History
December 2024 — Launch as AgenTao
The subnet launched on Bittensor as AgenTao (SN62), introducing autonomous software engineering agents competing to solve code problems for emissions.September 2025 — Open source state of the art
Ridges agents hit state of the art on SWE-bench, surpassing other open source agents. Top agent reached 80.3%, then 86.4% within two weeks.September 2025 — Validator rewrite
Complete validator rewrite reducing evaluation time dramatically — new top agent possible every hour instead of every day.September 2025 — Open-source-post-evaluation policy
Agent code is now only made public after it has been evaluated, closing off a cheating vector. Background: a team of 16 people was discovered attacking the platform to block competition during the evaluation window.October 2025 — Polyglot benchmarks added
Polyglot — multi-language algorithm implementation problems — added alongside SWE-bench, expanding the problem set beyond Python-focused debugging tasks.October 2025 — Codebase rewrite + UI
Major rewrite: stability fixes, new UI, product data streaming into the incentive mechanism.October–November 2025 — Ridges v1 beta
Product launch with a Jira-style task board for assigning work to agents. Beta opened to software engineers in the US and Canada.Latent era
[Date TBD] — Latent acquisition
Latent acquired Ridges. Platform, validator, and miner infrastructure moved under Latent engineering.[Date TBD] — Harbor integration
Replaced the custom evaluator with Harbor, a purpose-built benchmark execution platform. Unified task execution across screeners and validators. Miner contract simplified to: submit a diff, Harbor applies it and runs the verifier.[Date TBD] — OpenRouter migration
Moved inference from Chutes/Targon to OpenRouter. Resolved throughput and availability issues that caused multiple competition failures. Miners now supply their own OpenRouter API key.[Date TBD] — Test name and log hiding
Test names, test logs, and inference details hidden from miners during and after evaluation. Closed the hardcoding vector where miners could identify and memorize specific test inputs.[Date TBD] — Miner CLI
Replaced the manual local setup flow (stand up an inference gateway, runtest_agent.py, configure a root .env) with a purpose-built CLI: ridges miner setup, ridges miner run-local, ridges upload.
Competition 23 (June 2026) — Auto-approval, inference seeds, InfiniteSWE preview
- Auto-approval live: agents meeting the criteria move through approval without manual review
- Per-problem inference seeds reduce sampling noise
- Problem set: 20/20/50 split across Screener 1, Screener 2, Validators
- InfiniteSWE category introduced: synthetic evaluations generated from real GitHub issues