ScreenScryPublic betaPublic Beta model card
This confidence grade describes how well an exact device’s front geometry is supported. It is not a fit probability, a manufacturing tolerance, or a promise that two protectors are interchangeable.
device-evidence-v2screenscry-device-data-v6Same-brand matches scoring 95 or higher appear first; the remaining matches follow shape score, with brand familiarity breaking ties. They are not percentages of successful fit. Evidence grades describe completeness and remain visible in the dataset; a low grade alone does not exclude a useful slab-phone comparison.
Up to five close-shape alternatives with a known matching opening type, known matching edge type and no recorded front-fit evidence conflicts. Thickness disagreements remain visible in the full evidence record but do not affect front fit. Missing corner measurements are disclosed without blocking the comparison.
Larger coverage differences, uncertain openings, conflicting evidence and measured tighter corners require manual selection. Generic illustration radii do not affect scoring. Reviewed aperture estimates contribute a small advisory score adjustment and never exclude a candidate. Phone-body differences are not finished-glass measurements. Check opaque borders, camera and speaker placement, case clearance and model-specific installation trays.
Invalid dimensions, known flat-versus-curved edge conflicts, and alternatives beyond the directional size window. Different cutout types require the explicit ignore-cutout option. Foldables remain exact-model only.
Dimensional confidence measures support for the phone’s body width, body height and display dimensions. Its score is the weakest selected primitive, including identity and front surface. High confidence requires at least 85/100, two eligible independent sources for every dimensional primitive and no conflict in those features. One strong manufacturer source is labeled OEM supported; a complete editorial record is labeled Catalog supported.
Full fit evidence retains the A-D model below, including physical outline, corners and finished protector dimensions. A device can have high dimensional confidence while its exact protector outline remains unverified. Neither score is a probability of fit.
01 · Calculation
Each lane selects its strongest complete feature alternative. Known features are reconciled by source independence, authority, directness, exact-device identity, display surface, and freshness.
| Lane | Weight | What it represents |
|---|---|---|
| Identity | 15% | Exact device identity and OEM model-code evidence. |
| Chassis | 20% | Physical body width and height for the named device. |
| Display | 20% | Measured active area, or the documented diagonal-and-resolution alternative. |
| Outline | 15% | Glass dimensions, a vetted outline trace, or an eligible physical shape hash. |
| Corners and cutout | 10% | Corner radii or normalized corners, cutout family, shape hash, or vetted outline. |
| Protector template | 10% | Paired finished-protector dimensions from at least three reputable independent designs across two makers; catalog-only designs and manufacturer cross-fit classes are disclosed separately. |
| Display surface | 10% | Whether the evidence applies to the front, cover, or inner display. |
02 · Grade gates
Every listed condition must pass. A high numeric score alone cannot unlock a grade when identity, surface, independence, conflicts, or exact geometry remain unresolved.
| Grade | Score | Critical floor | Independence gate | Additional gates |
|---|---|---|---|---|
| A | ≥ 88 | ≥ 75 | ≥ 3 eligible evidence sets | Exact surface, no material conflicts, and an authoritative outline or measured-template consensus. One qualifying measured-template consensus counts as one evidence set, never as several source lineages. |
| B | ≥ 74 | ≥ 55 | ≥ 2 eligible source lineages | Complete body/display envelope, no material conflicts, and exact geometry or an authoritative outline. |
| C | ≥ 55 | - | No separate minimum | Complete body/display envelope and no material conflicts. |
| D | Otherwise | - | No separate minimum | Any record that does not clear a higher gate; all foldable and multi-surface records are capped here in v2. |
03 · Missingness and agreement
A missing feature contributes no invented value. It lowers lane coverage; a fully missing lane is omitted from the lane mean, while the overall coverage multiplier still lowers the device score. Derived children do not cast another vote.
A single lineage can carry confidence, but it cannot demonstrate independent agreement. Its agreement field remains unavailable; the public presentation labels the result not independently tested.
Copies that resolve to the same independence key count once. A retailer page, an affiliate copy, and a cached copy of the same underlying claim are not three independent sources.
Materially inconsistent independent values create a conflict. A and B require zero material conflicts; C also requires an internally complete, conflict-free body/display envelope.
04 · Source policy
This registry defines what a lineage is allowed to prove. Registration does not mean every device has that evidence; each device record shows the sources actually applied. Downstream copies share their upstream independence key and do not create another vote.
Explicit screen clip coordinates and equivalent circular radii. Native-aspect checks and exact-device runtime conflicts gate use. Pixels normalize to aperture width; millimeter estimates require a separate scale. Metadata, frame finishes and image masks count once. Undisclosed upstream measurements cannot unlock an additional grade lineage.
Official front-view apertures support normalized screen corner shape. Equivalent-circle radius is an approximation to the retained continuous contour. Shares Apple's engineering lineage; Apple artwork redistributed by a mockup site is never another independent source. Active-screen aperture and outer glass remain distinct.
Public straight-front thumbnails provide coarse shape corroboration. Exact editor masks require access; thumbnail estimates do not vote as precise geometry. Asset authorship and upstream independence remain unresolved.
Recovered from the permissioned source pool. Model-specific, undistorted masks and original asset provenance must be established before geometry use. Generic or generated phone scenes do not establish device dimensions.
Exact-model published body dimensions, display diagonal and resolution only. No inferred outline, corner, edge or cutout facts.
Exact-model published body dimensions, display diagonal and resolution only. No inferred outline, corner, edge or cutout facts.
Exact-model published body dimensions, display diagonal and resolution only. No inferred outline, corner, edge or cutout facts.
Primary OEM engineering evidence. Nominal body, cover-glass, active-area, and scaled front-outline facts may be used directly for the exact named model and front surface.
Primary OEM product specifications. Use exact-model identity, front-surface identity, body dimensions, and native display resolution only when stated literally. Product renders and emulator/device-skin assets do not establish physical glass, active-area, corner, cutout, or outline geometry.
Primary OEM published specifications. Use exact-model identity, front-surface identity, body dimensions, full-rectangle display diagonal, and native resolution. Samsung's 7R language describes unified chassis design curvature; it must not be imported as active-display, cover-glass, or protector corner radius.
Use exact-model body dimensions, display diagonal, resolution, and identity facts. Active display millimeters calculated from area/aspect are derived children and must not cast an additional vote.
Strong device-identifier, canonical-name, family, and release-date evidence. It is not an OEM geometry measurement and must not be relabeled as one.
Pinned simulator/private-API-derived static geometry in points. It may corroborate native corner and cutout families, but points are not millimeters and zero or square values require semantic review.
Direct physical-device runtime evidence when the build fingerprint, model identity, physical-hardware flag, surface, rotation, and coordinate space all validate. Repeated runs and builds remain one Android-framework lineage; native pixels are not millimeters without an independent scale.
The importer must replace the fallback independence key with the asset's named upstream credit and merge it with that upstream lineage. MockUPhone is a curator/distributor, not an additional independent vote. Transparent apertures and embedded coordinates may support a model-exact front trace; pixels require confirmed physical scaling.
Low-weight corroboration for diagonal, pixel resolution, CSS viewport, PPI, density class, and aspect ratio. It supplies neither a model-exact physical outline nor evidence that is sufficiently independent to unlock an A grade.
| Source | Review finding | Required next evidence |
|---|---|---|
| PureGear | The project owner supplied an exact PureGear Product Dimensions field for SKU 67136PG and confirmed permission. PureGear now has a supplier adapter, dimension parser, catalog discovery and paced collection. Cloudflare still prevents the first live catalog review from this environment. | Run a first complete collection using supplier-approved access, and review its product-page fields before enabling publication. The supplied quotation is retained with its provenance; it is not a fabricated live inventory record. |
| Custom Screen Protectors | The public dimensions file contains 20 seed records; 50,000 is future capacity. The linked compatibility script embeds the same records, with no larger public search dataset found. The sample TC52 DXF contains a plain rectangle without the stated radius or other claimed layers; the SVG is labeled as unsuitable for production tooling. | Require exact model and revision, measurement provenance, a validated outline and a stated coordinate surface before accepting any dimension. Published manufacturing tolerance alone does not establish measurement accuracy. |
05 · Protector evidence
Direct manufacturer records start as exact catalog evidence. When a product page or linked specification explicitly identifies the finished protector’s width and height, those values also become physical evidence. A number labeled as product, package, shipping, box, tray, frame, or applicator size never does.
Each accepted variant preserves its canonical URL, SKU, MPN or GTIN when available, exact named devices, parser version, observation time, and field provenance. Claims resolve through canonical model names, aliases, or OEM codes. Family-level prose and unresolved names are not silently promoted.
A numeric width and height must be tied to the exact protector variant and front display surface. ScreenScry keeps measurements separate by physical cut: regional, channel, and article copies count once; clear, privacy, or package SKUs from one maker with the same exact dimensions also count once; conflicting measurements among copies exclude that design. Published averages and sample sizes therefore describe qualifying independent physical designs, not listings.
One exact maker-owned variant explicitly names both devices. A family page, title similarity, or equal-looking products cannot qualify.
The maker publishes separate exact-device variants whose explicit finished-protector width and height agree within the conservative comparison tolerance.
At least two independent makers provide qualifying finished-protector dimension agreement for the same device pair.
06 · Multi-surface phones
Cover, inner, and active-window geometry cannot be combined into one front-surface profile. Model v2 caps every foldable or trifold at Grade D until each display surface is independently identified and modeled. Unscoped protector sizes, grades, and comparison outlines are withheld.
07 · Validation status
uncalibrated-shadow-hiddenThe experimental fit score stays hidden while independent evaluation is incomplete.
Independent, directional held-out labels are required before evaluation begins.
Independent held-out labels in each class, plus published performance metrics, are required.
08 · Changelog
puregear-reviewed-measurements-v1Integrated 26 reviewed supplier measurement records from the complete 191-page owner capture. Exact catalog identity, body-envelope validation, field provenance, and source/parser review gates apply at read time. Repeated sizes from one maker count once; 50 suspect copied dimension triplets are excluded. These are supplemental finished-protector measurements, not independently measured display, glass, or corner geometry. Source pages and observation dates remain visible; a newer committed supplier catalog supersedes the snapshot.
protector-evidence-v6Added staged ZAGG, OtterBox, Spigen, and Belkin product evidence with exact variant identity, field provenance, reviewed finished-protector dimensions, independent publication gates, and M1/M4/C1 cross-fit classes. Regional, channel, and article copies now count once by physical design; same-maker SKUs with one exact size pair count once; conflicting measurements within one design fail closed. Dimensional consensus still requires at least three reputable maker-keyed cuts across two normalized manufacturers.
device-evidence-v2Separated source lineages from protector designs, required paired multi-maker protector dimensions, made single-source agreement unavailable, admitted eligible physical runtime outlines, and published explicit failed-gate reasons without changing legacy match ordering.
device-evidence-v1Introduced the seven-lane evidence profile, critical-lane floor, conflict gates, and exact-device-only handling for multi-surface phones.
09 · Citation and reuse
Interim public-beta terms; a final dataset license is pending.
The derived ScreenScry beta export may be downloaded, analyzed, quoted, and cited for evaluation with ScreenScry, the dataset version, and an access URL identified.
Linked facts, drawings, images, mockups, repositories, and other source materials remain subject to their publishers' rights and terms.
The beta export does not sublicense raw source materials, authorize their redistribution, or turn an outline into a manufacturing or cutting template.
ScreenScry, “ScreenScry Device Data (Public Beta),” dataset screenscry-device-data-v6, model device-evidence-v2, https://screenscry.com/data.