# Rounds Each round: stills of every shot at 540x960 (4x jittered AA), scored by two fresh critic subagents that see only the stills and the rubric (`rounds/CRITIC_PROMPT.md`). Median per dimension below; colour measured with `src/measure.py`. No reference photos were given, so the "side-by-side" is the still alone against the brief read as the real place. ## Round 1 Stills: `rounds/r01/`. Changes before this round: first full shot list; strata LUT with ledges; stream-power erosion with layer erodibility; footprint-based LOD on detail. | shot | composition | light | colour | water | vegetation | far_detail | min | |---|---|---|---|---|---|---|---| | s0_river_did_this | 70 | 60 | 57.5 | 42.5 | 42.5 | 52 | 42.5 | | s1_flat_plateau | 57.5 | 52.5 | 60 | 35 | 27.5 | 57.5 | 27.5 | | s2_river_cuts | 72 | 62 | 55 | 47.5 | 33.5 | 54 | 33.5 | | s3_deepening | 55 | 60 | 53.5 | 37.5 | 25 | 52.5 | 25 | | s4_old_rocks | 51 | 56.5 | 52.5 | 87.5 | 30 | 42.5 | 30 | | s5_mile_deep | 65 | 50 | 47.5 | 68.5 | 27.5 | 45 | 27.5 | | s6_long | 76 | 65 | 62 | 52.5 | 45 | 59.5 | 45 | | s7_still_cutting | 55 | 56.5 | 49 | 45 | 42.5 | 52.5 | 42.5 | | s8_endcard | 64.5 | 54 | 46.5 | 85 | 30 | 50 | 30 | Weakest: s3 vegetation 25. Critics agreed on: tree "balls" and dark dots, plastic oversaturated rock, black inner gorge, blown blocky rim, river not reading as water. Measured: ground saturation 0.54–0.81 (target 0.35–0.60), shadows R/B up to 73 (blue crushed). ## Round 2 Fixed in depth: vegetation (weakest dimension, weakest shot). Shrub crowns on slopes removed and replaced with footprint-filtered scrub speckle; pinyon-juniper crowns rebuilt as three irregular lobes with ragged foliage, darker grey-green albedo, clumped density shared with the far-distance canopy texture; riparian band narrowed to the bank; crowns cast soft long shadows and occlude sky on the ground. Also from measurement: rock albedos desaturated, post saturation 1.10 → 0.94, s7/s8 sun raised and saturation 0.8, s1/s6 aerial perspective reduced. Bug found while checking s5: the river channel was cut as a flat slot into the V gorge, leaving 100 m vertical "carpet" walls at both banks; the gorge walls now start at the river edge after a flat bank (shader and CPU height). Distance field now blends smoothly into the river distance (bake seam removed). Round 2 median (critics A, B): | shot | composition | light | colour | water | vegetation | far_detail | min | |---|---|---|---|---|---|---|---| | s0_river_did_this | 70 | 62 | 63.5 | 40 | 45 | 55 | 40 | | s1_flat_plateau | 64 | 60 | 60 | 32.5 | 43.5 | 55 | 32.5 | | s2_river_cuts | 75 | 64 | 56.5 | 56.5 | 42.5 | 58 | 42.5 | | s3_deepening | 61 | 62 | 60 | 32.5 | 35 | 59 | 32.5 | | s4_old_rocks | 55 | 58 | 51 | 85 | 40 | 46.5 | 40 | | s5_mile_deep | 75 | 61 | 53.5 | 66 | 32.5 | 57.5 | 32.5 | | s6_long | 76 | 65 | 63 | 46.5 | 37.5 | 61 | 37.5 | | s7_still_cutting | 62 | 61 | 57.5 | 40 | 37.5 | 55.5 | 37.5 | | s8_endcard | 71 | 65 | 63 | 85 | 49 | 59 | 49 | Critics: 2. Weakest: s1_flat_plateau / water = 32.5. Mean of shot minimums 38.2. Weakest: s1 water 32.5 (river read as a cream levee / pink road; inner-gorge river invisible). ## Round 3 Fixed in depth: water (weakest dimension, weakest shot). Water rewritten as a single-scattering body (clear green today, silty red-brown during the time-lapse) with depth-dependent transmittance, flow-aligned waves whose amplitude follows a current field (glassy runs, choppy riffles, rapids with whitewater streaks), footprint-aware roughness so glints become a sheen at distance, a thin wet line at the waterline, and the wide bank foam removed. River width now varies along its course, with sand/gravel bars; riparian growth is patchy instead of continuous hedgerows; bank sand darkened. s1 reframed to follow the young river from 400 m up; gentle valley slopes on the plateau take soil cover (no cream levee). Round 3 median: | shot | composition | light | colour | water | vegetation | far_detail | min | |---|---|---|---|---|---|---|---| | s0_river_did_this | 72 | 64 | 61 | 42.5 | 50 | 56.5 | 42.5 | | s1_flat_plateau | 66 | 63 | 60 | 46.5 | 38 | 52.5 | 38 | | s2_river_cuts | 74 | 63 | 53.5 | 50 | 45 | 58 | 45 | | s3_deepening | 55 | 60 | 57.5 | 57.5 | 45 | 55 | 45 | | s4_old_rocks | 52 | 56.5 | 50 | 85 | 45 | 43.5 | 43.5 | | s5_mile_deep | 73 | 53.5 | 50 | 64 | 37.5 | 50 | 37.5 | | s6_long | 77 | 65 | 62 | 53.5 | 42.5 | 57 | 42.5 | | s7_still_cutting | 60 | 64 | 57.5 | 47.5 | 35 | 55 | 35 | | s8_endcard | 70 | 66 | 62 | 85 | 42 | 58 | 42 | Critics: 2. Weakest: s7_still_cutting / vegetation = 35. Mean of shot minimums 41.2. Water rose most (s1 32.5→46.5, s3 32.5→57.5). Weakest now: s7 vegetation 35. Every critic still flags "melted clay" rock with contour swirls, curtain-like vertical streaks, blocky white rim teeth and a mint-green Tonto. ## Round 4 Fixed in depth: vegetation (s7 weakest). Woodland density lowered to 6–36 % cover with soft clump edges and gaps (no carpets); scrub speckle lighter grey-green and sparser (no black specks); s7 reframed from 360–420 m above the rim so no orange mound fills the foreground. Rock (flagged by every critic): stretched varnish streaks replaced by triplanar vertical jointing and broad varnish; talus/rubble mantle on soft-layer slopes mutes the bedding; bedding ledges weaker in slope layers than in cliffs; Tonto recoloured grey-tan; rim breakup uses fbm instead of ridged noise (less blocky); Kaibab/Coconino albedo lowered. Round 4 median: | shot | composition | light | colour | water | vegetation | far_detail | min | |---|---|---|---|---|---|---|---| | s0_river_did_this | 72 | 64 | 63 | 47.5 | 45 | 55 | 45 | | s1_flat_plateau | 68 | 64 | 61 | 48 | 37.5 | 54 | 37.5 | | s2_river_cuts | 75 | 62.5 | 55 | 55 | 43.5 | 59 | 43.5 | | s3_deepening | 55 | 60 | 58 | 87.5 | 42.5 | 53.5 | 42.5 | | s4_old_rocks | 50 | 58 | 53.5 | 87.5 | 35 | 47.5 | 35 | | s5_mile_deep | 75 | 59 | 55.5 | 67 | 37.5 | 59 | 37.5 | | s6_long | 78 | 66 | 64 | 55 | 41.5 | 59 | 41.5 | | s7_still_cutting | 62 | 62 | 62 | 45 | 40 | 55 | 40 | | s8_endcard | 71 | 66 | 62.5 | 86.5 | 46.5 | 56.5 | 46.5 | Critics: 2. Weakest: s4_old_rocks / vegetation = 35. Mean of shot minimums 41.0. ## Outcome: film rendered below the bar (user decision) After four rounds the critic medians are 35–88 per dimension (shot minimums 35–46.5, mean 41.0), far below the 98 target, and rounds 3→4 were flat (41.2 → 41.0). The remaining gap is structural (walls read as "melted clay", hidden inner-gorge river, blocky rim bands, carpet-like trees), not tuning. The user watched the round-4 preview (`preview_540_round4.mp4`) and chose to render the final film at this quality. **`film.mp4` is below the photoreal bar set in BAR.md.** One change after the preview: the depth card now reads "a mile deep (1.6 km)" (was 1.8 km, which is the deepest point, not a mile); the preview still shows the old 1.8 km card.