Evidence A
Exported bitmap
- Full final-size one-bit PNG
- SHA-256 file identity
- Matched true 1:1 crop
- No claim about material response
Open protocol · version 1.0
How we will compare six dithers.
This published protocol holds the source, crop, tone, final size, machine process, and evidence capture constant so a dither comparison can be inspected rather than trusted on faith.
Protocol published; comparative physical results not yet published.
No benchmark runs, scores, winners, or simulated results appear on this page. The worksheet is an empty recording template. We will add comparative results only after real test pieces are engraved, physically photographed, blindly scored, and published with their complete records.
The comparison set
One controlled change.
Each comparison block uses one locked source-photo fixture and one locked physical process. It produces exactly six bitmap variants—one per EngravePrep algorithm. A new photo, material batch, machine, focus, interval, speed, or power setting starts a new block rather than being mixed into the old one.
- 01Floyd-Steinberg
- 02Jarvis-Judice-Ninke
- 03Stucki
- 04Atkinson
- 05Bayer ordered
- 06Threshold
Protocol
From source file to scored object.
These steps are completed in order for every comparison block. Deviations are recorded before analysis.
Lock the source
Use the same lossless source photograph for all six variants. Record its file name, SHA-256 hash, pixel dimensions, and subject description before processing.
Lock crop and tone
Freeze crop coordinates, aspect ratio, polarity, and every tone setting. Only the dither algorithm may change inside a comparison set.
Lock final size
Export every variant at the same physical width, height, DPI, and resulting pixel dimensions. Save each final bitmap and its SHA-256 hash.
Inspect at true 1:1
Publish the same detail region from every exported bitmap at one image pixel per displayed pixel—never enlarged, smoothed, or resampled.
Engrave a controlled block
Keep machine, material batch, focus method, line interval, speed, power, passes, and air-assist state fixed across the randomized six-sample block.
Photograph the objects
Capture actual physical engravings under one locked camera and lighting setup. Record the file and capture date; do not substitute renders, mockups, or simulations.
Score blind
Hide algorithm names behind randomized sample codes. Judges score detail, tonal separation, artifacts, and overall quality before the code key is opened.
Publish complete records
Release every valid row, exclusions with reasons, bitmap crops, physical photos, scoring data, and protocol deviations—not only the most flattering samples.
Fixed image fixture
The bitmap comparison is final-size and lossless.
The planned baseline uses real source photographs selected before any dither is generated: a portrait with fine facial and hair detail, a scene with long midtone transitions, and a high-contrast subject with highlight and shadow texture. Their original files and hashes define the fixtures. We will not quietly replace a difficult source after seeing an algorithm's output.
Within each photo set, crop rectangle, geometry, polarity, tone controls, sharpening, physical width and height, DPI, and output pixel dimensions are identical. The output is a one-bit PNG. The same region is extracted from each finished bitmap as a true 1:1 crop: one bitmap pixel maps to one display pixel, with no browser interpolation or editorial enlargement.
Bitmap inspection can show dot structure, missing source detail, or processing artifacts. It cannot prove how those dots will burn. That requires the separate physical process below.
Controlled physical process
A bitmap is not an engraving.
Laser type, optics, focus, material response, line interval, speed, power, and mechanics can change or erase differences visible in a file. We therefore report bitmap evidence and physical evidence separately.
Evidence B
Actual physical engraving
- Real engraved test piece, not a simulation
- Machine and same material batch disclosed
- Focus, interval, speed, power, passes, and date logged
- Locked camera, lighting, distance, exposure, and white balance
Settings are experimental records, not universal recommendations. Material safety, ventilation, fire supervision, focus procedure, and operating limits remain governed by the machine and material manufacturers.
Blind scoring
Judge the marks before naming the method.
A person who does not score the pieces assigns random sample codes and randomizes display order separately for each judge and evaluation target. Algorithm names, app labels, file names, and the code key stay hidden from judges until all scores are locked. Each session records its display order, viewing light, and viewing distance. Bitmap crops are reviewed in a separate session at true 1:1, never as substitutes for the objects.
Judges use whole numbers from 1 (poor) to 5 (strong) for each rubric. The artifact score is directionally aligned: 5 means the least objectionable artifacts. The worksheet stores bitmap-review scores separately from physical-process scores; they are never combined into a single “dither quality” number. Scores are retained by judge and sample and are not rewritten after unblinding.
Bitmap 1:1 rubric
DetailRecognition of fine, source-supported detail in the final one-bit bitmap.
Tonal separationReadability of highlight, midtone, and shadow regions in the final bitmap.
ArtifactsPatterning, banding, clumping, or moiré in the final bitmap; 5 means least objectionable.
OverallJudge preference for the bitmap at true 1:1 without making a claim about material response.
Physical engraving rubric
DetailRecognition of fine, source-supported detail on the actual engraved object.
Tonal separationReadability of highlights, midtones, and shadows in the physical engraving.
ArtifactsBanding, clumping, merged marks, or process artifacts; 5 means least objectionable.
OverallJudge preference for the physical result at the declared viewing distance.
Open data contract
What every CSV row records.
The downloadable file is header-only today. Each future row will describe one scored evaluation target—either a true 1:1 bitmap or an actual physical engraving—and point to the paired evidence files. Empty or non-applicable fields will remain empty rather than being inferred.
Image fixture
comparison block ID, engraving order and test-piece position, source file and SHA-256, source dimensions, crop coordinates and aspect ratio, final millimeters, DPI, output pixels, polarity, tone settings, app version
Bitmap artifact
dither name, bitmap file and SHA-256, true 1:1 crop file
Physical process
machine make/model, laser type, material and batch, thickness and surface preparation, focus method/offset, line interval, speed, power, passes, air assist, engraving date
Physical evidence
physical photo file, photo capture date, locked capture notes
Blind evaluation
evaluation target (bitmap 1:1 or physical engraving), viewing distance and light, judge-specific display order, sample code, judge ID, four 1–5 scores, notes, protocol version
Download the schema
CSV · 57 columns · header only · CC0
Publication rule
No winner without the evidence.
Results will identify the tested scope—not declare a universally best dither. Publication will include the complete valid dataset, source-fixture identities, all six bitmaps and matched 1:1 crops, actual physical photos, per-judge blind scores, exclusions, and deviations. Any aggregate will state its sample count and remain traceable to rows in the CSV.
Until that package exists, this page remains methodology only. For practical background, read the laser engraving dithering guide, the DPI and line interval guide, or the complete photo-preparation workflow.