Benchmaxxing world: credits =========================== What this is A single-page interactive view of the paper "Agents Catching Agents: Shortcut Cascades and Benchmark Gaming in Clinical Multi-Agent Systems". index.html is one self-contained file: it opens with no server and makes no network calls. It opens on an overview map, then an example conversation, then five stages: what was tested, the shortcut cascade, three referees, the hidden scoring note, how to catch it. Each stage is told one step at a time with Back and Next. Earlier versions of the page are kept in the folders round3/ to round7/ next to this file. Where the numbers come from Every number on the page that carries a gold chip is printed in the published MICCAI 2026 MedAgent workshop version of the paper (the camera-ready): https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_040.pdf The chip names the section or table and opens a note with the paper's own words. The arXiv posting (2608.03744, version 2 of 17 August 2026, CC BY 4.0) prints older MIMIC-CXR imaging cells; where the two differ, this page follows the published version. https://arxiv.org/abs/2608.03744 Headline numbers are written as "x of n" and detector rates as percentages. Where the paper prints a rate, "x of n" is that rate times the paper's n, rounded; the paper's own form is given in the Sources dialog and by hovering the number. What is simulated or drawn Each dot stands for one case. Which dot is which case, where it sits, how it moves, and the speech bubbles are schematic and tagged "illustration". The stage-03 boards (20 persuaded and 20 honest cases per detector) and the stage-05 stream are illustrations. In stage 04, the number of dots that return when the agent is told it is evaluated is taken from the summary files in the code repository; the page prints only the range given in the paper. The chest X-ray in the 3D scene, the exam card, the ICU table, the folded note, the model figures and the overview map are drawn in code. No report text and no patient record is on the page. One real chest X-ray is shown in stage 01 (credit below). The ICU record on the sheet of stage 01 is made up. On the sheet of stage 04, which option the model picks and which one the note rewards are an illustration. The conversation in the Example step (a question about fever, neck stiffness and altered mental status, two peer reviewers, a referee) is written for this page to show the idea. It is not a case from the paper's data and it carries no number. Data on the page MIMIC-CXR, CheXpert and SUPPORT2 cases appear only as dots and counts. So do the NIH ChestX-ray14 cases, except the one film credited below. One question is quoted in stages 01 and 04, with its five answer options: MedQA-USMLE, test split, item medqa-56 in the project's manifest. Jin D, Pan E, Oufattole N, Weng W-H, Fang H, Szolovits P. What disease does this patient have? A large-scale open domain question answering dataset from medical exams. Applied Sciences 11(14):6421, 2021. https://github.com/jind11/MedQA Licence: MIT (Copyright (c) 2022 Di Jin). Links on the page (each checked, HTTP 200, on 12 November 2026) Paper https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_040.pdf Code https://github.com/criticaldata/benchmaxxing (MIT) Data https://github.com/criticaldata/benchmaxxing/blob/main/docs/DATASETS.md Explainer https://criticaldata.github.io/benchmaxxing/ ("Honest, or just careful?") MIT Critical Data https://criticaldata.mit.edu/ Logos The MIT logo is a mark of the Massachusetts Institute of Technology, shown as on the project's explainer page. The benchmaxxing logo is the project's own, from the repository's assets/logo-animated.svg. The favicon is the same mark. Preview images og-image.jpg (1200 x 630) is the picture shown when the link is shared: a drawn chest X-ray and the headline, composed for this page. preview.gif is a short loop for sending as a picture next to the link; its frames are rendered from index.html. Layout The overview map and the way of moving through the page (one canvas, a path, a camera that dives into each station) are inspired by the task-world pages of Science's Last Exam, proton task: https://sciences-last-exam.org/assets/proton/ No code, text, data or image from that site is included; the markup, styles and script here were written for this page. Software included in index.html three.js r160 (three.min.js), unmodified. Copyright 2010-2023 Three.js Authors. MIT licence: The MIT License Copyright © 2010-2023 three.js authors Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. Chest X-ray shown in stage 01 One real film of the study, shown once as released and once with a corner tag ("PORTABLE") planted by the study's code: NIH ChestX-ray14, image 00000001_000.png, labelled cardiomegaly, scaled down to 336 pixels. Data provided by the NIH Clinical Center. Use of the dataset is unrestricted when it is credited as follows. Download site: https://nihcc.app.box.com/v/ChestXray-NIHCC Citation: X. Wang, Y. Peng, L. Lu, Z. Lu, M. Bagheri, R. M. Summers. ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases. IEEE CVPR 2017. Fonts System fonts only. No font files are bundled or fetched.