> ## Content Index
> Fetch the complete content index at: https://www.tao.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# Ridges Introduces Miner-Set Baselines With New Benchmark Phases
- URL: https://www.tao.media/ridges-introduces-miner-set-baselines-with-new-benchmark-phases/
- Published: 2026-10-28T13:39:06.000Z
- Updated: 2026-10-28T13:39:06.000Z
- Description: Qualifying agents will compete to become the initial leader before emissions begin in the next Ridges competition.
- Author: Bart Hillerich
- Tags: Ridges, Subnets, AI, News

Ridges, the AI software-engineering subnet on Bittensor SN62, is introducing a benchmark phase for its next competition, allowing miners to establish the starting performance baseline rather than having the team select one from a previous top agent.

Under the [new competition format](https://x.com/ridges%5Fai/status/2115210793969176732?ref=tao.media), qualifying agents compete for the highest score before any are approved for emissions, the subnet’s token rewards. The winner becomes the initial leader, and subsequent entrants must improve on that agent to qualify for rewards.

During the benchmark phase, submissions go through the usual evaluation and review process. Every agent that clears the judge’s review enters a pool. Once the phase closes, the highest-scoring agent wins, with ties decided first by lower average cost and then by earlier upload.

The winner’s starting reward score follows a different calculation from the usual improvement-based system. Under Ridges’ [benchmark-phase rules](https://docs.ridges.ai/incentive-mechanism?ref=tao.media#benchmark-phases), it receives 100 times its evaluation score: an agent scoring 0.16 earns 16 reward points. Those points set its relative share of the competition’s emissions rather than a fixed token payout.

Benchmark winners do not receive the time multiplier that normally increases the reward for beating a long-standing leader, regardless of how long the phase lasts. Ridges can also open a benchmark phase midway through an existing competition. In that case, a winner cheaper than the incumbent leader earns additional cost units, capped at about 16.7.

After the winner is approved, the competition resumes its usual rules. By default, new agents must score at least 3% higher than the leader, measured relative to the leader’s score, or be at least 6% cheaper while matching or exceeding its score. Individual competitions can set different thresholds and multipliers.

Approved agents’ reward scores decay with a 14-day half-life, and emissions are divided in proportion to their remaining scores. Later improvements can receive a larger initial reward when the leader has gone unbeaten for longer.

The change follows earlier work on evaluation costs. In [Competition 22](https://tao.media/ridges-shares-competition-22-results-after-introducing-cost-controls/?ref=tao.media), Ridges introduced a maximum cost of $0.29 per problem solved and reported nine agents approved for emissions out of 419 created.