DESIGN.md@docs/reward-modelsA
DESIGN.md@docs/reward-models from fergana-labs/stash is a DESIGN.md file: the design rules coding agents follow in that repo.
mdr add fergana-labs/stash/docs/reward-models/DESIGN.md@git:20261012.1a232d3Pins git:20261012.1a232d3 in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 14 of 14 checks passed
- Kind
- DESIGN.md file
- Repository
- fergana-labs/stash
- Path
docs/reward-models/DESIGN.md- Versions recorded
- 12
- Latest version
- git:20261012.1a232d3, committed Oct 12, 2026
- First version recorded
- Oct 5, 2026
- Size
- 51,600 bytes
- Repository stars
- 340
- License
- MIT
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Stash reward model training platform — design Stash automatically scores imported traces with a shared, versioned action evaluator. An explicitly permissioned corpus of corrections and independently reviewed action comparisons improves future candidates, with frozen benchmarks and gated releases. See [Shared evaluator and learning loop](SHARED_EVALUATOR.md) for the default product path, operator bootstrap, consent, evaluation policy and recurring training controls. The first checkpoint requires reviewed data and remote training; no pretrained shared model ships with the code. Teams can also train personal reward models from selected traces and annotations, the same kind of reward model used in RLHF. A second optimizer, GEPA, writes a skill for the agent (a SKILL.md it loads into its context) using the written comments as feedback and the trained reward model as the metric. This is a separate product area inside the Stash monorepo. For enabled personal accounts, native Codex and Claude Code uploads through `/api/v1/me/transcripts` also populate Traces. Existing recording hooks and `stash sessions import` use …
Read the whole file at its exact version.
How to install
mdr add fergana-labs/stash/docs/reward-models/DESIGN.md@sha256:5b4113bd236f79a9The line at the top of this page pins the label git:20261012.1a232d3, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_4bl3oewyvyenwhvv)
2 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 12 versions of this file, between Oct 5, 2026 and Oct 12, 2026, more than one a day.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20261012.1a232d3 latest | 2026-10-12 | 1a232d3 | 51,600 B | A | view · diff |
| git:20261012.97dcc27 | 2026-10-12 | 97dcc27 | 50,827 B | A | view · diff |
| git:20261009.9dec20f | 2026-10-09 | 9dec20f | 49,887 B | A | view · diff |
| git:20261009.22d0f61 | 2026-10-09 | 22d0f61 | 49,425 B | A | view · diff |
| git:20261009.5426ebe | 2026-10-09 | 5426ebe | 47,531 B | A | view · diff |
| git:20261009.ecc438d | 2026-10-09 | ecc438d | 45,917 B | A | view · diff |
| git:20261009.77fb61a | 2026-10-09 | 77fb61a | 45,905 B | A | view · diff |
| git:20261008.2552042 | 2026-10-08 | 2552042 | 33,025 B | A | view · diff |
| git:20261006.6dc2e6f | 2026-10-06 | 6dc2e6f | 31,812 B | A | view · diff |
| git:20261006.8a92e4c | 2026-10-06 | 8a92e4c | 31,197 B | A | view · diff |
| git:20261006.bf4cebb | 2026-10-06 | bf4cebb | 30,805 B | A | view · diff |
| git:20261005.89ee9d3 | 2026-10-05 | 89ee9d3 | 30,272 B | A | view |
Audit of the latest version
- pass: Size between 200 bytes and 200 KB (51600 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
fergana-labs/stash · 340 stars · license MIT · pushed 2026-10-22 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_4bl3oewyvyenwhvv GET https://markdownregistry.com/api/v1/resolve?ref=fergana-labs/stash/docs/reward-models/DESIGN.md GET https://markdownregistry.com/api/v1/blob/5b4113bd236f79a99d08bcea054d12f8f85a0f7486eaefb4d03738542ab52fe6
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: what DESIGN.md is and how agents use it · which agent file does what.