Expand description
Development kinetics.
The one dimension the reference model does not have: how long the film sat in the developer, and at what temperature. Development is a diffusion-limited reaction, so the useful approximations are all saturating functions of an effective time that itself follows an Arrhenius-like temperature law.
use kallo_core::DevelopmentParams;
let normal = DevelopmentParams::default();
let pushed = DevelopmentParams { time_min: 12.0, ..normal };
assert!(pushed.gamma_factor() > normal.gamma_factor());The constants below were chosen so the model behaves like a real process over
the usual 3–15 minute range. They are now fitted to a measured curve
family: ILFORD DELTA 400’s contrast-against-time graphs (publication
d400_a.pdf, page 5), two temperatures and four developers. Fitted jointly to
the ID-11 pair — 15 points, 20 °C and 24 °C — this form reproduces the
measured contrast to 0.005, and the fitted numbers are not the ones the
priors guessed:
| prior | fitted | |
|---|---|---|
gamma_inf | 1.00 | 2.20 |
gamma_0 | 0.15 | 0.18 |
tau_min | 4.5 | 39.0 |
q10 | 2.00 | 1.70 |
tau is the one that mattered. At 4.5 minutes the prior model reaches 94 % of
its final contrast by twelve minutes; the film is at 47 % at six and 71 % at
twelve. A push built on the prior was a saturated curve pushed further, not a
curve that is still building.
§The ladder, and the three quantities it measures
DELTA 400’s chart plots contrast against time and nothing else. TRI-X 400’s
printed page 12 plots the whole curve four times — D-76, large tank, 20 °C,
1-minute agitation, at 6, 8, 10 and 12 minutes — which is the first ladder in
the archive that can test everything the model does with time. Digitised and
fitted (see docs/measurements/2026-11-08-development-time.md), it gives:
| quantity | measured | this model |
|---|---|---|
| contrast, 6 to 12 min | 0.546 → 0.824 | +2.5 % at 8 and 10 min, +1.3 % at 12 |
| base plus fog | +0.078 per √min | fog_rate |
| ceiling (the highest density drawn, above base) | +0.91 per √min | d_max_gain |
| speed | −0.44 per ln(t / t_ref) | speed_gain |
| toe rate | +0.40 per minute | toe_gain |
The last three were priors until this sheet. The ceiling is the one that mattered most: the current model held it fixed while the film’s rises by nearly a whole density between six and twelve minutes, and a pushed render was therefore 0.2 to 0.6 of a density flat across its top. With all three fitted, the four curves of the ladder are reproduced to 0.03 of a density rms; before, the pushed rungs were off by 0.21, 0.41 and 0.62.
The toe sharpens as development goes on: the fitted corner rate rises from 3 to 8 between six and twelve minutes, and it is the last thing the ladder measures. Without a term for it the model’s 0.10-over-fog crossing — the ISO speed point, which lives in the toe — sat 0.10 to 0.27 of a decade left of the sheet’s even when the densities agreed to 0.03.
The individual rates are noisy: on a curve that stops climbing at the frame’s
edge the toe rate and the ceiling trade against each other, so the two-rate fit
can report 6 where another corner assignment reports 8. toe_gain is therefore
not fitted to those rates but to the densities: scanning the slope to
minimise the ladder’s density residual gives 0.40 per minute, and at that slope
the four rungs’ ISO speed points land within 0.035 of a decade — 0.020,
0.018, 0.008 and 0.034 — against 0.27 before.
The term is linear in the development time, which is all four rungs can say. It is unbounded, and nothing in the archive says what a 30-minute development does to a toe; the model is used between 3 and 20 minutes, and beyond that the term is an extrapolation like every other constant here.
Structs§
- Development
Params - How a development time and temperature move the characteristic curve.