Engine verification archive
Earlier engine verification records, with values, sources, tools, verification dates and limits.
Measured, digitized, fitted, estimated and counted values are recorded separately, with their verification dates and limits. Detailed raw curve data and the methods used to incorporate it into the engine are not published. Processing speed varies with device, image size and the processing used. Real performance is measured once the target device and input conditions are set.
Fitted
0.0036–0.0196 RMS
on-axis, two lenses at maximum aperture, against published MTF charts
Limits Field-wide residual is 0.0523–0.0866 — about 4.4× the on-axis figure for the f/0.95 lens and about 14.5× for the f/2 lens. The model misses worst around 8–13 mm field height at f/0.95.
Fitted
0.0523–0.0866 RMS
all field heights, same two lenses and charts
Limits Published charts are design representatives, not measurements of a production unit, so this excludes sample-to-sample decentring and tilt.
Counted
603 / 672 points
digitised data rows from the two official MTF charts — Bloom 50mm f/0.95 / Relief 50mm f/2
Limits Counts exclude the CSV column header; an earlier published count for the first lens mistakenly included it.
Digitized
1.5168
Schott BK7 refractive index at the d-line (587.6 nm), computed from published Sellmeier coefficients
Limits A catalogue constant the dispersion model must reproduce — quoted as a check, not as an engine measurement.
Fitted
16% steeper
blue-sensitive layer against green and red, one stock, from the published characteristic curve
Limits A ratio of the fitted straight-line gammas, not measured from film. One stock. The receipt's exact figures: +16.2% against the green layer, +15.8% against the red.
Fitted
B 0.639 / G 0.550 / R 0.552
per-layer gamma, one stock. Listed blue, green, red — the order the value is written in.
Limits Slopes of the straight-line region fitted to the digitised datasheet curve — not a measurement of developed film. Toe and shoulder lie outside the fit window.
Counted
1,600+ points across 7 stocks
characteristic curves read from manufacturer datasheets at 600 dpi
Limits Reconstruction residual ≈0.0015 logE.
Fitted
0.9997 / 0.987 R²
straight-line-region fit quality, floor over the worst layer — the featured stock / across all seven calibrated stocks
Limits R² of the straight-line region only — toe and shoulder are excluded, so this says nothing about fit quality there. The weakest fit is the tungsten-balanced stock's red layer; an earlier all-stocks figure copied the featured stock's neighbourhood and overstated it, which this receipt corrected.
Digitized
465 nm
blue-sensitive spectral peak of the consumer negative behind Amber 200, read from the official datasheet
Limits The other calibrated colour stocks peak at 440–451 nm; curve shapes around the peaks are the model's approximation.
Digitized
+0.1 logE
where the digitised green-layer slope of Amber 200 roughly halves (0.55 → 0.30)
Limits Read from the digitised curve; where the slope 'roughly halves' depends on how the two fit windows are placed.
Measured
74 71 59 → 144 140 125 → 201 198 186 sRGB
shadow → midtone → highlight neutral-cast swatches of the featured stock, engine output verified on 2026-07-28
Limits Engine output verified on 2026-07-28; retuning the develop path changes these values.
Measured
0.0031 / 0.0051 / 0.0031 encoded-luma σ
grain amplitude near black / at mid-grey / near white — medium grain preset, measured on the float renders before 8-bit quantisation
Limits One stochastic realisation (~37,000 pixels per patch). Encoded-signal amplitude — in linear light the peak sits nearer input 0.8. The preset's intensity and size constants are stylized; the density coupling is the measured part.
Measured
21 / 7 / 1 px — R / G / B
how far each channel's halation glow stays above a fixed 0.005 threshold beyond the edge of a 36 px synthetic disc — strongest vintage preset, engine output verified on 2026-07-29
Limits The reach depends on the threshold — a fainter cut grows every figure, and the R > G > B ordering held at every threshold the measurement tried; it is not claimed for arbitrary cuts (a threshold above every channel's glow reads zero for all three). The blue figure was 5 px when first published on 2026-07-16; re-measuring on 2026-07-29 gave 1 px, after intervening changes to how the additive glow is blurred. Synthetic highlight, not a point source; the ×3.4 display gain in the figure is presentation only, applied equally to all channels.
Estimated
60 / 90 / 45 seconds — R / G / B
characteristic times α in the layer-sensitivity model 1−exp(−α/t) behind the long-exposure article
Limits Literature-guided representative values for a typical colour negative — not fitted to a datasheet. Blue's α is smallest, so blue fails first.
Counted
41 samples
380–780 nm at 10 nm, the engine's internal spectral representation
Limits A design constant, not a measurement — spectral resolution is bounded by the 10 nm grid.
Counted
645 tests
library unit tests, build verified on 2026-07-28
Limits A count, not a coverage claim — it says nothing about what the tests assert, and it moves with every build.
Counted
56 binaries
audit and measurement binaries in the workspace
Limits A workspace count; it includes diagnostic binaries that never ship in the product.
Measured — Measured from engine output on the recorded verification date
Digitized — Read off a figure in published data
Fitted — A model fitted to data that was read off
Estimated — Estimated from representative values in the literature
Counted — A count
Lens MTF data read from the official technical data sheets published by Leica Camera AG. The two lenses are the Noctilux-M 50mm f/0.95 ASPH and the Summicron-M 50mm f/2. Glass dispersion from the Sellmeier coefficients in the Schott Optical Glass Data Sheets. Film characteristic curves read from official technical data published by Kodak and Ilford (E-4050 and others) at 600dpi with sub-pixel grid calibration, without smoothing. Product names are cited solely to identify public data sources. All trademarks belong to their respective owners; no affiliation or endorsement is implied.