Meta-harness optimization / 2026
AutoDesign
A research system for improving the harness that turns source material into faithful, editable artifacts.
Enter the engineNext / the optimization object
System state PosterHarness / observable / editable
The optimization object
Improve the system that makes the artifact.
Each rollout leaves more than a poster. It leaves source context, an executable artifact, render evidence, diagnostic traces, and a bounded candidate update to the reusable harness.
Accepted transition Evaluate the output. Repair one module. Retain the next system state.
Poster universe / AutoPosterBench
One harness.
Many research worlds.
Artifact studies / emerging formats
One source system.
More ways to communicate.
Academic posters are AutoDesign's validated domain. These research specimens explore the same source-to-artifact workflow across slides, the web, and motion.
10 slidesEditable deck
01 / Paper to slides
A research narrative, paced across ten editable frames.
LongCat-Next is reorganized into a presentation sequence with paper figures, benchmark evidence, and a consistent visual grammar.
- Source
- Paper, figures, tables
- Artifact
- HTML, PDF, and PPTX
- Status
- Exploratory generalization study
ResponsiveExecutable HTML
02 / Paper to web
An editorial research page built for reading, not just viewing.
The SAM 2 paper becomes a responsive long-form artifact with grounded figures, quantitative evidence, and verified project links.
- Source
- Paper, figures, project links
- Artifact
- Responsive HTML and CSS
- Status
- Exploratory generalization study
72 secondsSilent motion explainer
03 / Paper to video
Six scenes that move from problem to evidence.
The SAM 2 explainer turns the paper's interaction model, architecture, qualitative examples, and benchmark results into a timed visual sequence.
- Source
- Paper, figures, evaluation
- Artifact
- 1920 × 1080 H.264 video
- Status
- Exploratory generalization study
Representative optimization trace
The artifact changes once.
The harness keeps learning.
AutoDesign reads the evidence left by completed rollouts, finds failures that repeat, and turns them into bounded updates to the system behind future artifacts.
State 00Observed
Recurring signal
Weak hierarchy · uneven evidence · underfilled regionsRepresentative sequence reconstructed from the AutoDesign workflow and artifact traces. It explains the mechanism; reported benchmark results appear below.
Inside PosterHarness
A design system the model can operate.
PosterHarness is the poster-specific system optimized by AutoDesign. It grounds the Designer in the paper, executes an editable artifact, inspects the browser result, and routes localized repair.
Designer → render → evaluate → localized repair
Evaluation separation Training evidence → candidate update → independent development gate → 100-paper final benchmark
01 / Context & Memory
Ground the design before composition begins.
- Input
- Output
Compact method record
Evidence follows the harness.
PosterHarness is AutoDesign’s validated instantiation for academic posters. The artifact remains inspectable while system-level evidence accumulates across rollouts.
AutoPosterBench / reported evidence
Measured across source grounding, communication quality, and executable artifact reliability.
Three independent views
Harness benefit, final quality, and human preference.
01 / Harness transfer · 10-paper controlled subset
One PosterHarness, seven fixed configurations.
Every evaluated configuration improves when the optimized PosterHarness is attached.
02 / Final quality
AutoPosterBench
- 01 AutoDesign77.55
- 02 Claude Design70.68
- 03 OpenDesign69.15
100-paper AutoPosterBench main track · selected systems among 14 evaluated configurations.
03 / Blind human evaluation
- 11
- reviewers
- 55.2–77.8
- 95% confidence interval
- 67.6%
- tie-adjusted vs. Claude Design
Next research boundary Generalize the learned harness update beyond posters without overstating evidence from the validated poster domain.
Research resources