conversations listlengths 1 1 | chosen dict | rejected dict | images images listlengths 1 1 | ground_truth_corruption dict | meta dict |
|---|---|---|---|---|---|
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low visual contrast for interactive elements may hinder accessibility and discoverability.\",\n \"observed_feature\": \"Light gray text on a white background for 'Popular Searches' and search terms.\",\n \"likely_intent\": \"To create a clean, minimalist ... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low visual hierarchy and poor scanability due to lack of typographic distinction.\",\n \"observed_feature\": \"Uniform font weight, size, and color for both the 'Popular Searches' header and the search suggestions.\",\n \"likely_intent\": \"To maintain a ... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"diminished_visual_hierarchy",
"reduced_text_contrast",
"lost_item_separation"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [diminished_visual_hierarchy] Modify the 'Popular Searc... | {
"source_id": "screenshot_2024_4_16_16_1_2ac10256ecdc449eb565bc311a25615c-5",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/018_2024_4_16_16_1_2ac10256ecdc449eb565bc311a25615c-5",
"c... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The search input field lacks visual prominence and sufficient contrast against the white background, which may hinder discoverability.\",\n \"observed_feature\": \"Low-contrast search bar with light gray border and text on a white background.\",\n \"likel... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy for match status and scores is obscured by inconsistent text scaling and lack of clear container boundaries.\",\n \"observed_feature\": \"The live score updates (e.g., '118 runs to win from 19.2 overs with...') use a very large, bold fon... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"accessibility_contrast_degradation",
"visual_hierarchy_weakening",
"layout_stability_issues"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [accessibility_contrast_degradation] Cha... | {
"source_id": "screenshot_2024_4_25_20_38_1d4d22af88f240ceb1dbed0880ad8cab-9",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/022_2024_4_25_20_38_1d4d22af88f240ceb1dbed0880ad8cab-9",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The modal dialog obstructs critical navigation information and primary actions during the initial setup phase.\",\n \"observed_feature\": \"Large centered modal overlay with 'Welcome to Google Maps Navigation' text covering the route details and bottom actio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The modal dialog obstructs the primary user task (reviewing the route) and requires an extra interaction step to dismiss.\",\n \"observed_feature\": \"Full-screen overlay modal with 'Welcome to Google Maps Navigation' text.\",\n \"likely_intent\": \"To en... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"primary_action_degraded_hierarchy",
"route_info_legibility_loss",
"transport_mode_scanability_reduction"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [primary_action_degraded_hie... | {
"source_id": "screenshot_2024_5_6_19_47_f399bc507f27499b9a07b1cdc6d74785-12",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/024_2024_5_6_19_47_f399bc507f27499b9a07b1cdc6d74785-12",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low visual distinction between 'Featured Charts' and 'Weekly Song Charts' categories.\",\n \"observed_feature\": \"Repeated use of identical card styling and layout for different sections of the screen.\",\n \"likely_intent\": \"To maintain a consistent v... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low visual hierarchy and lack of differentiation between 'Featured' and 'Weekly' sections due to repetitive card design.\",\n \"observed_feature\": \"Identical card styling (color, typography, layout) for both 'Featured Charts' and 'Weekly Song Charts' secti... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"contrast_degradation_of_metadata",
"weakened_visual_hierarchy_of_tags",
"reduced_interaction_affordance_of_tags"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [contrast_degradatio... | {
"source_id": "screenshot_2024_4_9_1_23_c22cacecbf8e4349a41e778bd84fdce1-6",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/025_2024_4_9_1_23_c22cacecbf8e4349a41e778bd84fdce1-6",
"cho... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The text contrast ratio against the dark background appears low, which may impede legibility for users with visual impairments or in high-glare environments.\",\n \"observed_feature\": \"Grey text on a black background.\",\n \"likely_intent\": \"To create... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low text legibility due to insufficient contrast between foreground and background.\",\n \"observed_feature\": \"Dark grey/black text on a pure black background.\",\n \"likely_intent\": \"To create a high-contrast, 'dark mode' aesthetic that may reduce ey... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"contrast_degradation",
"layout_density_issue",
"legibility_degradation"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [contrast_degradation] Change the color of the main body text... | {
"source_id": "screenshot_2024_4_28_2_21_8f932a61650a44af9989eed51f55a064-14",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/027_2024_4_28_2_21_8f932a61650a44af9989eed51f55a064-14",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The pricing breakdown lacks immediate visual hierarchy, making the total cost feel disconnected from the individual line items.\",\n \"observed_feature\": \"Uniform font weight and size for Price, Fees, Delivery, and Tax items.\",\n \"likely_intent\": \"T... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The cost breakdown lacks transparency regarding the specific nature of the 'Fees', which may lead to cart abandonment during the final step.\",\n \"observed_feature\": \"Fees (2 x $38.92 each) listed as a line item.\",\n \"likely_intent\": \"To simplify t... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"visual_hierarchy_flattening",
"interaction_affordance_removal",
"accessibility_contrast_reduction"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [visual_hierarchy_flattening] Redu... | {
"source_id": "screenshot_2024_4_22_20_3_ef90b9135cdc4a258adce625999d70a1-14",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/030_2024_4_22_20_3_ef90b9135cdc4a258adce625999d70a1-14",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The text density is extremely high with minimal vertical rhythm, making the content difficult to scan and increasing cognitive load.\",\n \"observed_feature\": \"Tight line spacing (leading) and lack of paragraph separation in the body text.\",\n \"likely... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Poor text legibility due to insufficient line spacing and tight tracking.\",\n \"observed_feature\": \"High density of text blocks with minimal vertical rhythm between lines and characters.\",\n \"likely_intent\": \"To maximize the amount of information v... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"headline_de_emphasis",
"low_contrast_body_text",
"hidden_hyperlinks"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [headline_de_emphasis] Reduce the headline font size to match th... | {
"source_id": "screenshot_2024_4_29_12_39_22fbe291194e4d5381427cb4e6725252-16",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/031_2024_4_29_12_39_22fbe291194e4d5381427cb4e6725252-16",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The hierarchy of event details is somewhat flat, making it difficult to scan for core logistics like time and location quickly.\",\n \"observed_feature\": \"The event date/time and location are presented in a similar font weight and visual style as the secon... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low visual hierarchy for the event title due to lack of typographic contrast against the white background.\",\n \"observed_feature\": \"Light grey text color for the main event title 'Lululemon X Nu Fitness Vinyasa Yoga Class' on a white background.\",\n ... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"diminished_title_legibility",
"flattened_visual_hierarchy",
"lowered_call_to_action_prominence"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [diminished_title_legibility] Change ... | {
"source_id": "screenshot_2024_4_23_19_54_31528ae16c264b1b98bd4e1f25c5d9e5-13",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/033_2024_4_23_19_54_31528ae16c264b1b98bd4e1f25c5d9e5-13",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy for pricing information is competing, making it difficult to quickly distinguish the current price from the original price.\",\n \"observed_feature\": \"Large light-gray current price (HK$31.76) paired with smaller struck-through origina... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy for pricing information is cluttered, making it difficult to quickly discern the current price from the original price.\",\n \"observed_feature\": \"Juxtaposition of HK$31.76, HK$48.90, and '35% off sale for a limited time' in similar fo... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"primary_action_visual_weakening",
"price_contrast_reduction",
"trust_signal_obscuration"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [primary_action_visual_weakening] The 'Add t... | {
"source_id": "screenshot_2024_4_22_17_53_b40df5cc66774ed9926212a8aa955748-17",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/035_2024_4_22_17_53_b40df5cc66774ed9926212a8aa955748-17",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The primary Call to Action (CTA) directs users away from the current mobile context, creating significant friction in the conversion funnel.\",\n \"observed_feature\": \"Button labeled 'Sign up on desktop'\",\n \"likely_intent\": \"To ensure a seamless re... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The primary Call to Action (CTA) 'Sign up on desktop' creates a significant friction point by forcing users away from their current device.\",\n \"observed_feature\": \"Button labeled 'Sign up on desktop'\",\n \"likely_intent\": \"To redirect users to a p... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"weakened_primary_cta_visual_weight",
"low_contrast_interactive_links",
"reduced_nav_item_distinction"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [weakened_primary_cta_visual_we... | {
"source_id": "screenshot_2024_5_6_20_22_3052b8a651044a129575be958a91b3cd-15",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/036_2024_5_6_20_22_3052b8a651044a129575be958a91b3cd-15",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The primary call-to-action (CTA) is positioned at the very bottom of the screen, potentially outside the immediate thumb zone for one-handed mobile use.\",\n \"observed_feature\": \"Bottom-aligned 'Join the waitlist!' button.\",\n \"likely_intent\": \"To ... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low text legibility due to insufficient contrast between the white typography and the varied background imagery.\",\n \"observed_feature\": \"White text overlay on a photographic background with bright areas.\",\n \"likely_intent\": \"To create an immersi... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"hero_text_contrast_loss",
"cta_visual_hierarchy_degradation",
"interactive_card_affordance_removal"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [hero_text_contrast_loss] Change ... | {
"source_id": "screenshot_2024_3_20_19_8_0491d30277954c13be8efbdef2297be5-12",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/043_2024_3_20_19_8_0491d30277954c13be8efbdef2297be5-12",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The content hierarchy lacks clear distinction between the article metadata (author, timestamp) and the body text, which may lead to scanning fatigue.\",\n \"observed_feature\": \"Uniform font weight and similar color usage for the author name, timestamp, and... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The article title and body text lack sufficient visual hierarchy, making it difficult to distinguish between the headline and the reading content at a glance.\",\n \"observed_feature\": \"Uniform font weight and similar size/color for the title and the main ... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"weakened_visual_hierarchy",
"low_contrast_hyperlink",
"reduced_tag_affordance"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [weakened_visual_hierarchy] The main headline '9 Of Th... | {
"source_id": "screenshot_2024_4_29_12_53_f884f1149773424d8038613ec8c9736d-7",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/046_2024_4_29_12_53_f884f1149773424d8038613ec8c9736d-7",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The article text contains significant overlapping text layers that interfere with readability.\",\n \"observed_feature\": \"Semi-transparent text (e.g., 'MEERUB ANJUM') appearing directly on top of the main headline and body text.\",\n \"likely_intent\": ... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"Low visual hierarchy and poor text contrast in the header area.\",\n \"observed_feature\": \"The use of light grey text for 'Published on...' and secondary metadata over a white background.\",\n \"likely_intent\": \"To de-emphasize metadata and keep the u... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"diminished_headline_hierarchy",
"low_contrast_metadata",
"reduced_button_affordance"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [diminished_headline_hierarchy] Reduce the font ... | {
"source_id": "screenshot_2024_3_24_14_28_8fc710d1f03c4703bdcb462579b2c97f-8",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/050_2024_3_24_14_28_8fc710d1f03c4703bdcb462579b2c97f-8",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The global navigation bar lacks sufficient visual distinction from the content area, potentially leading to poor cognitive separation during scrolling.\",\n \"observed_feature\": \"A top navigation bar containing a hamburger menu, search icon, and brand logo... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The top navigation bar lacks sufficient visual distinction from the content area, which may impact wayfinding in long-form reading contexts.\",\n \"observed_feature\": \"Transparent or low-contrast navigation bar overlaying the top of the content area.\",\n ... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"low_contrast_text_color",
"flattened_visual_hierarchy",
"tightened_line_spacing"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [low_contrast_text_color] Change the color of the ma... | {
"source_id": "screenshot_2024_4_28_2_21_8f932a61650a44af9989eed51f55a064-25",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/052_2024_4_28_2_21_8f932a61650a44af9989eed51f55a064-25",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The content lacks visual hierarchy and structural scanning aids, making the long-form text appear as a monolithic block.\",\n \"observed_feature\": \"Uniform text weight, size, and color across the entire body of text.\",\n \"likely_intent\": \"To maintai... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The lack of visual hierarchy and structural landmarks makes the content difficult to scan and navigate.\",\n \"observed_feature\": \"Uniform text weight, color, and size throughout the entire block of text.\",\n \"likely_intent\": \"The design may aim to ... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"low_contrast_text",
"lost_quote_differentiation",
"cramped_typography_layout"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [low_contrast_text] Change the font color from bright w... | {
"source_id": "screenshot_2024_4_28_2_21_8f932a61650a44af9989eed51f55a064-15",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/053_2024_4_28_2_21_8f932a61650a44af9989eed51f55a064-15",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy of the typography makes it difficult to quickly scan the event details and differentiate the title from secondary information.\",\n \"observed_feature\": \"Large, bold title 'Music and Heritage: Bach, Cello Suites' overlaid on a high-con... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The primary call-to-action (CTA) 'Tickets' lacks sufficient visual weight and contrast against the background, making it difficult to identify as the main action.\",\n \"observed_feature\": \"Low-contrast 'Tickets' button on a light grey background at the bo... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"removed_image_overlay",
"diminished_cta_salience",
"low_contrast_promo_text"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [removed_image_overlay] Remove the dark gradient overlay... | {
"source_id": "screenshot_2024_4_25_17_12_e35f9572d14c41d4937430554345fe1c-15",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/055_2024_4_25_17_12_e35f9572d14c41d4937430554345fe1c-15",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy between headline and excerpt is undermined by the proximity of control elements to the text content.\",\n \"observed_feature\": \"Sliders for 'Headline' and 'Excerpt' font sizes are placed directly above the text they control.\",\n \"... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy between the 'Headline' and 'Excerpt' controls is ambiguous due to overlapping elements.\",\n \"observed_feature\": \"The 'Excerpt' label and its associated slider knob are visually colliding with the 'Headline' label and its slider knob.... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"weakened_visual_hierarchy",
"low_contrast_tags",
"reduced_slider_affordance"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [weakened_visual_hierarchy] The section headers 'Timelin... | {
"source_id": "screenshot_2024_4_28_10_49_320c5145851e4a5b9534e44d0631cde2-6",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/057_2024_4_28_10_49_320c5145851e4a5b9534e44d0631cde2-6",
... | |
[
{
"from": "human",
"value": "<image>\nYou are reviewing one UI screenshot.\n\nProduce trade-off-aware critiques. For each critique, include:\n- critique\n- observed_feature\n- likely_intent\n- positive_consequences (list)\n- negative_consequences (list)\n- tradeoff\n- contextual_conditions (list)\n- revisio... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy within the 'Drama' category makes it difficult to distinguish between the category header and the selectable tags.\",\n \"observed_feature\": \"The 'Drama' category header and the 'Drama' tag within the list share nearly identical typogr... | {
"from": "gpt",
"value": "[\n {\n \"critique\": \"The visual hierarchy between the 'Drama' category header and the 'Drama' search term/tag creates semantic ambiguity.\",\n \"observed_feature\": \"The screen contains a search bar with the text 'Drama' and a category header also labeled 'Drama'.\",\n \"lik... | {
"problem_id": "compound_01",
"compound_problem_ids": [
"low_contrast_tags",
"diminished_category_hierarchy",
"removed_selection_affordance"
],
"compound_size": 3,
"description": "Apply all of these simultaneous corruptions to the same screen:\n1. [low_contrast_tags] Change the text color within ... | {
"source_id": "screenshot_2024_4_20_13_54_7ac0c41a22b04b6e9654ab31edad0cfa-14",
"source_root": "/scratch/gilbreth/jasonwu/designreasoner/outputs/info_eval_end_to_end_package/outputs/mcinfo_table_repro_variations/run_diff_bo4_imggen2_20260530_180022/samples/062_2024_4_20_13_54_7ac0c41a22b04b6e9654ab31edad0cfa-14",
... |
End of preview. Expand in Data Studio
MC_Info Pexels RLHF-V (Clean)
A preference dataset for fine-tuning a VLM to produce trade-off-aware UI design critiques.
519 rows from best-of-4 MC_Info runs on Gilbreth and Gautschi clusters with real Pexels stock photos (no broken image placeholders).
Dataset Structure
Each row has the ShareGPT+ranking format used by LLaMA-Factory for DPO training:
| Field | Type | Description |
|---|---|---|
conversations |
[{"from": "human", "value": "<image>\\ncritic prompt..."}] |
The critic's prompt with embedded image |
chosen |
{"from": "gpt", "value": "critique JSON"} |
Winning bo4 round critique (preferred) |
rejected |
{"from": "gpt", "value": "critique JSON"} |
Non-selected bo4 round critique (dispreferred) |
images |
list[PIL.Image] |
Corrupted UI screenshot |
ground_truth_corruption |
dict | Metadata about injected UI problems |
Preference Signal
- chosen: The best of 4 critique generations, selected by the MC_Info judge
(Gemma 4 26B). These are proper trade-off critiques from
rN/corrupted_critique_gemma4-26b.jsonl. - rejected: A different round's critique that the judge ranked lower.
Both chosen and rejected are genuine critiques (no corruption plan text).
See training/TRAINING.md in the source repo for the full data quality documentation.
Data Quality
Important: The earlier dataset (
designreasoner-mcinfo-pexels-rlhf-v) used oracle critiques that contained corruption plan text ("Known injected corruption") instead of proper critiques. Do NOT use that dataset for training — the preference signal is meaningless. This clean dataset uses actual bo4 round critiques for both chosen and rejected.
Source
Generated from:
- Gilbreth run:
run_diff_bo4_imggen2_20260606_213235(276 rows) - Gautschi runs: 4 runs from June 6-7, 2026 (243 rows)
- Downloads last month
- 56