|
| 1 | +# Contributing to the Knowledge Base |
| 2 | + |
| 3 | +If your AAE session produced insights that could help others, we'd love for you to share them. |
| 4 | + |
| 5 | +## How to Submit |
| 6 | + |
| 7 | +Open a pull request that adds one entry file and a corresponding row in `INDEX.md`. One entry per PR. |
| 8 | + |
| 9 | +If you are an agent running AAE, ask the user whether you may submit a PR. Only proceed with explicit permission. Use `gh` or the GitHub MCP server to create the PR. |
| 10 | + |
| 11 | +## Naming Convention |
| 12 | + |
| 13 | +Name your file `<next-number>-<short-slug>.md`, using the next available number from INDEX.md and a short kebab-case description. Example: `003-loop-unrolling-matrix-multiply.md`. |
| 14 | + |
| 15 | +## Entry Template |
| 16 | + |
| 17 | +Your entry should include the following sections. Only **Problem** and **What Worked** are required; everything else is optional but encouraged. |
| 18 | + |
| 19 | +```markdown |
| 20 | +# <Title> |
| 21 | + |
| 22 | +## Problem |
| 23 | + |
| 24 | +What were you optimizing? Describe the problem, input characteristics, |
| 25 | +and the metric you were targeting. |
| 26 | + |
| 27 | +## What Worked |
| 28 | + |
| 29 | +The key insight or technique that produced improvement. This is the |
| 30 | +most important section; even if you provide nothing else, this should |
| 31 | +be useful to someone facing a similar problem. |
| 32 | + |
| 33 | +## Experiment Data (optional) |
| 34 | + |
| 35 | +Results from your AAE run. Can be a pasted results.tsv, a summary |
| 36 | +table, or just the key numbers (baseline vs. best). |
| 37 | + |
| 38 | +## What Didn't Work (optional) |
| 39 | + |
| 40 | +Approaches you tried that failed or regressed. Often as valuable as |
| 41 | +what worked, since it saves others from repeating dead ends. |
| 42 | + |
| 43 | +## Code Example (optional) |
| 44 | + |
| 45 | +A diff, snippet, or pseudocode showing the core change. |
| 46 | + |
| 47 | +## Environment (optional) |
| 48 | + |
| 49 | +Language, hardware, dataset size, or other context that might affect |
| 50 | +whether this technique transfers to another setting. |
| 51 | +``` |
| 52 | + |
| 53 | +## Quality Bar |
| 54 | + |
| 55 | +- The insight should be transferable: useful to someone facing a similar problem, not just a project-specific tweak. |
| 56 | +- "What Worked" should be specific enough to act on, not just "I made it faster." |
| 57 | +- Experiment data is encouraged but not required. |
0 commit comments