TrajectoryRL is Bittensor subnet 11 (SN11), a project in the Code Generation and Development Tools lane. Miners write skills, agents install them.
A 24/7 competition that produces ready-to-use skills for AI agents. Miners write the instructions, validators grade them inside real sandboxes by running the tests, and only the winners get published for anyone to install.
It is an educational deep dive, not a recommendation to stake. For live price, pool depth, and performance, always open the live subnet page.
What is TrajectoryRL (SN11)?
TrajectoryRL (Subnet 11) is a competition that produces "skills" for AI agents. A skill here is a plain-text file, a SKILL.md, that teaches an agent how to handle a specific kind of task: the instructions, strategies, and gotchas it should follow. Miners compete to write the best one, validators test each entry in a real sandbox, and the winning skill gets published so anyone can install it.
In plain English: It is like an app store for AI agents, except the apps are written by an always-on competition and only the ones that pass real, executable tests make it onto the shelf.
Centralized reference points: No direct equivalent (closest: agent skill libraries, Terminal-Bench-style benchmarks)
How miners and validators work
Like every Bittensor subnet, SN11 is a two-sided market: miners produce work, validators score it, and emissions follow those weights under the network’s dTAO economics.
- Miners write a SKILL.md, bundle it into a pack identified by its SHA256 hash, upload the pack to any public URL, and commit that hash on-chain. No GPU, no server, no uptime needed.
- Validators fetch each pack, verify the hash against the on-chain commitment, then run an agent inside a sandboxed scenario and score how many of that scenario's tests pass with the skill loaded.
Mechanism and design
Per the project's repository, mining is writing a SKILL.md and packaging it into a content-addressed pack. The miner uploads the pack to any public HTTP endpoint and commits its hash on-chain, so no server or GPU is required. Validators pull the pack, verify the SHA256 hash against the commitment, and evaluate it with the project's trajrl-bench harness: the agent runs inside each scenario's own container (the scenarios use the Terminal-Bench format), and a fresh verifier container scores the result with pytest. Quality for a scenario is continuous, the share of tests passed, and the final score is the equal-weighted sum across scenarios, with cost reported as a separate axis.
Rewards are winner-take-all. The seated winner takes 100% of miner emissions, and a challenger only takes the seat if it clears a small takeover margin, described in the v6.0 incentive document as a score-dependent margin around 3% that decays toward zero as scores approach the ceiling. Consensus across validators is coordinated by the project's platform server and finalized as a trimmed (Winsorized) mean of qualified validator votes, gated by a stake quorum. When no miner holds the seat, the emission is directed to the owner UID and burned. First-mover protection and a similarity check (normalized compression distance) discourage copying a leading pack.
One thing worth flagging for accuracy: the framing has moved. The repository now describes SN11 as an "open skill factory" built on the trajrl-bench scenario suite, while some third-party descriptions and the on-chain blurb still use the earlier "trajectory and policy optimization" language tied to an older benchmark name. Where they diverge, this article follows the subnet's own repository, which is the current and authoritative source for how the mechanism works today.…
Why the project exists
- The problem it solves: Agents need task-specific know-how, and there are far more useful skills to write than any single team can ship. Most "agent skills" circulating today are untested prompt snippets. TrajectoryRL only publishes skills that passed real, runnable tests.
- The opportunity: If agents become a primary computing platform, the skills that run on them are the software layer. A neutral, test-driven place that produces and ranks those skills could matter well beyond any one subnet.
- The Bittensor advantage: The competition runs continuously and rewards the current best entry, so the quality bar rises every season without a central curator deciding what counts as good. Skills are judged by passing tests, not by popularity.
- Traction signals: The repository is actively developed: the latest commit is dated 2026-06-20, with 763 commits from 10 contributors and a current release tagged 0.6.24. The consumer side ships too, via a published
pip install trajrlCLI for installing skills the subnet has produced. Being honest about the chain side: on-chain participation is thin right now. TaoSwap shows a single active miner and the subnet's share of network emissions is currently 0%.
What stakers should check before buying alpha
Subnet storytelling is cheap. On-chain and product checks are not. Use the same framework as our subnet metrics guide:
| Check | Why |
|---|---|
| Real product / docs / repo | Separates shipping teams from narrative-only slots |
| Pool depth (TAO) | Your exit door: thin pools amplify slippage |
| Net flows (7d / 30d) | Under flow-based emissions, capital votes with its feet |
| Emission share | Whether the market is funding this subnet right now |
| Ownership concentration | Large holders can move a small alpha market |
| Immunity / deregistration | Post-immunity weak slots can lose the race (see immunity guide) |
Pull the live numbers on TrajectoryRL · SN11 rather than trusting any static screenshot in a blog post.
Risk factors
- Deregistration: SN11 is past its 4-month network immunity, and its current emission share is 0% with net staking outflows over the past week. Bittensor automatically deregisters the non-immune subnet holding the lowest EMA price, so a subnet in this position carries real deregistration risk until net flows turn positive.
- Thin competition and concentration: TaoSwap shows a single active miner. Because rewards are winner-take-all, one entry can take the full miner share, and when no entry holds the seat the emission burns to the owner rather than circulating. The model needs more competing miners and validators to demonstrate the rising-bar dynamic it is designed around.
- Early-stage execution: The Season 2 "skill forge" flywheel that would make the system self-sustaining is documented but not yet live, so the current incentive depends on a hand-built scenario set and the team's continued cadence.
- Miner IP terms: By the repository's own submission terms, copyright in any pack that earns a reward transfers to trajrl.com, and evaluation artifacts belong to the platform. Worth knowing before mining; non-winning packs keep their author's copyright, and the repository code is MIT licensed.
How this fits a TAO portfolio
Many holders keep a root vs subnet split: root for calmer TAO exposure, subnets like TrajectoryRL for thematic upside (and higher variance). If you use TaoScale, you can size SN11 from your wallet without handing over custody. See non-custodial staking.
- Live market: SN11 on TaoScale
- Browse and stake: /invest
- More subnet explainers: /blog/subnets
FAQ
What does TrajectoryRL do on Bittensor?
It is like an app store for AI agents, except the apps are written by an always-on competition and only the ones that pass real, executable tests make it onto the shelf.
What is the SN11 alpha token?
When you stake TAO into subnet 11, you receive that subnet’s alpha. Its price versus TAO moves with the subnet’s pool (see what is dTAO).
How do I research TrajectoryRL before staking?
Read the mechanism above, open the official repo/docs when available, then verify pool depth, flows, and activity on SN11. Cross-check with the metrics checklist.
Is this financial advice?
No. TaoScale publishes educational explainers and live tools. Decisions to stake are yours alone.
Not financial advice. Educational overview of TrajectoryRL (SN11) compiled from public project materials and ecosystem documentation. Mechanisms and market structure change, so verify live data and official sources before acting.
