Visual Rubric — grading against Total War: Rome II

The standard for this project. Critics score rendered frames against these criteria, not against source code. Every item is something a Rome II frame actually does, chosen because it is diagnostic — the specific tells that separate a shipped AAA strategy game from a competent WebGL demo.

Scoring per criterion: 0 = absent, 1 = attempted but wrong, 2 = acceptable, 3 = matches Rome II, 4 = exceeds it. A subsystem passes at mean ≥ 3.0 with no criterion below 2.

A. Lighting & atmosphere (the highest-signal category)

#CriterionThe specific tell
A1Warm/cool splitSunlit surfaces are warm; shadowed surfaces take a distinctly cool blue-grey sky bounce. Flat grey shadow = instant fail.
A2Aerial perspectiveDistant terrain/city progressively desaturates and shifts toward sky colour. Should be strong — Rome II's distance haze is very pronounced.
A3Shadow qualityTight and dark at contact points (feet, wall bases), softening with distance from the occluder. No acne, no peter-panning, no crawling while the camera pans.
A4Sun direction legibilityYou can tell where the sun is from the shading alone, without seeing it.
A5Sky depthGradient from horizon haze to zenith, cloud form with real silhouette and silver lining toward the sun. Not a flat blue dome.
A6Exposure & contrastFilmic S-curve. Highlights roll off rather than clipping to white; shadows retain detail rather than crushing to black.
A7Ambient occlusionContact darkening in crevices, under eaves, between men in a rank. Subtle — not a dirty grey outline around everything.
A8No flat ambientNothing looks lit by a uniform hemisphere. Every surface has directional information.

B. Terrain & ground

#CriterionThe specific tell
B1No visible tilingFrom a high camera, no repeating texture grid. Requires macro variation, not just a small tile.
B2Material varietyDry grass, trampled dirt, mud, gravel, stone — blended by slope/height/curvature, interlocking rather than cross-fading.
B3Close-up fidelityAt eye level in the melee, the ground still reads as ground: detail normal, small stones, grass tufts.
B4Erosion plausibilityValleys have V-profiles and connected drainage; ridges have coherent lines. Not obviously fBm noise.
B5Grass integrationGrass fades into terrain colour with no visible density ring or LOD pop; moves with wind.
B6Vegetation placementTrees clustered as they would actually grow — along water, in copses, following boundaries. Not uniform sprinkle.
B7Human landscapeRoads, field boundaries, terraces. Evidence of habitation.

C. Units & crowds (where hobby projects fail hardest)

#CriterionThe specific tell
C1Individual variationNo two adjacent men identical: height, kit, skin, beard, cloak, shield emblem. Scan a rank — if you spot a repeated pair, fail.
C2Animation desyncIdle and march cycles visibly out of phase across a unit. Synchronised breathing is a fatal tell.
C3Kit specificityYou can identify the unit type from its silhouette: segmentata bands, scutum curve, galea cheek-guards, spangenhelm, bare-chested fanatics.
C4Formation densityMen nearly shoulder to shoulder (~0.86 m). Rome II formations are dense; sparse ranks read as a tech demo.
C5Formation legibilityRanks and files readable from above; the shape of the formation is obvious.
C6Weight in motionFootfall, hip counter-rotation, lean into acceleration. No sliding feet, no floating.
C7LOD invisibilityNo popping as the camera moves. Silhouette preserved at every tier.
C8CorpsesFallen men lie conformed to the ground, varied poses, persistent, piling where fighting was heaviest.
C9Scale truthA man is 1.75 m against 6.5 m walls and 43 m domes. Everything reads at correct relative size.

D. City & architecture

#CriterionThe specific tell
D1Mass and thicknessThe wall reads as metres of solid masonry, not a plane. Visible depth at gates, embrasures, wall-walk.
D2Material honestyBrick courses read as courses. Travertine looks like travertine. Romans painted their buildings — ochres, reds, whites, not bare grey.
D3Silhouette recognitionThe skyline is unmistakably ancient Rome without a caption.
D4Density without repetitionInsulae are varied in height, width, roofline, colour. No copy-paste grid.
D5WeatheringDirt at bases, sun-bleaching on top, staining below drains.
D6Period accuracyNothing medieval or fantastical. Correct proportions for known buildings.

E. Combat & VFX

#CriterionThe specific tell
E1DustMoving formations trail sunlit dust; a cavalry charge rolls a wall of it. Rome II's most recognisable effect.
E2Clash readabilityThe moment lines meet is visually distinct — compression, dust, debris.
E3Front-rank fightingOnly men in contact are swinging. A whole block flailing is wrong.
E4Blood restraintPresent and physical; not a red firework.
E5Ground persistenceBlood pools, trampled earth, spent shafts accumulate over the battle.
E6Particle lightingParticles lit by the sun, depth-faded against geometry. No flat unlit billboards.
E7BannersLegible, wind-driven cloth. The landmarks that let you read the field.
E8Missile plausibilityVolleys ragged not synchronised; arcs correct; misses scatter believably.

F. Presentation & UI

#CriterionThe specific tell
F1RegisterBronze/gold/oxblood, Trajan capitals, engraved metal. Reads as a historical AAA title, not a web dashboard.
F2Information densityUnit cards carry strength, morale, fatigue, ammo, status at a glance.
F3FinishNothing misaligned, clipped, low-contrast, or default-browser-styled. Crisp on HiDPI.
F4RestraintHUD does not fight the battle for attention.

G. Frame-level craft

#CriterionThe specific tell
G1CompositionA still frame is photographic — it has a subject and depth layers.
G2Colour gradeWarm, rich and saturated, with a coherent grade. Corrected after checking real Rome II frames: an earlier draft of this rubric said "sun-bleached, slightly desaturated", and that was wrong as a target. Rome II puts gold helmets, white linen, red and ochre shields, green pasture and a deep blue-grey sky in one image. A monochrome dust-beige frame is a far worse error than mild over-saturation. Still avoid orange-teal.
G2bDynamic rangeDeep near-black shadows and clipping highlights coexist in one frame. If nothing in the image is below ~15% or above ~90% luminance, the range is too narrow.
G3Anti-aliasingNo stair-stepping on wall edges or spear shafts; no shimmer on distant ranks.
G4Post restraintBloom on genuine highlights only. Subtle vignette and grain. No lens-flare spam.
G5BelievabilityThe instinctive reaction is "screenshot from a game I could buy", not "impressive for a browser".

Critic instructions

  1. Look at the frame before reading anything about it. Form a first impression and write it down. First impressions catch what analysis rationalises away.
  2. Score every applicable criterion with a one-line justification citing what you see.
  3. Be harsh. Grade inflation makes you useless. If the mean is above 3.0 you are claiming this matches a game with a multi-million-dollar art budget — hold that bar. Most first passes should score 1.5–2.5.
  4. Name the single highest-leverage fix: the one change that would most improve the frame. Be specific and actionable ("distance haze is far too weak, raise density ~4× and tint it toward sun colour"), never vague ("improve the lighting").
  5. Do not credit intent. If code exists for an effect but the effect is not visible in the frame, it scores 0.
  6. State the verdict explicitly: PASS (mean ≥ 3.0, nothing below 2) or FAIL, with the mean and the list of sub-2 criteria.