Skip to content

Antique atlas look: the universe engraved as a 1690 Dutch atlas - #1132

Draft
rulkens wants to merge 304 commits into
mainfrom
feat/antique-atlas-look
Draft

rulkens wants to merge 304 commits into
mainfrom
feat/antique-atlas-look

Conversation

@rulkens

@rulkens rulkens commented Nov 13, 2026

Copy link
Copy Markdown
Owner

Draft. Not ready for review: the review panel loop is still running, and this branch will be split into reviewable PRs before landing.

What

An antique atlas look for Skymap, toggled with ?antique. It renders the live frame as a Dutch engraved plate of about 1690 (Cellarius / Valk & Schenk for the sky, Blaeu for the Earth). It is still a chart of the real data: star glyphs come from the catalogue by magnitude, contours and washes from the rendered frame, and land and sea from each body's ocean mask.

  • Plate pipeline: a stylize frame step with paper, plate shaders, ornaments, cartouches and the weathered chrome theme.
  • Sky: engraved star glyphs on their own path, constellation lines and figures, the Milky Way seen from outside and named, the cosmic web charted (spines and isobaths), and engraved galaxies.
  • Bodies:
    • Earth c.1690, with coasts, borders, towns and mountains (molehills and range-gated hachures);
    • Mars terrain;
    • the Moon after Riccioli (names and craters);
    • Jupiter's Great Red Spot.
  • Lettering: IM Fell (OFL) as MSDF atlases, period name ranking, and a paper knock-out behind names.
  • Tuning: a debug section for tuning the look live.

Design: docs/superpowers/specs/2026-11-07-antique-atlas-look-design.md. Plan, tasks 1–38: docs/superpowers/plans/2026-11-07-antique-atlas-look.md.

The bug fixes found on this branch already landed on their own, as #1104–#1113. The last one, item 8 (DPR labels and line widths), is tracked in docs/backlog/2026-11-10-antique-branch-fix-prs.md.

Before landing

  • Review panel graded high enough
  • Perf check, including the figure lattice cost and about 360 MB of GPU memory
  • Split into reviewable PRs
  • Remove the session trailers from older commits, the __pycache__ folder and the Python sprite tools
  • Ship a digits font: 'Antique Figures' currently falls back to whatever Baskerville or Times the OS has
  • /feature-done

🤖 Generated with Claude Code

rulkens and others added 30 commits November 7, 2026 13:21
…e star italic

Owner ruling 2026-11-07: the starCatalog rule's hide entries go; period names
stay where they exist, catalog numbers still hide unless focused. Spec and
plan Task 11 updated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ipped atlas

Cetus, the dragon and the sea serpent pack at source resolution with a gentler
ink push; the atlas is mipped and decor.wesl turns a figure's sampled coverage
into the plate's density, replacing the re-cut outline and hatching.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…cut's records

The visual and pick stage decoded the record words inline; the antique print stage
is about to read the same records. leafRecord.wesl now owns the cut and record
bindings, the node lookup, the box rebuild and the word layout, so every stage
places and ranks a star alike.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the look on, passes listed in ANTIQUE_MOVED_PASSES leave their line for a
row of their own; the constellations move off the hdr roster onto antique-lines,
so the quarter-res reduce no longer averages them away. The plate inks that row
like the rulings, masked by the sky and knocked out under star glyphs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e image

Image peaks could not rank bright stars (a halo raises the local mean, so Polaris and
Fomalhaut vanished) and their count followed device pixels (4-5x the stars at 2x). The
star Layer now prints each naked-eye star (m <= 5.75) of the cut's leaves and the seeded
roster as one point into antique-print-points, one texel per 4 print units, its depth
keeping the brightest per texel. The glyph pass reads that row: class 5 - round(m), and a
star on a body never prints. peaks.wesl, the peak half of reduce and the peaks/stars rows
are gone. A render target's scale function now also sees the canvas, so a row can be
sized in print units.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Thirteen c.1690 names (Blaeu/Visscher spellings: Hollandia Nova, Terra
Australis Incognita, Oceanus Aethiopicus...) from a lat/lon table, drawn by
a second body-Layer screen-label row under label:body. Lands in spaced
capitals, seas in IM Fell italic, at 16 px under the planets' capitals.
A name shows when the globe is 250 px or more, its box fits inside the
rim and its anchor is in daylight or twilight; with the look off the
producer returns before any work. A name carries Earth as subject for
focus emphasis, and the director never sets it as Earth's title.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nes, Saturn's ring as paper ellipses, water lines off Earth's coasts

The plate ray-casts up to four of the frame's widest globes (PlateBodies uniform,
packed by packAntiquePlateBodies from ctx's body states, radii and the Sun), so
its hatching reads N.L and the arc from the subsolar point instead of the
quarter-res light. Lines come on a ladder of four burin widths in rank order and
bow half-way toward the globe; a 27 degree second set enters only in the deepest
shade; the night side keeps reflected light at the limb. Saturn's ring is paper
with four ellipses and the Cassini line, the ball's shadow on it hatched along
the ring and its shadow on the ball one solid stroke. Water lines come from a
probit of the land share blurred in two quarter-res passes. The land and sea
washes apply only to a record flagged as Earth with its ocean mask, which ends
Saturn's yellow limb fringe. Recipe: the five-round offline model.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… the top rule

The generator leaves the frame's top edge open and two generated rule strips
flank the cartouche, so the gap is always the cartouche's width. Plan included.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e one-Mpc seam test

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…its real horizon

At 4 km over Jezero the inner-bound sphere was still cast: a comb of curtain lines
converged on a false horizon under the skyline, its outline left a stray line, and
the sphere behind the rover overrode the rover's tone. A globe is now recorded only
from beyond its highest peak's radius times 1.02; nearer, the plain path draws it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e they print nothing

The first family below tone 0.06, the second below 0.8, the ball's phase gradient
(two extra casts) where it is not hatched, and the four water lines off the open
sea or under shade. Earth filling the plate at 500 km: stylize 6.8-9.4 ms before,
6.4 after (interleaved A/B, 90 frames).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lines row, still hidden behind bodies

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tes, a field-adaptive dotted outline, a solid core and dust lanes left as paper

A quarter-res nebula pass blurs the wash into the fields the plate engraves from,
and a one-workgroup compute takes the glow's 30th-percentile luma each frame so
the galaxy gate and outline level rise with a dense field. Reduce's floor factor
drops to 1.15 so foreground stars stop growing bubbles in a bright disc. The old
glow contours and outline colour stay for light the gate rejects.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…, scale of miles, engraved thumbnail

All under data-look=antique through var(--hook, <plain value>) hooks; with the
look off the computed styles of the HUD are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ghost buttons stay text

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…drawn shape

The control's own background showed past the incurved corners and the scroll
ends; framed controls now take --framed-control-ground (transparent in the
antique look). Primary buttons keep their accent fill instead of a tablet.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… the plate's iso-line gate

A structure tensor over each 4×4 wash block: a galaxy's isophotes run one way
locally, a star field's speckle every way (measured on the app's frames: the
band seen from Sgr A* stays under 0.4, M51, M87 and the MW over 0.45).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ay sky is its own field and stays paper

Ground blocks hold no sky and used to count as dark field: over a horizon the
30th percentile fell to zero and the bright aerial sky passed the galaxy gate,
printed as iso-line hatching (Mars surface views). The glow marks a block with
no sky by alpha -1; the histogram spans the whole 0..1 range.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…isc gets a land map

Inside an atmosphere the day sky fills foreground:0's alpha, so the look took the
whole view for body and lost the rulings, stars and horizon. There the look now binds
the row's depth instead (a flag in the uniform record, read in cover.wesl by reduce,
glyphs and plate), and a second difference of that depth inks occluding edges: the
rover, crater rims, skylines. A sea-mask pass draws Mars's dark albedo into the land
row (threshold in ANTIQUE_SEA_ALBEDO, mean − 0.7σ of its map); reduce drops the
colour guess, and the plate reads land with a body-masked bilinear so no false coast
rings the limb.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…markings as engraved tone

The plate's body block becomes a dispatcher: each cast record carries a print
convention code (ANTIQUE_BODY_PRINTS, mirrored as PRINT_* in plateBodies.wesl), and
Earth, Mars and the Moon each own a module under antiqueLook/bodies/. Earth's coasts,
water lines and washes moved there unchanged. Mars has no water edges, so its dark
albedo is no longer sea: the mask (now body-albedo-mask, a lightness ramp) feeds a
burin family of its own and a thin red-lead wash, with no coast and no smalt. The Moon
is a stub that engraves as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rail boxes

Under data-look="antique" the Splash's dark vignette becomes a paper wash and its white
type ink; the phone ScaleBracket's dark halo becomes a 1px paper knock-out; the ScaleBar
and TimeBar pill are drawn as kit tablets (the engraved frame's 43.8px corner slice does
not fit a 40px pill). Every hook falls back to today's value, so the plain look is
unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ines only where the isophotes are coherent, the wash held to its block's dimmest texel

- The structure-marker passes move to a half-res antique-structures row
  (ANTIQUE_MOVED_PASSES); structureGlow.wesl box-means it to 1/16 with the
  rings eroded away, and the plate draws the old outline colour and dotted
  contour from its alpha. The print carries galaxies only, so a cluster's,
  group's or void's disc is never engraved as one.
- The plate reads the glow's coherence (glow.wesl) and falls back from
  iso-lines to tone stipple where a star field's speckle would wander them
  into fingerprints.
- reduce's floor factor drops to 1.0 and the line field smooths (Lxh 2 cells,
  Lsm 2.2, unsharp 1.5): on the app's real M87 frame the recipe's values
  (1.15, 1.3, 1.6, 2.5) left field points wobbling the isophotes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RRpRLq7tupBVTs2Mc6EeFC
… galaxy engraving

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RRpRLq7tupBVTs2Mc6EeFC
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RRpRLq7tupBVTs2Mc6EeFC
…s the Mars and Moon views

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RRpRLq7tupBVTs2Mc6EeFC
Nine period names on the generated disc: BRACHIVM PERSEI, SAGITTARII,
SCVTI-CENTAVRI, NORMAE and EXTERIVS on the generator's own arm ridges
(with the lopsided bend the sprites carry), RAMVS ORIONIS at Reid et al.
2019's Local arm, Sol hic under the Sun, TRABS and NVCLEVS by the bar and
bulge. A new spacedItalic cut sets them as area lettering. A second
milkyWay label row fades them in with the disc's size on screen, the eye
outside the disc and the view off edge-on; You are here yields to Sol hic
while they show. A test holds the arm order against where each generated
arm crosses the Sun-centre line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RRpRLq7tupBVTs2Mc6EeFC
rulkens and others added 25 commits November 8, 2026 06:16
…e-full grew to 6 layers, and a cube view of its canvas-sized texture failed validation on every allocation); the default view names its dimension
…ty registry (a hand list missed the blackHoles and missions Layers, so the Galactic Centre and Voyager 2 stayed captioned)
… ReadyFrameContext requires (marker-lines draw saw an undefined dpr)
… ring-and-dot sign and its name (the old no-glint gate dropped them); the sign list and the label rule read one names table
…e took them off with the figures and rulings, so the sky round the Moon lost its lines)
…the sky and projected as an 8x8 lattice, so a corner stretches them as it stretches the stars (the screen-space similarity grew them off-axis); facesNorth and screenNorthAngle go
… the hidden corners pulled along the sphere to where they just project (it popped out whole before)
…uche (Cetus overlapped it by 15-24% at many window sizes and popped out); only a beast keeps that rule

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GuaxX7BKw6sAJQ3n5zk6hD
… (the sign draw sat inside its gate)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GuaxX7BKw6sAJQ3n5zk6hD
…e first cut had dropped the thin contour and half the crown hatching); the crown star and the plate's ruled line are filled from neighbouring engraving
… superclusters): the pass sat in the look's hidden list; it now prints on the line row, where a 1-px ring survives
…nt (dotted contours from the structures row), washed per kind: void smalt, cluster carmine, supercluster verdigris; the void takes a halo there
…lim double-ruled band (no checkered frame or corners), title centred in small caps with a disclosure arrow, carmine on hover
…-width frame, checkered top bar and title plate; only the body folds away and the frame closes under the bar (the band design is reverted)
…wner ruling): StepInputs.northUp, northUpArm and its tests are gone, so roll works in the OpenSpace scheme as in the plain look
…spike

Conflicts:
- produceMilkyWayLabel.ts (+ test): kept both sides (antique name alpha and antiqueLookOn imports plus main's MILKY_WAY_CENTER_WORLD pick gate; both tests kept).
- label2DDirector.ts (+ test): main's text-rect declutter (declutterPadPx) is the only policy; the branch's antique rules (rectOf with dpr, rim lift, body discs, fallback, farIds) re-applied on it; the screenSeparation arm and ANTIQUE_LABEL_DECLUTTER (pad 8, now the default) deleted; both test sets kept.
- starCatalog/vertex.wesl: kept the branch's leafRecord imports, starTint now from lib::starTint (#1092).

Ripple: the three printVertex.wesl renamed vertexPrint.wesl for the shader-folder ratchet (#1091); antiqueLook and bodies/earth/antiqueMap added to its ALLOWED debt list (their files hold both stages).
…to passes/, shared/ and one folder per vs+fs program (io, vertex, fragment); their shaderFolderShape ALLOWED rows are gone
…ery level, 1,000 m relief in the 80 km window; rebaked earth-1690.bin
…ates the relief pass over a 200 m ramp, same measure as the molehills
@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 13, 2026 •

Copy link
Copy Markdown

Deploying with  Cloudflare Workers  Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status Name Latest Commit Preview URL Updated (UTC)
✅ Deployment successful!
View logs
skymap 887a3ba Commit Preview URL

Branch Preview URL
Nov 17 2026, 03:23 AM

earthPass.ts: kept the branch's earthSurfaceUniforms call; main's only change there (cloudCoverageHere comment) moved into earthSurfaceUniforms.ts.
renderFrame/renderFrame.timing/runFrame tests: took main's registry-derived fixtures (#1137); the antique settings and target rows now come from INITIAL_SETTINGS, and the two hand-built spots (FrameInputs.antique, gpu.fontAtlases null for loadLookFonts) were added.
filamentsPass.test.ts: kept both the branch's dpr line-width test and main's fade-gate test.
renderFrameSplitBaseline.test.ts: deleted as on main (#1137 dropped it; its only unique check moved to filamentsPass.test.ts); the branch's change there was the new antique/dpr args in the fixture.
docsCommands: union of fetch-mpcorb/sbdb-diameters and the branch's fetch scripts.
ToneMap.d.ts: kept both printShoulder and film.
App.tsx: leftStack ref wraps NavigationPanelContainer (sky mode's container).
DisplaySection, container, test: both antique and film switches and props.
ScaleBarContainer: hidden while standing (sky), still reports its row to the view cover.
compositor.ts, io.wesl, fragment.wesl: film lanes 8-17, print shoulder moved to lanes 18-19 (80-byte buffer); main's `uniforms` name.
renderTargets.ts: both look-target imports.
labels/vertex, starCatalog/vertex, structureMarker/ring: branch logic in main's renamed identifiers.
coreSettingsSlices, selectors, tokens.css, rawDataRegistry, docsReference and frame tests: union of both sides.
Branch-added shader uses of `u.` in markerLines/texturedBody renamed to `uniforms.`.
Pure rename of one- and two-letter shader names, verified by identical shots.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant