5279EMULSION PROJECT
/
PUBLIC VISUAL ARCHIVE · V4—V49

Every improvement records
why the earlier result was wrong.

Only releases with comparable image or movie evidence appear here. Current V49 returns randomness to negative-density formation; V48 remains a reproducible witness of the display-RGB residual error.

CURRENT VISUAL RELEASEV49latest release with reviewable media on this site
RESEARCH CYCLE08organized by topic; experiment IDs live in the evidence appendix
NEXT VISUAL RELEASEmeasurement-dependentnumbered only after new measurement or a verifiable correction
V49DENSITY-DOMAIN CORRECTION

Do not suppress chroma noise—prevent the wrong RGB noise from forming

V49 fixes the remaining ownership error in V48. V48 gave 2383 ownership of deterministic colour but still added a formed-minus-mean residual from another observer graph in display RGB, producing lifted-sensor-like red, green and blue points in shadows. Kodak publishes only each record's marginal 48-micrometre Status-M RMS, not their cross-covariance. V49 therefore does not invent a coloured joint law: it publishes one symmetric common-density component while the image is still a negative and lets both material observers see that same formed negative directly.

2383 PROJECTION
T020 · same V49 formed negative → 2383 xenon / CIE observer
SCAN / DI OBSERVER
T020 · same V49 formed negative → Cineon / Scan-DI observer
Full changelog

Improvements

  • Removed display-RGB formed-minus-mean stochastic reinjection
  • Randomness completes inside the 5279 negative before both observers
  • Scaled the symmetric normalized common component by the smallest local Kodak marginal RMS
  • Raised web motion to 1920x1440 CRF 11 and added opponent-error gates against the still

Errors / limitations

  • V48's stochastic residual and deterministic mean belonged to different observer graphs
  • The public granularity curves do not identify the real 5279 cross-record spectrum
  • V49 common density is a conservative hypothesis boundary, not a claim of perfectly registered real record events

What we learned

  • The coloured-point defect was an image-formation ownership error, not merely excessive grain
  • Projection opponent RMS falls 62.8% while luma activity remains nearly unchanged
  • Unmeasured opponent variance should remain explicit uncertainty rather than being silently decided by an RGB safety filter
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[70] ISO 10505:2009 — Root mean square granularity of photographic films
V48FIRST-PRINCIPLES BASELINE

Let 2383 own its colour: separate mean image from stochastic delta

V48 adds no new style and does not continue the Silver Efex route. It corrects one V46 observer-ownership error: projection borrowed scan low-frequency opponent colour to contain speckles caused by unmeasured three-record covariance. The direct 5279-to-2383-to-xenon/CIE chain now owns the projection mean; the old safety transform contributes only the formed-minus-mean stochastic delta. Negative formation, grain, H-D, 48-micrometre RMS, MTF, 2383 and scan remain frozen.

2383 PROJECTION
T020 · direct 5279 → 2383 deterministic colour + managed stochastic delta
SCAN / DI OBSERVER
T020 · same V48 negative → Cineon / Scan-DI observer
Full changelog

Improvements

  • Direct 2383 observer owns the projection mean
  • V46 containment acts only on the formed-minus-mean stochastic delta
  • Scan and the complete negative remain unchanged
  • One second of paired native 5.7K 12-bit ProRes 4444 XQ

Errors / limitations

  • V46 wrote scan opponent colour into the projection mean
  • Removing all management would reopen speckles because the 5279 cross-spectrum remains unmeasured
  • Native NPS, speed-population recipe and exact DIR topology remain unknown

What we learned

  • Observer differences need no exaggerated hue
  • A safety filter that changes the mean has become a grade
  • Same-negative V48−V46 linear-RGB MAE is only 0.002200
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[58] Realistic Film Grain Rendering[68] CIE 1931 colour-matching functions, 2 degree observer — 1 nm data table[69] KODAK VISION Color Print Film 2383 / 3383 Technical Data, 2005
V47SILVER-HALIDE MORPHOLOGY

Beyond grain strength: reconstruct complex, heterogeneous silver-halide organization

V47 advances the Silver Efex evidence from an architectural clue to a controlled measurement. A 2048-square 16-bit flat-field probe was exported through the locally installed Nik 8 Kodak Tri-X 400 model, measuring RMS, spatial correlation, skew and excess kurtosis across sixteen tones. The first candidate matched strength and correlation length but was stopped mid-render because its near-zero kurtosis was still too orderly. The accepted SHM comparator uses multiscale populations, a slow occupancy field, asymmetric clusters/voids and a thick-tail Hermite population. It remains an independent same-class comparator: it does not replace V46's three-record 5279 or claim that a monochrome still product measures colour motion-picture film.

2383 PROJECTION
T020 · V46 deterministic 2383 observer → V47 SHM finite silver-halide organization
SCAN / DI OBSERVER
T020 · V46 deterministic scan/DI observer → V47 SHM finite silver-halide organization
Full changelog

Improvements

  • Measure Tri-X 400 RMS, lag-1, skew and kurtosis across sixteen tones through a controlled local Nik 8 export
  • Replace FSD's single correlated Gaussian copula with three independently seeded spatial populations
  • Add a slow occupancy field that varies fine/coarse population balance without creating a low-frequency brightness cloud
  • Use a second Hermite population for asymmetric clusters and voids and a third Hermite population for measured thick tails
  • Refit finite-site strength to N=1250 and reject the N=176 candidate that was about twice too strong and read closer to 16 mm
  • Form density through an inverse-binomial CDF at every tone instead of adding coloured noise to finished RGB
  • Hold the deterministic opponent field fixed and constrain only scalar-density travel at gamut boundaries
  • Renew organization per frame without translating or looping a grain plate

Errors / limitations

  • The first SHM candidate audited only the latent field against a broad stock envelope; formed-density kurtosis was slightly negative and still resembled orderly correlated noise
  • The N=176 prototype measured about 0.0286 RMS on the real frame—roughly twice the controlled Tri-X result and liable to read as 8/16 mm
  • Silver Efex is a monochrome still-image product and cannot identify 5279 cross-record covariance, pre-DIR layer randomness or motion-picture temporal law
  • The controlled export covers only 2048 square at Grain Size 1; automatic resolution/format scaling remains unmeasured
  • SHM operates after V46's deterministic observer, so it is a morphology experiment rather than a replacement physical-negative branch

What we learned

  • Equal RMS and grain radius do not guarantee silver-halide character; positive skew, positive excess kurtosis and local spectral variation govern rare clusters and voids
  • Controlled Tri-X lag-1 stays similar from shadows to highlights; tone mainly changes participation amplitude rather than manufacturing dramatic grain-size breathing
  • Varying population balance at constant local mean creates heterogeneous organization without an overlay-like mottle
  • The older FSD is genuinely a different route: it tests finite-density formation, while SHM additionally tests stock-owned complex spatial morphology
  • The earlier impression that N=176 read as 16 mm was not merely subjective; the controlled black-box test quantifies it
  • The honest current boundary is to retain physical V46 and keep SHM as a removable comparator on the same deterministic observers
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[58] Realistic Film Grain Rendering[63] Nik Silver Efex User Guide — Film Grain (Branded)[64] Nik Color Efex User Guide — Grain engine and calibrated branded films[65] DxO — The science of film: calibrated grain matrices by tone region[72] Simulating Film Grain using the Noise-Power Spectrum
V46CERTIFIED SPECTRAL INVERSE

Keep finite silver-halide events and spectral density inside their measurable boundaries

V46 is the first complete consolidation after the source-loss rebuild and a genuine numerical image correction. The former model held target RMS beyond Kodak's published granularity support while finite-site activation probability approached zero, creating very rare but enormous dye-density impulses. V46 holds the complete stochastic state at the measured endpoint. It also replaces the clipped iterative Status-M inverse with an exact nonnegative active-set solution. A 129-cubed base atlas plus local exact 5-cubed microbricks bounds maximum real-frame printer-density error to 0.0005094 D. Reliable findings from legacy internal V46–V86 studies are absorbed here, while those labels remain immutable experiment IDs rather than being rewritten as public releases.

2383 PROJECTION
T020 · V46 endpoint-stable 5279 → 2383 xenon projection
SCAN / DI OBSERVER
T020 · V46 same negative → Period 2K / Cineon scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T020 · independent FSD finite-density control · not part of V46
CAMERA BASELINERETAINED / SHARED
T020 · official Panasonic V-709 original camera witness · no film pipeline
ADDITIONAL SOURCENJARAW_S001_S001_T032 · Frame 0–23

T020 tests mixed foliage and bark; T032 tests rainy cyan-green shadows and low contrast; T007 tests water, fine green detail and local saturation. Projection and scan are newly rendered from the same V46 negative, while FSD and Panasonic V-709 remain independent controls.

2383 PROJECTION
T032 · V46 certified spectral inverse · 2383 projection
SCAN / DI OBSERVER
T032 · same V46 negative · Period 2K / Cineon scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T032 · FSD finite-site density
CAMERA BASELINERETAINED / SHARED
T032 · official Panasonic V-709 camera original
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T020 tests mixed foliage and bark; T032 tests rainy cyan-green shadows and low contrast; T007 tests water, fine green detail and local saturation. Projection and scan are newly rendered from the same V46 negative, while FSD and Panasonic V-709 remain independent controls.

2383 PROJECTION
T007 · V46 certified spectral inverse · 2383 projection
SCAN / DI OBSERVER
T007 · same V46 negative · Period 2K / Cineon scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T007 · FSD finite-site density
CAMERA BASELINERETAINED / SHARED
T007 · official Panasonic V-709 camera original
Full changelog

Improvements

  • Hold the complete stochastic state outside Kodak's measured granularity support, eliminating the catastrophic tail produced as activation probability approached zero
  • Replace the clipped projected Status-M spectral inverse with an exact nonnegative active-set/KKT solution
  • Build a power-2 129-cubed base atlas and load exact 5-cubed microbricks only at active-set boundaries and interpolation-disagreement regions
  • Discover cache demand from every real pixel before/after MTF and in mean/formed density for T020, T032 and T007; all 25,333 final risk cells are covered
  • Compile the adaptive observer as a parallel CPU kernel with fast-math disabled and bit-identical NumPy-reference output
  • Share one negative printer-density observation between scan and projection, removing duplicate spectral-inverse work
  • Keep evidence-minimal identity record formation; do not invent RGB cross-covariance from three 48-micrometre marginal curves
  • Render each scene as a 24-frame 5760×4320 12-bit ProRes 4444 master with a scale-honest web witness

Errors / limitations

  • The old model held fixed RMS outside published exposure support while activation probability tended to zero, causing calibration amplitude to diverge into rare broken-TV-like dye events
  • The former Status-M inverse clipped a projected iteration and did not guarantee the KKT optimum; worst shadow error had reached about 0.01399 D
  • A fixed 27-probe cell test missed active-set boundaries reached by real pixels; V46 uses the exact runtime predicate on complete pipeline states
  • Presenting legacy V46–V86 audits as dozens of visual releases confused research evidence with rendered releases; their historical IDs remain but the site consolidates them thematically
  • V46 still lacks a closed measured loop of same-batch 5279, same-batch 2383, measured printer light and a characterized scanner, so it is not an absolute colour-reproduction claim

What we learned

  • A granularity endpoint must constrain the complete finite-event state, not only macro RMS; otherwise a mathematically matched RMS can hide an unphysical microscopic tail
  • Sparse synthetic probes cannot represent real-frame risk; the final cache must cover pre/post-MTF and mean/formed density states
  • Exact active-set structure is locally sparse, favouring a base atlas plus risk microbricks over an expensive exact solve at every pixel
  • One negative printer-density result can derive both scan coordinates and projection input; branch differences belong after the shared negative
  • V70–V85 show that shared record events trade opponent grain for stronger luma grain rather than removing noise for free; identity record formation is the honest boundary without cross-spectral measurements
  • V46 improves numerical correctness and evidence ownership. It adds no creative grade and does not exaggerate hue merely to make projection and scan look different
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[21] The Essential Reference Guide for Filmmakers[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[27] ITU-R BT.1886 — HDTV studio display EOTF[48] LAD for KODAK VISION Color Print Film — H-61B[68] CIE 1931 colour-matching functions, 2 degree observer — 1 nm data table[69] KODAK VISION Color Print Film 2383 / 3383 Technical Data, 2005
V45OFFICIAL OBSERVER

Let 2383 be seen through a standard observer—not guessed through an approximation

V45 is a single-variable spectral measurement revision. Earlier versions sampled an analytical CIE 1931 approximation directly on the Kodak 2383 graph's 20 nm nodes. V45 uses the official CIE 1931 2-degree 1 nm table, linearly interpolates the same 2383 dye-density and xenon graph samples, and integrates them from 380 to 780 nm with explicit trapezoidal endpoint weights. The 5279 negative, H-D curves, three speed layers, DIR, MTF, grain, black, contrast, gamma, scanner and delivery are frozen. White changes by less than 4×10⁻⁷, so this is not a global white-balance move; only specific dye combinations are observed more accurately.

2383 PROJECTION
T020 · V45官方CIE 1 nm · 5279 → 2383 projection
SCAN / DI OBSERVER
T020 · V45冻结scan分支 · Period 2K / Cineon
FSD FINITE-SITE DENSITYRETAINED / SHARED
T020 · 独立FSD finite-site density对照 · 不并入V45
CAMERA BASELINERETAINED / SHARED
T020 · official Panasonic V-709 original camera witness · no film pipeline
ADDITIONAL SOURCENJARAW_S001_S001_T032 · Frame 0–23

T020 tests mixed foliage and dark bark; T032 tests rainy cyan-green low contrast; T007 tests water, green detail and local saturation. Each case includes projection, frozen scan, independent FSD and the Panasonic V-709 camera witness.

2383 PROJECTION
T032 · V45官方CIE 1 nm · 2383 projection
SCAN / DI OBSERVER
T032 · 冻结Period 2K / Cineon scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T032 · FSD finite-site density
CAMERA BASELINERETAINED / SHARED
T032 · official Panasonic V-709 camera original
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T020 tests mixed foliage and dark bark; T032 tests rainy cyan-green low contrast; T007 tests water, green detail and local saturation. Each case includes projection, frozen scan, independent FSD and the Panasonic V-709 camera witness.

2383 PROJECTION
T007 · V45官方CIE 1 nm · 2383 projection
SCAN / DI OBSERVER
T007 · 冻结Period 2K / Cineon scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T007 · FSD finite-site density
CAMERA BASELINERETAINED / SHARED
T007 · official Panasonic V-709 camera original
Full changelog

Improvements

  • Add and integrity-check the official CIE 1931 2-degree 1 nm observer table
  • Linearly interpolate Kodak's plotted 2383 dye density and xenon relative SPD from 20 nm to the 1 nm axis
  • Replace the analytical 20 nm approximation with trapezoidal integration over 380–780 nm
  • Rebuild and hash-lock a separate V45 193³ monitor lattice so an old runtime cache cannot be reused silently
  • Freeze V44 5279 formation, DIR, MTF, grain, scan, black, contrast, gamma and delivery
  • Validate T020, T032 and T007 for 24 native-resolution frames each; decode each web still from its final movie
  • Add a same-negative observer ablation, six-movie motion-colour gates and a two-transfer delivery-light consistency audit

Errors / limitations

  • V44 and earlier used an analytical CIE approximation rather than the CIE's published one-nanometre standard-observer table
  • Changing build_2383_projection_lut while still loading V30's 193³ cache would create a completely unchanged false upgrade; V45 binds profile and lattice hash
  • One-nanometre integration cannot invent unpublished Kodak measurements: the source 2383 dye and xenon graphs remain 20 nm plots
  • V45 still lacks a closed loop of same-batch 5279, 2383, measured illuminant and characterized scanner, so it is not an absolute reproduction claim

What we learned

  • The official one-nanometre observer leaves the dye-free white almost unchanged while producing measurable changes for specific dye mixtures
  • Across the complete 25³ spectral cube, linear-RGB RMS versus the old observer is 0.00456917 and maximum absolute node change is 0.0398455
  • After LAD, neutral-scale and normal-process monitor boundaries, the complete 193³ output-lattice RMS is 0.000215819 with mean RGB shift near 10⁻⁶ and a localized maximum of 0.03810
  • On one already-formed T020 negative, the scan remains bit-identical and projection linear-RGB RMS is only 0.000037904, confirming an observer correction rather than a grade
  • Peak-normalizing a single 2383 dye does not automatically increase accuracy; Status-A inversion cancels much arbitrary scale, making spectral shape more important than peak height
  • Spectral sampling, runtime lattice and delivered image must be one versioned evidence object
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[21] The Essential Reference Guide for Filmmakers[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[27] ITU-R BT.1886 — HDTV studio display EOTF[48] LAD for KODAK VISION Color Print Film — H-61B[68] CIE 1931 colour-matching functions, 2 degree observer — 1 nm data table[69] KODAK VISION Color Print Film 2383 / 3383 Technical Data, 2005
V44OBSERVER INTEGRITY

Let the negative, observer and display scale carry only their own physical facts

V44 is not another guessed grain profile. It answers the cheap coarse texture seen during native 5.7K playback. The unmeasured V43H NPS, Spirit and 2383-grain candidates are withdrawn; the accepted V42 negative returns. A rejected V44 candidate also proves that fully direct analytical projection colour produces unsafe dark opponent tails, so the validated V31 normal-process colour boundary remains instead of inventing a stronger projection difference. Native 5.7K masters remain untouched, while review derives display light from the encoded master, integrates it over actual review pixels and only then applies sRGB. The still decodes the final movie's same frame.

2383 PROJECTION
T020 · V42负片模型 → 2383亮度/纹理 · 正常工艺监看颜色边界
SCAN / DI OBSERVER
T020 · V42负片模型 → Period 2K / Cineon scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T020 · 独立FSD finite-site density对照 · 不并入V44
CAMERA BASELINERETAINED / SHARED
T020 · official Panasonic V-709 camera witness · no film pipeline
Full changelog

Improvements

  • Withdraw all three unmeasured V43H candidates and restore V42 negative morphology and the accepted period scanner
  • Retain the validated V31 normal-process monitor boundary: 2383 lightness/texture with low-frequency scan-referenced dye chroma
  • Keep stochastic 2383 grain at zero until a measured three-record NPS/covariance exists
  • Preserve the native 5760×4320 12-bit XQ master instead of blurring it to solve playback scaling
  • Add a 1920 review path: BT.1886 decode, linear-light pixel-area integration, then sRGB encoding
  • Derive the still from the same frame of the final encoded movie, removing the pre/post-encode split authority
  • Separate theatrical-print evidence, telecine/Blu-ray transfer and web display as three explicit evidence boundaries

Errors / limitations

  • V43H constrained a guessed NPS with official 48 µm RMS, but one aperture integral cannot identify a spatial spectrum
  • V43H added common-mode 2383 texture without public three-record statistics; ablation shows it explains only about 0.33% of projection high-frequency energy
  • The first V44 candidate disabled V31 completely; across 24 frames its projection dark opponent p99.99 reached 0.04882 with about 127 isolated >0.06 impulses per million dark pixels, so the candidate was rejected
  • A player using sharp resize on native 5.7K stochastic structure can fold energy above display Nyquist into coarse false texture
  • V44 is still not a measured closed loop of 5279 capture, same-batch 2383 and a characterized scanner; it cannot claim absolute reproduction

What we learned

  • The coarse result was mainly neither Wavefront error nor the new print grain; image structure and playback scaling acted together
  • On the matched frame, Lanczos review raises projection high-frequency energy to 1.71× and scan to 1.21× relative to linear-light area integration
  • The defensible fix is to retain the native master and provide a scale-defined review derivative—not arbitrarily soften the film model
  • A theatrical print, telecine/Blu-ray transfer and modern reference still have different light sources, white points, resolutions and finishing decisions; none is the other's colour truth
  • Current evidence supports a scan-referenced normal-process projection monitor; similar branch colour is a declared limitation and is more accurate than an invented theatrical colour difference
  • Projector flicker and development streakiness remain future measurable modules, not baseline effects added merely because they read as filmic
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[7] US 5,314,793 — multilayer speed and granularity architecture[8] Spirit DataCine / Spirit HD technical data[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[27] ITU-R BT.1886 — HDTV studio display EOTF[29] CSS Color Module Level 4 — sRGB and web colour[60] Evaluating video using QuickTime test pattern files[67] How Hollywood Fakes the 90s Film Look Today — Walter Volpatto interview
V43HHYPOTHESIS EDITION

Isolate the most likely unmeasured pieces as one reversible experiment

V43H asks a bounded question: what might 5279 look like if the most probable—but still unmeasured—negative grain spectrum, period Spirit observer and subordinate 2383 texture were completed? V42 colour, H-D, DIR, MTF, 48 µm RMS, black, gamma and RAW interpretation remain frozen. An isolated profile narrows the 35 mm cloud spectrum, moves one quarter toward a documented-family Spirit candidate and tests weak, spectrally neutral common-density 2383 texture. Projection and scan share one realized V43H negative; FSD remains independent; Panasonic V-709 is only a camera witness.

2383 PROJECTION
T020 · V43H 5279 → 2383 xenon projection
SCAN / DI OBSERVER
T020 · V43H 5279 → Period 2K / Cineon scan
FSD FINITE-SITE DENSITY
T020 · 独立FSD finite-site density对照
CAMERA BASELINE
T020 · official Panasonic V-709 camera witness · no film pipeline
ADDITIONAL SOURCENJARAW_S001_S001_T032 · Frame 0–23

T032 tests a rainy cyan-green scene, dark columns and low contrast; T007 tests water, green fine detail, local saturation and the sharpness/grain relationship. Every example includes projection, period scan, independent FSD and the unfilmed Panasonic V-709 witness.

2383 PROJECTION
T032 · V43H 2383 projection
SCAN / DI OBSERVER
T032 · V43H Period 2K scan
FSD FINITE-SITE DENSITY
T032 · FSD finite-site density
CAMERA BASELINE
T032 · official Panasonic V-709 camera original
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T032 tests a rainy cyan-green scene, dark columns and low contrast; T007 tests water, green fine detail, local saturation and the sharpness/grain relationship. Every example includes projection, period scan, independent FSD and the unfilmed Panasonic V-709 witness.

2383 PROJECTION
T007 · V43H 2383 projection
SCAN / DI OBSERVER
T007 · V43H Period 2K scan
FSD FINITE-SITE DENSITY
T007 · FSD finite-site density
CAMERA BASELINE
T007 · official Panasonic V-709 camera original
Full changelog

Improvements

  • Create a V43H-only profile with explicit hypothesis_not_measurement provenance
  • Keep official 48 µm RMS amplitude while narrowing and densifying the candidate 35 mm spatial spectrum
  • Move the period scanner only 25% toward a candidate bounded by DFT architecture and Kodak's generic telecine plot
  • Test weak, spectrally neutral common-density 2383 texture, estimate its amplitude from the three-record mean, and prohibit independent RGB print impulses
  • Produce projection and scan from the same V43H negative realization and reuse one spectral integration for the deterministic observer
  • Keep FSD as an independent finite-density route rather than promoting it into physical 5279
  • Render all three requested sources as projection, scan, FSD and official Panasonic V-709 camera witness
  • Write native 5.7K 12-bit XQ masters first, then derive sRGB companions, stills and hover media from the encoded files

Errors / limitations

  • V43H's grain NPS is not a Kodak measurement; 48 µm RMS cannot uniquely identify a spatial spectrum
  • The Spirit centres and bandwidths are not disclosed DFT responses—only a quarter-step toward a synthetic candidate
  • Public 2383 data do not identify three-record grain covariance or exposure-conditioned NPS, so the common-mode term remains subordinate
  • The outdoor T003/T005 charts do not authorize a new white balance, complete GH7 matrix or global saturation correction
  • The first discrete-spike gate mistook a Poisson expected count for a hard maximum; 17 real green-edge candidates on T007 failed only because ceil(14.7) was 15
  • Passing delivery gates proves internal consistency, not that predicted parameters have become 5279 facts

What we learned

  • V39's broken-television colour noise came directly from unidentified independent RGB print-Poisson tails; common density does not create isolated primary impulses
  • One observer integration can return both the physical realization and deterministic mean, so FSD does not require a second 193³ spectral graph
  • The matched T032 V42→V43H mean-channel change remains below 0.001, preserving a predictive difference without adding a grade
  • Grain fineness can change while official 48 µm RMS stays fixed because aperture amplitude and spatial NPS are different constraints
  • A stochastic event rate needs a statistical acceptance bound: 432 tests now use a Bonferroni one-percent family-wise false-rejection rate, while V39's thousands-per-million failure remains orders of magnitude outside it
  • The central product boundary of a Hypothesis Edition is reversibility: every unmeasured degree of freedom must remain independently removable
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[7] US 5,314,793 — multilayer speed and granularity architecture[8] Spirit DataCine / Spirit HD technical data[21] The Essential Reference Guide for Filmmakers[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[49] Panasonic V-Log to V-709 3D LUT[58] Realistic Film Grain Rendering[59] A Reproduction Model of Film Grain Texture for Digital Movies
V42ARCHIVED CALIBRATION

Make the engine actively defend the conclusions of the research

V42 is not a new grade and does not claim a new Kodak measurement. It freezes V41 colour, density, grain, DIR, MTF, black, gamma and both observers, then turns the accepted V37–V41 conclusions into startup gates the engine must pass. The validated Philox-u32 Bernoulli Metal graph is now the Production default. Archive CPU remains a reproducible reference, but a different stochastic implementation is no longer described as the same grain realization. Delivery has one picture authority: encode the 12-bit BT.1886 master first, then derive the sRGB QuickTime companion and still from that delivered file.

2383 PROJECTIONRETAINED / SHARED
T031 · V41同帧画面见证 · V42冻结图像模型 / 2383观察
SCAN / DI OBSERVERRETAINED / SHARED
T031 · V41同帧画面见证 · V42冻结图像模型 / Period 2K scan
FSD FINITE-SITE DENSITYRETAINED / SHARED
T031 · V41 FSD对照见证 · V42研究边界
CAMERA BASELINERETAINED / SHARED
T031 · Panasonic V-709 As Shot witness
Full changelog

Improvements

  • Name the explicit engine V42 so a software 'V2' cannot be confused with the image-version history
  • Assert V37 stable integration, V40 colour-grain repair and V41 colour/record boundaries at runtime
  • Make the validated Philox-u32 Bernoulli Metal graph the default Production execution
  • Retain Archive CPU and Reference NumPy as research references without demanding particle-for-particle identity
  • Freeze +0.45 stop, grain 1.0, oversample 1 and salt 0 for the baseline; any override must be marked experimental
  • Write only the BT.1886 professional picture during formation and derive sRGB/JPEG from the encoded master
  • Correct the recovery record: byte identity proves the Archive refactor, not identical Metal and NumPy emulsions
  • Record the V41 engine-directory loss explicitly and protect 214 authored source, test and research files with a SHA-256 inventory checked by GitHub CI

Errors / limitations

  • Before recovery, the complete engine lived only in an unversioned local experiment directory; its disappearance required reconstructing 199 files from 895 successful edit records
  • No surviving evidence attributes the deletion trigger to Claude, the Python crash, the macOS watchdog event or a cleanup command; the trigger remains unknown
  • The public hero currently retains the matched V41 Production witness; the formal V42 one-second three-source rerender is not yet published
  • Executable gates prevent known research drift but cannot replace missing 5279 NPS, coating or scanner measurements
  • V41's 12.5% colour residual remains reversible outdoor-chart evidence; the V42 name does not promote it to a complete GH7 characterization

What we learned

  • Version correctness should be defined by formation equations, statistical contracts and delivery authority—not one fortunate random pixel hash
  • Archive and Production may form different emulsion instances while obeying the same H-D, 48 µm RMS, NPS and temporal-independence boundaries
  • Runtime research assertions prevent profile leakage or optimization code from silently rewriting the image model
  • Deriving every viewing file from the delivered 12-bit master structurally prevents the still and movie from becoming different pictures
  • The established failure cause was a single unversioned source copy; the exact deletion mechanism cannot be reconstructed from the surviving logs
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[47] CVProResRawMetadata[58] Realistic Film Grain Rendering[63] Nik Silver Efex User Guide — Film Grain (Branded)[66] Complete Guide to Using the DKC-Pro Color Chart — colourimetry data
V41ARCHIVED CALIBRATION

Let the chart reveal direction without turning one shoot into a grade

V41 uses T003 to estimate the direction of the remaining input-chroma residual, then treats the closer, mildly defocused T005 as a true holdout. Both clips repeat the same under-chroma and hue-error direction across synthetic and natural colours, so the boundary is probably real. But both were captured outdoors at 5500 K under directional light, which is not enough to characterize the GH7. V41 therefore applies only 12.5% of the fitted correction, preserves scene luminance and the neutral axis, and leaves white balance, exposure, black, contrast and gamma untouched. It also replaces V40's hard intermediate-basis clip with record-safe signed transport: signed values survive only when all three 5279 record exposures remain non-negative; unsafe pixels fall back to V40.

2383 PROJECTION
T031 · Frame 132–155 · V41物理5279 → 2383正常工艺监看
SCAN / DI OBSERVER
T031 · Frame 132–155 · V41物理5279 → Period 2K scan
FSD FINITE-SITE DENSITY
T031 · Frame 132–155 · V41颜色 → FSD有限位点密度
CAMERA BASELINE
T031 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T002 · Frame 0–23

T002 tests dark low-chroma tails and black stability. T007 tests water, green fine detail, saturation and 35 mm sharpness. T031 tests organic shadow texture through both observers.

2383 PROJECTION
T002 · V41物理5279 → 2383正常工艺监看
SCAN / DI OBSERVER
T002 · V41物理5279 → Period 2K scan
FSD FINITE-SITE DENSITY
T002 · V41颜色 → FSD有限位点密度
CAMERA BASELINE
T002 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T002 tests dark low-chroma tails and black stability. T007 tests water, green fine detail, saturation and 35 mm sharpness. T031 tests organic shadow texture through both observers.

2383 PROJECTION
T007 · V41物理5279 → 2383正常工艺监看
SCAN / DI OBSERVER
T007 · V41物理5279 → Period 2K scan
FSD FINITE-SITE DENSITY
T007 · V41颜色 → FSD有限位点密度
CAMERA BASELINE
T007 · Panasonic V-709 As Shot witness
CONTROLLED PIPELINE COMPARISON

Three density models, one observer boundary

These are not three grades. Physical 5279 is the V41 model; FSD is a finite-site control informed by the Silver Efex study; the deterministic baseline disables stochastic density.

T031 · Frame 132–155

Same V41 colour, 5279 mean and 2383 observer; only stochastic density formation changes.

V41 PHYSICAL 5279
Full three-record 5279 emulsion formation
FSD FINITE-SITE DENSITY
Independent inverse-binomial density control
DETERMINISTIC NO-GRAIN
Stochastic density disabled; colour, MTF and observer retained
T002 · Frame 0–23

Same V41 colour, 5279 mean and 2383 observer; only stochastic density formation changes.

V41 PHYSICAL 5279
Full three-record 5279 emulsion formation
FSD FINITE-SITE DENSITY
Independent inverse-binomial density control
DETERMINISTIC NO-GRAIN
Stochastic density disabled; colour, MTF and observer retained
T007 · Frame 276–299

Same V41 colour, 5279 mean and 2383 observer; only stochastic density formation changes.

V41 PHYSICAL 5279
Full three-record 5279 emulsion formation
FSD FINITE-SITE DENSITY
Independent inverse-binomial density control
DETERMINISTIC NO-GRAIN
Stochastic density disabled; colour, MTF and observer retained
Full changelog

Improvements

  • Add the closer T005 chart as an independent holdout that was never used to fit the transform
  • Estimate a cross-group residual in Bradford-adapted D50 chroma while restoring exact D65 scene luminance
  • Reject the visibly excessive 100% result and the still-too-large 25% result; retain a conservative 12.5% step
  • Reduce median hue error in synthetic and natural groups on both T003 and T005
  • Improve natural-colour chroma error on both clips
  • Use non-negative 5279 record exposure as the safety condition for signed intermediate transport
  • Freeze all V40 grain, DIR, MTF, 2383, scan, black, contrast and gamma parameters
  • Feed physical 5279, FSD and deterministic controls from the same V41 colour boundary

Errors / limitations

  • One outdoor directional-light chart cannot identify a complete camera matrix, illuminant SPD or stock-specific interimage response
  • The first 100% matrix visibly over-corrected foliage and yellow patches
  • The 25% candidate passed chart gates but still raised final 2383 median chroma by about 15%, beyond the evidence
  • T005 defocus is acceptable for patch medians, but local glare and gradients limit individual-patch precision
  • V41 remains a reversible colour-boundary experiment, not a final GH7 characterization before uniform D65 and tungsten controls

What we learned

  • A defocused chart can still witness large-patch statistics when sampling stays away from edges and within-patch dispersion is reported
  • Repeating the direction on a disjoint clip is more informative than increasing fit order on one chart
  • A 100% correction in chart space can be amplified by negative and 2383 nonlinearities, so the final formed image needs its own gate
  • A 12.5% step lowers hue error in both groups on both clips while keeping input luminance change below 1e-5
  • V40's non-negative intermediate basis was not a RAW clip; the physical safety condition is non-negative exposure in the combined records
  • FSD and physical 5279 should share a colour boundary while remaining distinct density-formation hypotheses
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[47] CVProResRawMetadata[58] Realistic Film Grain Rendering[63] Nik Silver Efex User Guide — Film Grain (Branded)[66] Complete Guide to Using the DKC-Pro Color Chart — colourimetry data
V40ARCHIVED CALIBRATION

Accurate grain constrains energy, covariance and extreme tails

V40 withdraws three unsupported V39 inferences: treating Kodak's processed 48 µm RMS measurement as a pre-DIR source-layer target, passing marginal record RMS to the observer without the validated opponent-colour integration, and inventing independent RGB/record Poisson grain for 2383 without stock-specific covariance or NPS evidence. V40 preserves density-domain 5279/2383 MTF and the accepted colour, black and gamma. It restores the RMS constraint to the published post-process boundary, restores opponent integration inside both observers, and prevents the final V31 adapter from reintroducing the high-frequency opponent residual. Intrinsic stochastic 2383 grain remains withheld until it can be measured. This is not chroma denoising; it refuses to create colour degrees of freedom that the evidence does not identify.

2383 PROJECTION
T031 · Frame 132–155 · V40 5279 → 2383正常工艺监看
SCAN / DI OBSERVER
T031 · Frame 132–155 · V40 5279 → Period 2K scan
CAMERA BASELINE
T031 · Frame 132–155 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T002 · Frame 0–23

T002 stresses dark low-chroma tails. T007 tests water, green fine detail and edge integration. T031 tests organic shadow texture through both observers.

2383 PROJECTION
T002 · Frame 0–23 · V40 2383正常工艺监看
SCAN / DI OBSERVER
T002 · Frame 0–23 · V40 Period 2K scan
CAMERA BASELINE
T002 · Frame 0–23 · Panasonic V-709 As Shot
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T002 stresses dark low-chroma tails. T007 tests water, green fine detail and edge integration. T031 tests organic shadow texture through both observers.

2383 PROJECTION
T007 · Frame 276–299 · V40 2383正常工艺监看
SCAN / DI OBSERVER
T007 · Frame 276–299 · V40 Period 2K scan
CAMERA BASELINE
T007 · Frame 276–299 · Panasonic V-709 As Shot
CONTROLLED PIPELINE COMPARISON

Three density models, one observer boundary

These are not three grades. Physical 5279 is the V40 model; FSD is a finite-site control informed by the Silver Efex study; the deterministic baseline disables stochastic density.

T031 · Frame 132–155

Same mean image and 2383 observer; only the density-formation mechanism changes.

V40 PHYSICAL 5279
Full three-record 5279 emulsion formation
FSD FINITE-SITE DENSITY
Independent inverse-binomial density control
DETERMINISTIC NO-GRAIN
Stochastic density disabled; colour, MTF and observer retained
T002 · Frame 0–23

Shadows and low-chroma surfaces test tails, opponent colour and overlay-like noise.

V40 PHYSICAL 5279
Full three-record 5279 emulsion formation
FSD FINITE-SITE DENSITY
Independent inverse-binomial density control
DETERMINISTIC NO-GRAIN
Stochastic density disabled; colour, MTF and observer retained
T007 · Frame 276–299

Water, fine edges and green detail test whether grain scale and 35 mm sharpness coexist.

V40 PHYSICAL 5279
Full three-record 5279 emulsion formation
FSD FINITE-SITE DENSITY
Independent inverse-binomial density control
DETERMINISTIC NO-GRAIN
Stochastic density disabled; colour, MTF and observer retained
Full changelog

Improvements

  • Restore the 48 µm granularity constraint to Kodak's measured post-process density boundary
  • Restore high-frequency opponent-colour integration inside the scan and 2383 observers
  • Stop the final V31 adapter from re-adding an already-integrated high-frequency opponent residual
  • Withdraw independent stochastic 2383 record populations until covariance/NPS evidence exists
  • Withdraw V39's underidentified signed intermediate film-RGB cancellation
  • Gate opponent energy and isolated 3×3 primary-colour impulses on every native delivered frame
  • Make the first picture authority 12-bit ProRes 4444 XQ, then derive every viewing and website image from it
  • Add FSD finite-site density and a deterministic no-grain baseline as controlled comparisons without changing physical V40
  • Withdraw FSD's random gamut-boundary chroma scaling and reform its density in the post-observer signal domain with a fixed opponent field
  • Audit the Apple-standard RAW input with the T003 DKC-Pro chart; its warm as-shot evidence does not justify a global magenta trim or automatic white balance

Errors / limitations

  • V39 matched each record's marginal RMS but did not constrain cross-record covariance or distribution tails
  • Kodak's 48 µm number describes processed film and cannot uniquely invert pre-DIR speed-layer yields
  • The public 2383 material supplies no exposure-conditioned three-record grain covariance or NPS; independent RGB Poisson was false precision
  • V39's signed intermediate film basis introduced underidentified channel cancellation in dark green regions
  • The final V31 adapter was re-adding high-frequency opponent colour after the observer had already integrated it
  • An 8-bit JPEG representative frame can smooth colour impulses and cannot substitute for a 12-bit movie gate

What we learned

  • Grain identity is not only RMS and size: record covariance, skew, extreme tails and observer integration decide whether it reads as silver or digital chroma noise
  • Density remains the image variable, but a published posterior granularity measurement cannot be moved arbitrarily earlier in the chemistry
  • Without 2383 stochastic statistics, retaining 5279 structure transferred by print MTF is more accurate than inventing print grain
  • Native-resolution every-frame tail audits catch sparse failures that representative stills and downscaled proxies miss
  • The local Silver Efex engine confirms G=inverse-CDF(Binomial(N,p),u)/N followed by Y'=(1-alpha)Y+alpha G, not display-space additive noise
  • Every Silver Efex B&W stock owns a separate measured 1000-square morphology; this supports stock-specific modelling but cannot become 5279 three-record data
  • FSD's first linear-RGB colour transport failed the native dark-tail gate; the corrected post-observer signal-domain definition holds opponent colour fixed and removes the sparse primary impulses without blur or desaturation
  • Without copying a stock texture, FSD at N=176 and sigma 0.597 px matches physical V40's calibration-frame luma RMS, high-pass energy and spatial correlation while retaining much lower opponent residual by design
  • DKC-Pro rows are not equal height and a printed title strip precedes patches 7–12; the first chroma sampling was withdrawn, and the corrected grid leaves an input matrix as a plausible boundary that still needs controlled-light validation
  • A synthetic D65 gray ramp stays below 0.00018 maximum delta u-prime/v-prime through both observers, rejecting a neutral-green crossover; a constant warm ramp produces 0.00253/0.00220 exposure-dependent crossover whose true 5279 magnitude remains unmeasured
  • T003 neutral patches 2–5 average R/G=1.172 and B/G=0.748, rejecting a fixed shared green-decode hypothesis; real daylight still cannot identify a new white balance, black offset or camera matrix
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[5] LAD — Laboratory Aim Density, KODAK H-61[7] US 5,314,793 — multilayer speed and granularity architecture[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[45] Apply built-in camera LUTs in Final Cut Pro[47] CVProResRawMetadata[58] Realistic Film Grain Rendering[63] Nik Silver Efex User Guide — Film Grain (Branded)[64] Nik Color Efex User Guide — Grain engine and calibrated branded films[65] DxO — The science of film: calibrated grain matrices by tone region[66] Complete Guide to Using the DKC-Pro Color Chart — colourimetry data
V39WITHDRAWN EXPERIMENT

Grain is not a residual on the image; density is the image

V39 resolves structural misplacements found in the complete research and code audit. V38 formed a finite-site negative but still applied MTF to a display positive, represented 2383 grain as a luminance ratio, normalized granularity after stochastic DIR, and clipped signed wide-gamut basis components before the three physical film records were formed. V39 applies 5279 MTF to processed negative density, constrains developed dye yield before stochastic DIR, and lets the scanner or 2383 observe that single realized density. 2383 MTF and three-record dye clouds are formed in Status-A density with no display grain operation. V38 colour, H-D, black, gamma and dual delivery remain frozen. This is not a prettier grade; it puts image structure back in its measured domains.

2383 PROJECTION
T031 · Frame 132–155 · V39 5279密度 → 2383密度 → 正常工艺监看
SCAN / DI OBSERVER
T031 · Frame 132–155 · V39 5279密度 → Period 2K scan
CAMERA BASELINE
T031 · Frame 132–155 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T002 · Frame 0–23

T002 tests toe and low-chroma texture. T007 tests water, grass and high-frequency greens. T031 tests dark organic texture and the full density-to-print path.

2383 PROJECTION
T002 · Frame 0–23 · V39 2383正常工艺监看
SCAN / DI OBSERVER
T002 · Frame 0–23 · V39 Period 2K scan
CAMERA BASELINE
T002 · Frame 0–23 · Panasonic V-709 As Shot
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T002 tests toe and low-chroma texture. T007 tests water, grass and high-frequency greens. T031 tests dark organic texture and the full density-to-print path.

2383 PROJECTION
T007 · Frame 276–299 · V39 2383正常工艺监看
SCAN / DI OBSERVER
T007 · Frame 276–299 · V39 Period 2K scan
CAMERA BASELINE
T007 · Frame 276–299 · Panasonic V-709 As Shot
Full changelog

Improvements

  • Move processed 5279 MTF from display-linear RGB into negative record density
  • Apply the published 48 µm RMS constraint to developed dye yield before stochastic DIR/interimage transport
  • Move 2383 MTF and independent three-record finite dye clouds into Status-A print density
  • Remove the final display-luminance-ratio 2383 grain operation
  • Preserve signed BT.2020-to-film basis components and clamp once after physical record exposure is formed
  • Make the active profile own the 0.38-pixel phase radius and projection opponent crossover
  • Make V39→V38 switching restore every Archive-domain setting explicitly
  • Remove duplicate analytical projection work and share dual-observer intermediates with zero pixel error
  • Derive the sRGB viewing copy from the encoded BT.1886 master and use ProRes 4444 XQ to retain V39's fine density structure

Errors / limitations

  • V38 applied the published processed-negative MTF to an already formed display positive
  • V38 scaled the combined residual after stochastic DIR, so chemistry never saw the calibrated developed event
  • V38's fine 2383 term was a post-projection luminance ratio and remained structurally overlay-like
  • V38 clipped signed wide-gamut basis values before the film-record sensitivity matrix
  • A pointwise 193³ negative-to-output cache cannot represent a spatially formed 2383 density image
  • Independently compressing BT.1886 and sRGB copies from floating-point light ceased to be transparent once V39 carried finer density structure

What we learned

  • Density is the image variable, but density magnitude is not sharpness; MTF describes the spatial transfer of density modulation
  • MTF and granularity must share one physical geometry and density chain while remaining separately measurable properties
  • Moving the RMS constraint before chemistry fixes layer order but cannot reveal Kodak's unpublished coating recipe
  • Public 2383 material does not identify exposure-conditioned record granularity, so print grain remains subordinate and explicitly bounded
  • The released projection is still a normal-process Rec.709 monitor proof with a period-scan low-frequency colour boundary; physical projection remains available but is not mislabelled as a theatre measurement
  • Quality-first spatial 2383 formation moves the bottleneck from grain sampling to the accurate analytical observer
  • Both transfers must share the delivered master's compression structure; the master-derived XQ companion limits the worst per-channel mean light error to 0.001092
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[5] LAD — Laboratory Aim Density, KODAK H-61[7] US 5,314,793 — multilayer speed and granularity architecture[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[45] Apply built-in camera LUTs in Final Cut Pro[47] CVProResRawMetadata[58] Realistic Film Grain Rendering
V38PREVIOUS BASELINE

One observer light, encoded explicitly for each display target

V38 corrects the natural-still / dense-QuickTime split in V37. The completed projection and scan observers already produced display-linear light, but the release then applied the BT.709 camera OETF and allowed QuickTime, JPEG and the browser to invert or approximate it differently. Shadows, contrast and colour density diverged even though the film calculation was identical. V38 freezes the complete V37 image model and encodes one observer-linear result two ways: an inverse-BT.1886 gamma-2.4 professional Rec.709 master, and a 12-bit sRGB-transfer QuickTime companion for direct viewing on this Mac. JPEG and web motion derive only from the companion. P3 and HDR are not used as unmeasured saturation or brightness controls.

2383 PROJECTION
T031 · Frame 132–155 · V38 2383观察 · sRGB Mac viewing chain
SCAN / DI OBSERVER
T031 · Frame 132–155 · V38 Period 2K scan · sRGB Mac viewing chain
CAMERA BASELINE
T031 · Frame 132–155 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T002 · Frame 0–23

T002 stresses toe and dark neutral texture. T007 holds water and green fine detail. T031 exposes the original natural-still / dense-video discrepancy.

2383 PROJECTION
T002 · Frame 0–23 · V38 2383观察 · sRGB Mac viewing chain
SCAN / DI OBSERVER
T002 · Frame 0–23 · V38 Period 2K scan · sRGB Mac viewing chain
CAMERA BASELINE
T002 · Frame 0–23 · Panasonic V-709 As Shot
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T002 stresses toe and dark neutral texture. T007 holds water and green fine detail. T031 exposes the original natural-still / dense-video discrepancy.

2383 PROJECTION
T007 · Frame 276–299 · V38 2383观察 · sRGB Mac viewing chain
SCAN / DI OBSERVER
T007 · Frame 276–299 · V38 Period 2K scan · sRGB Mac viewing chain
CAMERA BASELINE
T007 · Frame 276–299 · Panasonic V-709 As Shot
Full changelog

Improvements

  • Freeze the V37 negative, 2383, scan, colour, grain, MTF, DIR, black and gamma
  • Define delivery input as completed display-linear Rec.709 observer light—not scene-linear camera exposure
  • Encode the professional master with inverse BT.1886 gamma 2.4 in 12-bit Rec.709 ProRes 4444
  • Add a separate 12-bit sRGB-transfer QuickTime companion that matches JPEG review in the Mac default display mode
  • Generate JPEG and web motion from the same companion and the same representative frame
  • Audit both files by decoding them back to one display-linear result

Errors / limitations

  • Since V25, the release comments treated the BT.709 camera OETF and a BT.1886 reference display as a reversible pair even though they are not inverses
  • The previous web gate allowed channel MAE up to 2.5% and median-luma error up to 1%, enough for a visible shadow mismatch to pass
  • V37 JPEG used an exact BT.709 inverse before sRGB encoding while QuickTime applied Apple's video-gamma interpretation; the two paths never shared one ODT

What we learned

  • The still was closer to the intended observer-linear image; much of the video's extra density was a delivery/playback artifact rather than 5279 character
  • This MacBook Pro's Liquid Retina XDR includes a dedicated HDTV Video BT.709–BT.1886 reference mode, which is more appropriate than a speculative P3 or HDR expansion
  • A P3-capable panel cannot recover colour already bounded by the Rec.709 observer; changing container gamut would add no evidence
  • Professional and QuickTime files may carry different code values, but must decode to the same display-linear light
  • Release validation must cross the master, still, web proxy and an actual managed player—not only inspect file tags
Evidence: [26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[29] CSS Color Module Level 4 — sRGB and web colour[60] Evaluating video using QuickTime test pattern files[61] Use presets and reference modes with your Apple display[62] MacBook Pro (16-inch, 2024) Technical Specifications
V37ARCHIVED CALIBRATION

Every frame is new film; the sampler no longer breathes as one field

V37 answers the overlay-like boil visible in the local QuickTime master. Successive film frames expose different pieces of emulsion, so silver-halide sites should remain independent from frame to frame. The error was not independence itself: V36 also rotated one whole-frame bilinear subpixel sampling phase every frame, adding a second numerical animation. V37 keeps all 45 new Philox emulsion identities per frame and changes only the integration kernel to a 30-degree stable-balanced phase. In the T031 ablation, whole-frame high-frequency variation in projection falls by about 60% and directional variation by about 71%, while the average orientation remains neutral. Colour, H-D, black, gamma, MTF, DIR, grain amplitude, grain size and both observers remain frozen.

2383 PROJECTION
T031 · Frame 132–155 · V37 stable emulsion → 2383 cinema observer
SCAN / DI OBSERVER
T031 · Frame 132–155 · V37 stable emulsion → Period 2K scan
CAMERA BASELINE
T031 · Frame 132–155 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T002 · Frame 0–23

T002 controls walls, shadows and fine texture. T007 verifies that water and foliage are not frozen or softened. T031 is the phase-selection scene for dark bark, fungi and surrounding green.

2383 PROJECTION
T002 · Frame 0–23 · V37 2383 cinema observer
SCAN / DI OBSERVER
T002 · Frame 0–23 · V37 Period 2K scan
CAMERA BASELINE
T002 · Frame 0–23 · Panasonic V-709 As Shot
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T002 controls walls, shadows and fine texture. T007 verifies that water and foliage are not frozen or softened. T031 is the phase-selection scene for dark bark, fungi and surrounding green.

2383 PROJECTION
T007 · Frame 276–299 · V37 2383 cinema observer
SCAN / DI OBSERVER
T007 · Frame 276–299 · V37 Period 2K scan
CAMERA BASELINE
T007 · Frame 276–299 · Panasonic V-709 As Shot
Full changelog

Improvements

  • Continue to form independent silver-halide/dye-cloud sites on every frame; add no temporal smoothing, advection or frozen grain plate
  • Replace whole-frame per-frame subpixel rotation with a 30-degree stable-balanced integration phase
  • Screen native T031 candidates at 0, 30 and 90 degrees; reject the fixed bias at 0 and over-correction at 90
  • Keep the 0.38-pixel radius, five size classes, three speed populations and three colour records unchanged
  • Freeze every V36 colour, density, sharpness, black, gamma and observer parameter
  • Render one native 5.7K 12-bit second for T002, T007 and T031 in projection and period-scan branches

Errors / limitations

  • The initial V35 T031 tail audit compared frames 0–23 with V34 frames 132–155 and overstated extreme temporal differences; that conclusion is formally withdrawn
  • A fixed 0-degree phase greatly stabilized temporal energy but left measurable horizontal/vertical preference and could not be released
  • Reducing grain strength or correlating it over time would hide the symptom while violating frame-independent motion-picture emulsion

What we learned

  • Independent random fields do not require whole-frame statistics to breathe; a globally changing numerical integration kernel creates an extra animation
  • The 30-degree balanced phase retains the V36 mean orientation on T031 while reducing projection high-pass CV to 0.400×
  • Scanner aperture and scene structure mask part of the phase benefit, so projection and scan must be measured separately
  • Organic grain motion comes from independent density formation under a stable imaging operator—not from moving one noise texture with the scene
  • Continuous grain centres and density-domain 2383 formation remain valuable future research, but stay outside the baseline until they pass equivalent gates
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[21] The Essential Reference Guide for Filmmakers[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[58] Realistic Film Grain Rendering[59] A Reproduction Model of Film Grain Texture for Digital Movies
V36ARCHIVED CALIBRATION

Match the frame before judging 35 mm grain and sharpness

V36 does not make the grain finer or hide it with extra softness. The audit found that V35 T007 and T031 began at frame 0, while V34 used the curated frame-276 and frame-132 windows. Source motion and texture were therefore presented as a film-model change. V36 locks camera, projection, scan, still and hover video to the same absolute source frames, then rechecks processed-stock MTF and 48 μm diffuse RMS granularity on one physical scale. At the correct T031 window, Production differs from V34 by only about 0.1% in high-frequency, temporal and grain-to-edge measures, so the film model stays frozen.

2383 PROJECTION
T031 · Frame 132–155 · V36 5279 → 2383 cinema observer
SCAN / DI OBSERVER
T031 · Frame 132–155 · V36 5279 → Period 2K scan
CAMERA BASELINE
T031 · Frame 132–155 · Panasonic V-709 As Shot witness
ADDITIONAL SOURCENJARAW_S001_S001_T002 · Frame 0–23

T002 is the unchanged-window control. T007 restores frame 276 for water and fine foliage; T031 restores frame 132 for dark bark, fungi and surrounding green.

2383 PROJECTION
T002 · Frame 0–23 · V36 2383 cinema observer
SCAN / DI OBSERVER
T002 · Frame 0–23 · V36 Period 2K scan
CAMERA BASELINE
T002 · Frame 0–23 · Panasonic V-709 As Shot
ADDITIONAL SOURCENJARAW_S001_S001_T007 · Frame 276–299

T002 is the unchanged-window control. T007 restores frame 276 for water and fine foliage; T031 restores frame 132 for dark bark, fungi and surrounding green.

2383 PROJECTION
T007 · Frame 276–299 · V36 2383 cinema observer
SCAN / DI OBSERVER
T007 · Frame 276–299 · V36 Period 2K scan
CAMERA BASELINE
T007 · Frame 276–299 · Panasonic V-709 As Shot
Full changelog

Improvements

  • Lock T002 0–23, T007 276–299 and T031 132–155 as absolute source windows
  • Expose source frames in every web branch and fail mismatched comparisons
  • Re-run Philox and spatial-kernel ablations at the correct T031 frame
  • Add a joint physical-scale audit for 5279 MTF and 48 μm granularity
  • Freeze V35 colour, black, gamma, MTF, DIR, grain amplitude and spectrum
  • Use a shorter-GOP, higher-fidelity hover proxy so delivery encoding does not exaggerate grain boil

Errors / limitations

  • The first V36 salt screen repeated the frame-0 mistake; all four-salt results were invalidated once the frame contract was found
  • V35 recorded 24-frame trials but did not make absolute start frame a release-blocking comparison field
  • MTF or 48 μm RMS alone cannot prove a 35 mm impression; both must share film geometry and observer scale

What we learned

  • The released V35 T031 frame is pixel-identical to a new frame-0 Production render, proving a segment-selection error rather than a hidden model change
  • At frame 132, Philox/V34 median temporal-difference RMS is 1.00139 and grain-to-edge ratio is 1.00131
  • Grains constitute the realized density image, but absolute density is not sharpness; MTF describes spatial density-modulation transfer
  • Kodak E-58 makes noise frequency, negative and print granularity, both MTF stages and magnification joint determinants of visible graininess
  • Quality first includes refusing to retune a correct film model to compensate for an invalid comparison
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[21] The Essential Reference Guide for Filmmakers[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[55] Metal Performance Shaders tuning hints[56] Parallel Random Numbers: As Easy as 1, 2, 3 — Random123 / Philox
V35ARCHIVED CALIBRATION

Do not change the film—change the computation: an auditable Production graph

V35 is not a new grade and does not rewrite the film. V34 colour, black, gamma, MTF, DIR, grain amplitude and grain spectrum are frozen. What changes is finite-site execution. Pixel, frame, record, speed layer and size class receive deterministic Philox4x32-10 identities; complete uint32 words are compared directly with a 2^32 fixed-point threshold derived from the float32 probability. Asynchronous Metal sampling overlaps CPU expectation filtering and the V31 colour boundary reuses memory. Twenty-four-frame, five-region colour, clipping, RGB high-frequency correlation, grain-energy and temporal-difference gates all pass while both masters render 23.65% faster than V34.

2383 PROJECTION
T031 · V35正常5279 → 2383 cinema observer · Philox-u32 Production
SCAN / DI OBSERVER
T031 · V35 5279 → Period 2K scan · Philox-u32 Production
CAMERA BASELINE
T031 · official Panasonic V-709 · As Shot 0.00 stop见证
ADDITIONAL SOURCENJARAW_S001_S001_T002

T002 carries the complete five-region temporal gate; T007 stresses water, grass and near-zero hard black; T031 stresses dark neutral texture, warm fungi and green surroundings.

2383 PROJECTION
T002 · V35正常5279 → 2383 cinema observer
SCAN / DI OBSERVER
T002 · V35 Period 2K scan
CAMERA BASELINE
T002 · Panasonic V-709 · As Shot 0.00 stop
ADDITIONAL SOURCENJARAW_S001_S001_T007

T002 carries the complete five-region temporal gate; T007 stresses water, grass and near-zero hard black; T031 stresses dark neutral texture, warm fungi and green surroundings.

2383 PROJECTION
T007 · V35正常5279 → 2383 cinema observer
SCAN / DI OBSERVER
T007 · V35 Period 2K scan
CAMERA BASELINE
T007 · Panasonic V-709 · As Shot 0.00 stop
Full changelog

Improvements

  • Replaced the 24-bit inverse-CDF candidate with direct Philox-u32 Bernoulli trials; maximum observed probability representation error is 2.269e-10
  • Asynchronous Metal finite-site submission exposes shared output memory and overlaps CPU expected-density filtering
  • All 45 record/speed/size identities per frame are decoded, deduplicated and persisted in provenance
  • Every result records source, algorithm, profile, LUT, bridge, command and stochastic-identity hashes
  • The V31 final chroma adapter reuses full-frame buffers; the V34 photographic model and creative boundary stay frozen
  • One native 5.7K 12-bit second for T002, T007 and T031, each with projection and period-scan masters

Errors / limitations

  • Calling the first 24-bit inverse-CDF candidate exact-distribution was too absolute; statistical agreement is not infinite mathematical precision
  • Same-process parallel observers reproduced a Numba workqueue SIGABRT, so V35 rejects observer_workers=2 before decoding
  • Shared-memory observer subprocesses were byte-identical but memory-bandwidth pressure worsened about 10.94 seconds of serial work to roughly 25 seconds
  • Single-Gaussian and full-residual convolution saved 0.65–0.9 s/frame but let ~5e-6 density reorder reach isolated 900–960/65535 projection-code differences through 2383 thresholds, so both are rejected
  • The Python Metal bridge still owns process-level device/queue state; it is a research tool and cannot become the OFX boundary unchanged

What we learned

  • Quality-first validation must include rare threshold events and tails, not only mean colour or PSNR
  • Direct uint32 Bernoulli is faster than floating inverse CDF in the 30-site native-frame microbenchmark
  • Four formed-negative seeds give layer standard-deviation ratios 0.999918/1.000264/0.999852; NPS difference stays below ordinary reference seed variation
  • Twenty-four frames across five regions show no systematic green, blue or magenta shift; scan and projection grain/temporal-energy departures stay below 0.3%
  • OFX v1 should be full-frame, serial per instance, supportsTiles=false, with host-owned Metal queues and per-instance resource rings
Evidence: [50] Numba Threading Layers — extra notes[54] OpenFX Image Effect Plug-in Rendering[55] Metal Performance Shaders tuning hints[56] Parallel Random Numbers: As Easy as 1, 2, 3 — Random123 / Philox[57] Metal Best Practices — Command Buffers
V34ARCHIVED CALIBRATION

Let developer adjacency happen once—and let each master encode once

V34 follows a full algorithm and render audit, not a new grade. Kodak's 5279 MTF was measured after recommended ECN-2 processing, and Kodak explains that MTF above 100% commonly comes from developer adjacency. The old path first added V21 deterministic DIR acutance and then applied a kernel already fitted to the complete processed-stock MTF, creating roughly 1–3.5% local overlap. V34 removes only that duplicate. Grain, interimage DIR, three speed populations, 48 μm RMS, colour, black and gamma are not tuned by taste. The accepted V31 colour boundary now runs in memory, so projection and scan each receive one ProRes generation.

2383 PROJECTION
T031 · V34正常5279 → 2383 cinema observer · 单世代
SCAN / DI OBSERVER
T031 · V34 5279 → Period 2K scan · 单世代
CAMERA BASELINE
T031 · official Panasonic V-709 · As Shot 0.00 stop见证
ADDITIONAL SOURCENJARAW_S001_S001_T002

T002 stresses endpoints; T007 stresses water, grass and fine edges; T031 stresses dark neutral surfaces, warm fungi and green surroundings.

2383 PROJECTION
T002 · V34正常5279 → 2383 cinema observer
SCAN / DI OBSERVER
T002 · V34 Period 2K scan
CAMERA BASELINE
T002 · Panasonic V-709 · As Shot 0.00 stop
ADDITIONAL SOURCENJARAW_S001_S001_T007

T002 stresses endpoints; T007 stresses water, grass and fine edges; T031 stresses dark neutral surfaces, warm fungi and green surroundings.

2383 PROJECTION
T007 · V34正常5279 → 2383 cinema observer
SCAN / DI OBSERVER
T007 · V34 Period 2K scan
CAMERA BASELINE
T007 · Panasonic V-709 · As Shot 0.00 stop
Full changelog

Improvements

  • Processed-stock 5279 MTF becomes the sole owner of deterministic adjacency acutance
  • The accepted V31 colour boundary runs after both complete observers and before delivery encoding
  • V30 intermediate-master decoding and the projection's second ProRes generation are removed
  • The idle extended-linear BT.2020→V-Gamut→Rec.709 round trip is fused as the product of the same matrices
  • Nine zero-contribution native Gaussian passes are skipped with identical output SHA-256
  • Partial-range manifests now report PCM trim/re-encode and regenerated timecode accurately
  • Three native 24-frame 5.7K 12-bit dual-master trials: T002, T007 and T031

Errors / limitations

  • Earlier research already identified MTF as a total processed response, but V21 deterministic adjacency was never cross-checked against the fitted MTF
  • V31 encoded both V30 observers, decoded them for the chroma boundary, then encoded projection again; repeatable did not mean lossless
  • Two workers reached about 28.85 s/frame on 48 GiB but created about 6.6 GiB of swap and were rejected by the quality/stability gate
  • The partial-range manifest said stream copied even when PCM was correctly decoded, trimmed and losslessly re-encoded
  • An uncalled pre-V21 DIR helper remained in source and created ambiguity for a future OFX port

What we learned

  • Official processed MTF and DIR chemistry are not two independent sharpness overlays that may simply be multiplied
  • Removing a ProRes generation improves both speed and image integrity more safely than aggressive process parallelism
  • The fused matrix keeps 99.9926% of native clipped 12-bit channel codes identical; the remainder are one-code rounding boundaries
  • Scan median luma and highlight endpoints stay effectively fixed; the model correction is concentrated around formerly double-counted local edges
  • The next major speedup belongs to a resident Metal/OpenFX graph with resource reuse and host-queue asynchrony—not more Python processes on 48 GiB
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[21] The Essential Reference Guide for Filmmakers[46] Adjust camera settings in Final Cut Pro[54] OpenFX Image Effect Plug-in Rendering[55] Metal Performance Shaders tuning hints
V33ARCHIVED CALIBRATION

Do not grade away real green light: lock the input, exposure and black boundaries first

V33 applies no global magenta compensation and changes no 5279, grain, DIR, 2383 or scan pixel. It separates an FCP Standard witness, a 0-stop As Shot camera view and the explicit +0.45-stop virtual film EI. Across three scenes it adds failing gates for display black, toe occupancy, robust contrast, effective gamma, native masters, partial-range audio/timecode and safe scheduling on the 48-GiB reference machine. Technical Neutral exists as a disabled boundary until a gray card or ColorChecker supplies repeatable evidence.

2383 PROJECTION
T031 · V31正常5279 → 2383 cinema observer · V33边界复验
SCAN / DI OBSERVER
T031 · V31 5279 → Period 2K scan · V33黑场复验
CAMERA BASELINE
T031 · official Panasonic V-709 · As Shot 0.00 stop见证
ADDITIONAL SOURCENJARAW_S001_S001_T002

T002 stresses sky and dark completion; T007 stresses water highlights, fine grass and an almost zero-hard-black scene.

2383 PROJECTION
T002 · 正常5279 → 2383 cinema observer · 图像冻结
SCAN / DI OBSERVER
T002 · Period 2K scan · 图像冻结
CAMERA BASELINE
T002 · Panasonic V-709 · As Shot 0.00 stop
ADDITIONAL SOURCENJARAW_S001_S001_T007

T002 stresses sky and dark completion; T007 stresses water highlights, fine grass and an almost zero-hard-black scene.

2383 PROJECTION
T007 · 正常5279 → 2383 cinema observer · 图像冻结
SCAN / DI OBSERVER
T007 · Period 2K scan · 图像冻结
CAMERA BASELINE
T007 · Panasonic V-709 · As Shot 0.00 stop
Full changelog

Improvements

  • Native 24-frame 0.00-stop As Shot V-709 witnesses for T002, T007 and T031
  • The +0.45-stop virtual film EI is explicitly labelled and no longer confused with untouched camera exposure
  • SHA-locked FCP Standard T031 source-frame-144 witness
  • Automated display-black, toe, p05–p95 contrast, 32-bin monotonic tone and effective log-luma-power gates
  • Technical Neutral boundary retained but disabled until neutral-card evidence exists
  • Sample-accurate PCM trim and absolute-source-frame timecode for partial deliveries
  • One native Archive-Exact worker on the 48-GiB reference machine; pixels and random seeds unchanged

Errors / limitations

  • The first three-way concurrent 5.7K float attempt created unacceptable system memory pressure; the panic report showed exhausted compressor segments and near-exhausted swap
  • Defining black as all three channels near zero missed pixels whose luma had reached black but retained tiny chroma; the final gate uses the same encoded-luma threshold as the FCP audit
  • Without a neutral target under the scene illumination, foliage bounce, As Shot white balance and local V-709 residual cannot be uniquely separated

What we learned

  • A neutral mathematical input remains neutral through BT.2020→V-Gamut→V-Log→official V-709; maximum channel spread is 0.000589
  • The finished scan's hard-black fraction is strongly scene dependent: nearly zero in T007 and roughly 1–2% in darker T002/T031
  • Both observers retain monotonic robust tone mappings across all three scenes; no hidden gamma reversal appears
  • Worker count changes scheduling only; avoiding swap on 48 GiB is both safer and more predictable than overcommitting memory
  • Technical Neutral is now a falsifiable gray-card question rather than a subjective global tint decision
Evidence: [44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[45] Apply built-in camera LUTs in Final Cut Pro[46] Adjust camera settings in Final Cut Pro[47] CVProResRawMetadata[49] Panasonic V-Log to V-709 3D LUT
V32ARCHIVED CALIBRATION

Freeze the accepted image; turn the next step from taste into repeatable measurement

V32 changes no V31 image-forming parameter. Two new GH7 ProRes RAW scenes independently test the 5279 emulsion, normal 2383 projection and period-2K scan without per-shot tuning. Native format, luminance preservation, highlight clipping, temporal texture, neutral-axis stability, OFX tile regions and cinema X′Y′Z′ delivery now become automated release gates. V32 is not a new look; it makes the current credible look portable, reproducible and falsifiable.

2383 PROJECTION
T007 · V31正常5279 → 2383 cinema observer · V32测量验证
SCAN / DI OBSERVER
T007 · V31 5279 → Period 2K scan · V32测量验证
CAMERA BASELINE
T007 · official Panasonic V-709camera baseline
ADDITIONAL SOURCENJARAW_S001_S001_T031

T031 uses neutral stone, warm mushrooms, moss and dark leaves to stress neutral-axis, chroma and fine-texture stability.

2383 PROJECTION
T031 · 正常5279 → 2383 cinema observer
SCAN / DI OBSERVER
T031 · 5279 → Period 2K scan
CAMERA BASELINE
T031 · official Panasonic V-709camera baseline
Full changelog

Improvements

  • Two new 24-frame native-resolution trials: T007 and T031
  • All V31 image formation frozen; no per-shot exposure, colour, grain or contrast adjustment
  • Frame-wise luma, p99 highlight, hard-clip, texture-power and near-neutral a/b gates
  • SMPTE ST 428-1 12-bit X′Y′Z′ lossless DCDM cinema test sequence
  • P3-D65/gamma-2.6 ProRes transport explicitly rejected because MOV, frame-header and player interpretations are ambiguous
  • Numerical OFX tile/ROI parity contract for future Resolve migration
  • Stage timing proves that stochastic emulsion and observers—not ProRes encoding—dominate runtime

Errors / limitations

  • The first P3 ProRes probe repeated V25's cross-player metadata ambiguity and was removed from delivery
  • A first DCDM round-trip compared against an already reduced web preview and therefore measured a second resampling; final QA decodes the native representative frame
  • The first T007 V31-boundary command passed a directory where the encoder expected a file; it failed before producing a valid frame and was rerun with the correct path

What we learned

  • Once an image is credible, freezing it can advance the model more than another subjective colour move
  • One parameter set surviving water highlights, fine grass, neutral stone and warm mushrooms is stronger evidence than one attractive still
  • DCDM can preserve the completed projection observer but cannot invent gamut already absent from its Rec.709 monitor result
  • Stochastic emulsion and spectral observation dominate native 5.7K runtime; encoding is a small fraction
  • An OFX halo must use full output width or proxy and tile sizes will silently change the crossover scale
Evidence: [26] ITU-R BT.709 — HDTV production and programme exchange[28] Digital Cinema System Specification[32] DCI Compliance Test Plan — calibrated screen luminance 48 cd/m²[44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[49] Panasonic V-Log to V-709 3D LUT[53] SMPTE ST 428-1:2019 — D-Cinema Distribution Master image characteristics
V31ARCHIVED CALIBRATION

Normal-process colour no longer acquires an accidental retained-silver signature

V31 answers the mild bleach-bypass character in V30 projection. Normal ECN-2 and ECP-2D remove the picture silver image. V30 had no residual-silver term, but it preserved C/L while applying a steeper 2383 lightness curve, so darker regions automatically lost absolute chroma and combined with strong luminance texture to read like retained silver. V31 keeps V30's film, grain, DIR, MTF, black, gamma, LAD and luminance. At the final delivery boundary, Period 2K supplies low-frequency dye colour while the 2383 projection retains its high-frequency opponent texture and exact per-pixel Rec.709 Y.

2383 PROJECTION
T002 · 正常5279 → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T002 · 5279 → Period 2K / Rec.709Blu-ray
CAMERA BASELINE
T002 · official Panasonic V-709camera baseline · no film pipeline
ADDITIONAL SOURCENJARAW_S001_S001_T020

T020 tests dark bark colour; T032 tests whether genuine rainy cyan-green atmosphere survives the correction.

2383 PROJECTION
T020 · 正常5279 → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T020 · V30逐像素不变的Period 2K scan
CAMERA BASELINE
T020 · official Panasonic V-709camera baseline
ADDITIONAL SOURCENJARAW_S001_S001_T032

T020 tests dark bark colour; T032 tests whether genuine rainy cyan-green atmosphere survives the correction.

2383 PROJECTION
T032 · 正常5279 → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T032 · V30逐像素不变的Period 2K scan
CAMERA BASELINE
T032 · official Panasonic V-709camera baseline
Full changelog

Improvements

  • Normal ECN-2/ECP-2D baseline explicitly excludes retained silver, skip bleach, ENR and bleach bypass
  • Period 2K supplies low-frequency OKLab a/b; 2383 retains its high-frequency opponent residual
  • Constant-hue Rec.709 gamut compression around exact projection luminance
  • V30 Kodak LAD and all 5279/texture parameters locked
  • Three native 24-frame 5.7K 12-bit dual-master trials
  • V30 scan masters verified by file SHA-256 regression

Errors / limitations

  • V30 said tone and colour were separated but preserved C/L, so replacing L still changed absolute colour
  • A cached-LUT placement was bypassed by the legacy hybrid branch and rejected by regression
  • A deterministic mean correction was pulled back by the grain mean-colour stage and also rejected
  • A global saturation lift or black adjustment would be a grade, not a baseline repair

What we learned

  • A viewing transform can create a bleach-bypass impression without a chemical silver term
  • Chroma and saturation are not interchangeable when lightness changes
  • Normal bleaching is a falsifiable process boundary, not a taste preference
  • Any future bypass toggle must explicitly model residual silver density
  • The accepted organic texture does not need to be weakened to correct colour
Evidence: [3] Processing KODAK Motion Picture Films, Module 7: ECN-2[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[48] LAD for KODAK VISION Color Print Film — H-61B[51] Process ECP-2D Specifications, H-24 Module 9A[52] Motion Picture Film Processing Information — skip bleach / ENR
V30PREVIOUS BASELINE

Official LAD aims remove the 2383 cast; three camera baselines isolate what film changes

V30 traces the residual blue-magenta projection veil to an equal-RGB LAD shortcut and to unsupported hue authority assigned to a vendor D60 LUT and digitized dye curves. Kodak's published 2383 aims—1.09/1.06/1.03 D—replace the shortcut. T002, T020 and T032 each receive one native-resolution second plus an official Panasonic V-709 camera baseline, so source colour, 5279, 2383 projection and period scanning can be compared directly.

2383 PROJECTION
T002 · 5279 negative → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T002 · 5279 negative → Period 2K / Rec.709Blu-ray
CAMERA BASELINE
T002 · official Panasonic V-709camera baseline · no film pipeline
ADDITIONAL SOURCENJARAW_S001_S001_T020

T020 tests neutral bark, fungi and dark texture; T032 tests genuine cyan-green rain haze and low-contrast detail against the same official camera baseline.

2383 PROJECTION
T020 · 5279 → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T020 · 5279 → Period 2K / Rec.709Blu-ray
CAMERA BASELINE
T020 · official Panasonic V-709camera baseline
ADDITIONAL SOURCENJARAW_S001_S001_T032

T020 tests neutral bark, fungi and dark texture; T032 tests genuine cyan-green rain haze and low-contrast detail against the same official camera baseline.

2383 PROJECTION
T032 · 5279 → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T032 · 5279 → Period 2K / Rec.709Blu-ray
CAMERA BASELINE
T032 · official Panasonic V-709camera baseline
Full changelog

Improvements

  • Kodak H-61B 2383 LAD aims: 1.09/1.06/1.03 D
  • Removed vendor D60 LUT control over Kodak physical hue
  • Digitized net-dye curves no longer force final hue or saturation
  • Three 24-frame native 5.7K 12-bit dual-master trials
  • Official Panasonic V-709 camera baseline for every source
  • Matched frame-12 sRGB stills and 24-frame hover loops
  • Safe sequential observers with pixel-identical output
  • Neutral-chroma, observer-hue, endpoint and format gates

Errors / limitations

  • V29's equal 1.00/1.00/1.00 LAD ignored the official channel aims and introduced a blue-magenta shift
  • A third-party D60 LUT is not measured Kodak 2383 spectral evidence
  • Digitization uncertainty cannot justify a high-weight global hue control
  • An initial camera baseline double-applied legal-range scaling; it was rejected before release
  • Two Python observer threads could enter Numba's workqueue concurrently and abort

What we learned

  • The blue-purple filter dyes in unprocessed 2383 wash out in processing and must not become a uniform projected veil
  • A warmer scan does not prove projection should be absolutely blue; each observer needs its own neutral-axis test
  • Official unequal LAD aims explain the correction better than a subjective blue reduction
  • T032 genuinely contains cyan-green rainy haze; the camera baseline separates scene colour from observer error
  • Sequential observation is pixel-identical here and avoids contention
  • A useful original comparison needs a necessary camera display transform; flat Log is not the scene appearance
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[3] Processing KODAK Motion Picture Films, Module 7: ECN-2[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[48] LAD for KODAK VISION Color Print Film — H-61B[49] Panasonic V-Log to V-709 3D LUT[50] Numba Threading Layers — extra notes
V29PREVIOUS BASELINE

From one-second trials to complete motion validation

V29 turns the remaining work from subjective film flavour into falsifiable validation. Public 5279 evidence constrains neutral H-D, MTF and 48 µm RMS, but does not publish a stock-specific NPS, DIR matrix, layer recipe or Spirit spectrum. Those unknown parameters remain V28. The complete 165-frame T002 source receives a new finite-site emulsion at every absolute source frame and is delivered with both observers, source 24-bit four-channel audio and timecode.

2383 PROJECTION
T002 · 完整5279 negative → 2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T002 · 完整5279 negative → Period 2K / Rec.709Blu-ray
Full changelog

Improvements

  • Complete 165-frame T002 motion stress test
  • Absolute source-frame seeds across exact parallel ranges
  • Shared per-frame negative for the 2383 and Period 2K observers
  • Full-motion black, highlight, high-frequency, temporal and segment-boundary validation
  • Source 24-bit/48-kHz four-channel PCM and 12:04:05:23 timecode retained
  • V28 H-D, MTF, 48 µm RMS, colour, black, gamma and observers locked

Errors / limitations

  • A 48 µm single-aperture RMS value is not a complete grain NPS
  • A Kodak patent diffusion example is not a 5279-specific DIR matrix
  • Public Spirit hardware information is not its proprietary spectral calibration
  • Scene-by-scene tuning without physical comparison would mix creative grading into the baseline

What we learned

  • The code-completable portion of the last 15–20% is validation and delivery; stock identification now requires measurement
  • Successive film frames form new emulsion mosaics rather than translating or looping one texture
  • Parallel rendering can preserve absolute-frame randomness and exact pixels
  • Full motion exposes flicker, grain swimming, black bias and highlight discontinuity better than a still
  • Audio, timecode, colour signalling and frame count are part of an industry-standard master
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[3] Processing KODAK Motion Picture Films, Module 7: ECN-2[7] US 5,314,793 — multilayer speed and granularity architecture[8] Spirit DataCine / Spirit HD technical data[21] The Essential Reference Guide for Filmmakers[22] Noise Power Spectra of Photographic Dye Images[23] US 4,536,472 — dye-cloud diffusion, Wiener spectrum and low-frequency mottle[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[34] US 6,190,847 — Kodak DIR diffusion-factor assay[35] EP 1,016,902 — green-only ECN-2 exposure as an interimage exclusion control[36] Process ECN-2, H-24 processing modules[37] ISO 5-3 density spectral conditions and aperture dependence
V28PREVIOUS BASELINE

Correcting the boundary between ProRes RAW conversion and Panasonic's Camera LUT

V28 traces the remaining green veil to stage order before the negative: AVFoundation already supplies extended-linear BT.2020/D65, but V27 interpreted that RGB buffer again as Panasonic RAW Gamut. V28 uses a linear BT.2020-to-XYZ-to-V-Gamut primary conversion with no second white balance or creative grade; every V27 film, grain, tone and observer parameter stays locked.

2383 PROJECTION
T020 · 修正输入契约后的2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T020 · 修正输入契约后的Period 2K / Rec.709Blu-ray
ADDITIONAL SOURCENJARAW_S001_S001_T032

The genuinely cyan-green rainy forest separates source colour from the extra veil produced by the misplaced Camera LUT.

2383 PROJECTION
T032 · 修正输入契约后的2383 cinema observerRec.709监看
SCAN / DI OBSERVER
T032 · 修正输入契约后的Period 2K / Rec.709Blu-ray
Full changelog

Improvements

  • Verified the decoded Core Video buffer as extended-linear BT.2020/D65
  • Removed the second RAW-Gamut Camera-LUT interpretation
  • Linear BT.2020 → XYZ D65 → Panasonic V-Gamut primary conversion
  • Retained AVFoundation standard conversion and as-shot metadata with no second white balance
  • Recomputed both observers from the same corrected negative
  • Added a hash-validated analytical 2383 cache and bit-identical sampler

Errors / limitations

  • V27 treated an already converted BT.2020 buffer as Panasonic RAW Gamut
  • A misplaced nonlinear 3D camera separation created scene-dependent green and blue errors
  • Scan gray-axis calibration could not remove an error introduced before film exposure
  • A global magenta, saturation or gamma adjustment would be a grade and would damage real foliage green

What we learned

  • Camera-LUT validity depends on the input stage, not merely the camera model
  • The rainy T032 source is genuinely cyan-green; the goal is to remove only the extra fluorescent veil
  • Near-neutral green/red ratios improve while highlight anchors and clipping remain stable
  • Uniform synthetic gray stays neutral and the V27 Spirit neutral calibration remains necessary
  • The accelerated final 12-bit decoded pixels are bit-identical to the reference implementation
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[8] Spirit DataCine / Spirit HD technical data[26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[44] Panasonic Apple ProRes RAW Output LUT — RAW Gamut to V-Log/V-Gamut[45] Apply built-in camera LUTs in Final Cut Pro[46] Adjust camera settings in Final Cut Pro[47] CVProResRawMetadata
V27RESEARCH BASELINE

Separating a scanner's green veil from film colour

V27 identifies the pale green veil in the Blu-ray branch as a level-dependent scanner neutral-axis error, not 5279 colour or grain. A 2049-level neutral exposure scale corrects RGB balance while preserving each pixel's Rec.709 luminance, black, contrast and gamma.

2383 PROJECTION
T020 · 2383 cinema observer的Rec.709监看(same as V26)
SCAN / DI OBSERVER
T020 · neutral-scale-constrained Period 2K / Rec.709Blu-ray
ADDITIONAL SOURCENJARAW_S001_S001_T032

T020 tree bark and fungi separate neutral dark texture from real foliage green; T032 tests the correction in rainy cyan-green detail.

2383 PROJECTION
T032 · 2383 cinema observer的Rec.709监看(same as V26)
SCAN / DI OBSERVER
T032 · neutral-scale-constrained Period 2K / Rec.709Blu-ray
Full changelog

Improvements

  • 2049-level scanner neutral-scale calibration
  • Level-dependent RGB balance after the completed scan observer
  • Exact per-pixel Rec.709 luminance preservation
  • Byte-identical V26 projection master and locked V26 emulsion
  • Review of the certified April 2003 5279 provisional: identifier provenance confirmed, numerical parameters still absent
  • Complete Chinese / English website switch with remembered preference

Errors / limitations

  • V26 balanced only 18% gray and one dense-negative anchor
  • An unsupported 18% residual scanner correction left a green hump in shadows and lower mids
  • A global magenta offset cannot fix an error that changes direction over the gray scale
  • 2K aperture softness and the RGB-axis error appeared together as one green haze

What we learned

  • Maximum neutral-channel residual falls from 0.01820 to 0.00236
  • Maximum green-opponent residual falls from 0.02172 to 0.00242
  • Maximum per-pixel luminance drift is below 1.8×10⁻⁷
  • The April 2003 provisional names 5279, May 2003 JVT-H022 switches the same identifier to 5218, and the later patent returns to 5279: document-branch drift, not a calibration payload
  • The latest hourly audits add evidence boundaries, not invented 5279 parameters
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[8] Spirit DataCine / Spirit HD technical data[11] US 5,500,316 — colour negative contrast adjusted for electronic scanning[26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[34] US 6,190,847 — Kodak DIR diffusion-factor assay[35] EP 1,016,902 — green-only ECN-2 exposure as an interimage exclusion control[36] Process ECN-2, H-24 processing modules[38] US 7,899,113 — film-grain simulation identifiers and model database[39] JVT-H022 — SEI message for film grain encoding[40] JVT-I013r2 — film grain encoding syntax and results[41] US provisional 60/462,389 — A Method for Simulating Film Grain on Encoded Video Sequences
V26EXPOSURE-CONDITIONED GRAIN

Exposure selects a grain spectrum, not only grain loudness

Colour, black, contrast, gamma and Rec.709 delivery stay locked while fast, medium and slow populations receive different five-class cloud distributions.

2383 PROJECTION
T020 · 2383 cinema observer的Rec.709监看
SCAN / DI OBSERVER
T020 · Period 2K / Rec.709Blu-ray
ADDITIONAL SOURCENJARAW_S001_S001_T032

T032 tests exposure-conditioned grain in rain, low contrast and cyan-green detail.

2383 PROJECTION
T032 · 2383 cinema observer的Rec.709监看
SCAN / DI OBSERVER
T032 · Period 2K / Rec.709Blu-ray
Full changelog

Improvements

  • Separate cloud-size weights for fast/mid/slow layers
  • A wider large-cloud tail in shadows and a finer slow-layer highlight tail
  • Independent finite-site development every frame
  • NPS, activation, mean-drift and temporal-correlation diagnostics

Errors / limitations

  • The public 48 µm RMS does not uniquely identify an NPS
  • The per-sublayer distributions remain bounded hypotheses
  • The period scanner used only two gray anchors and retained an invented residual

What we learned

  • Fast sensitive crystals dominate shadow grain power
  • Slow populations dominate highlight texture
  • The V26 colour branch remained numerically V25
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[7] US 5,314,793 — multilayer speed and granularity architecture[21] The Essential Reference Guide for Filmmakers[23] US 4,536,472 — dye-cloud diffusion, Wiener spectrum and low-frequency mottle[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58[33] US 6,815,153 — improved speed and granularity in high-speed colour negative film
V25OUTPUT-STANDARD FIX

Viewing conditions and interchange masters are not the same thing

An initial output error baked inverse display EOTFs into files. The corrected release restores both monitor masters to Rec.709 OETF, full 1-1-1 metadata and consistent sRGB web conversion.

2383 PROJECTION
T020 · 2383 cinema observer的Rec.709监看
SCAN / DI OBSERVER
T020 · Period 2K / Rec.709Blu-ray
ADDITIONAL SOURCENJARAW_S001_S001_T032

Dark posts and low-contrast rainy texture test black, gamma and metadata consistency.

2383 PROJECTION
T032 · 2383 cinema observer的Rec.709监看
SCAN / DI OBSERVER
T032 · Period 2K / Rec.709Blu-ray
Full changelog

Improvements

  • Corrected the accidental brightness lift
  • Rec.709 OETF/1-1-1 for both monitor masters
  • BT.1886 kept only as a reference-display EOTF
  • Exact seeded parallel acceleration

Errors / limitations

  • The first export mixed BT.1886, P3 gamma and Rec.709 tags
  • Midtones and shadows became 20–28% too bright
  • Public evidence still could not identify every interlayer parameter

What we learned

  • The lift was measurable, not subjective
  • A reference-display EOTF is not a source-file transfer function
  • Negative research results prevent unjustified colour changes
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[26] ITU-R BT.709 — HDTV production and programme exchange[27] ITU-R BT.1886 — HDTV studio display EOTF[28] Digital Cinema System Specification[29] CSS Color Module Level 4 — sRGB and web colour[30] ACES 2 Output Transform parameters and display encodings[31] ITU-R BT.1886-0 — Reference electro-optical transfer function[32] DCI Compliance Test Plan — calibrated screen luminance 48 cd/m²
V2435 MM SPECTRUM

35 mm grain spectrum and luma/chroma separation

Cloud power shifts toward finer scales while the two observers integrate chroma grain separately. Mean colour, tone, MTF and sensitometry remain V23.

2383 PROJECTION
T020 · 5279 → 2383 xenon projection
SCAN / DI OBSERVER
T020 · Period 2K / CineonBlu-ray
ADDITIONAL SOURCENJARAW_S001_S001_T032

The rainy T032 scene checks that reduced chroma texture does not return as cyan-green CCD noise.

2383 PROJECTION
T032 · 5279 → 2383 xenon projection
SCAN / DI OBSERVER
T032 · Period 2K / CineonBlu-ray
Full changelog

Improvements

  • Finer cloud-size distribution
  • Reduced large-cloud tail and correlation scale
  • Observer-specific chroma-grain integration
  • One-second live web previews

Errors / limitations

  • The selected spectrum remains bounded, not uniquely measured
  • 48 µm RMS alone cannot determine low-frequency clustering
  • Creative grading stayed deliberately outside the model

What we learned

  • Gauge appearance depends on magnification and low-frequency power as well as RMS
  • Signed grain deltas can be managed without changing mean colour
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[7] US 5,314,793 — multilayer speed and granularity architecture[8] Spirit DataCine / Spirit HD technical data[21] The Essential Reference Guide for Filmmakers[22] Noise Power Spectra of Photographic Dye Images[23] US 4,536,472 — dye-cloud diffusion, Wiener spectrum and low-frequency mottle[25] Print Grain Index — An Assessment of Print Graininess from Color Negative Films, E-58
V23CROSS-SCENE HOLDOUT

Continuous dye-cloud populations and colour generalization

Five cloud classes approximate a continuous distribution, new T020/T032 RAW scenes test colour holdout, and a dense analytical print lattice accelerates the observer.

2383 PROJECTION
T020 · 5279 → 2383 xenon projection
SCAN / DI OBSERVER
T020 · Period 2K / CineonBlu-ray
ADDITIONAL SOURCENJARAW_S001_S001_T032

Rainy cyan-green foliage and low-contrast texture test whether colour and grain generalize beyond the first scene.

2383 PROJECTION
T032 · 5279 → 2383 xenon projection
SCAN / DI OBSERVER
T032 · Period 2K / CineonBlu-ray
Full changelog

Improvements

  • Five-class cloud quadrature
  • Two new one-second native RAW trials
  • 193³ exact analytical projection cache

Errors / limitations

  • Texture still evoked early CCD or 16 mm
  • Chroma grain remained too prominent
  • The public RMS curve did not identify a full spectrum

What we learned

  • A colour chain can generalize while texture scale remains wrong
  • Most compute time lies in negative formation, not encoding
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[7] US 5,314,793 — multilayer speed and granularity architecture[21] The Essential Reference Guide for Filmmakers[22] Noise Power Spectra of Photographic Dye Images[23] US 4,536,472 — dye-cloud diffusion, Wiener spectrum and low-frequency mottle[24] EP 0,905,561 — speed-layer coupler coverage and spectrally differentiated dye records
V22ANALYTICAL PRINT DYES

Analytical dyes, interimage coupling and relative-white monitoring

Status-A principal curves are inverted to dye amounts; print interimage operates around LAD and D60 contributes relative chroma only.

2383 PROJECTION
V22 · 5279 → 2383 xenon projection
SCAN / DI OBSERVER
V22 · Period 2K / CineonBlu-ray
Full changelog

Improvements

  • Analytical 2383 dye inversion
  • LAD-centered interimage coupling
  • Neutral-subtracted D60 chroma

Errors / limitations

  • Exact spectral print rendering was expensive
  • One source scene could not prove generalization
  • Grain morphology was still discrete

What we learned

  • Integral density is not dye amount
  • White-point calibration must not tint neutral pixels
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[5] LAD — Laboratory Aim Density, KODAK H-61[16] Digital Color Management for Motion Picture Film[17] US 2002/0118211 — analytical dye density and interimage measurement[18] US 8,654,192 — LAD-anchored print-exposure matrix[19] Common LUT Format implementation guide[20] ISO 5-3:2009 — Photography and graphic technology: Spectral conditions
V21DEVELOPMENT REBUILD

Development, grain and observers work in their own domains

DIR transport moves into nine sub-emulsions; record-specific cloud morphology and distinct Status-M, scanner and printer observers establish the modern structure.

2383 PROJECTION
V21 · 2383 xenon projection显示适配
SCAN / DI OBSERVER
V21 · Period 2K / CineonBlu-ray
Full changelog

Improvements

  • Population-domain DIR reaction–diffusion
  • Record-specific fast/mid/slow morphology
  • Separated measurement, scan and print observers

Errors / limitations

  • 2383 Status-A densities were still treated as dye amounts
  • D60 adaptation could contaminate neutrals
  • Print computation was too slow

What we learned

  • Uniform sensitometry already includes uniform chemistry, so local DIR must vanish on a flat field
  • Measurement observers are not viewing observers
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[3] Processing KODAK Motion Picture Films, Module 7: ECN-2[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[5] LAD — Laboratory Aim Density, KODAK H-61[6] US 5,298,376 — DIR inhibitor transport and colour saturation[7] US 5,314,793 — multilayer speed and granularity architecture[8] Spirit DataCine / Spirit HD technical data[11] US 5,500,316 — colour negative contrast adjusted for electronic scanning[12] US 5,705,327 — nonlinear curve shape for telecine transfer
V20PREVIOUS BASELINE

Exposure-dependent sublayer dye contribution

Fast, medium and slow layers alter record mixing through their marginal activation; thin negatives separate less completely than slow-layer highlights.

2383 PROJECTION
V20 · 2383 projection
SCAN / DI OBSERVER
V20 · 2K DI / Blu-ray
Full changelog

Improvements

  • Marginal-activation record mixing
  • First coupling of grain and colour formation
  • More natural scanner-shoulder release

Errors / limitations

  • DIR remained a post-density approximation
  • Status-M and scanner observers were conflated
  • Sublayer sizes were not record-specific

What we learned

  • Period Kodak patents pre-compensated some films for telecine
  • Final colour-negative images are dye clouds after silver removal
  • The next step required reordering the algorithm
Evidence: [2] Exploring the Color Image[3] Processing KODAK Motion Picture Films, Module 7: ECN-2[6] US 5,298,376 — DIR inhibitor transport and colour saturation[11] US 5,500,316 — colour negative contrast adjusted for electronic scanning[12] US 5,705,327 — nonlinear curve shape for telecine transfer
V19ORGANIC GRAIN

Polydisperse dye clouds and frame-by-frame boiling

Each speed population is split into multiple sizes with subpixel phases and sparse larger clouds, reducing the digital chroma-noise impression.

2383 PROJECTION
organic grain · projection
SCAN / DI OBSERVER
organic grain · Blu-ray
Full changelog

Improvements

  • Polydisperse sizes
  • Subpixel phases
  • Small-cloud majority with a sparse large tail

Errors / limitations

  • Sublayer colour contribution was static
  • DIR still followed layer merging
  • Colour records shared representative ECDs

What we learned

  • Organic structure comes from irregular statistics, not larger noise
  • Fresh sampling must preserve mean density
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[7] US 5,314,793 — multilayer speed and granularity architecture
V18DISPLAY ADAPTATION

Physical 2383 projection versus its monitor presentation

A separate Rec.709 viewing result is derived from the physical print while GH7 sensor noise is separated from virtual-emulsion granularity.

2383 PROJECTION
2383显示器观看适配
SCAN / DI OBSERVERRETAINED / SHARED
Blu-ray分支retainedV17
Full changelog

Improvements

  • Projection-monitor branch
  • Multi-anchor neutral adaptation
  • Sensor-noise separation

Errors / limitations

  • Slight blue and saturated-hue shifts remained
  • Coarse populations still looked regular

What we learned

  • Perceived blue can come from nonlinear contrast/chroma ratios
  • Camera noise must not be preserved as film grain
Evidence: [7] US 5,314,793 — multilayer speed and granularity architecture[8] Spirit DataCine / Spirit HD technical data
V17SCAN CORRECTION

Integrating a Spirit 2K aperture in transmission

The scanner averages light through the negative at 2K before returning to density and Cineon, fixing grain-lifted blacks.

2383 PROJECTIONRETAINED / SHARED
Projection branch retained from V16
SCAN / DI OBSERVER
V17透射域scan
Full changelog

Improvements

  • 2K transmission-domain aperture
  • Correct display-boundary ordering
  • Controlled high-frequency chroma grain

Errors / limitations

  • Monitor projection still needed adaptation
  • The scan observer remained approximate
  • Cloud morphology remained regular

What we learned

  • A scanner measures transmitted light, not display-clipped RGB
  • Correct operation order can repair floating blacks without a grade
Evidence: [8] Spirit DataCine / Spirit HD technical data[10] Charlie’s Angels: Full Throttle — technical specifications
V16VIEWING CALIBRATION

Projection black, scan black and stable hue

Flare-free projection, typical cinema projection and Blu-ray finishing are separated; dark gamut compression becomes hue-stable.

2383 PROJECTION
典型16 ft-Lprojection
SCAN / DI OBSERVER
2K DI / Blu-ray完成
Full changelog

Improvements

  • Typical 1% projection flare
  • OKLab constant-hue compression
  • Blu-ray lower-scale gamma and black anchor

Errors / limitations

  • Display black created one-sided grain bias
  • Monitor projection still felt heavy
  • Scan chroma grain exceeded film references

What we learned

  • Projection and Blu-ray blacks come from different chains
  • A finished movie can constrain the observer, not measure bare 5279
Evidence: [8] Spirit DataCine / Spirit HD technical data[9] Digital Intermediates, August 2003[10] Charlie’s Angels: Full Throttle — technical specifications
V15COMPLETE PRINT PATH

Orange base, printer lights, 2383 and gamut mapping

The full spectral path retains the orange base, balances it with printer lights and forms a 2383 positive before viewing.

2383 PROJECTION
完整光谱印片链
SCAN / DI OBSERVER
scan链校准
Full changelog

Improvements

  • Full D-min transmission
  • Printer-light neutral balance
  • H-61/TAF-style separation calibration

Errors / limitations

  • Projection remained too deep and blue
  • Dark red and green hues rotated
  • Scan black/grain ordering was wrong

What we learned

  • Some print richness is physical 2383 slope; some was viewing adaptation error
  • Gamut compression should preserve hue
Evidence: [3] Processing KODAK Motion Picture Films, Module 7: ECN-2[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[5] LAD — Laboratory Aim Density, KODAK H-61
V14THREE SPEED LAYERS

Finite fast, medium and slow sensitive sites

Each colour record receives three finite populations; binomial development produces exposure-dependent variance and is normalized to Kodak's 48 µm RMS curves.

2383 PROJECTION
multilayer grain · projection
SCAN / DI OBSERVER
multilayer grain · scan
Full changelog

Improvements

  • Nine finite populations
  • p(1−p) variance
  • 48 µm aperture calibration

Errors / limitations

  • R/G/B shared representative grain sizes
  • Each population still had one circular size
  • Colour covariance felt synthetic

What we learned

  • Variance falls again as a population becomes fully developed
  • Coarse shadows and fine highlights arise from population crossover
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[7] US 5,314,793 — multilayer speed and granularity architecture
V13PROJECTION SOURCE

From a black-body white to a cinema xenon spectrum

A structured xenon SPD, CIE observer and restrained projection scatter replace a generic illuminant.

2383 PROJECTION
Corrected 2383 xenon projection
SCAN / DI OBSERVER
Spirit 2K scan
Full changelog

Improvements

  • Xenon spectral power distribution
  • CIE XYZ observation
  • Conservative Callier/projection scatter

Errors / limitations

  • The first print result was severely magenta
  • A physical projection looked too dense on a monitor

What we learned

  • Physical projection and monitor presentation are distinct targets
  • 2383's steep curves amplify small hue errors
Evidence: [2] Exploring the Color Image[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data[7] US 5,314,793 — multilayer speed and granularity architecture
V12COLOURED MASK

Preserving negative net-density lobes from masking couplers

The 21-point 5279 spectral curves are treated as signed net-density changes, giving the coloured mask its correct direction.

2383 PROJECTION
After mask correction · projection
SCAN / DI OBSERVER
After mask correction · scan
Full changelog

Improvements

  • 21-wavelength spectral LUT
  • Signed negative-density lobes
  • First local DIR adjacency

Errors / limitations

  • The scan observer remained narrow-band
  • DIR was applied after total density
  • Colour separation lacked sublayer dependence

What we learned

  • Negative values represent masking-coupler consumption
  • An average-flat correction still leaves stock-specific spectral residuals
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[3] Processing KODAK Motion Picture Films, Module 7: ECN-2[6] US 5,298,376 — DIR inhibitor transport and colour saturation
V11GRAY-SCALE CALIBRATION

Calibrate the full gray scale, not only 18% gray

LAD and a multi-step neutral scale establish the first Cineon/Spirit-style scanning baseline.

2383 PROJECTION
2383 xenon projection
SCAN / DI OBSERVER
Cineon / Spirit 2K scan
Full changelog

Improvements

  • Multi-point neutral calibration
  • Cineon code 95/445 anchors
  • Both branches delivered as Rec.709

Errors / limitations

  • Local colour still used a coarse matrix
  • Scanner spectra were approximated as Status-M
  • Blu-ray black finishing was incomplete

What we learned

  • A neutral mid-gray does not guarantee neutral shadows or highlights
  • Black is a viewing-chain decision, not a reason to alter negative D-min
Evidence: [5] LAD — Laboratory Aim Density, KODAK H-61[8] Spirit DataCine / Spirit HD technical data
V10FIRST 2383

Kodak 2383 becomes a second real material

2383 sensitometry, dyes and printer lights replace the idea of applying one display LUT to the negative.

2383 PROJECTION
5279 → 2383 初版
SCAN / DI OBSERVERRETAINED / SHARED
Scan branch retained from V9
Full changelog

Improvements

  • Second H-D curve set
  • LAD density aim
  • 2383 dyes and print grain

Errors / limitations

  • Net negative dye curves were misread as positive-only dyes, causing severe magenta
  • Orange base was removed or compensated twice
  • The steep print curve amplified small spectral errors

What we learned

  • 5279 D-min must remain in the optical printing path
  • Scan de-masking and printer-light compensation are different operations
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[3] Processing KODAK Motion Picture Films, Module 7: ECN-2[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data
V9TWO-BRANCH ORIGIN

5279 curves, spectral dyes and two viewing results

The three Status-M curves are sampled separately and scanning is split from the print interpretation for the first time.

2383 PROJECTION
Early print interpretation
SCAN / DI OBSERVER
5279Negative scan interpretation
Full changelog

Improvements

  • Independent H-D curves
  • Early net spectral-density model
  • Separate scan and print branches

Errors / limitations

  • Print dyes still used an empirical positive matrix
  • Orange-mask negative lobes were incomplete
  • The two branches lacked a common colour reference

What we learned

  • The data-sheet dye curves are D-min-subtracted net changes
  • Masking couplers must be counted exactly once
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[4] KODAK VISION Color Print Film 2383 / 3383 Technical Data
V835 MM REFINEMENT

Pulling a raw 16 mm texture back toward fine 35 mm

Smaller dye clouds, more effective sites and reduced shadow excursions move the spatial scale toward 35 mm 5279.

2383 PROJECTIONRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
SCAN / DI OBSERVERRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
Full changelog

Improvements

  • Cloud radii reduced to roughly 0.5–1.3 native pixels
  • More effective dye-cloud sites
  • Less shadow colour separation

Errors / limitations

  • Mean response still used a shared curve
  • Scan and projection observers were not physical
  • The colour mask was not spectral

What we learned

  • Fine grain can still boil organically when sites are smaller and denser
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image
V7FILM OPTICS

Silver halide, scatter and anti-halation in one chain

Layer scatter, anti-halation and local edge response turn uniform noise into an exposure-dependent density structure.

2383 PROJECTIONRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
SCAN / DI OBSERVERRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
Full changelog

Improvements

  • Exposure-dependent grain
  • Controlled halation
  • Coupled grain and edge response

Errors / limitations

  • Morphology still resembled pushed 16 mm
  • Speed layers were only broad segments
  • Black and viewing conditions were conflated

What we learned

  • Shadow/highlight grain changes come from different speed populations entering development
  • Resolution matching matters as much as grain amplitude
Evidence: [2] Exploring the Color Image[3] Processing KODAK Motion Picture Films, Module 7: ECN-2
V6RAW HIGHLIGHTS

Recover sensor information before shaping a film shoulder

The GH7 ProRes RAW linear interpretation was corrected so camera clipping and negative shoulder compression became separate operations.

2383 PROJECTIONRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
SCAN / DI OBSERVERRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
Full changelog

Improvements

  • Linear RAW path
  • Per-record highlight shoulders
  • Bounded long-range red scatter

Errors / limitations

  • Grain remained too strong
  • Halation was initially too generic
  • Panasonic colour handling was incomplete

What we learned

  • A negative shoulder is not a soft clip
  • 5279 rem-jet suppresses exaggerated rear-surface halation
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image[3] Processing KODAK Motion Picture Films, Module 7: ECN-2
V55279 CALIBRATION

From generic film look to one named stock

Kodak VISION 500T 5279 documentation begins to constrain scale, record response and grain instead of treating every fast colour negative alike.

2383 PROJECTIONRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
SCAN / DI OBSERVERRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
Full changelog

Improvements

  • 5279 curve and sensitivity targets
  • 35 mm image scale
  • Separate R/G/B records

Errors / limitations

  • RAW/display clipping still affected highlights
  • Shadow texture resembled digital chroma noise
  • Speed layers were not explicit

What we learned

  • The published blue-record granularity is higher than green and red
  • Stock character is a joint property of curves, dyes and granularity
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t
V4STRUCTURE PROTOTYPE

Sharpness and grain become one medium

The first version to stop overlaying noise on a finished image. Detail and randomness are formed together in a virtual emulsion-density field.

2383 PROJECTIONRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
SCAN / DI OBSERVERRETAINED / SHARED
The viewing branches had not yet been separated; both retain the same experiment image
Full changelog

Improvements

  • Exposure-domain emulsion prototype
  • Channel-dependent MTF
  • Independent structure per frame

Errors / limitations

  • Cloud radii and observer blur were too large
  • Closer to 8/16 mm than 35 mm 5279
  • Colour still relied on empirical matrices

What we learned

  • Coarse grain must be paired with lower high-frequency resolution
  • Grain belongs to image formation, not a post layer
Evidence: [1] KODAK VISION 500T 5279 / 7279 Technical Data, H-1-5279t[2] Exploring the Color Image