Lith printing

[marcinz606]

Lith printing as its own pipeline stage, B&W only, off by default.

The model

Two phases, following Rudman's "New Rules" and Moersch's Lessons 1-6:

• a long low-gamma highlight branch fixed by exposure alone
• a near-vertical infectious knee whose position is the snatch point
• past it, the flat Dmax "lith band" with no tonal separation at all

D_h = f_max · (1 − exp(−f_rate · (D0 + 0.301·E·v) / f_max))
D   = D_h + (D_max − D_h) · σ((D0 + 0.301·E − K) / w)

Controls are Exposure (0 to 5 stops, default 2), Snatch Point (development-time proxy, sets K) and Abruptness (sets w). "abrupt" is Moersch's own word for the blackening, and his additive table is what maps it to the hydroquinone-to-alkali ratio.

Colour comes from the paper

PaperProfile gains a lith_path: four (a*, b*) anchors at output-density fractions 0.10 / 0.35 / 0.65 / 1.00, running peach, ochre, olive, neutral. The olive knot is the point, since on warmtone paper the transition between warm highlights and cold blacks really does go green.

This finally gives the four B&W profiles a reason to differ from each other. Fomatone liths warm and colourful; Neutral and the Ilford papers stay nearly colourless, which matches Rudman's lithability test reading an incorporated accelerator in Multigrade.

Toner interaction

Lith silver is fine, high-surface-area silver sitting on the steep part of the tone-vs-particle-size curve (Kong & Shore 2007), so the same bath moves it much further than normal filamentary silver. Selenium keeps its silver-proportional shape but reaches into the midtones and turns the green-black shadows magenta. Gold drops the bleach-limited weighting for a flat conversion, matching Moersch: "gold toner attacks all densities evenly". Sepia, iron blue, copper and vanadium are inert or redundant on lith and grey out.

Two latent bugs fixed

Both flattened the new colour away, and both were pre-existing assumptions that only a chromatic B&W print could expose:

• The silver ledger rebuilt all three channels from the pixel's mean density. Harmless when the input is grey, but it meant a trace of any toner wiped the lith colour and replaced it with the toner's own. Now each channel uses its own density; the mean still drives susceptibility. Bit-identical for a grey print, so no existing toning moved.
• toning.wgsl re-collapsed B&W to luma as a safety net, after the lith pass had put real colour on the print.

Verification

make all green: 3750 passed, 4 skipped.

CPU/GPU parity is bit-exact (max delta 0.0000) across disabled, enabled, snatch/abruptness extremes, lith+selenium and lith+gold.

Driven headlessly through the real app on a B&W scan: later snatch deepens the blacks and widens the band (0.5% → 8.2% → 20.5% of frame), the paper drives the colour (Fomatone 0.203 vs Ilford 0.072 mean R−B), and selenium bites harder under lith than without.

Docs updated in the same change: USER_GUIDE.md (new panel section with the ⓘ marker, plus the Paper profile bullet), PIPELINE.md (new stage section, the per-channel covering-power correction, the lith toner constants) and CLAUDE.md stage order.

Not modelled

Semiquinone/bromide diffusion halos, pepper fog, snowballs and the other fault modes.