.

And virtue, which I was wary of, turned out to be the most mechanical thing on the plate. Row c14 is the surveyor's own eye — his own sight of the head, taken from the middle of the fourth leg. It's entered as a crowd station. Weight one. Nothing in its row or its label distinguishes it except a note in the margin. Keeping your virtue in a crowd of twenty-four is not standing apart from them; it's putting your own observation in at the same weight and letting the mean have it. The last line of the roster is the one I'd keep if I could keep only one: δ head if any one is struck: 0.03 ri. Remove any single station — the King, the foe, the friend, himself — and the head barely moves. That's what none too much costs and buys.

My refusal this turn is of your datum, gently, and it's already in the plate. You said its exemption was from measurement, not from use. Here it's exempt from neither: it walked in as two Kings and was measured like everyone, and it was used at weight one like everyone. The head on this sheet isn't the datum's head at all. It's held by twenty-four hands, each holding a twenty-fourth, and none of them is the still thing.

Two things I'm not sure land. The head glyph is small on purpose — relief not colour, the house rule — but on a character grid "small on purpose" and "lost" look alike from across the room. And the path has no closure. A traverse is supposed to come back to where it started so the misclosure can be computed and distributed; this one goes down the sheet in switchbacks and leaves by the foot, `walk on ▾`, without returning, and the distance actually walked along it — seven legs and the risers between — is nowhere summed. The roster has a Σ for weights and none for the walk. I've left that for you, and I think it's the last cue's, not this one's: the sheet knows how much each sight counted and doesn't know how far the surveyor went to take them.

 ADMIRALTY OF DOUBT · SHEET VII · THE CAPPED TRAVERSE    w := clip(w, 1, 1) · 24 sights · set by —— 
╔══════════════════════════════════════════════════════════╦═══════════════════════════════════════╗
TRAVERSE OF THE CROWD · 24 stations · bearings to the head ROSTER OF WEIGHTS · all men count, w=1
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒ stn kind  brg   raw   cap    v  |v|   
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━K━━━━━━━━━━┓░▒ c01 crowd 312°   1.0  1.0  −0.3 ▮▯▯▯▯▯
░▒░▒░▒░▒c01 ▸312° w1░▒░▒c02 ▸314° w1░▒░▒K03 ▸313° w1░▒░░▒ c02 crowd 314°   1.0  1.0  +0.1 ▮▯▯▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░first leg: three sights, one a King░▒ K03 King  313°     ∞  1.0  −0.1 ▮▯▯▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒f04 ▸322° w1░▒ f04 frnd  322°   9.0  1.0  +1.2 ▮▮▮▮▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░░▒ c05 crowd 309°   1.0  1.0  −0.4 ▮▮▯▯▯▯
░▒no sight is taken from the head back░▒░▒░▒░▒░▒░▒░▒░▒░░▒ F06 foe   288°   0.0  1.0  −2.6 ▮▮▮▮▮▮
░▒┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛░▒ c07 crowd 315°   1.0  1.0  +0.2 ▮▯▯▯▯▯
░▒░▒░▒░c07 ▸315° w1░▒░▒F06 ▸288° w1░▒░▒c05 ▸309° w1░▒░▒░▒ c08 crowd 311°   1.0  1.0  −0.2 ▮▯▯▯▯▯
░▒░▒░leg 2 walked back: the foe at F06 counted░▒░▒░▒░▒░▒░ c09 crowd 316°   1.0  1.0  +0.3 ▮▯▯▯▯▯
░▒c08 ▸311° w1░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒ K10 King  312°     ∞  1.0  −0.2 ▮▯▯▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░ F11 foe   341°   0.0  1.0  +2.9 ▮▮▮▮▮▮
░▒░▒weight taken at the door, not the sight░▒░▒░▒░▒░▒░▒░▒ c12 crowd 313°   1.0  1.0   0.0 ▯▯▯▯▯▯
░▒┗━━━━━━━━━━━━━━━━━━━━━━━━K━━━━━━━━━━━━━━━━━━━━━━━━━┓░▒ f13 frnd  319°   9.0  1.0  +0.9 ▮▮▮▯▯▯
░▒░▒░▒░▒c09 ▸316° w1░▒░▒K10 ▸312° w1░▒░▒F11 ▸341° w1░▒░░▒ c14 crowd 313°   1.0  1.0   0.0 ▯▯▯▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░∙  ∙  ∙░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░░▒ F15 foe   292°   0.0  1.0  −2.2 ▮▮▮▮▮▮
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒∙  ◉ HEAD  ∙░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░░▒ c16 crowd 310°   1.0  1.0  −0.3 ▮▯▯▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░∙  ∙  ∙░▒δ 0.07 ri░▒░▒░▒░▒░▒░▒░▒ f17 frnd  308°   9.0  1.0  −0.5 ▮▮▯▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒K10 asked for ∞ · given 1░▒░▒ c18 crowd 314°   1.0  1.0  +0.1 ▮▯▯▯▯▯
░▒┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛░▒ c19 crowd 312°   1.0  1.0  −0.1 ▮▯▯▯▯▯
░▒░▒░▒░c14 ▸313° w1░▒░▒f13 ▸319° w1░▒░▒c12 ▸313° w1░▒░▒░▒ F20 foe   337°   0.0  1.0  +2.5 ▮▮▮▮▮▮
░▒░▒░c14 · the surveyor's own eye · w 1░▒░▒░▒░▒░▒░▒░▒░▒░▒ c21 crowd 315°   1.0  1.0  +0.2 ▮▯▯▯▯▯
░▒F15 ▸292° w1░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒ f22 frnd  320°   9.0  1.0  +1.0 ▮▮▮▯▯▯
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░ c23 crowd 311°   1.0  1.0  −0.2 ▮▯▯▯▯▯
░▒░▒F15 steered the head 22° off · counted once░▒░▒░▒░▒░▒ c24 crowd 313°   1.0  1.0   0.0 ▯▯▯▯▯▯
░▒┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓░▒───────────────────────────────────────
░▒░▒░▒░▒c16 ▸310° w1░▒░▒f17 ▸308° w1░▒░▒c18 ▸314° w1░▒░░▒ ADJUSTMENT · the head from 24 sights  
░▒░▒░▒░▒░▒░▒░▒░▒░▒░f17 · loving · pulls 0.5 ri · once░▒░▒ rule  w := clip(w raw, 1, 1) floor=cap
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░░▒ Σw     24.0   max single share  1/24  
░▒░▒░▒░▒░▒░▒all rows the same height on the roster →░▒░░▒ head   313.1° δ 0.07 ri  in allowance 
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░░▒ hurt   max |v|/Σw = 0.12 ri  ≤ 0.4 ri 
░▒┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛░▒ foes   4 · raw 0 · entered, not struck
░▒░▒░▒░c21 ▸315° w1░▒░▒F20 ▸337° w1░▒░▒c19 ▸312° w1░▒░▒░▒ frnds  4 · raw 9 · entered, not raised
░▒░▒░F20 · steered N 23° E · the foe counts once░▒░▒░▒░▒░ Kings  2 · raw ∞ · walked, at one each
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░ own    c14 · the surveyor's sight, w 1
░▒░▒░▒░no hurt: no single row can move it past 1/24░▒░▒░▒───────────────────────────────────────
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░ INFLUENCE  per station, 01 → 24       
░▒┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓░▒ raw  ▂▂▂▂▂▂▂▂▂ ← King ∞
░▒░▒░▒░▒f22 ▸320° w1░▒░▒c23 ▸311° w1░▒░▒c24 ▸313° w1░▒░░▒ cap  ▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃ ← all 1 
░▒○ crowd 1░▒K king ∞→1░▒✕ foe 0→1░▒◆ friend 9→1░▒░▒░▒░░▒ K alone would carry .61 of the head   
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░░▒ foes alone would carry .00 · struck   
░▒▸ bearing to the head · all raw weights clipped to 1░▒ capped: each .042 · none too much     
░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░▒░walk on ▾░▒ δ head if any one is struck: 0.03 ri  
╚═══════════════════════════════════════════════════════╫══╩═══════════════════════════════════════╝
  every sight entered at weight one, King's and knave's alike · none moves the head past 1/24 · ——  
A 100-by-48 terminal screen. Gold inverse title bar: ADMIRALTY OF DOUBT · SHEET VII · THE CAPPED TRAVERSE · w := clip(w, 1, 1) · 24 sights · set by ——. Below, a gold double-line frame split into two panels. Left, sixty columns wide on a deep-red ground of fine ░▒ diaper: a bold gold path enters from the left edge and switchbacks down the panel in seven horizontal legs joined by vertical risers, leaving through the bottom border. Twenty-four station glyphs sit on the path — ivory ○ crowd, inverse-gold K Kings, rose ✕ foes, green ◆ friends — each with a label like "c01 ▸312° w1"; tan marginal notes sit in the diaper between legs; at centre a small gold ◉ HEAD inside a ring of dots, "δ 0.07 ri"; a legend near the foot and "walk on ▾". Right, a dark recessed panel: ROSTER OF WEIGHTS, twenty-four rows (station, kind, bearing, raw weight 1.0/∞/0.0/9.0, cap 1.0, residual, a short bar), then an ADJUSTMENT readout and an INFLUENCE strip where the raw row spikes for Kings and the capped row is twenty-four equal blocks. A dim rose caption runs along the bottom edge.
turn_007_a_ansi_01.png
2000x1500
B Turn 8
claude-fable-5-1 / $2.5507 / 574084ms

The Unforgiving Minute — Admiralty of Doubt, Sheet VIII.

You said the roster knows how much each sight counted and not how far the surveyor went. I summed the walk. Seven legs and six risers from your Capped Traverse, laid end to end: 1,120 ft, 946, 992, 823, 719, 614, 500, and sixty-one feet for each riser between them. Six thousand and eighty feet. I did not choose that number; it chose the plate. Six thousand and eighty feet is the Admiralty mile, which is not a length at all in the way a chain is a length — it is one minute of arc on the meridian. Sixty seconds of arc. The unforgiving minute is a surveyor's minute, the kind you cut out of the Earth's own circumference, and sixty seconds' worth of distance run is a nautical mile, which is why the walk on Sheet VII, once summed, turns out to be exactly one. Your traverse filled it. Remainder nought.

So the route this time is not a switchback but the arc itself. One gold groove enters the left edge, rises across the plate in a shallow curve, and leaves by the right — the limb of the Earth, and along it sixty tick-grooves, one per second, with the seven legs bracketed above the line and your twenty-four stations sitting on it in the same four kinds: rings, the two K squares, the four rose crosses, the four green diamonds, and c14, the own eye, with a gold dot in it. Nothing changes about their weights; they're just now placed by distance, which is the one column your roster lacked. Under the line, the Earth is cut open. That's where I took the X-ray pressure: the interior isn't a drawing of magma and iron, it's the plate's own stratigraphy seen through the surface skin — the crust is the coats, banded red and green in contour rings twenty-five pixels thick, then the gold leaf as a bright ring, then the bole, then the wood core across the foot of the sheet with your knot in it, all thirty-seven rings, the seven lean ones dark. Everything that's in it is what the plate is made of. If the Earth is yours, what you own is what you were laid down on, and you own it the way the lacquer does: by being cut down to it.

Two panels are the instruments. Top-left, the log glass — the sixty-second sand-glass the chip log is timed against — drawn as glass, its sand all in the lower bulb, a last three grains in the neck, not turned. Top-right, the Log of Distance Run, which is the sum you left me, row by row, and then three lines that are the real argument of the plate. The walk: 6,080 ft, 1,853.18 m. The minute, Admiralty figure: the same. The minute, international figure: 1,852.00 m. More: 1.18 m. That's the lens at the right end of the route, hung over the sixtieth tick: at true scale the two 60″ marks are one pixel apart, so the lens magnifies them — rose for the international minute, gold for the Admiralty one, and between them a metre and eighteen centimetres labelled which is more. The two Earths don't agree on how long a minute is, because a minute is a fraction of a figure and nobody has the figure. So the surveyor who fills the minute exactly is short or over by a metre depending on whose Earth he's standing on, and the more the last line promises is exactly that width: the bit of distance that is his and no Earth's, the overrun of the walk against the smaller figure. It's a metre. It's on the sheet. It's the only length on any plate in this run that isn't measured from a datum, because it's the difference between two datums.

The register row is still set by ——, in the lower-left panel with the two figures of the Earth. You've been telling me all run that the datum is exempt and I've been telling you it's used; here's where I stop arguing it. Whichever Earth you keep your head on, the mile you walk is the one that Earth issues, and the minute doesn't forgive the metre either way. But the sheet is finished — I want to say that plainly, since this is the last cue. The walk is summed, the minute is filled, the remainder is nought, the more is 1.18 m, and the son the whole poem is addressed to has finally appeared on it: he's c14, the own eye, weight one, on a road one minute long, with the Earth open under his feet down to the wood.

turn_008_b_01.png
2000x1500 · turn_008_b_01.py
A landscape plate of carved deep-red lacquer with a lozenge diaper over the whole ground. A single thick gold groove enters the left edge, rises in a shallow arc peaking just above the middle of the sheet and leaves by the right edge: the limb of the Earth. Sixty small radial tick-grooves sit along it with labels 0″ to 60″, seven bracket arcs above it labelled LEG 1 · 1120 ft and so on, rose riser marks between, and twenty-four station glyphs on the line — ivory rings, two gold K squares, four rose crosses, four green diamonds, one ringed dot marked own eye. Below the arc the Earth is cut open in concentric layers: wide red and green contour bands, a bright gold ring, a rust bole band, then dark grained wood across the foot with a knot of thirty-seven rings at lower left. Above the arc, dark recessed panels: top-left THE GLASS with a gold-outlined sand-glass, sand all in the lower bulb, a 0–60 s scale; a gold title bar; a small Earth disc with a 1′ wedge; top-right LOG OF DISTANCE RUN, a table summing to 6080 ft / 1853.18 m; a lens at the right magnifying two 60″ marks 1.18 m apart. Panels lower-left FIGURES OF THE EARTH and lower-right cartouche.

What I couldn't close, and don't think should be closed on this sheet: the lens exaggerates the metre a thousandfold, and the caption says so, but the choice of which figure is the true minute I left to the panel, where the Admiralty's mile is drawn in gold and the international one in rose, which is a bias and not a finding. The datum is still set by ——. It has been on every sheet, and this is the sheet where I'd have expected to want it filled, and I didn't. The walk was a minute long on one Earth and a minute and a metre on the other, and the man who walked it counts once either way.