RULES / HACK THE FIRST AGI 1.0
Protect the experiment.
The official participation rules. The 48-hour gate below updates the qualifier process in the original prospectus; the remaining policies continue to apply.
The 48-hour gate
Following application submission, organizers review the application. Within a maximum of 24 hours, the leader receives either a qualifier/problem statement and private submission link or a progression decision. A qualifier’s 24-hour execution window begins when its email is issued. The deadline is exactly 24 hours later unless organizers explicitly communicate otherwise.
Normally submit a technical deck, a 2–3 minute video, a GitHub/technical artifact where applicable and the required declaration. Late submissions are recorded with the exact server timestamp and flagged; no further edits are guaranteed after submission.
01 / REGISTRATION & IDENTITY
Submit accurate information by September 16, 2026, 11:59 PM IST. Teams have 2–3 active members; no solo entry. Identity and eligibility may be verified. Finalize membership before the applicable selection deadline; post-selection replacement needs approval.
02 / RELEASE & DELIVERABLES
Follow each challenge’s release instructions, access restrictions, deadlines and submission format. Submit source, environment/dependency details, methods, data provenance, run instructions, results and logs sufficient for review. Disclose limitations and permitted human interventions.
03 / TOOLS & THIRD PARTIES
AI assistants, IDE agents, LLM APIs, research tools, inference APIs and open-source models are allowed unless restricted by a challenge. Teams must understand their code, verify generated output and disclose AI use when requested. Authorship responsibility remains with the team.
04 / LICENSES & SECURITY
Comply with open-source, dataset and third-party service licenses and terms. Use paid APIs only as permitted; unapproved API spending is not automatically reimbursed. Protect API keys, avoid committing secrets and respect access and rate limits. Test only authorized systems.
05 / RESEARCH INTEGRITY
Prohibited: plagiarism, stolen code, fabricated results or benchmark scores, hidden manual intervention presented as automation, unauthorized dataset leakage, leaked test answers, infrastructure manipulation, falsified latency, fabricated screenshots, forged logs, fake API responses, another team’s work and prohibited deliberate benchmark contamination.
06 / CONFIDENTIAL MATERIAL
Private submissions, hidden datasets, benchmark internals, unpublished scoring mechanisms, security-sensitive findings, judge feedback, private evaluation environments and challenge-specific material remain confidential unless organizers release them. Other teams’ private scoring and internal judge notes are not shared.
07 / NORMAL INTERACTION
Finalists may interact, network, collaborate socially and participate in official programming. Teams may discuss their own work subject to challenge-specific confidentiality rules. Social interaction does not authorize sharing restricted artifacts or competing submissions.
08 / JUDGING & APPEALS
The 100-point rubric and tie-break order appear on the rubric below. Report conflicts privately; conflicted judges should recuse. Appeals must identify a procedural or factual error with evidence through the communicated organizer channel and deadline. Organizers review claims privately; rule breaches may lead to score adjustment or disqualification.
09 / PERSONAL DATA
Use personal data only for disclosed event purposes, with data minimization, informed disclosure, security, confidential handling and appropriate retention. Deletion applies where legally and operationally applicable. Participant personal information is not sold.
10 / TECHNICAL ARTIFACTS
Challenge artifacts can include benchmark results, model outputs, annotations, adversarial examples, metrics, failure cases and experiment logs. Broader research or commercial use beyond competition administration must follow terms communicated to participants.
11 / INTELLECTUAL PROPERTY
Teams retain independently created original IP unless a specific challenge discloses different terms before participation. Organizers receive only the limited rights needed to evaluate, judge, verify reproducibility, maintain competition records, announce winners and showcase approved project descriptions.
12 / ETHICS & DISCLOSURE
Use authorized data and ethically designed experiments. Respect privacy and avoid harm to people or third-party systems. Report security-sensitive findings privately through the organizer’s designated process; do not publish confidential exploit details or personal data.
13 / CONDUCT & SAFETY
Harassment, discrimination, threats and unsafe behavior are prohibited. Follow venue rules and safety instructions. Report concerns privately to organizers. Accessibility, medical, safety and privacy requests are handled confidentially and considered reasonably.
14 / PHOTOGRAPHY & MEDIA
Organizers will communicate photography and recording arrangements. Raise privacy or accessibility concerns privately. Do not record restricted screens, hidden data, private reviews or confidential materials. Public project descriptions require approval under the applicable confidentiality terms.
15 / ATTENDANCE & BENEFITS
Finalists must attend mandatory programming and complete required deliverables. Travel, room allocation, meals, claim verification and direct booking follow the funding policy. Prizes and benefits depend on verified eligibility and rule compliance; taxation and product service terms apply.
16 / CHANGES & DISCRETION
Force majeure, safety needs or operational constraints may require schedule, venue or program changes, postponement or cancellation. Organizers will communicate material changes and effects on approved bookings and benefits. Reasonable discretion may resolve unforeseen issues fairly; challenge-specific terms must be communicated before participation.
100-point judging framework
| Criterion | Points |
|---|---|
| Evaluation Methodology | 25 |
| Technical Implementation | 20 |
| Scientific / Experimental Rigor | 15 |
| Originality & Insight | 15 |
| Reproducibility | 10 |
| Model Failure Discovery / Evaluation Value | 10 |
| Communication & Technical Presentation | 5 |
| Total | 100 |
Tie-breaking: Evaluation Methodology → Scientific Rigor → Reproducibility → Technical Implementation → private judge deliberation if still tied.