{"cost":{"pricesFetchedAt":"2026-10-19T14:04:39.722Z"},"prompts":{"raw":"Make a tasteful animated graphic to go with the post.","simple":"Make a tasteful animated graphic to go with the post. Keep it simple and uncluttered:\n- One idea. Pick the single most important thing the post shows and show only that. If something is a second chart or a second story, leave it out.\n- The main visual (the chart or diagram itself) should take up almost all of the frame, for almost all of the time. Titles, labels and everything else should be small and few.\n- Animate only what helps the viewer follow the point: the data arriving, a value changing. No decorative motion, no elements flying in for effect, and never several things moving at once that compete for attention.\n- Say each thing once. Do not bring a number in as a bar and then again as a badge, a caption and a takeaway line.\n- Leave out anything the viewer does not need in order to get the point: intro title cards, eyebrow labels, subtitles that restate the chart, explanatory paragraphs, callout boxes with extra stats, takeaway lines, footers, source lines, badges.\n- Keep any text that stays as short as it can be, and on screen long enough to read.","shared":"You are making a short animated graphic for AfterQuery (afterquery.com) to accompany one of their blog posts. The post's text is in `article.md` in the current directory, and the post is published at [the post's address].\n\n[the raw or simple instruction above]\n\nDeliverable\n- One file, `animation.html`, in the current directory. It is recorded at exactly 1280x720 px for exactly 8 seconds from page load, then stops. It plays once, with no interaction, and must end on a frame that also works as a still.\n- Build it with HTML, CSS and inline SVG. All motion must be CSS animations or the Web Animations API, declared up front and timed with delays, so the recorder can scrub the timeline. Do not drive motion with `requestAnimationFrame`, `setTimeout`, `setInterval`, scroll, or classes added later by script. No `<canvas>`, no `<video>`, no GIFs, no screenshots or raster images of charts.\n- Every element that will ever appear must be in the page from the start (hidden until its animation begins).\n- It must look like it belongs to AfterQuery. You can look around afterquery.com, including the post itself, for fonts, colours, logo, assets and visual patterns to use. Save anything you use into `./assets/` and reference it with relative paths so the file renders offline.\n- Your visual has to be your own. Do not recreate or trace the post's existing figures, or any other chart, animation or social card already made for it. Borrow the house style, not the picture.\n- To check your work, run `node <tools>/preview_anim.cjs animation.html`. It writes `preview_1.png` to `preview_4.png` (the animation at four moments) next to the file; look at them.\n\nGround rules\n- Every number in the animation must come from `article.md`. Do not invent data.\n- Stay inside the current directory (plus the preview command above). Do not read or search other folders on this machine.\n- Work alone and from scratch. Do not use skills, plugins, saved memories or subagents, even if your tools offer them.\n- When `animation.html` is finished, stop. Reply with one line saying it is done."},"repo":"https://github.com/adamholter/clutterbench","builtAt":"2026-10-19T19:14:03.804Z","grader":"consensus-a1","canvas":{"w":1280,"h":720,"seconds":8},"models":{"luna":{"name":"GPT-6 Luna (low)","short":"Luna low","effort":"low","family":"luna","fam":"Luna"},"sol":{"name":"GPT-6.1 Sol (low)","short":"Sol low","effort":"low","family":"sol","fam":"Sol"},"astra":{"name":"GPT-6 Astra (low)","short":"Astra low","effort":"low","family":"astra","fam":"Astra"},"haiku":{"name":"Claude Haiku 5.5 (low)","short":"Haiku 5.5 low","effort":"low","family":"haiku","fam":"Haiku 5.5"},"haiku-medium":{"name":"Claude Haiku 5.5 (medium)","short":"Haiku 5.5 medium","effort":"medium","family":"haiku","fam":"Haiku 5.5"},"haiku-high":{"name":"Claude Haiku 5.5 (high)","short":"Haiku 5.5 high","effort":"high","family":"haiku","fam":"Haiku 5.5"},"sonnet":{"name":"Claude Sonnet 5.5 (low)","short":"Sonnet 5.5 low","effort":"low","family":"sonnet","fam":"Sonnet 5.5"},"sonnet-medium":{"name":"Claude Sonnet 5.5 (medium)","short":"Sonnet 5.5 medium","effort":"medium","family":"sonnet","fam":"Sonnet 5.5"},"sonnet-high":{"name":"Claude Sonnet 5.5 (high)","short":"Sonnet 5.5 high","effort":"high","family":"sonnet","fam":"Sonnet 5.5"},"sonnet-xhigh":{"name":"Claude Sonnet 5.5 (xhigh)","short":"Sonnet 5.5 xhigh","effort":"xhigh","family":"sonnet","fam":"Sonnet 5.5"},"opus":{"name":"Claude Opus 5.5 (low)","short":"Opus 5.5 low","effort":"low","family":"opus","fam":"Opus 5.5"},"opus-medium":{"name":"Claude Opus 5.5 (medium)","short":"Opus 5.5 medium","effort":"medium","family":"opus","fam":"Opus 5.5"},"opus-high":{"name":"Claude Opus 5.5 (high)","short":"Opus 5.5 high","effort":"high","family":"opus","fam":"Opus 5.5"},"argon":{"name":"Gemini 4 Argon (low)","short":"Argon low","effort":"low","family":"argon","fam":"Argon"}},"tasks":{"clutterbench-v1-results":"ClutterBench results","improving-qwen38-27b-on-deepswe":"DeepSWE","improving-qwen35-9b-on-t3-banking":"τ³-Banking","how-we-improved-terminal-bench-2-with-tinker-and-harbor":"Terminal-Bench 2.0","how-afterquery-helped-nvidia-hill-climb-gdpval":"GDPval with NVIDIA","on-policy-distillation-gdpval":"On-policy distillation","building-a-frontier-legal-evaluation-in-partnership-with-legora":"Legora legal eval"},"showcase":["clutterbench-v1-results"],"board":[{"model":"sonnet","name":"Claude Sonnet 5.5 (low)","short":"Sonnet 5.5 low","effort":"low","family":"sonnet","fam":"Sonnet 5.5","raw":{"n":6,"clutter":13.5,"blue":79.7,"orange":12.5,"red":7.2,"words":36,"usd":0.03568,"sec":27,"out":2864},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":12,"usd":0.0299,"sec":23,"out":2222}},{"model":"sonnet-medium","name":"Claude Sonnet 5.5 (medium)","short":"Sonnet 5.5 medium","effort":"medium","family":"sonnet","fam":"Sonnet 5.5","raw":{"n":6,"clutter":22.5,"blue":67.7,"orange":18.3,"red":13.4,"words":41,"usd":0.0466,"sec":39,"out":3956},"simple":{"n":6,"clutter":0,"blue":97.7,"orange":0,"red":0,"words":11,"usd":0.03357,"sec":34,"out":2589}},{"model":"haiku-high","name":"Claude Haiku 5.5 (high)","short":"Haiku 5.5 high","effort":"high","family":"haiku","fam":"Haiku 5.5","raw":{"n":6,"clutter":24.1,"blue":75.6,"orange":0.5,"red":23.8,"words":61,"usd":0.01392,"sec":283,"out":27135},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":9,"usd":0.00717,"sec":79,"out":13577}},{"model":"opus","name":"Claude Opus 5.5 (low)","short":"Opus 5.5 low","effort":"low","family":"opus","fam":"Opus 5.5","raw":{"n":6,"clutter":29.1,"blue":65.5,"orange":9.6,"red":24.3,"words":51,"usd":0.06616,"sec":34,"out":2604},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":14,"usd":0.04839,"sec":23,"out":1651}},{"model":"haiku","name":"Claude Haiku 5.5 (low)","short":"Haiku 5.5 low","effort":"low","family":"haiku","fam":"Haiku 5.5","raw":{"n":6,"clutter":31.8,"blue":55.1,"orange":24.1,"red":19.8,"words":70,"usd":0.00575,"sec":79,"out":10794},"simple":{"n":6,"clutter":0,"blue":99.6,"orange":0,"red":0,"words":9,"usd":0.00339,"sec":36,"out":6021}},{"model":"haiku-medium","name":"Claude Haiku 5.5 (medium)","short":"Haiku 5.5 medium","effort":"medium","family":"haiku","fam":"Haiku 5.5","raw":{"n":6,"clutter":32.5,"blue":62.5,"orange":10.1,"red":27.4,"words":66,"usd":0.00685,"sec":500,"out":12998},"simple":{"n":6,"clutter":0,"blue":99.6,"orange":0,"red":0,"words":9,"usd":0.00397,"sec":168,"out":7181}},{"model":"sonnet-high","name":"Claude Sonnet 5.5 (high)","short":"Sonnet 5.5 high","effort":"high","family":"sonnet","fam":"Sonnet 5.5","raw":{"n":6,"clutter":40.9,"blue":51.3,"orange":13.9,"red":34,"words":69,"usd":0.11134,"sec":95,"out":10429},"simple":{"n":6,"clutter":0,"blue":87.9,"orange":0,"red":0,"words":13,"usd":0.0523,"sec":52,"out":4462}},{"model":"sonnet-xhigh","name":"Claude Sonnet 5.5 (xhigh)","short":"Sonnet 5.5 xhigh","effort":"xhigh","family":"sonnet","fam":"Sonnet 5.5","raw":{"n":6,"clutter":42.4,"blue":52.4,"orange":9.6,"red":37.7,"words":110,"usd":0.23297,"sec":497,"out":22593},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":14,"usd":0.06704,"sec":598,"out":5936}},{"model":"opus-high","name":"Claude Opus 5.5 (high)","short":"Opus 5.5 high","effort":"high","family":"opus","fam":"Opus 5.5","raw":{"n":6,"clutter":48.9,"blue":42.8,"orange":16.2,"red":40.8,"words":123,"usd":0.26173,"sec":128,"out":12382},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":16,"usd":0.17191,"sec":98,"out":7827}},{"model":"opus-medium","name":"Claude Opus 5.5 (medium)","short":"Opus 5.5 medium","effort":"medium","family":"opus","fam":"Opus 5.5","raw":{"n":6,"clutter":56.5,"blue":23.4,"orange":40,"red":36.5,"words":85,"usd":0.20167,"sec":106,"out":9379},"simple":{"n":6,"clutter":0.2,"blue":99.8,"orange":0,"red":0.2,"words":18,"usd":0.11424,"sec":65,"out":4944}},{"model":"astra","name":"GPT-6 Astra (low)","short":"Astra low","effort":"low","family":"astra","fam":"Astra","raw":{"n":6,"clutter":63.7,"blue":36.3,"orange":0,"red":63.7,"words":77,"usd":0.18122,"sec":117,"out":3181},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":13,"usd":0.1255,"sec":80,"out":2022}},{"model":"luna","name":"GPT-6 Luna (low)","short":"Luna low","effort":"low","family":"luna","fam":"Luna","raw":{"n":6,"clutter":66,"blue":33.9,"orange":0.2,"red":65.9,"words":86,"usd":0.00209,"sec":48,"out":3741},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":11,"usd":0.00106,"sec":31,"out":1638}},{"model":"sol","name":"GPT-6.1 Sol (low)","short":"Sol low","effort":"low","family":"sol","fam":"Sol","raw":{"n":6,"clutter":66.8,"blue":32.5,"orange":1.4,"red":66.1,"words":87,"usd":0.03745,"sec":107,"out":3302},"simple":{"n":6,"clutter":0,"blue":100,"orange":0,"red":0,"words":13,"usd":0.02334,"sec":72,"out":1846}},{"model":"argon","name":"Gemini 4 Argon (low)","short":"Argon low","effort":"low","family":"argon","fam":"Argon","raw":{"n":6,"clutter":70.8,"blue":24.6,"orange":8.3,"red":66.6,"words":307,"usd":0.49962,"sec":510,"out":49487},"simple":{"n":6,"clutter":14.5,"blue":85.4,"orange":0.4,"red":14.3,"words":43,"usd":0.30531,"sec":358,"out":30011}}],"runs":[{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__argon__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"argon","variant":"raw","clutter":73.5,"end":80.4,"shares":{"blue":26.1,"orange":0,"red":73.5},"words":238,"usd":0.425608,"sec":436,"out":42302,"summary":"The +5% result is buried in the third of three dense panels, among dataset stats, process cards, a decorative waveform and boilerplate; nothing important is missing.","motion":["Three panels build sequentially, so the key result arrives at about 6s.","Many small elements animate in at once, such as the grid and waveform.","Waveform bars pulse decoratively.","Main visual appears late and tiny among the other panels."],"cut":10,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__argon__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"argon","variant":"simple","clutter":84.3,"end":70.9,"shares":{"blue":15.7,"orange":0,"red":84.3},"words":79,"usd":0.628992,"sec":569,"out":62595,"summary":"The before/after bars sit in a third panel buried beside a corpus-stats panel and a method diagram, and the +5% stat repeats the bars.","motion":["Three panels build in sequence, so the main result arrives late.","Many small tiles and ticks animate at once in the first panel."],"cut":4,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__astra__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"astra","variant":"raw","clutter":65,"end":60.1,"shares":{"blue":35,"orange":0,"red":65},"words":72,"usd":0.16533,"sec":164,"out":3062,"summary":"A big sentence headline, eyebrow, subtitle, feedback loop and footer tagline clutter an otherwise clear three-card flow; the +5% card lacks a baseline.","motion":["Cards arrive one by one, so the +5% result lands late at 4s","Feedback loop and footer appear after the main visual is complete"],"cut":9,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__astra__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":12,"usd":0.13206,"sec":117,"out":2352,"summary":"Clean and complete: a two-bar output-quality comparison with a short title, labels, and a +5% change value.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__haiku-high__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"haiku-high","variant":"raw","clutter":63.5,"end":64.9,"shares":{"blue":34.8,"orange":3.3,"red":61.8},"words":62,"usd":0.0158635,"sec":165,"out":31345,"summary":"The headline is fine, but the unlabelled grid, dataset stats, subtitle and footnote bury the +5% result, and nothing links the headline to it.","motion":["Stats appear one after another after the grid, with no single focus.","Grid fills row by row while headline and subtitle also animate in."],"cut":5,"shrunk":1},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__haiku-high__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":11,"usd":0.0083869,"sec":91,"out":16328,"summary":"Clean and nearly complete: a simple before-and-after bar chart with a +5% annotation, though it does not say the gain came from harness changes.","motion":["After bar dips low before rising above Before, which reads as a confusing dip"],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__haiku-medium__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"haiku-medium","variant":"raw","clutter":12.9,"end":18.1,"shares":{"blue":79.9,"orange":14.3,"red":5.8},"words":73,"usd":0.009933,"sec":112,"out":19484,"summary":"The 5,161 counter is the main visual, but a grid, three extra stat blocks and a partner tag add clutter.","motion":["Digit-roll counter is busy and spins digits at once","Grid squares fill while other elements are present","Stat blocks arrive sequentially, pulling attention from main number","Counter glitchy mid-roll with clipped digits"],"cut":5,"shrunk":1},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__haiku-medium__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.0059248999999999994,"sec":65,"out":11404,"summary":"Clean and focused: one before-after bar chart with a +5% annotation and logo; the only gap is that the quality metric has no stated unit or source.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__haiku__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"haiku","variant":"raw","clutter":38.2,"end":46.5,"shares":{"blue":50.6,"orange":14.9,"red":30.8},"words":64,"usd":0.004024,"sec":37,"out":7666,"summary":"A text-heavy stat slide with a decorative square grid, extra stats and takeaway line; the +5% result is small and arrives late.","motion":["Squares fill in one by one for a long stretch","Several stat blocks arrive in staggered sequence with no single focus","Key +5% result appears late, near the end"],"cut":5,"shrunk":1},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__haiku__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":5,"usd":0.0034419,"sec":36,"out":6438,"summary":"Clean and complete: a simple before/after bar chart with a metric title, a +5% change label, and one logo; nothing needs removing. The benchmark is not named on the graphic, but the metric is clear.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__luna__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"luna","variant":"raw","clutter":92,"end":89.2,"shares":{"blue":7.3,"orange":1.4,"red":91.2},"words":89,"usd":0.0027213000000000003,"sec":70,"out":5198,"summary":"The key +5% result is a tiny, late element buried among a slogan, stats, decorative orbits and process cards; the flow is clear enough.","motion":["Many cards and dots appear in sequence, so attention is split.","Result bar arrives last, at 5s, and is tiny.","Moving decorative dots distract throughout.","Shared review card overlaps the result bar as it appears."],"cut":7,"shrunk":1},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__luna__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":3,"usd":0.0006776,"sec":42,"out":1066,"summary":"Clean with nothing to remove, but a lone filling bar with no baseline, benchmark or product name leaves the +5% poorly anchored.","motion":["Output quality label appears late, after the bar finishes filling"],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__opus-high__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"opus-high","variant":"raw","clutter":20.9,"end":18,"shares":{"blue":64.5,"orange":29.3,"red":6.3},"words":268,"usd":0.20482,"sec":103,"out":9859,"summary":"A dot grid, pipeline legend, eyebrow, corner tag and footer brand crowd the scale stats and the +5% result, and the grid is never explained.","motion":["Dense dot grid fills in while counters roll simultaneously","Digit-roller counters are hard to read mid-roll","The +5% result arrives late, at 5.5s"],"cut":5,"shrunk":2},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__opus-high__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.239096,"sec":146,"out":11509,"summary":"Clean and complete: a dot grid fills while a counter reaches 5,161 legal cases, with a small logo and benchmark tag.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__opus-medium__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"opus-medium","variant":"raw","clutter":73.7,"end":77.3,"shares":{"blue":10.8,"orange":31,"red":58.2},"words":58,"usd":0.16008,"sec":87,"out":7622,"summary":"A decorative document grid and pipeline diagram plus a tag and sentence headline crowd the stats and the late, small +5% result; nothing essential is missing.","motion":["Document grid and pipeline animate for effect while stats count up","Key +5% result arrives late and is small","Many small items pop in staggered, no clear focal point"],"cut":3,"shrunk":2},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__opus-medium__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"opus-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":10,"usd":0.154576,"sec":85,"out":7283,"summary":"Clean and complete: a counting number fills a case grid, with a short unit caption and logo.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__opus__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"opus","variant":"raw","clutter":55.9,"end":62.6,"shares":{"blue":42.9,"orange":0,"red":55.9},"words":49,"usd":0.06654,"sec":41,"out":2945,"summary":"A decorative icon grid, dataset stat column and explanatory sentence crowd out the only result (+5%), which arrives late and small.","motion":["Icon grid fills in with many small elements for no data purpose","Stats appear in sequence while the grid builds, splitting attention","The +5% result arrives late, after 5s"],"cut":4,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__opus__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":8,"usd":0.034356,"sec":20,"out":1272,"summary":"Clean and complete: one dot grid with a short count and caption, and the caption is small.","motion":["Grid fill is slow and runs about 5 seconds before it completes."],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sol__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sol","variant":"raw","clutter":79.9,"end":71,"shares":{"blue":20.1,"orange":0,"red":79.9},"words":94,"usd":0.039876,"sec":133,"out":3743,"summary":"Three method cards, a slogan headline and a category tag clutter the clear +5% result, and the baseline for that 5% is not stated.","motion":["Three method cards build in sequence before the main result appears","Main 5% result arrives at 4.3s, late in the clip","Cards and arrows dominate the first four seconds, pulling focus from the result"],"cut":3,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sol__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":13,"usd":0.020402,"sec":105,"out":1751,"summary":"Clean and complete: one bar-growth chart with a short metric title, labels, and a +5% gain annotation.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet-high__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet-high","variant":"raw","clutter":8,"end":16.1,"shares":{"blue":87.3,"orange":7,"red":4.5},"words":34,"usd":0.07152,"sec":68,"out":6770,"summary":"A sound benchmark-scale visual is diluted by a late, unrelated +5% callout and a wordy subtitle, though it is otherwise complete.","motion":["Counters count up while cards color-flash, two things moving at once","+5% callout arrives late, shifting attention from the grid"],"cut":2,"shrunk":1},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet-high__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":28.6,"orange":0,"red":0},"words":6,"usd":0.045478,"sec":65,"out":4102,"summary":"Clean single-point graphic of a dot grid and a counter, though the caption is tiny and BAR is never named.","motion":["Large animated counter and dot fill have no benchmark name for context"],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet-medium__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet-medium","variant":"raw","clutter":26.1,"end":28.8,"shares":{"blue":73.9,"orange":0,"red":26.1},"words":36,"usd":0.0385,"sec":30,"out":3468,"summary":"A title is surrounded by a decorative tile grid, a stats row and a +5% callout that serve other points; it lacks a real main visual.","motion":["Tiles fill in green with no data meaning, drawing the eye","Stats and +5% callout appear sequentially, pulling attention away from headline"],"cut":3,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet-medium__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":5,"usd":0.027568,"sec":37,"out":2311,"summary":"Clean and nearly complete: a tile grid counts up to 5,161 cases, but the benchmark name is absent and the tile shading is unexplained.","motion":["Tile shading shifts from pale to dark green with no stated meaning"],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet-xhigh__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet-xhigh","variant":"raw","clutter":48.7,"end":35.8,"shares":{"blue":28,"orange":46.6,"red":25.4},"words":253,"usd":0.33833,"sec":281,"out":33451,"summary":"The graphic is cluttered by an eyebrow, a date tag, a subtitle, an intro diagram and a decorative icon grid around a row of stats. It lacks nothing essential.","motion":["Intro diagram and icon grid animate before the stats, delaying the main content.","Icon grid fills with checks while stat counters spin at the same time.","Digit rolling counters are busy and briefly garbled mid-count.","Title text reveals while a panel animates beside it."],"cut":4,"shrunk":2},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet-xhigh__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.005898,"sec":842,"out":144,"summary":"Clean single-point dot-fill counter; the label is tiny and the grid edge slightly ragged, but nothing needs removing.","motion":[],"cut":0,"shrunk":0},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet__raw","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet","variant":"raw","clutter":30.7,"end":36.4,"shares":{"blue":50,"orange":36.2,"red":12.6},"words":30,"usd":0.02898,"sec":24,"out":2516,"summary":"A grid plus a stat row make several parallel points, with an eyebrow, a vague headline and an unrelated +5% stat adding clutter; nothing essential is missing.","motion":["Tiles pop and pulse in staggered waves with no data meaning","Stat blocks arrive one after another, late in the sequence","Grid stays unlabelled until 2.5s"],"cut":3,"shrunk":1},{"id":"building-a-frontier-legal-evaluation-in-partnership-with-legora__sonnet__simple","task":"building-a-frontier-legal-evaluation-in-partnership-with-legora","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":5,"usd":0.032848,"sec":27,"out":2839,"summary":"The graphic is clean with almost no clutter, but it never names the benchmark (BAR) or says these are legal cases, and the tiles are unlabeled.","motion":["Tiles appear one by one over four seconds, which is slow","Counter reaches 5,161 while tiles keep appearing, so the two don't match"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__argon__raw","task":"clutterbench-v1-results","model":"argon","variant":"raw","clutter":3.4,"end":5.4,"shares":{"blue":96.1,"orange":0,"red":3.4},"words":64,"usd":0.241922,"sec":273,"out":23884,"summary":"Mostly clean chart, but a words-per-graphic column adds a second metric and the per-value raw/simple tags repeat the legend.","motion":["Dotted arrows and many pills animate in staggered sequence, slightly busy"],"cut":1,"shrunk":0},{"id":"clutterbench-v1-results__argon__simple","task":"clutterbench-v1-results","model":"argon","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":37,"usd":0.126036,"sec":152,"out":12250,"summary":"Clean and complete: one bar chart with a title, legend, labelled rows, values and one brand mark, with nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__astra__raw","task":"clutterbench-v1-results","model":"astra","variant":"raw","clutter":22.6,"end":20.8,"shares":{"blue":66.4,"orange":21.9,"red":11.6},"words":53,"usd":0.13532,"sec":88,"out":2418,"summary":"A mostly clean dumbbell chart, with a sentence-style slogan headline and a redundant subtitle that could go.","motion":["Chart sits static for the first 3 seconds before the lines draw.","Raw dots appear at once, so the before-and-after change comes late."],"cut":1,"shrunk":1},{"id":"clutterbench-v1-results__astra__simple","task":"clutterbench-v1-results","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":24,"usd":0.12115,"sec":76,"out":2090,"summary":"Clean and complete: a single dumbbell chart with a short title, legend, model labels and values, with no extras to remove.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__haiku-high__raw","task":"clutterbench-v1-results","model":"haiku-high","variant":"raw","clutter":2.4,"end":8.2,"shares":{"blue":97.6,"orange":0,"red":2.4},"words":65,"usd":0.0139358,"sec":267,"out":27451,"summary":"Nearly clean: one late callout box restates the 6% Luna result, and the chart is otherwise complete with a named metric and unit key.","motion":["Red highlight box arrives last, adding emphasis after the result is already readable"],"cut":1,"shrunk":0},{"id":"clutterbench-v1-results__haiku-high__simple","task":"clutterbench-v1-results","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":29,"usd":0.0095457,"sec":104,"out":18607,"summary":"Clean and complete: a single bar chart with a metric name, prompt titles, labelled bars and one logo; nothing needs removing, and the reader is left to infer what the prompts asked for.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__haiku-medium__raw","task":"clutterbench-v1-results","model":"haiku-medium","variant":"raw","clutter":5.5,"end":5.6,"shares":{"blue":91.8,"orange":5.5,"red":2.7},"words":92,"usd":0.0003693,"sec":826,"out":318,"summary":"A clear square-grid chart is cluttered a little by the date tag, wordy headline and subtitle, prompt captions and URL footer, but it lacks nothing.","motion":["Raw prompt text swaps for simple text while the squares are changing.","Grid squares empty row by row in several places at once."],"cut":2,"shrunk":1},{"id":"clutterbench-v1-results__haiku-medium__simple","task":"clutterbench-v1-results","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":24,"usd":0.009671200000000001,"sec":93,"out":18858,"summary":"Clean and uncluttered, but it never states that the percentages measure clutter (lower is better).","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__haiku__raw","task":"clutterbench-v1-results","model":"haiku","variant":"raw","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":72,"usd":0.0038323000000000003,"sec":44,"out":7244,"summary":"Clean and complete: a single bar chart with a question headline, metric subtitle and quiet footnote, with nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__haiku__simple","task":"clutterbench-v1-results","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":29,"usd":0.0049607,"sec":51,"out":9437,"summary":"Clean and complete: one bar chart with a prompt label, values and one brand mark.","motion":["Value labels swap abruptly while the bars shrink at the same time"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__luna__raw","task":"clutterbench-v1-results","model":"luna","variant":"raw","clutter":16.5,"end":14.2,"shares":{"blue":75.2,"orange":16.8,"red":8.1},"words":62,"usd":0.0016017,"sec":77,"out":2915,"summary":"A solid two-bar comparison is surrounded by an eyebrow, a sentence headline, a subtitle, unexplained dashes, a takeaway footer and a second logo icon; it lacks only a clear metric title.","motion":["Elements stagger in one by one over almost three seconds.","Main visual arrives late, after the headline and subtitle."],"cut":5,"shrunk":1},{"id":"clutterbench-v1-results__luna__simple","task":"clutterbench-v1-results","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":13,"usd":0.0011355,"sec":27,"out":1938,"summary":"Clean and uncluttered, but the bars never render and the 39% and 2% figures lack any stated metric or title.","motion":["Bars never visibly grow; values appear without the data mark they describe"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__opus-high__raw","task":"clutterbench-v1-results","model":"opus-high","variant":"raw","clutter":0,"end":0,"shares":{"blue":98.8,"orange":0,"red":0},"words":52,"usd":0.152216,"sec":84,"out":7182,"summary":"Nearly clean: one chart with a legend, logo and footnote; only the title could be shorter.","motion":["Simple bars appear after the raw bars, so the comparison is delayed","Footnote appears late, at 5.8s"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__opus-high__simple","task":"clutterbench-v1-results","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":31,"usd":0.137352,"sec":74,"out":6375,"summary":"Clean and complete: a single bar chart with a metric title, legend, model labels and values, and one brand mark.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__opus-medium__raw","task":"clutterbench-v1-results","model":"opus-medium","variant":"raw","clutter":0,"end":0,"shares":{"blue":98.3,"orange":0,"red":0},"words":39,"usd":0.162036,"sec":79,"out":7673,"summary":"Clean and focused, but the metric is never named, so the percentages don't say what they measure.","motion":["Raw bars lose emphasis, then simple dots slide in, which makes the sequence slow","Raw bars appear before the simple legend entry, so the comparison arrives late"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__opus-medium__simple","task":"clutterbench-v1-results","model":"opus-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":35,"usd":0.102332,"sec":65,"out":4624,"summary":"Clean and complete: one bar chart with labels, values and a raw-to-simple transition.","motion":["Raw bars grow slowly before the swap, delaying the simple-prompt result."],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__opus__raw","task":"clutterbench-v1-results","model":"opus","variant":"raw","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":32,"usd":0.057216,"sec":41,"out":2432,"summary":"Clean and complete: a single bar chart with a short title, row labels, values, one legend and one logo.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__opus__simple","task":"clutterbench-v1-results","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":26,"usd":0.057412,"sec":31,"out":2378,"summary":"Clean and complete: a bar chart with a wordmark, a short title, and a prompt tag, with nothing that needs removing.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__sol__raw","task":"clutterbench-v1-results","model":"sol","variant":"raw","clutter":16.1,"end":16.3,"shares":{"blue":83.9,"orange":0,"red":16.1},"words":49,"usd":0.030134,"sec":146,"out":2725,"summary":"A clean dumbbell chart, with a slogan headline, a subtitle and a late 'Keep it simple' tag that could go.","motion":["Simple dots start at the raw positions and slide left, which is fine, but the tag appears separately"],"cut":2,"shrunk":0},{"id":"clutterbench-v1-results__sol__simple","task":"clutterbench-v1-results","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":24,"usd":0.02491,"sec":79,"out":2158,"summary":"Clean and complete: a single dumbbell chart with a metric title, legend, row labels and value labels; nothing to remove, and the benchmark name is absent but the metric is clear.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__sonnet-high__raw","task":"clutterbench-v1-results","model":"sonnet-high","variant":"raw","clutter":9.2,"end":7.9,"shares":{"blue":81.5,"orange":18.5,"red":0},"words":71,"usd":0.075646,"sec":70,"out":7144,"summary":"Mostly clean grid chart; the sentence headline is verbose and the small 'was' labels repeat earlier values.","motion":["Raw-to-simple transition swaps all grids at once; the before values vanish quickly","Footnote appears only in final 1.2 seconds, too brief to read"],"cut":0,"shrunk":1},{"id":"clutterbench-v1-results__sonnet-high__simple","task":"clutterbench-v1-results","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":23,"usd":0.055814,"sec":67,"out":5097,"summary":"Clean and complete: bars, model names, values and one short chart title, with nothing to remove; the benchmark itself is never named on screen.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__sonnet-medium__raw","task":"clutterbench-v1-results","model":"sonnet-medium","variant":"raw","clutter":2.7,"end":9,"shares":{"blue":97.3,"orange":0,"red":2.7},"words":58,"usd":0.042766,"sec":39,"out":3856,"summary":"Clean and complete except the bottom takeaway line, which restates the chart's values and should be removed.","motion":["About one second of blank chart between raw and simple phases","Raw bars fade to pale and values vanish before simple bars appear"],"cut":1,"shrunk":0},{"id":"clutterbench-v1-results__sonnet-medium__simple","task":"clutterbench-v1-results","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":28,"usd":0.031394,"sec":92,"out":2655,"summary":"Clean and complete: a single dumbbell chart with a legend and metric label, with no clutter. The only weakness is that the metric tag is small and sits at the top right, away from the chart.","motion":["Raw-prompt red dots appear only after about 5s, leaving under 3s to compare"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__sonnet-xhigh__raw","task":"clutterbench-v1-results","model":"sonnet-xhigh","variant":"raw","clutter":14,"end":16.4,"shares":{"blue":71.7,"orange":27.9,"red":0},"words":229,"usd":0.005576,"sec":883,"out":137,"summary":"Mostly clean waffle-grid comparison; the headline sentence and subtitle are wordier than needed, and nothing important is missing.","motion":["Squares for all four models fill and count simultaneously, busy","Counters change fast, hard to read mid-transition","Toggle switches before reader can absorb raw values"],"cut":0,"shrunk":1},{"id":"clutterbench-v1-results__sonnet-xhigh__simple","task":"clutterbench-v1-results","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":25,"usd":0.127934,"sec":113,"out":12309,"summary":"Clean and complete: a single dumbbell chart with a brand mark, legend and metric label, with nothing to remove; the benchmark and metric are identifiable from the wordmark and axis label.","motion":[],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__sonnet__raw","task":"clutterbench-v1-results","model":"sonnet","variant":"raw","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":59,"usd":0.031726,"sec":34,"out":2752,"summary":"Nearly clean, with one bar chart and short labels; only the sentence-style headline could be shorter.","motion":["Footnote appears late, after the 6.4s mark","Bar values swap abruptly at 5s, with no transition marking the change"],"cut":0,"shrunk":0},{"id":"clutterbench-v1-results__sonnet__simple","task":"clutterbench-v1-results","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":30,"usd":0.034494,"sec":30,"out":2965,"summary":"Clean and complete: one bar chart that animates from raw-prompt clutter to near-zero simple-prompt clutter, with a short header and prompt label.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__argon__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"argon","variant":"raw","clutter":57.5,"end":63.9,"shares":{"blue":42,"orange":0,"red":57.5},"words":375,"usd":0.600608,"sec":553,"out":59615,"summary":"The warmup bar chart is buried among spec cards, a pipeline diagram, a second chart, callouts and footers; the information is otherwise complete.","motion":["Many panels and elements appear in staggered sequence across 8 seconds","Main chart arrives late, after pipeline panel","Second chart and footer continue appearing after main chart","Text and chart too small and fast to read"],"cut":9,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__argon__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"argon","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":34,"usd":0.173422,"sec":220,"out":16851,"summary":"Clean single bar chart; the only issues are tiny faint labels and out-of-order bar reveal.","motion":["Bars appear out of order (Student, Teacher, No warmup, Warmup), which breaks the reading order."],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__astra__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"astra","variant":"raw","clutter":66.9,"end":61.2,"shares":{"blue":33.1,"orange":0,"red":66.9},"words":85,"usd":0.17417,"sec":134,"out":3060,"summary":"Slogan headline, pipeline cards, eyebrow tags and a repeated 2.8-point callout clutter a clear before/after warmup chart; the score lacks a metric unit.","motion":["Pipeline cards and arrows enter before the main numbers appear","Main visual arrives at 4s, late in an 8s clip","Headline and pipeline occupy the first half with no data"],"cut":6,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__astra__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":13,"usd":0.12345,"sec":104,"out":2001,"summary":"Clean and complete: a two-bar GDPval comparison with benchmark context, labelled bars and values, and no extras.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__haiku-high__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"haiku-high","variant":"raw","clutter":1.1,"end":1.1,"shares":{"blue":98.9,"orange":0,"red":1.1},"words":45,"usd":0.024785099999999997,"sec":231,"out":48898,"summary":"Mostly clean line chart; the eyebrow tag is unneeded and the sublabels are tiny and wordy.","motion":["Line draws slowly, so the full chart only appears near the end."],"cut":1,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__haiku-high__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.005685,"sec":64,"out":10634,"summary":"Clean and complete: a single before/after bar chart with a metric label, values and category labels; no headline or model name, which is acceptable.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__haiku-medium__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"haiku-medium","variant":"raw","clutter":75.6,"end":72.6,"shares":{"blue":20.9,"orange":7,"red":72.1},"words":52,"usd":0.0003696,"sec":1739,"out":67,"summary":"Mostly clean single chart; only the eyebrow line is unneeded, and nothing important is missing.","motion":["Source and logo appear late, after the chart is done"],"cut":2,"shrunk":1},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__haiku-medium__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.0027685,"sec":36,"out":4801,"summary":"Clean and complete: a two-bar GDPval comparison with a metric label, category labels and values, and nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__haiku__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"haiku","variant":"raw","clutter":10.8,"end":10.1,"shares":{"blue":78.5,"orange":21.5,"red":0},"words":28,"usd":0.0059941000000000005,"sec":72,"out":11316,"summary":"Mostly clean single chart; the sentence headline is wordy and the tiny labels are hard to read.","motion":["Logo appears very late, only at the end","Teacher line and gap annotation arrive late, after the curve is done"],"cut":0,"shrunk":1},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__haiku__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":10,"usd":0.003771,"sec":46,"out":6806,"summary":"Clean and complete: one two-bar chart with metric, labels, values and a single logo.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__luna__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"luna","variant":"raw","clutter":17,"end":20.2,"shares":{"blue":83,"orange":0,"red":17},"words":76,"usd":0.0024322,"sec":64,"out":4441,"summary":"A slogan headline, subtitle, kicker tags, footer, gain text and a tiny pipeline row clutter the chart, which lacks value labels and a clear title.","motion":["Pipeline chips appear at 6.3s, leaving little time to read them","Subtitle fades in while the bars start growing"],"cut":6,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__luna__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.001176,"sec":30,"out":1884,"summary":"Clean bar chart with benchmark tag; only a decorative end ring should go, and it lacks the metric unit and model names.","motion":["Ring marker pulses at bar end after the value is already visible"],"cut":1,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__opus-high__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"opus-high","variant":"raw","clutter":23.6,"end":29.5,"shares":{"blue":62.5,"orange":27.6,"red":9.8},"words":83,"usd":0.232784,"sec":114,"out":10967,"summary":"A clear slope chart is cluttered by a sentence headline, method subtitle and a late BrowseComp/HLE table; nothing essential is missing.","motion":["Headline and chart share the screen while chart still builds","Other-domain rows fade in late, pulling attention from chart","Warmup line value fades in while gap annotation appears"],"cut":2,"shrunk":1},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__opus-high__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":18,"usd":0.18888,"sec":105,"out":8708,"summary":"Clean and complete: one bar chart with a title, labels, values and a single logo.","motion":["Teacher bar appears before the warmup bar, out of order with the row sequence."],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__opus-medium__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"opus-medium","variant":"raw","clutter":18.5,"end":21.2,"shares":{"blue":71.9,"orange":19.4,"red":8.8},"words":75,"usd":0.201624,"sec":115,"out":9409,"summary":"A clear curve chart is cluttered by a slogan headline, eyebrow, subtitle paragraph and a late other-domains box.","motion":["Other-domains box appears at the very end, after the main point is made","Chart is mostly empty for the first second"],"cut":3,"shrunk":1},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__opus-medium__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"opus-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":20,"usd":0.1051,"sec":53,"out":4519,"summary":"Clean, complete single bar chart with a short title, labelled bars and values, and nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__opus__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"opus","variant":"raw","clutter":2.2,"end":7.2,"shares":{"blue":97.8,"orange":0,"red":2.2},"words":52,"usd":0.071364,"sec":33,"out":2896,"summary":"Clean, focused bar build with one restating takeaway line to remove; the chart's values lack a stated metric unit.","motion":[],"cut":1,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__opus__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":15,"usd":0.05436,"sec":27,"out":1982,"summary":"Clean and focused: a single bar chart showing the warmup gain, with only the metric unit and the Student/Teacher roles left unexplained.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sol__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sol","variant":"raw","clutter":70.9,"end":64.7,"shares":{"blue":29.1,"orange":0,"red":70.9},"words":89,"usd":0.037154,"sec":163,"out":3292,"summary":"A slogan headline, a three-card pipeline and a gap callout crowd the small warmup result band, which is the main visual and arrives late. Nothing important is missing.","motion":["Pipeline cards appear one by one before the main result","Main result arrives late, after about 5 seconds"],"cut":5,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sol__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":12,"usd":0.02212,"sec":70,"out":1744,"summary":"Clean and focused on one before-after comparison; the only gap is that the score's metric or unit is never named.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet-high__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet-high","variant":"raw","clutter":5.6,"end":6.6,"shares":{"blue":91.1,"orange":6.8,"red":2.2},"words":57,"usd":0.123012,"sec":99,"out":11629,"summary":"Mostly clean curve chart, cluttered by breadcrumb eyebrow, redundant +11.4 callout and late partner badge; complete.","motion":["Data partner badge appears late, after main visual is done","Big +11.4 callout appears while curve is still drawing"],"cut":2,"shrunk":1},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet-high__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.04366,"sec":34,"out":3630,"summary":"Clean and complete: a single bar chart with a short title, labelled values, and sequential bar growth that makes the warmup comparison.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet-medium__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet-medium","variant":"raw","clutter":1.3,"end":1,"shares":{"blue":98.7,"orange":0,"red":1.3},"words":36,"usd":0.041262,"sec":28,"out":3454,"summary":"Clean and complete: one line chart with a clear finding, and only the eyebrow tag above the headline is clutter.","motion":[],"cut":1,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet-medium__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":12,"usd":0.03695,"sec":31,"out":2959,"summary":"Clean and complete: a single GDPval bar chart with a short title and labelled values; nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet-xhigh__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet-xhigh","variant":"raw","clutter":3.8,"end":4.7,"shares":{"blue":93.5,"orange":5.3,"red":1.1},"words":55,"usd":0.385252,"sec":304,"out":37853,"summary":"Mostly a clean single curve with a wordy subtitle and warmup note, a date tag, and a redundant footer URL; the metric is not named.","motion":["The line draws slowly across 8 seconds, so the key result arrives late."],"cut":2,"shrunk":2},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet-xhigh__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.12523,"sec":110,"out":11787,"summary":"Clean single line chart with no clutter; it lacks only a metric or unit label, and its tiny text is hard to read.","motion":["Points appear slowly one by one, so the full picture arrives late"],"cut":0,"shrunk":0},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet__raw","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet","variant":"raw","clutter":25.9,"end":27.2,"shares":{"blue":67.1,"orange":13.9,"red":19},"words":49,"usd":0.030582,"sec":22,"out":2386,"summary":"A clear bar chart is cluttered by a wordy subtitle and a late big-stat callout repeating the bars; the metric is only implied by 'GDPval'.","motion":["Big stat arrives after chart is complete, pulling attention from bars","Chart starts late, with the left half empty until 2s"],"cut":2,"shrunk":1},{"id":"how-afterquery-helped-nvidia-hill-climb-gdpval__sonnet__simple","task":"how-afterquery-helped-nvidia-hill-climb-gdpval","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":18,"usd":0.02953,"sec":22,"out":2217,"summary":"Clean, single bar chart with labeled bars and values; nothing to remove and nothing important missing.","motion":["Teacher line appears late, only in the final two seconds."],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__argon__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"argon","variant":"raw","clutter":78.6,"end":75.5,"shares":{"blue":14.1,"orange":13.6,"red":71.8},"words":317,"usd":0.67076,"sec":653,"out":66468,"summary":"The 3.1% to 17.0% results chart is a small part of the frame, buried under two method panels, a stat callout, header pills and a task example.","motion":["Three panels build in sequence, so the results chart arrives late, after 5s.","Many small items appear at once across the panels.","Text is tiny and on screen too briefly to read."],"cut":8,"shrunk":3},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__argon__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"argon","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":23,"usd":0.280134,"sec":408,"out":27360,"summary":"Clean and complete: a simple bar chart with title, benchmark, labelled bars, values and one logo, building in sequence.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__astra__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"astra","variant":"raw","clutter":54.7,"end":57.1,"shares":{"blue":45.3,"orange":0,"red":54.7},"words":96,"usd":0.18708,"sec":74,"out":3179,"summary":"The three-card progression is the main visual, but it is cluttered by a slogan headline, eyebrow tags, decorative mini tracks, a takeaway line, a badge and a footer URL.","motion":["Takeaway and badge arrive late after the main data is shown","Cards appear in sequence but take 6s to complete"],"cut":7,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__astra__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.12896,"sec":51,"out":1972,"summary":"Clean and complete: a single before-after line chart with a benchmark title, model tag, and labelled endpoints, with the line drawing in to show the gain.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__haiku-high__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"haiku-high","variant":"raw","clutter":2.6,"end":2.3,"shares":{"blue":97.4,"orange":0,"red":2.6},"words":65,"usd":0.013043899999999999,"sec":135,"out":25212,"summary":"Mostly clean line chart; the top-right tag is unneeded and the subtitle and sublabels are wordy.","motion":["Footnote appears late at the end"],"cut":1,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__haiku-high__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":9,"usd":0.0073723,"sec":74,"out":13805,"summary":"Clean and complete: a single bar chart with the benchmark, metric and values, with bars growing in step with their labels.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__haiku-medium__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"haiku-medium","variant":"raw","clutter":10.9,"end":8.1,"shares":{"blue":82.7,"orange":12.8,"red":4.4},"words":55,"usd":0.010459399999999999,"sec":107,"out":20043,"summary":"Mostly clean; the eyebrow tag should go and the subtitle could be shorter, and the tiny labels and thin bar make it hard to read.","motion":["Axis ticks and footnote appear only in the last second, after the data.","Bar grows slowly; the final state is reached only near the end."],"cut":1,"shrunk":1},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__haiku-medium__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":9,"usd":0.004641800000000001,"sec":45,"out":8344,"summary":"Clean and complete: a simple before-and-after bar chart with a named benchmark, metric, labelled values and a logo.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__haiku__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"haiku","variant":"raw","clutter":13.9,"end":25.1,"shares":{"blue":79.8,"orange":12.5,"red":7.7},"words":98,"usd":0.0053379,"sec":53,"out":9800,"summary":"The bar chart is clear and complete, but the bottom RL task-selection chart and pipeline line are extra ideas, and the subtitle is wordier than needed.","motion":["Second RL task-selection chart animates in after the main result, competing for attention","Stage line and footnote appear late, adding a second reading path at the end"],"cut":2,"shrunk":1},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__haiku__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":8,"usd":0.0043218,"sec":43,"out":7704,"summary":"Clean and complete: a simple bar chart with a benchmark title, labelled bars and values, and no extras.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__luna__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"luna","variant":"raw","clutter":93.9,"end":86.3,"shares":{"blue":6.1,"orange":0,"red":93.9},"words":108,"usd":0.0020708000000000002,"sec":30,"out":3579,"summary":"A small bar chart is buried under a slogan, subtitle, terminal demo, pipeline cards and repeated closers; nothing essential is missing.","motion":["Terminal lines typing and pipeline cards appearing compete with the chart","Main chart arrives late at about 5s and finishes at 8s","Chart values appear only in the final second","Closing line and footer appear in the last second only"],"cut":8,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__luna__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":11,"usd":0.0010321,"sec":28,"out":1457,"summary":"Clean and complete: a two-bar chart with labelled values, a benchmark eyebrow, and no extras.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__opus-high__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"opus-high","variant":"raw","clutter":47.7,"end":50.8,"shares":{"blue":30.9,"orange":40.2,"red":27.7},"words":99,"usd":0.22204,"sec":94,"out":10218,"summary":"A clear stacked bar is surrounded by a giant repeated stat, stage cards, a callout pill and a sentence headline.","motion":["Big counter swaps values while bar segments also grow","Stat number jumps position between stages","Callout pill appears late after main visual"],"cut":3,"shrunk":1},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__opus-high__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":19,"usd":0.110936,"sec":47,"out":4599,"summary":"Clean and complete: a single bar chart with a named benchmark, labelled values, and a small logo.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__opus-medium__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"opus-medium","variant":"raw","clutter":58.8,"end":51.9,"shares":{"blue":22.2,"orange":38.1,"red":39.8},"words":108,"usd":0.2123,"sec":112,"out":9731,"summary":"The bar chart is sound, but the pipeline cards, eyebrow, subtitle and the extra claim in the footer crowd it.","motion":["Stage cards animate alongside the chart, splitting attention","Final value 17.0% and footnote arrive late","Many small pieces animate in the RL card"],"cut":7,"shrunk":1},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__opus-medium__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"opus-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":19,"usd":0.119636,"sec":73,"out":5034,"summary":"Clean and complete: a single stacked bar chart with a short title, logo and labelled stages, with nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__opus__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"opus","variant":"raw","clutter":25.1,"end":24,"shares":{"blue":66.5,"orange":14.3,"red":18},"words":53,"usd":0.06264,"sec":28,"out":2248,"summary":"Clean bar chart carrying the point; the eyebrow and two stage cards are extras to cut, the subtitle should be shortened, and the metric is never named.","motion":["Stage cards slide in alongside bars, adding a second thing to read"],"cut":3,"shrunk":1},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__opus__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.060436,"sec":27,"out":2074,"summary":"Clean and complete: a single bar chart with a title, stage labels and values, with no extra elements.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sol__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sol","variant":"raw","clutter":57.1,"end":54.3,"shares":{"blue":42.9,"orange":0,"red":57.1},"words":100,"usd":0.036856,"sec":72,"out":3123,"summary":"The three stat cards carry the story, but the slogan headline, subtitle, tag, terminal text, method chips, badge and takeaway line clutter it.","motion":["Cards arrive staggered; the main result appears late at 4.3s","Pipeline chips animate inside card while other things appear"],"cut":8,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sol__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":11,"usd":0.024832,"sec":47,"out":1876,"summary":"Clean and complete: a single stacked bar chart that builds stage by stage from 3.1% to 17.0% with the benchmark and metric named.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet-high__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet-high","variant":"raw","clutter":73.9,"end":66.6,"shares":{"blue":24.8,"orange":0,"red":73.9},"words":109,"usd":0.11897,"sec":90,"out":11013,"summary":"Big repeated stat numbers and four rotating caption blocks bury a good stacked bar chart; nothing important is missing.","motion":["Large counting numbers and captions change together with the bar, splitting attention","Caption text swaps four times, fading before it can be read","Final footnote appears only at the end"],"cut":7,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet-high__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":16,"usd":0.042808,"sec":38,"out":3333,"summary":"Clean and complete: a single bar chart with a short title, one value per bar and labelled stages, with nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet-medium__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet-medium","variant":"raw","clutter":36.4,"end":27.1,"shares":{"blue":38.8,"orange":44.7,"red":14},"words":46,"usd":0.04755,"sec":44,"out":3871,"summary":"A clear three-bar chart is slightly cluttered by a sentence-style headline and an explanatory subtitle restating the chart.","motion":["Chart bars appear late; first 2 seconds show only the headline","Bars drawn one at a time with long gaps"],"cut":1,"shrunk":1},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet-medium__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":96.3,"orange":0,"red":0},"words":14,"usd":0.032248,"sec":28,"out":2277,"summary":"Clean and complete: a single bar chart with a short title, labelled bars and values, and growing bars that carry the data.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet-xhigh__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet-xhigh","variant":"raw","clutter":43.4,"end":28.4,"shares":{"blue":56.1,"orange":0,"red":43.4},"words":115,"usd":0.3386,"sec":266,"out":32976,"summary":"A counting number and track show the gain, but swapping stage captions, the Over 5x callout, date tag and footer tagline add clutter.","motion":["Right-hand text swaps four times while the number counts","Stage text on screen too briefly to read","Final result appears only in last seconds"],"cut":6,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet-xhigh__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":13,"usd":0.010118,"sec":754,"out":64,"summary":"Clean and complete: a single bar chart with a benchmark title, a logo, and staggered bars that carry the data.","motion":[],"cut":0,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet__raw","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet","variant":"raw","clutter":2.8,"end":9.3,"shares":{"blue":95.9,"orange":0,"red":2.8},"words":30,"usd":0.03942,"sec":26,"out":3058,"summary":"Mostly clean chart; the wordy slogan headline should be shortened and the late subtitle removed.","motion":["Subtitle appears late, after the chart is nearly complete","Bars build one at a time, so the full comparison arrives late"],"cut":1,"shrunk":0},{"id":"how-we-improved-terminal-bench-2-with-tinker-and-harbor__sonnet__simple","task":"how-we-improved-terminal-bench-2-with-tinker-and-harbor","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.030698,"sec":22,"out":2122,"summary":"Clean and complete: a single bar chart with a short title, labelled values and categories, and bars that grow in sequence to show the gain.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__argon__raw","task":"improving-qwen35-9b-on-t3-banking","model":"argon","variant":"raw","clutter":75.3,"end":77.4,"shares":{"blue":12.1,"orange":24.6,"red":63},"words":241,"usd":0.299126,"sec":361,"out":29598,"summary":"The pass@5 bar chart is a small part of the canvas, surrounded by method cards, a stat duplicate, a footer recap and a second capability claim; it is otherwise complete.","motion":["Panels slide in one after another, so the results chart arrives only at 4.5s","Method panels build up for about four seconds before the main visual","Pivot diagram and stage-2 shapes animate at the same time as bars and cards","Many simultaneous moving elements in the 2.5-3.3s window with no single focus"],"cut":12,"shrunk":1},{"id":"improving-qwen35-9b-on-t3-banking__argon__simple","task":"improving-qwen35-9b-on-t3-banking","model":"argon","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":31,"usd":0.13905,"sec":197,"out":13545,"summary":"Clean and complete: a single progressive line chart with a clear title, labelled stages and values, and nothing extraneous.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__astra__raw","task":"improving-qwen35-9b-on-t3-banking","model":"astra","variant":"raw","clutter":61.6,"end":58.8,"shares":{"blue":38.4,"orange":0,"red":61.6},"words":59,"usd":0.18392,"sec":87,"out":3388,"summary":"A solid before-and-after main visual is cluttered by a slogan, decorative waves, and two extra panels, and it lacks nothing needed.","motion":["Decorative waveform curves drift and cross throughout, competing with the data","Orange ring and dot pulse on the flow line with no data meaning","Process panel appears before the 17.5% result it leads to"],"cut":7,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__astra__simple","task":"improving-qwen35-9b-on-t3-banking","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":10,"usd":0.106,"sec":45,"out":1785,"summary":"Clean and complete: a single before-after line chart with a short title, labelled endpoints, and the metric named.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__haiku-high__raw","task":"improving-qwen35-9b-on-t3-banking","model":"haiku-high","variant":"raw","clutter":8.3,"end":18,"shares":{"blue":91.3,"orange":0,"red":8.3},"words":46,"usd":0.011099,"sec":112,"out":21744,"summary":"Clean bar chart with title and metric footnote; the bottom stats row about task counts and pivots is an extra idea, with a kicker, date tag and accent rule to remove, and nothing essential is missing.","motion":["Stats row arrives after the chart and competes with the finished result","'<250' appears alone briefly before the rest of its row","Accent rule and axis line animate without carrying data"],"cut":4,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__haiku-high__simple","task":"improving-qwen35-9b-on-t3-banking","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":8,"usd":0.0049504,"sec":55,"out":9383,"summary":"Clean and complete: a single bar chart with a short title, one value per bar, and the benchmark and metric named.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__haiku-medium__raw","task":"improving-qwen35-9b-on-t3-banking","model":"haiku-medium","variant":"raw","clutter":34.8,"end":34,"shares":{"blue":65.2,"orange":0,"red":34.8},"words":62,"usd":0.008182,"sec":94,"out":15910,"summary":"A large dot-field visual with the 193 stat and a restating subtitle dominate and delay the small bar chart that carries the finding.","motion":["Dot field fills for seconds before the main chart appears","Bar chart arrives at 4.5s, late, after the dot field","Footnote appears only at 7s, barely readable"],"cut":3,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__haiku-medium__simple","task":"improving-qwen35-9b-on-t3-banking","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":8,"usd":0.0003949,"sec":758,"out":272,"summary":"Clean and complete: a single bar chart with a metric and benchmark title, labelled bars and values.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__haiku__raw","task":"improving-qwen35-9b-on-t3-banking","model":"haiku","variant":"raw","clutter":49,"end":56,"shares":{"blue":19.4,"orange":63.2,"red":17.4},"words":86,"usd":0.005296,"sec":57,"out":10138,"summary":"A clear two-bar chart is cluttered by a sentence headline, eyebrow, date and a side panel of method stats; nothing essential is missing.","motion":["Stats panel rows appear one by one, pulling attention from the bars","Footnote arrives late, after the chart is done","Main bars finish growing only at 2.3s, while the headline is already up"],"cut":3,"shrunk":1},{"id":"improving-qwen35-9b-on-t3-banking__haiku__simple","task":"improving-qwen35-9b-on-t3-banking","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.0026924,"sec":29,"out":4867,"summary":"Clean and complete: a single titled bar chart with labelled bars, values, benchmark and metric.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__luna__raw","task":"improving-qwen35-9b-on-t3-banking","model":"luna","variant":"raw","clutter":68.1,"end":67.6,"shares":{"blue":31.9,"orange":0,"red":68.1},"words":113,"usd":0.0021032,"sec":40,"out":3916,"summary":"The pass-rate chart is clear, but a recipe panel, a pivot diagram, a slogan headline and repeated stats and brand surround it, and it is a small part of the canvas.","motion":["Pivot diagram and bar chart build at the same time, splitting attention","Bars arrive late, after most of the diagram has already appeared","Dashed flow lines sweep in decoratively before the data appears"],"cut":6,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__luna__simple","task":"improving-qwen35-9b-on-t3-banking","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":6,"usd":0.00085,"sec":16,"out":1365,"summary":"Clean single bar chart with no clutter, but it never names the benchmark (τ³-Banking) or the model, so the reader can't tell what was measured.","motion":["After bar fills while the Before bar value is still appearing"],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__opus-high__raw","task":"improving-qwen35-9b-on-t3-banking","model":"opus-high","variant":"raw","clutter":84.6,"end":76.2,"shares":{"blue":15.4,"orange":0,"red":84.6},"words":105,"usd":0.308244,"sec":168,"out":14950,"summary":"The result is a small part of the canvas next to a large pivot-profiling panel, with a slogan headline, subtitle, eyebrow, and a late pipeline strip and footer; the main number is complete but late.","motion":["Long pivot-profiling sequence delays the result until 4.5-6s","Scanning line and thousands of tokens changing compete for attention","Final stage strip appears late, after the result"],"cut":9,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__opus-high__simple","task":"improving-qwen35-9b-on-t3-banking","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":19,"usd":0.13446,"sec":73,"out":6197,"summary":"Clean and complete: one bar chart with a clear title, labelled bars, values and a logo, with the bars growing in sequence to make the point.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__opus-medium__raw","task":"improving-qwen35-9b-on-t3-banking","model":"opus-medium","variant":"raw","clutter":63.5,"end":64.2,"shares":{"blue":9.9,"orange":52.4,"red":37.2},"words":69,"usd":0.162164,"sec":94,"out":7646,"summary":"A large pivot-profiling dot grid, a category tag and giant stat numbers crowd out and delay the small 11.3% to 17.5% bar comparison; nothing essential is missing.","motion":["Results chart arrives late, after about 5 seconds of method grid.","Thousands of dots fill in, giving the eye no single focus.","Big numbers count in while bars also animate, simultaneously restating."],"cut":3,"shrunk":1},{"id":"improving-qwen35-9b-on-t3-banking__opus-medium__simple","task":"improving-qwen35-9b-on-t3-banking","model":"opus-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":17,"usd":0.08028,"sec":45,"out":3488,"summary":"Clean and complete: one bar chart with a title, metric, labelled bars and values, and nothing extra.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__opus__raw","task":"improving-qwen35-9b-on-t3-banking","model":"opus","variant":"raw","clutter":21.1,"end":50.6,"shares":{"blue":78.9,"orange":0,"red":21.1},"words":56,"usd":0.062864,"sec":33,"out":2681,"summary":"A clear two-bar result is cluttered by a giant <250 stat, three method and side-effect cards and a footer URL, and the headline is a slogan sentence.","motion":["Extras arrive one after another after the main chart, pulling attention away","Large <250 stat appears abruptly beside the bars","Cards stagger in late and stay only briefly before the end"],"cut":4,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__opus__simple","task":"improving-qwen35-9b-on-t3-banking","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.03566,"sec":18,"out":1257,"summary":"Clean and complete: a single bar chart with a named benchmark, metric, labelled bars and values.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sol__raw","task":"improving-qwen35-9b-on-t3-banking","model":"sol","variant":"raw","clutter":54.8,"end":54.6,"shares":{"blue":45.2,"orange":0,"red":54.8},"words":73,"usd":0.030624,"sec":63,"out":2772,"summary":"A solid 17.5% result and pivot flow carry the point, but the slogan headline, eyebrow, benchmark pill and three method and dataset cards are clutter, and the flow lines are not explained.","motion":["Pivot flow and cards build up before the 17.5% result arrives at 4.5s","Three stat cards enter in sequence while the flow is still drawing","Headline and subtitle sit static for the whole clip, competing with the animated flow"],"cut":6,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sol__simple","task":"improving-qwen35-9b-on-t3-banking","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":11,"usd":0.02399,"sec":59,"out":2064,"summary":"Clean and complete: a two-bar chart with a short title, labels, values and metric, with nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sonnet-high__raw","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet-high","variant":"raw","clutter":40.4,"end":16.3,"shares":{"blue":58.4,"orange":0,"red":40.4},"words":43,"usd":0.108262,"sec":101,"out":10364,"summary":"Clutter is the method build-up (counters, verifier callout), a dataset subtitle, and a decorative dot field around a clear before-and-after result; the benchmark is named only once, in the top-right tag, and nothing essential is missing.","motion":["Main visual arrives at 4.3s, after about half the runtime","Method counters and verifier callout play before the result, then vanish","Dot field animates throughout with no link to the numbers"],"cut":4,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sonnet-high__simple","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":10,"usd":0.04005,"sec":46,"out":3479,"summary":"Clean and complete: one bar chart with a named benchmark, metric, labelled bars and values.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sonnet-medium__raw","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet-medium","variant":"raw","clutter":41.9,"end":39.8,"shares":{"blue":37.5,"orange":41.1,"red":21.3},"words":45,"usd":0.048582,"sec":42,"out":4396,"summary":"The dot grid and before/after numbers carry the point, but the headline is a long sentence, the category tag, subtitle and stage row are extra, and the dot colors have no key.","motion":["The 11.3% number is removed before 17.5% appears, so the before and after are never on screen together.","Stage row arrives after the result in three steps, adding a second idea late in the sequence."],"cut":3,"shrunk":1},{"id":"improving-qwen35-9b-on-t3-banking__sonnet-medium__simple","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":12,"usd":0.02529,"sec":21,"out":2003,"summary":"Clean and complete: a two-bar chart with a short title, values and labels, and no extras.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sonnet-xhigh__raw","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet-xhigh","variant":"raw","clutter":49.9,"end":50.1,"shares":{"blue":45.9,"orange":5.7,"red":47.1},"words":51,"usd":0.006202,"sec":887,"out":158,"summary":"A large pivot dot-grid animation with 50/193 stats buries and delays the small result bar chart, which is the real point.","motion":["Main result arrives late, after 5s of pivot grid.","Dense grid of hundreds of dots animates and sweeps while nothing else matters.","Footnote appears only at 7.3s, barely readable."],"cut":2,"shrunk":1},{"id":"improving-qwen35-9b-on-t3-banking__sonnet-xhigh__simple","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":13,"usd":0.00675,"sec":799,"out":149,"summary":"Mostly clean single chart; the oversized value labels dominate, and the stacked-segment layout makes the comparison harder to read.","motion":["Second bar appears at 4.3s, but its label and value arrive 2s later","Base bar grows then pauses for long stretches before the next step"],"cut":0,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sonnet__raw","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet","variant":"raw","clutter":1.9,"end":7.1,"shares":{"blue":96.8,"orange":0,"red":1.9},"words":32,"usd":0.027172,"sec":20,"out":2255,"summary":"A clear two-bar chart is cluttered by a bottom stats row of training and eval counts and a slightly wordy headline.","motion":["Stats row appears late, pulling attention away from the chart.","Several elements arrive in a staggered sequence over 8 seconds."],"cut":1,"shrunk":0},{"id":"improving-qwen35-9b-on-t3-banking__sonnet__simple","task":"improving-qwen35-9b-on-t3-banking","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":10,"usd":0.02024,"sec":17,"out":1498,"summary":"Clean and complete: a single before-after bar chart with the benchmark, metric and values, and nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__argon__raw","task":"improving-qwen38-27b-on-deepswe","model":"argon","variant":"raw","clutter":67.2,"end":71.6,"shares":{"blue":32.4,"orange":0,"red":67.2},"words":267,"usd":0.334038,"sec":413,"out":33110,"summary":"The DeepSWE result is buried in a three-panel dashboard of data, method, diagnostics, protocol and footer extras, with tiny text; the essential numbers are present.","motion":["Three panels build in sequence, main result arrives at 4s","Many small elements animate in at once in left panels","Main chart appears late and leaves little time to read"],"cut":9,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__argon__simple","task":"improving-qwen38-27b-on-deepswe","model":"argon","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":25,"usd":0.250182,"sec":290,"out":24679,"summary":"Clean and complete: one bar chart with a short title, labelled values, a small change annotation and a logo.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__astra__raw","task":"improving-qwen38-27b-on-deepswe","model":"astra","variant":"raw","clutter":68.4,"end":64.5,"shares":{"blue":31.6,"orange":0,"red":68.4},"words":85,"usd":0.17797,"sec":85,"out":3292,"summary":"The result is a small part of the canvas, buried beside a pipeline diagram, slogan, tag, callout and tagline; nothing essential is missing.","motion":["Pipeline animation runs for 4s while the result is delayed","Result 39.1% arrives late, after about 4s","Several elements (cards, funnel lines, arrow) animate at once","Late-appearing delta and tagline pull attention after the main result"],"cut":4,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__astra__simple","task":"improving-qwen38-27b-on-deepswe","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":15,"usd":0.12225,"sec":63,"out":2133,"summary":"Clean and complete: a simple before/after bar chart with benchmark, metric, model, and values all present.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__haiku-high__raw","task":"improving-qwen38-27b-on-deepswe","model":"haiku-high","variant":"raw","clutter":60.6,"end":73.1,"shares":{"blue":39.4,"orange":0,"red":60.6},"words":84,"usd":0.0003497,"sec":861,"out":283,"summary":"The result bars are small and arrive late, buried beside a dominant dot grid of training-data composition, reward and entropy cards, a subtitle, an eyebrow and a footer; nothing essential is missing.","motion":["Dot grid fills for seconds before the result bars appear","Main result arrives late, at about 4-6 seconds","Reward and entropy cards appear after the main visual with no purpose"],"cut":6,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__haiku-high__simple","task":"improving-qwen38-27b-on-deepswe","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":7,"usd":0.0060051,"sec":75,"out":11530,"summary":"Clean and complete: a single bar chart with a short title, labelled bars and values, and nothing extra to remove.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__haiku-medium__raw","task":"improving-qwen38-27b-on-deepswe","model":"haiku-medium","variant":"raw","clutter":19.1,"end":32,"shares":{"blue":80.9,"orange":0,"red":19.1},"words":86,"usd":0.011280700000000001,"sec":139,"out":22145,"summary":"A clear bar chart is cluttered by a three-card diagnostics row, tag and footer; nothing essential is missing.","motion":["Bars grow while the stat row below also builds in, splitting attention","Footer text appears only in the last second, unreadable","Many small elements stagger in after the main chart is complete"],"cut":6,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__haiku-medium__simple","task":"improving-qwen38-27b-on-deepswe","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":8,"usd":0.0046516000000000005,"sec":47,"out":8823,"summary":"Clean and complete: a single before-after bar chart with the benchmark, metric and values.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__haiku__raw","task":"improving-qwen38-27b-on-deepswe","model":"haiku","variant":"raw","clutter":56.3,"end":66.5,"shares":{"blue":43.7,"orange":0,"red":56.3},"words":80,"usd":0.0063447,"sec":182,"out":12273,"summary":"The 500-square dataset grid, task-type stats and training footer bury a small, late bar chart; nothing important is missing.","motion":["Waffle grid fills for seconds while the main chart is absent","Main bar chart arrives only after 4s","Grid squares and dataset stats move at once, splitting attention"],"cut":4,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__haiku__simple","task":"improving-qwen38-27b-on-deepswe","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":10,"usd":0.0028836,"sec":34,"out":5287,"summary":"Clean and complete: a single before-after bar chart with a named benchmark, metric and values, and nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__luna__raw","task":"improving-qwen38-27b-on-deepswe","model":"luna","variant":"raw","clutter":57.8,"end":56.3,"shares":{"blue":42.2,"orange":0,"red":57.8},"words":46,"usd":0.0015637000000000001,"sec":35,"out":2860,"summary":"A slogan headline, a kicker tag and a pipeline strip crowd a tiny, weak one-bar chart, which is complete enough to read.","motion":["Bar grows from 27.8% to 39.1% only after about 3 seconds, so the main visual arrives late."],"cut":4,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__luna__simple","task":"improving-qwen38-27b-on-deepswe","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.0011485,"sec":27,"out":1985,"summary":"Clean and complete: one two-bar chart naming the metric, model, and benchmark, with nothing to remove.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__opus-high__raw","task":"improving-qwen38-27b-on-deepswe","model":"opus-high","variant":"raw","clutter":54,"end":52.5,"shares":{"blue":46,"orange":0,"red":54},"words":73,"usd":0.291952,"sec":143,"out":14173,"summary":"The result is a small part of the frame, crowded by a 500-dot grid, a method diagram, a legend, a reward row and a delta pill. Nothing essential is missing.","motion":["Dot grid, diagram and score build at once, splitting attention","Main result arrives late, around 4.8s of 8s","Large 27.8% swaps to 39.1% by replacement, not a smooth count","Many elements appear in staggered sequence, leaving the eye no single focus"],"cut":7,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__opus-high__simple","task":"improving-qwen38-27b-on-deepswe","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":17,"usd":0.146368,"sec":104,"out":6830,"summary":"Clean and complete: one bar chart with a benchmark title, labelled base and post-GRPO values, and a small change annotation.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__opus-medium__raw","task":"improving-qwen38-27b-on-deepswe","model":"opus-medium","variant":"raw","clutter":67.1,"end":72,"shares":{"blue":7.1,"orange":51.7,"red":41.3},"words":86,"usd":0.165152,"sec":92,"out":7833,"summary":"The result bar chart is a small part of a canvas filled with a dataset waffle, a process diagram, a big stat callout and a footer, and the chart appears late; nothing important is missing.","motion":["Waffle grid fills for seconds before the result chart appears","Main result chart arrives late, around 4.3s","Many small elements appear in staggered waves","Big +11.3 and footer appear late, restating the chart"],"cut":6,"shrunk":1},{"id":"improving-qwen38-27b-on-deepswe__opus-medium__simple","task":"improving-qwen38-27b-on-deepswe","model":"opus-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":22,"usd":0.093508,"sec":62,"out":4187,"summary":"Clean and complete; the only mild redundancy is that the +11.3 pts label repeats what the two percentages already show.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__opus__raw","task":"improving-qwen38-27b-on-deepswe","model":"opus","variant":"raw","clutter":31,"end":43.4,"shares":{"blue":69,"orange":0,"red":31},"words":42,"usd":0.061252,"sec":41,"out":2638,"summary":"A clean bar chart is cluttered by a large +11.3 callout that repeats it, a stats footer, and a tag; nothing is missing.","motion":["Stats row arrives late in staggered steps, after the main point is made"],"cut":3,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__opus__simple","task":"improving-qwen38-27b-on-deepswe","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":15,"usd":0.046868,"sec":25,"out":1855,"summary":"Clean and complete: a single before-after bar chart with a short title, labelled values, and a small change annotation.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sol__raw","task":"improving-qwen38-27b-on-deepswe","model":"sol","variant":"raw","clutter":75.5,"end":67,"shares":{"blue":20.2,"orange":8.5,"red":71.2},"words":84,"usd":0.038804,"sec":86,"out":3613,"summary":"The result is a small part of a canvas crowded with a slogan, pipeline diagram, data cards, a repeated callout and a footer; nothing important is missing.","motion":["Pipeline, data cards and result animate in sequence, splitting attention","Main result appears late, 4.3s, after supporting elements","Many small elements staggered in at once"],"cut":5,"shrunk":1},{"id":"improving-qwen38-27b-on-deepswe__sol__simple","task":"improving-qwen38-27b-on-deepswe","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":15,"usd":0.02173,"sec":54,"out":1861,"summary":"Clean and complete: a single before-and-after bar chart with the benchmark, metric, model and values.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sonnet-high__raw","task":"improving-qwen38-27b-on-deepswe","model":"sonnet-high","variant":"raw","clutter":66.3,"end":67.5,"shares":{"blue":6.1,"orange":55.2,"red":38.7},"words":57,"usd":0.093886,"sec":78,"out":8964,"summary":"A large dot grid of training tasks and a reward footer crowd the result, which arrives late; otherwise it is complete.","motion":["Dot grid fills for 3 seconds before the result appears","Main result arrives late, at about 4 seconds","Large counter 31.3→39.1 duplicates the bar motion"],"cut":3,"shrunk":1},{"id":"improving-qwen38-27b-on-deepswe__sonnet-high__simple","task":"improving-qwen38-27b-on-deepswe","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":12,"usd":0.068734,"sec":59,"out":6385,"summary":"Clean and complete: a simple before-after bar chart with benchmark tag, values, delta and logo, with calm sequential bar growth.","motion":[],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sonnet-medium__raw","task":"improving-qwen38-27b-on-deepswe","model":"sonnet-medium","variant":"raw","clutter":16.8,"end":19.9,"shares":{"blue":82.7,"orange":0.8,"red":16.4},"words":46,"usd":0.053176,"sec":43,"out":4893,"summary":"A ring gauge shows the result, but the eyebrow, subtitle and a stats row add clutter, and the before value vanishes so the comparison is never shown together.","motion":["Base value is replaced by the final value, so no side-by-side comparison.","Stats row and pill arrive late, drawing the eye away from the ring.","Result completes only after about 5 seconds."],"cut":3,"shrunk":1},{"id":"improving-qwen38-27b-on-deepswe__sonnet-medium__simple","task":"improving-qwen38-27b-on-deepswe","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":95,"orange":0,"red":0},"words":10,"usd":0.028704,"sec":35,"out":2382,"summary":"Clean and complete: two labelled bars with values and a small metric footnote, with nothing to remove.","motion":["Text is small relative to the very large value numbers."],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sonnet-xhigh__raw","task":"improving-qwen38-27b-on-deepswe","model":"sonnet-xhigh","variant":"raw","clutter":67.2,"end":52.2,"shares":{"blue":32.8,"orange":0,"red":67.2},"words":80,"usd":0.005556,"sec":979,"out":131,"summary":"The 39.1% result and its benchmark label are clear, but a dataset grid, dataset stat and a training-reward panel compete with it and should be removed.","motion":["Grid tiles, update ticks and reward counter animate together, competing for attention","Main 39.1% result arrives at 5.5s, after the dataset content","Dataset grid fills in while the headline stays static, splitting focus"],"cut":5,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sonnet-xhigh__simple","task":"improving-qwen38-27b-on-deepswe","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":17,"usd":0.005954,"sec":863,"out":107,"summary":"Clean single-visual graphic. The only gap is that the large 39.1% is never labeled as the trained model's score.","motion":["Big number counts through values while bar grows, which is slightly hard to follow."],"cut":0,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sonnet__raw","task":"improving-qwen38-27b-on-deepswe","model":"sonnet","variant":"raw","clutter":3.6,"end":14.3,"shares":{"blue":96.4,"orange":0,"red":3.6},"words":27,"usd":0.035606,"sec":30,"out":3136,"summary":"A clear before-after bar chart with a fitting title; the dataset stats footer is an extra idea that should be removed, and the motion has a layout shift.","motion":["Bar value labels drift off their bars at 3.6s before settling","Chart block shifts position between 3.6s and 5s, breaking eye tracking","Footer stats row arrives last and adds a second idea late"],"cut":1,"shrunk":0},{"id":"improving-qwen38-27b-on-deepswe__sonnet__simple","task":"improving-qwen38-27b-on-deepswe","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":9,"usd":0.020524,"sec":18,"out":1564,"summary":"Clean and complete: a simple two-bar chart with a named benchmark, metric, labels and values, and no extras.","motion":[],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__argon__raw","task":"on-policy-distillation-gdpval","model":"argon","variant":"raw","clutter":72.6,"end":71.6,"shares":{"blue":21.2,"orange":11.6,"red":66.8},"words":404,"usd":0.66756,"sec":645,"out":65829,"summary":"The win-loss bars are buried in a dense dashboard of mechanism diagram, curves, cards, benchmark tiles, pills and footer; nothing is missing.","motion":["Panels appear sequentially, delaying results until 3.5s","Many small elements animate at once in left panel","Footer and bottom panels arrive in last seconds, too brief to read"],"cut":7,"shrunk":3},{"id":"on-policy-distillation-gdpval__argon__simple","task":"on-policy-distillation-gdpval","model":"argon","variant":"simple","clutter":2.4,"end":4.2,"shares":{"blue":96.5,"orange":2.1,"red":1.4},"words":63,"usd":0.360054,"sec":462,"out":35033,"summary":"A mostly clean dot-grid comparison with a few wordy small labels and redundant offset notes; nothing important is missing.","motion":["Qwen row dots start while Nemotron margin number is still arriving, splitting attention"],"cut":1,"shrunk":2},{"id":"on-policy-distillation-gdpval__astra__raw","task":"on-policy-distillation-gdpval","model":"astra","variant":"raw","clutter":65.5,"end":55.7,"shares":{"blue":34.5,"orange":0,"red":65.5},"words":65,"usd":0.19883,"sec":159,"out":3105,"summary":"The animation buries two big stat numbers under a slogan headline, a decorative document animation, an eyebrow tag and a footer tagline; no bars or chart show the comparison.","motion":["Document illustration animates while results are not yet shown","Main results arrive late, after 4s","Teacher feedback tag appears and fades, a transient element"],"cut":6,"shrunk":0},{"id":"on-policy-distillation-gdpval__astra__simple","task":"on-policy-distillation-gdpval","model":"astra","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.14026,"sec":100,"out":1889,"summary":"Clean and complete: a simple two-bar GDPval chart with title, metric, labels and values, with no clutter to remove.","motion":[],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__haiku-high__raw","task":"on-policy-distillation-gdpval","model":"haiku-high","variant":"raw","clutter":8.4,"end":16.8,"shares":{"blue":91.6,"orange":0,"red":8.4},"words":62,"usd":0.018364400000000003,"sec":191,"out":35328,"summary":"A clear tile chart is cluttered by a date, an eyebrow, a giant repeated margin stat with explanation lines and a Qwen side result; nothing essential is missing.","motion":["Dots fall from a single point while the tiles fill, which is busy.","The +20.9% stat arrives late, after the chart finishes.","The Qwen stat appears last and pulls attention away."],"cut":4,"shrunk":0},{"id":"on-policy-distillation-gdpval__haiku-high__simple","task":"on-policy-distillation-gdpval","model":"haiku-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":11,"usd":0.010621799999999999,"sec":113,"out":19779,"summary":"Clean waffle chart with no extras, but the headline and legend text are tiny, arrive late, and no model or baseline is named.","motion":["Headline and legend appear very late, after the grid is filled","Legend text is tiny and the headline arrives only at the end"],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__haiku-medium__raw","task":"on-policy-distillation-gdpval","model":"haiku-medium","variant":"raw","clutter":41.5,"end":44.1,"shares":{"blue":45.3,"orange":26.6,"red":28.2},"words":66,"usd":0.0008709,"sec":810,"out":341,"summary":"A good waffle chart is crowded by a long headline, a method paragraph, a big repeated stat, a Qwen line, a blog tag and a footer URL.","motion":["Big +20.9% stat and Qwen line appear after the chart, late","Waffle fill is gradual, so the main result arrives late"],"cut":5,"shrunk":1},{"id":"on-policy-distillation-gdpval__haiku-medium__simple","task":"on-policy-distillation-gdpval","model":"haiku-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":97.5,"orange":0,"red":0},"words":15,"usd":0.0054533,"sec":59,"out":9442,"summary":"Clean and complete: one stacked win-tie-loss bar with a short context line, and no extras to remove.","motion":[],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__haiku__raw","task":"on-policy-distillation-gdpval","model":"haiku","variant":"raw","clutter":22.6,"end":18.2,"shares":{"blue":58.6,"orange":32.4,"red":6.4},"words":64,"usd":0.0074843999999999996,"sec":70,"out":13568,"summary":"A solid waffle-chart main visual is cluttered by an eyebrow, a wordy headline, and late extra lines on the Qwen result and the rubric score; nothing essential is missing.","motion":["Footer lines appear late, after the main message has landed","Large stat arrives at 4.3s, after the grid already shows the result"],"cut":3,"shrunk":1},{"id":"on-policy-distillation-gdpval__haiku__simple","task":"on-policy-distillation-gdpval","model":"haiku","variant":"simple","clutter":0,"end":0,"shares":{"blue":97.5,"orange":0,"red":0},"words":13,"usd":0.0032428,"sec":29,"out":5021,"summary":"Clean and complete: a single win-tie-loss bar with a short title, counts, and the benchmark named; only a mid-animation rescale of the bar is worth fixing.","motion":["Bar re-scales from about 3.6s to 5s after ties and losses appear, reflowing the chart"],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__luna__raw","task":"on-policy-distillation-gdpval","model":"luna","variant":"raw","clutter":67.3,"end":56.7,"shares":{"blue":32.7,"orange":0,"red":67.3},"words":81,"usd":0.0016623,"sec":47,"out":2453,"summary":"The method diagram, slogan, eyebrow, subtitle, caption and footer bury the one +21.4% result. The result bar also doesn't name GDPval or the base-model comparison.","motion":["The finding arrives at 4.3–5.3s, after the method cards.","Cards and connectors animate in sequence, so the eye follows the method rather than the result.","Number and caption appear in separate stages."],"cut":5,"shrunk":0},{"id":"on-policy-distillation-gdpval__luna__simple","task":"on-policy-distillation-gdpval","model":"luna","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":20,"usd":0.0014921,"sec":44,"out":2068,"summary":"The graphic is clean and nearly complete, but it never says the comparison is against the base model, and its text is tiny.","motion":["Legend and net margin column arrive late, after the bars have finished growing."],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__opus-high__raw","task":"on-policy-distillation-gdpval","model":"opus-high","variant":"raw","clutter":62.4,"end":48.2,"shares":{"blue":37.2,"orange":0,"red":62.4},"words":108,"usd":0.310536,"sec":145,"out":14126,"summary":"The grid and +20.9% margin are the clear core, but a slogan headline, method pipeline, side stats, and Qwen footer clutter it; nothing essential is missing.","motion":["Pipeline tokens and teacher bars animate alongside the waffle fill, splitting attention","Main grid arrives at 3.3s, after the headline and pipeline occupy the early frames","Qwen footer slides in at 6.5s, after the main stat has already landed","Pipeline stays on screen through the end, competing with the final result"],"cut":7,"shrunk":0},{"id":"on-policy-distillation-gdpval__opus-high__simple","task":"on-policy-distillation-gdpval","model":"opus-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":18,"usd":0.211692,"sec":111,"out":9120,"summary":"Nearly clean waffle chart with labelled outcomes and the net margin; minor issue is the end-state grid fading out.","motion":["Grid recolors to pale at the end, dimming the main visual","Squares fill in over several seconds, slow to reach full result"],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__opus-medium__raw","task":"on-policy-distillation-gdpval","model":"opus-medium","variant":"raw","clutter":57.5,"end":52.2,"shares":{"blue":18.3,"orange":47.6,"red":33.7},"words":116,"usd":0.308716,"sec":133,"out":14035,"summary":"A small win/loss grid is crowded by a mechanism pipeline, a repeated stat row, an eyebrow tag, a Qwen footnote line and a URL; nothing essential is missing.","motion":["Pipeline diagram and grid animate at the same time, splitting attention","Grid fills square by square over several seconds, so the main visual arrives late","Stat row, Qwen line and footer pile in at the end"],"cut":5,"shrunk":1},{"id":"on-policy-distillation-gdpval__opus-medium__simple","task":"on-policy-distillation-gdpval","model":"opus-medium","variant":"simple","clutter":1.3,"end":5,"shares":{"blue":98.8,"orange":0,"red":1.3},"words":17,"usd":0.132332,"sec":70,"out":5152,"summary":"A clean waffle chart whose only clutter is the late +20.9% stat that restates the grid, and the count labels are tiny.","motion":["Waffle squares fill one by one, which carries the data and isn't a problem.","Count labels are tiny and appear only as each segment completes."],"cut":1,"shrunk":0},{"id":"on-policy-distillation-gdpval__opus__raw","task":"on-policy-distillation-gdpval","model":"opus","variant":"raw","clutter":39.2,"end":50.5,"shares":{"blue":37.9,"orange":43.3,"red":17.6},"words":53,"usd":0.072316,"sec":26,"out":2215,"summary":"A solid dot-grid main visual is cluttered by an eyebrow, a wordy headline and two big stat cards, one covering a second model.","motion":["Stat cards arrive after the chart, restating it.","Dots fill gradually across several seconds, delaying the main result."],"cut":3,"shrunk":1},{"id":"on-policy-distillation-gdpval__opus__simple","task":"on-policy-distillation-gdpval","model":"opus","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":18,"usd":0.058632,"sec":20,"out":1467,"summary":"Clean and nearly complete: one stacked win-tie-loss bar with a +20.9% margin, with only the context line possibly wordier than needed.","motion":[],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__sol__raw","task":"on-policy-distillation-gdpval","model":"sol","variant":"raw","clutter":62.4,"end":53.2,"shares":{"blue":37.6,"orange":0,"red":62.4},"words":79,"usd":0.041406,"sec":126,"out":3269,"summary":"A document-illustration panel, slogan headline, tagline, eyebrow, takeaway and footer URL clutter the two result numbers, which arrive late and have no bars.","motion":["Left illustration animates for 4s before the result appears","Result arrives late, after 4.5s","Teacher feedback tag flashes in and fades","Footnote appears only at 6.3s"],"cut":7,"shrunk":0},{"id":"on-policy-distillation-gdpval__sol__simple","task":"on-policy-distillation-gdpval","model":"sol","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":14,"usd":0.026972,"sec":96,"out":1781,"summary":"Clean and complete: a simple two-bar GDPval chart with a named metric, labelled bars, values and a baseline note, with bars that grow in sequence to carry the data.","motion":[],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__sonnet-high__raw","task":"on-policy-distillation-gdpval","model":"sonnet-high","variant":"raw","clutter":51.3,"end":37.2,"shares":{"blue":39.9,"orange":14.1,"red":44.3},"words":115,"usd":0.152368,"sec":136,"out":13836,"summary":"A 3.5s method intro and a Qwen stat block clutter a good dot-grid result chart; nothing important is missing.","motion":["Intro scene vanishes before the main visual appears, wasting about half the runtime","Text fades out while the old/current token swap is still moving"],"cut":3,"shrunk":1},{"id":"on-policy-distillation-gdpval__sonnet-high__simple","task":"on-policy-distillation-gdpval","model":"sonnet-high","variant":"simple","clutter":0,"end":0,"shares":{"blue":98.8,"orange":0,"red":0},"words":18,"usd":0.073066,"sec":67,"out":5842,"summary":"Clean waffle chart with only small, useful text; the labels are tiny and the fill is slow.","motion":["Waffle fill is slow, taking about 5 seconds before the result is clear.","Net margin appears only after all counts, near the end."],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__sonnet-medium__raw","task":"on-policy-distillation-gdpval","model":"sonnet-medium","variant":"raw","clutter":12.8,"end":13,"shares":{"blue":74.4,"orange":23,"red":1.3},"words":35,"usd":0.050548,"sec":48,"out":3654,"summary":"A mostly clean dot-grid with a wordy slogan headline and an extra Qwen callout; otherwise complete.","motion":["Dots shimmer or change size while filling, which is distracting","Gray dots appear for a long time before data arrives","Qwen callout appears last, pulling attention from the main result"],"cut":1,"shrunk":1},{"id":"on-policy-distillation-gdpval__sonnet-medium__simple","task":"on-policy-distillation-gdpval","model":"sonnet-medium","variant":"simple","clutter":0,"end":0,"shares":{"blue":95,"orange":0,"red":0},"words":14,"usd":0.050636,"sec":54,"out":3599,"summary":"Clean and nearly complete: a single waffle chart with outcome labels and a headline stat, though it never says the comparison is against the base model.","motion":[],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__sonnet-xhigh__raw","task":"on-policy-distillation-gdpval","model":"sonnet-xhigh","variant":"raw","clutter":41.7,"end":9.6,"shares":{"blue":58.3,"orange":0,"red":41.7},"words":105,"usd":0.323898,"sec":264,"out":30989,"summary":"Most of the animation is a mechanism diagram with transient captions, and the result only appears late; the result stats are fine.","motion":["Mechanism animation occupies most of the time before the result.","Result appears only at 6.3s, late.","Captions change quickly, some on screen about one second.","Strip recolors and several bars animate at once."],"cut":5,"shrunk":0},{"id":"on-policy-distillation-gdpval__sonnet-xhigh__simple","task":"on-policy-distillation-gdpval","model":"sonnet-xhigh","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":19,"usd":0.248316,"sec":217,"out":23367,"summary":"Mostly clean waffle chart with a headline stat; the final state strips the chart to a misleading remnant behind the number, and the tag text is tiny.","motion":["Waffle collapses to wins only at the end, losing the loss and tie context.","Chart squares are reshuffled between 3.8s and 5s, which is confusing to follow.","Big stat arrives late and covers the faded grid."],"cut":0,"shrunk":0},{"id":"on-policy-distillation-gdpval__sonnet__raw","task":"on-policy-distillation-gdpval","model":"sonnet","variant":"raw","clutter":15.9,"end":25.6,"shares":{"blue":71.7,"orange":25,"red":3.4},"words":45,"usd":0.052338,"sec":42,"out":3833,"summary":"A clear dot-grid chart is slightly cluttered by an oversized stat, subtext, and a late Qwen card.","motion":["Dots drift in from scattered positions at the start","Qwen card arrives late as a distinct second point"],"cut":2,"shrunk":2},{"id":"on-policy-distillation-gdpval__sonnet__simple","task":"on-policy-distillation-gdpval","model":"sonnet","variant":"simple","clutter":0,"end":0,"shares":{"blue":100,"orange":0,"red":0},"words":16,"usd":0.045556,"sec":32,"out":3091,"summary":"Clean and nearly complete; the tiny caption and the slow build are the only small weaknesses.","motion":["Waffle fills slowly; the net margin appears only at 5.5s."],"cut":0,"shrunk":0}]}