Methodology · v1
Analysis Methodology
This document specifies the scoring framework Steam Analyser applies to Steam capsule artwork. It defines the rubric, viewing conditions, scoring scale, and aggregation method.
Scope
The capsule rubric evaluates the visual asset on six dimensions: genre clarity, title readability, contrast and colour, uniqueness and polish, brand consistency, and composition.
Each dimension is scored as a 1–10 integer against published anchors. Dimensions are evaluated at three rendered contexts, full (920 × 430), small (≈231 × 87), and tiny (120 × 45), on Steam's #1b2838 background. Dimension scores aggregate to a 0–100 overall score.
The framework is versioned. Material revisions advance the version stamp on this page and are recorded with a dated description of the change.
Viewing conditions
Each dimension is evaluated against three rendered contexts on the background colour Steam displays, under attention conditions representative of typical browsing. A capsule that performs at full size but degrades at thumbnail size will fail in production, where the majority of impressions occur at smaller renders.
| Test | What it means |
|---|---|
| Full size | 920 × 430, the header capsule on the store page. |
| Small | Around 231 × 87, the home-page rail render. Small enough that ornate fonts collapse. |
| Tiny | 120 × 45, the auto-generated thumbnail in search and discovery. Postage-stamp size. The brutal test. |
| Background | Steam dark #1b2838. The capsule sits on this colour, surrounded by other capsules sitting on this colour. |
| Squint test | If the design collapses when squinting, dimensions that rely on fine detail get reduced. |
| Greyscale test | Run for contrast and silhouette. Strong silhouettes survive desaturation; muddy mid-tone designs collapse. |
| Quick scroll | Slight blur, low attention, under one second to parse. The realistic browsing condition. |
The 120 × 45 thumbnail is auto-generated from the small capsule (462 × 174). The full 920 × 430 render accounts for an estimated 5–10% of total capsule impressions on Steam; the remaining majority occur at smaller sizes.
Scoring anchors (1–10)
Every dimension shares the same anchor scale. Each dimension begins at 6 (competent baseline) and moves up or down based on observable evidence in the asset. Scores of 9 and 10 are reserved for execution that holds up at all rendered sizes, not at full size alone.
| Score | Label | What it means |
|---|---|---|
| 9 – 10 | Excellent | Stands out clearly. Strong hierarchy, clean craft, memorable. |
| 7 – 8 | Good to solid | Normal range for a well-executed asset. Reads well with minor gaps. |
| 6 | Competent baseline | Functional. Does its job without notable strengths or weaknesses. The starting point for every score. |
| 4 – 5 | Below average | Clarity, readability, or execution issues that hurt discoverability. |
| 1 – 3 | Significant problems | Confusing, illegible, or actively misleading. |
Capsule rubric: six dimensions
The full criteria for each capsule dimension. The same criteria are applied to every capsule scored by the framework.
1. Genre Clarity
JudgesWhat kind of game this is at TINY (120 × 45) size. The first second of attention.
Rewards
- Theme-specific cues a player can name without context
- Familiar genre iconography (deck fans, dungeon tiles, base grids, isolated horror subject)
- Pose, setting, and gameplay-implying composition
- Readable fantasy that signals subgenre, not just genre
Penalises
- Mixed messaging where visuals could imply two unrelated genres
- Generic theme with no gameplay type implied
- Misleading signals (cozy aesthetic on a horror game and vice versa)
- Pure character portrait with no environmental or genre context
Anchors
| Score | What it looks like |
|---|---|
| 10 | At tiny size, you can name the genre or subgenre from visuals alone. |
| 5 | You can tell it's a game; genre is ambiguous. |
| 1 | Communicates the wrong genre, or actively misleads. |
2. Title Readability
JudgesLetterforms, spacing, contrast, and legibility at full size and at tiny size. Whether the logo collapses when Steam scales it down.
Rewards
- Strategic placement on a controlled background zone (not noisy texture)
- Strong letterform weight and outline that holds up at 120 × 45
- High contrast against the surrounding artwork at the actual rendered size
Penalises
- Decorative or ornate fonts that lose legibility small
- Small taglines or extra subtitle text that becomes unreadable
- Title placed on busy texture or active subject competing for attention
- Insufficient contrast between letterforms and background at tiny size
Anchors
| Score | What it looks like |
|---|---|
| 9–10 | Reads cleanly at every rendered size, including 120 × 45. |
| 6 | Reads at full and small; collapses at tiny. |
| ≤4 | Title is not readable at tiny size (almost always a 4 or below unless the title is intentionally minimal). |
3. Contrast & Color
JudgesHow the capsule pops against Steam's #1b2838 dark background, in a quick scroll, evaluated in colour and in greyscale.
Rewards
- Strong light/dark separation between subject and background
- Clean silhouette that survives desaturation
- Saturation control that doesn't muddy the image
- Lighting separation between subject and background plane
Penalises
- Limited value range and muddy mid-tones
- Subject blending into background
- Unclear edges or low silhouette readability in grayscale
- Dark-on-dark composition that disappears against #1b2838
Anchors
| Score | What it looks like |
|---|---|
| 9–10 | Strong silhouette and value contrast that pops at every size, including grayscale. |
| 6 | Reads cleanly at full size; blends at tiny size. |
| 1–3 | Subject is lost in grayscale or against #1b2838. |
4. Uniqueness & Polish
JudgesHow premium and distinct the capsule feels relative to common capsules in the genre. The dimension that decides whether it's memorable.
Rewards
- A clear visual hook (character silhouette, motif, distinctive prop, unusual palette)
- Clean craft: crisp edges, intentional typography, coherent effects
- Visual storytelling that communicates a unique selling point
- A distinctive art style that signals 'this game, not that one'
Penalises
- Cheap asset vibe, template look, generic stock-pack aesthetic
- Random effects, default-feeling gradients, default-pack flares
- Generic theme with no standout idea
- Competent-but-forgettable execution that vanishes in a feed
Anchors
| Score | What it looks like |
|---|---|
| 9–10 | Premium craft + distinctive hook a player remembers later. |
| 5–6 | Competent but generic; nothing wrong, nothing memorable. |
| 1–3 | Reads as cheap or template-feel; visually indistinguishable from neighbours. |
5. Brand Consistency
JudgesInternal cohesion only. Whether the capsule reads as one coherent piece, consistent rendering style, palette, and identity, or as a collage of mismatched assets.
Rewards
- Consistent rendering register across every element
- Coherent palette and lighting language
- Recognisable identity cues (character, motif, signature treatment)
- Genre-appropriate visual conventions used well without feeling derivative
Penalises
- Photoreal element on a flat-shaded environment under a hand-painted sky
- Pasted-in elements that don't share lighting or rendering
- Generic presentation with no memorable identity cue
- Style inconsistency that the brain catches before the brand registers
Anchors
| Score | What it looks like |
|---|---|
| 9–10 | Single coherent image with a recognisable identity hook. |
| 6 | Mostly consistent; one element feels slightly off. |
| 1–3 | Visually mismatched assets; reads as a collage. |
6. Composition
JudgesHierarchy, focal point, balance, clutter, safe margins, and crop resilience across the four capsule aspect ratios.
Rewards
- One clear primary subject at small and tiny sizes
- Supporting elements that guide the eye, don't compete with it
- Effective use of space; no dead-centre voids unless intentional
- Background → midground → foreground depth that creates a clear read
- Crop resilience across header (920 × 430), small (462 × 174), main (1232 × 706), and vertical (748 × 896)
Penalises
- Two or three competing focal points that all collapse at tiny size
- Edge-hugging title or subject that gets cropped on a different size
- Awkward empty gaps or scattered attention
- Equal emphasis on every element with no hierarchy
Anchors
| Score | What it looks like |
|---|---|
| 9–10 | Single focal point reads at every size; composition survives every crop. |
| 6 | Composition works at full size; weakens at smaller crops. |
| 1–3 | Multiple competing focal points; critical elements cropped on at least one size. |
How the 0–100 score is computed
The six dimension scores (each 1–10) combine into a single 0–100 overall score. The aggregation is weighted toward the dimensions that drive discoverability hardest at small sizes, genre clarity, title readability, and contrast, since a capsule that fails at thumbnail size fails before any other dimension can contribute.
A 9 across the board lands above 90. A 5 across the board lands around 50. A capsule that scores 9 on five dimensions and 3 on one is rarely above 75, failing hard on a single dimension affects the whole.
The aggregation reflects the load-bearing role of small-size legibility. A capsule that fails at thumbnail size fails before the remaining dimensions can contribute, and is scored accordingly.
Score bands
The label associated with each overall score range. We use these on result pages and in aggregate reports.
| Overall score | Label |
|---|---|
| 90 – 100 | Exceptional |
| 85 – 89 | Standout |
| 80 – 84 | Very Strong |
| 70 – 79 | Strong |
| 65 – 69 | Solid |
| 50 – 64 | Average |
| Below 50 | Weak |
Non-fabrication rules
Per-dimension explanations are constrained by the following rules. They prevent the framework from describing elements not present in the asset and require each judgement to be grounded in observable evidence.
- ·Only describe elements you can clearly see. Do not guess unreadable text, logos, characters, or details.
- ·If the title text is not readable at a given size, explicitly state it is not readable at that size.
- ·Brand consistency must be scored from internal cohesion cues only, not assumed brand recognition.
- ·Detail must reference specific visible elements (top-left logo, centre character silhouette, warm orange gradient, busy particle field), not generic adjectives.
- ·Detail must include at least one explicit statement about what works or fails at small or tiny size.
FAQ
What does Steam Analyser actually measure?
Six dimensions of the visual capsule asset: genre clarity, title readability, contrast and colour, uniqueness and polish, brand consistency, and composition. Each dimension is scored on a 1–10 integer scale against published anchors and aggregated into a 0–100 overall score.
What sizes does the capsule get scored at?
Three rendered contexts. Full size (920 × 430 for the header capsule), small (around 231 × 87, the home-page rail render), and tiny (120 × 45, the auto-generated thumbnail). Every dimension is mentally re-tested at the smaller sizes; a capsule that scores 9 at full size and 4 at tiny size is a capsule that fails in production. The Steam dark background (#1b2838) is assumed throughout.
Is the rubric biased toward a particular art style?
No. The rubric scores execution, not aesthetic preference. A pixel-art horror capsule and a hand-painted cozy capsule are both judged against the same six dimensions: does the genre read at thumbnail size, does the title hold up, is contrast strong against #1b2838, is the craft clean, does the composition work. The dimension that comes closest to taste is uniqueness and polish, and even there we score whether the capsule has a distinctive hook, not whether the hook is to our personal taste.
Where do the scores come from?
Each asset is evaluated by an automated scoring engine that applies the criteria documented on this page to every submission. The criteria are the system of record. The engine ensures the criteria are applied consistently across all analyses; the methodology document is the human-readable specification of what the engine evaluates.
Can two different scoring runs return different scores for the same capsule?
Slight variation between repeated scorings is possible. The framework constrains this through integer-only scores against explicit anchors, mandatory references to specific visible elements in each explanation, and per-dimension rules that limit interpretive latitude. Across repeated scorings of the same asset, dimension scores typically vary by ≤1 and overall scores by ≤3 points. A score that feels significantly off can be re-run; persistent disagreement is signal worth investigating.
Why don't you publish capsules that score below 60?
Two reasons. First, brand: cautionary-tale lists of named games are unfair to the developers and counterproductive, they don't help readers improve their own capsule. Second, statistical: capsules below the eligibility floor often have data-quality issues (missing image, malformed cache, edge cases) that would pollute the aggregate. Below-threshold capsules contribute to the anonymised pattern frequencies in our reports, never to named examples.
How is the methodology revised?
The framework is versioned. Material revisions advance the version stamp on this page and are recorded with a dated description of the change. The methodology is not silently altered.
Can I rerun a score after I've updated my capsule?
Yes. Upload the new version, or paste the updated Steam URL once it is live, and a fresh score is produced against the same rubric. Existing scores are cached and free to re-fetch. To compare versions side-by-side, save both screenshots or use the upload flow, which preserves prior attempts.
Last reviewed: 2026-05-08. Sourced from Steam's official Steamworks documentation and the Steam Analyser scoring methodology.