` links separated by ", ", `*` marks after equal-contribution authors, then `
` lines for `(* equal contribution)` and the affiliations separated by ` `.
**Button badges.** Five `for-the-badge` shields with emoji labels on #555555 and messages #007ec6, #000000, #ea7233, #ea7233 and #d8b800, one per source line inside ``. Model and dataset badges share the 🤗 label and are told apart by color (orange for models, yellow for the dataset).
**Figure plus justified text.** `
` followed by one or two `` paragraphs with `` for the key terms and `` for a research question. Used three times in a row.
**Reference tables.** Pipe tables with `|------|-------------|` delimiters and no alignment colons; GitHub centres header cells and left-aligns body cells. File and script names sit in inline code; the metrics table bolds each abbreviation (`**NSE** (Normalized Score Error)`). No numeric results.
**Command blocks.** `bash` fences with `#` comment lines naming each variant ("# Qwen2.5-VL-3B (Single GPU)"), a leading `cd` into the working directory, and long commands broken with trailing backslashes, one environment variable per line.
**Citation.** "If you find this work useful, please cite our paper:" then an `@article{zhang2026progresslm, ...}` block that still lists the arXiv preprint as the journal.
## Imagery
| Image | Type | Source size | Markdown | Rendered at 920px | Placement |
|---|---|---|---|---|---|
| icon.png | PNG mascot, transparent background | 812×851 | `![]()
` inside the h1 | about 29×30 | start of the title |
| teaser.png | PNG task figure, opaque white | 3000×1470 | `![]()
` in `
` | 856×419 | directly after the badge row |
| bench_curation.png | PNG data-construction figure, opaque white | 2000×986 | `![]()
` in `
` | 856×422 | after the two justified paragraphs |
| statistics.png | PNG statistics panel, opaque white | 2000×639 | `![]()
` in `
` | 856×273 | under `### Overview of ProgressLM-Dataset` |
The teaser is two rounded panels drawn like a worked example: "First see how the task is being done:" over six robot-arm frames labelled No.1 to No.6 and 0% to 100%, then "Then, tell me how much of this task has been completed so far?" with an observation photo, a grey "Direct Prediction" box ending in a red "About 70%", and a gold-framed "Progress Reasoning" box with `` and `` text, orange and blue braces for "Episodic Retrieval" and "Mental Simulation", and a green "46%" score. Titles are large bold sans-serif, so they read at body size even at 856px. The curation figure has three labelled sub-panels ((a) Demonstration Setup in a gold frame, (b) Observation Sampling in purple, (c) Answerability Augmentation in blue). The statistics panel combines three sunburst charts (Bench, SFT, RL), a legend, a grid of robot photos and a word cloud. All three figures are opaque white PNGs with no `` dark variant, so the page is designed for the light theme. There is no GIF or video.
## Adapting to Other Content
- **A figure-heavy paper repo.** Keep this order exactly: icon title, trophy h3, credits, button badges, then each figure at `width="100%"` (856px) followed by a justified paragraph. Export figures at 2000 to 3000px wide so they stay sharp at 856px on high-density screens.
- **A paper without an oral slot.** Drop the session-time paragraph and keep the h3 for the venue alone ("🏆" only for awards or orals).
- **A library.** Drop the credit block and the session note, keep the icon title and the badge row with Docs / PyPI / License messages in the same `for-the-badge` emoji form, and replace the three figures with one architecture figure and a Quickstart.
- **A dataset.** Lead with the statistics panel (it renders 856×273, short enough to sit in the first screen with the teaser), make the 🤗 DATASET badge first, and promote the File / Description tables to the first `##`.
- **A benchmark with scores.** Add a results table after the statistics panel with numeric columns right-aligned (`|------:|`) and the best value in bold; ProgressLM's README has none.
- **A repo with only one figure.** Keep the teaser and the two justified paragraphs, and use `##` for the dataset overview so the outline does not start with stray `###` entries.
## Do's and Don'ts
| Do | Don't |
|---|---|
| Give every figure `width="100%"` and a source at least 2000px wide | Mix figure widths or upscale small images |
| Keep all five badges in one style (`for-the-badge`, emoji label, #555555) | Mix flat and `for-the-badge` badges in one row |
| Tell same-label badges apart by color and message (orange models, yellow dataset) | Use five different colors with no grouping |
| Follow each figure with a short paragraph that explains it | Stack three large figures with no text between them |
| Put affiliation numbers on author names when affiliations are numbered | Number affiliations but leave the names unmarked |
| Use `##` for top-level sections from the first one on | Start the outline with several `###` headings before any `##` |
| Remove the session-time line once the conference is over, or move it into a News list | Leave a dated room and time in the header indefinitely |
## Agent Prompt Guide
1. "Write a README header in GitHub HTML: `` with a 30px-high icon `![]()
` before the full paper title, an `` '🏆 Venue Oral Presentation', a centred paragraph with the session time, and one `
` credit block with linked names, `*` equal-contribution marks, an italic note and ` `-separated affiliations."
2. "Add one centred row of five shields.io badges with `?style=for-the-badge` and an emoji as the label (label #555555): 🌐 Website #007ec6, 📄 Paper #000000, 🤗 SFT Model #ea7233, 🤗 RL Model #ea7233, 🤗 Dataset #d8b800, each with an `alt` equal to its message."
3. "After the badges, place three figures, each as `
` (856px at a 920px column), each followed by one or two `` paragraphs with `` key terms. Then write `## Installation`, `## Datasets`, `## SFT Training`, `## RL Training (GRPO)`, `## Evaluation` and `## Citation`, using `###` subsections, File / Description tables with names in inline code, and `bash` blocks with `#` comment labels."
## Similar Pages
- readme-donaldssh-k-merge: also an ACL 2026 oral, but restrained: a Markdown-in-div header, flat badges and one compact 660px diagram where ProgressLM uses tall button badges and three full-width figures.
- readme-deepmancer-hairport: another header built on a row of `for-the-badge` shields under a centred title, with a bold venue line inside a centred paragraph.
- readme-chuan-10-agile: a centred title and `###` subtitle over a row of `for-the-badge` buttons (Project Page, arXiv, Paper, Hugging Face Data), the same button-row idea with text labels instead of emoji.
- readme-shjo-april-trace: block-letter `for-the-badge` badges under a centred title, followed by numbers rather than figures.
## Quick Start
````markdown
ProjectName: Full Paper Title
🏆 Venue Year Oral Presentation
We will present it on Day, Month DD at Room during Session X, HH:MM-HH:MM. Welcome to join us!
Author One*, Author Two*, Author Three
(* equal contribution)
1Institution A 2Institution B
Task definition with the key idea in bold.
How the data were built, with (i) dimension one and (ii) dimension two.
### Overview of ProjectName-Dataset
## Installation
### SFT Environment
```bash
conda create -n env python=3.11 -y
conda activate env
pip install -e .
```
## Datasets
| File | Description |
|------|-------------|
| `split.jsonl` | What it holds |
## Training
## Evaluation
| Metric | Description |
|--------|-------------|
| **ABC** (Full Name) | What it measures |
## Citation
If you find this work useful, please cite our paper:
```bibtex
@article{key2026,
title={...},
}
```
````
## Provenance
- Repo: https://github.com/ProgressLM/ProgressLM (`README.md`, branch `main`, last README commit 047770c on 2026-05-18).
- Paper: "PROGRESSLM: Towards Progress Reasoning in Vision-Language Models", ACL 2026 (oral, per the README) and ICLR 2026 Workshop on World Models, arXiv 2601.15224; first author Jianshu Zhang. Project page https://progresslm.github.io/ProgressLM/.
- Selection: a presentation banner line and uniform button-style link badges sit over large, high-resolution explanatory figures and a dataset statistics panel, showing how to lead with figures for a figure-heavy paper.
- Evidence: raw Markdown and GitHub-rendered HTML (saved 2026-10-28), 1440×900 and 2700px captures at a 920px README column, shields.io SVGs for each badge URL with capture pixels as a check, and the four PNGs fetched from the repo for pixel sizes, transparency and accent colors.
- Written 2026-10-28.
