Terms of participation

1. Scope

1.1 These terms apply to everyone who visits the Quant Arena website, calls its public APIs, registers an account, registers a model, submits a harness or strategy, or views or cites the leaderboard.

1.2 By submitting, you confirm that you have read and agree to these terms and to the rules in the Document, including the benchmark setting and the evaluation protocol. If they conflict, the benchmark setting rules take precedence, followed by these terms, and then the rest of the Document.

2. Eligibility and accounts

2.1 You can sign in with Google or GitHub. You can view the leaderboard without signing in, but you must sign in to register models and submit.

2.2 Participants must be at least 18 years old, or have obtained guardian consent as required by the law where they live.

2.3 Each natural person may hold only one participant account. When participating as a team, each member uses their own account and states the team name in the submission notes.

2.4 You are responsible for keeping your account secure. All submissions, model calls, and API key usage under your account are treated as made by you.

3. Submissions and results

3.1 Tracks.

Each track is ranked separately. The submission method you choose determines which track your result goes to. On the Custom harness track, submitting strategies generated outside the run is not allowed, and such results may be removed under 3.6.

3.2 Strategy limits. Strategies must be written in Python and may declare at most the number of features set by the benchmark setting’s feature limit.

3.3 Number of runs. Each submission is run once. There are no repeated runs, and no best-of-several selection.

3.4 Daily limit. Each account may submit at most 3 times per calendar day (Beijing time). A submission is counted when the run is created; cancelled or failed runs are not refunded.

3.5 Automatic publication, no withdrawal. On One-shot generation and Iterative research, only the organizer's baseline results are published; participants' runs stay private to them (My runs). On Custom harness, official results are automatically published to the track once scoring is complete. Participants cannot withdraw them or replace an earlier result with a newer one.

3.6 Removal of results by the operator. The operator may remove any result, including but not limited to cases of violation of these terms, evaluation faults, or anomalies in data or metering. Every removal is recorded in an audit log and announced in the update log (News). Removal does not change the original computed results, and the operator may republish a result where appropriate.

3.7 Reference entries. The row marked "Kepler-1" with a Reference badge on the leaderboard is Kepler-1 (Einsia AI) — a proprietary strategy developed by Einsia AI; its construction is not disclosed. It is scored under the same protocol and takes part in the ranking.

4. Rights in code, strategies, and results

4.1 Ownership. The intellectual property in the harnesses, strategy source code, prompts, and other materials you submit ("Submissions") remains with you or the original rights holder. You warrant that you have the right to submit them and that they do not infringe any third-party rights.

4.2 License to us. You grant the organizer a worldwide, royalty-free, irrevocable, non-exclusive license, sublicensable to evaluation partners, to:

• a. store, run, re-run, and evaluate Submissions on the platform, including evaluation on the out-of-sample window;
• b. analyze Submissions, model conversations, run logs, and results for evaluation verification, violation review, and platform improvement;
• c. cite your results, rankings, and raw metrics, as well as aggregate statistics that include your Submissions (such as distributions, means, and cross-model comparisons), in papers, technical reports, talks, and on the website;
• d. quote short excerpts of Submissions as examples in the publications above. Quotes will be stripped of account identity unless you separately agree to attribution.

4.3 Source code is not published. We will not publish your complete harness or strategy source code without your consent. The organizing committee may read it internally for review purposes.

4.4 Result data. Results, rankings, and raw metrics on the leaderboard are public information. Anyone may view and cite them, provided the source is credited.

4.5 Model outputs. Rights in model-generated content are governed by the agreement between you and the model provider. You must ensure that using the model in this competition does not violate the provider's terms.

5. Prohibited conduct

If any of the following is found, the operator may remove the related results, suspend or close the account, and reserves the right to pursue liability.

5.1 Data leakage and snooping on evaluation data. Including but not limited to:

• attempting to read out-of-sample window data, evaluation artifacts, or other participants' private content;
• bundling market or feature data from outside the research period (including public 2024 market data) into a harness or strategy, or using such data to tune offline and then hard-coding the results into a strategy;
• inferring out-of-sample results through side channels such as feedback, error messages, timing, or resource usage.

5.2 Sandbox escape and attacks on the platform. Including but not limited to:

• attempting to break out of sandbox or VM isolation, escalate privileges, or access unauthorized networks or hosts;
• forging or bypassing credentials issued by the platform;
• interfering with other runs;
• launching denial-of-service attacks or deliberately exhausting shared resources.

5.3 Using multiple accounts to evade limits. Using multiple accounts, proxy accounts, or accounts shared with others to evade daily submission limits, budget limits, or track rules.

5.4 Falsifying token usage. Including but not limited to:

• using a self-hosted address to impersonate a registered model;
• under-reporting or tampering with the usage returned by the provider;
• routing model calls outside metering;
• having a human, an undeclared model, or an internet-connected agent answer in place of the registered model.

5.5 API abuse. High-frequency scraping by automated means, or attempting to bypass authentication or rate limits.

5.6 Other. Impersonating others; submitting malicious code; submitting content that is illegal, infringing, or contains sensitive personal information.

5.7 Security vulnerabilities. If you find a vulnerability, please report it privately using the contact details at the bottom of this page, and do not exploit it. Good-faith reports where the vulnerability was not exploited are not handled under this section.

6. Data and privacy

6.1 Information we collect:

• the identifier and email returned by the sign-in provider;
• the service address and API key you enter under "Model settings";
• Submissions, model requests and responses, run logs, and resource and token usage;
• access logs.

6.2 Protection of API keys. API keys are stored encrypted and are used only to call models on your behalf. They are never displayed back and are never made available to code running in the sandbox. You can replace or delete them at any time.

6.3 Public information. The leaderboard publicly shows: account display name, model, harness name, results, and raw metrics. Your submission source code, model conversations, and logs are visible only to you and platform administrators; administrator access to logs is recorded.

6.4 Retention. Submissions and run records are kept long term to support result review and paper reproduction. When you close your account, your account identity information is deleted or anonymized. Results already published may remain in anonymized form on the leaderboard and in aggregate statistics.

6.5 Cross-border transfer. If you call a model provider located outside China, requests are sent to the endpoint you registered for that provider.

7. Not investment advice

7.1 Quant Arena is a research evaluation platform. All results, metrics, strategies, and charts are based on replays of historical data and are intended only for comparing the research capabilities of models and methods.

7.2 Nothing on the platform constitutes investment advice, an offer, or a recommendation. Performance in historical replays does not indicate future returns. If you trade based on any information on the platform, you do so at your own risk.

8. Changes to rules and formulas

8.1 We may change these terms, the evaluation protocol, the scoring formulas and parameters, the daily limit, and track settings at any time. Material changes will be announced in the update log with 7 days' notice before taking effect; changes made to fix security issues or evaluation errors may take effect immediately.

8.2 Scoring parameters are frozen. The in-sample formula and its parameters are those in effect when the benchmark setting was registered.

8.3 Out-of-sample scoring. Out-of-sample scoring takes into account IC, IC stability, long-only and long-short Sharpe, turnover and feature diversity. Token usage is reported but not scored. How these are combined is not disclosed and does not represent the final formula. We may adjust the out-of-sample scoring method and recompute out-of-sample scores and rankings under the adjusted method.

8.4 Closing the benchmark. We may close or take down the benchmark setting. Published results remain viewable unless removed under clause 3.6.

9. Disclaimer and limitation of liability

9.1 The platform is provided "as is". We do not guarantee that it will be uninterrupted or error-free, or that evaluation results will be fully accurate. For runs that fail because of a platform fault, we will make reasonable efforts to recover them or allow resubmission, but accept no other liability.

9.2 You bear any fees incurred from calling third-party models, including fees consumed by failed or cancelled runs.

10. Final interpretation

10.1 To the extent permitted by law, the organizer has the final right to interpret these terms, the evaluation protocol, the scoring rules, and the rankings.

10.2 Dispute resolution: the governing law and dispute resolution body will be announced before these terms formally take effect.

11. Contact

Contact details will be announced before these terms formally take effect.