
Rules
The full rules.
Everything that decides your score, in order. Read the gate first. It comes before the rubric, and it disqualifies wrappers. Registration closed Aug 31. The build window runs Sep 1 to 10, 2026; submissions close Sep 10, 23:59 UTC.
01Timeline
- Aug 16 to 31
Registration
Registration ran Aug 16 to 31 and is closed.
- Sep 1 to 10
Build window
Ten days to build and submit an agent with persistent memory. Submissions close Sep 10, 23:59 UTC.
- Sep 5 to 7
Partner workshops
Live sessions with Base and Virtuals Protocol.
- Sep 11 to 12
Judging
Projects are reviewed and scored on one leaderboard.
- Sep 13 to 15
Winners
The top five are announced, with winner spotlights on X.
All deadlines are in UTC.
02Two stages, in order
Judging runs in order and the stages do not blend. First a pass/fail gate. Then a 100-point rubric, a PMF bonus, and a partner multiplier.
Final score = (rubric + PMF bonus) × multiplier. The leaderboard shows all three.
Every project is reviewed by a panel of judges. The gate is decided by a majority of that panel, and a tie fails the gate. Rubric and PMF scores are the mean of the judges who passed you through the gate.
03The gate: memory must be load-bearing
Sibyl Memory is load-bearing when your core function fails or materially degrades without it. It is on a critical path, not decorative.
The litmus test judges apply: delete the Sibyl Memory layer. Does the project still do what it claims? If yes, it is not load-bearing, and it is disqualified. If no, it passes.
- Cold-start recall in the demo video. A fresh session recalls state written earlier, as one continuous unedited segment with an on-screen timestamp or commit hash.
- Critical-path calls in the repo. The README points to where memory is written and read; a judge can find them in under two minutes.
- The deletion test. Remove the memory calls and the core function breaks.
Load-bearing is about dependence, not volume. One entity your whole build hinges on passes. Ten thousand rows you never read does not.
04The rubric (100 points)
Only projects that pass the gate are scored. Each criterion is scored to its max; the sum is the rubric score.
Memory is load-bearing · 40%
How central and sophisticated the memory use is. Recall is competitive; coordination and dynamic-storage patterns top the band.
Innovation & originality · 25%
A novel idea and real usefulness. Clever but useless caps out low.
Technical execution · 20%
Clean and robust, no smoke and mirrors. Survives a second run and a curious judge.
Pitch & presentation · 15%
A tight 2 to 5 minute story where the load-bearing moment is unmistakable.
05The PMF bonus (up to +10)
A build that solves a real pain point for a real audience earns bonus points on top of the rubric. Genuine means evidenced: a named audience with a validated pain point, a waitlist or design partners, real usage, or pilots. A market-size slide or a claim that users will love it earns nothing.
The default is 0, and that is fine for most prototypes. Any nonzero bonus needs a publicly verifiable artifact a judge can check in five minutes. Fabricated evidence is a disqualification, including after payout.
06The partner multiplier
Verified partner stacks lift your score on a diminishing schedule: +15% for the first, +10% for the second, capped at x1.25. The second stack adds less on purpose, so a two-partner build cannot swamp the rubric or the PMF bonus. Sibyl Memory is mandatory, so it is never a multiplier. A stack counts only when a judge can see it doing real work in the demo, serving the product’s actual function. A claimed stack that is not exercised loses that bonus.
- Base. Deployment is the eligibility floor. An executed onchain action earns the bonus: a wallet operation, an x402 payment, a B20 read, or a contract interaction shown in the demo.
- Virtuals. An ACP job, a registered or transacting agent, or another Virtuals-native integration exercised in the demo.
07Prizes
$10,000 in USDC across the top five.
- 01$4,000USDC on Base + Network School residency
- 02$2,500USDC on Base + one entry into a Base Support Program
- 03$1,500USDC on Base
- 04$1,000USDC on Base
- 05$1,000USDC on Base
08What to submit
- A public GitHub repo under an OSI-approved license (MIT or Apache-2.0), with real commit history.
- A 2 to 5 minute demo video: the problem and who has it, the product, how it works, and how it uses Sibyl Memory. It must include the fresh-session recall beat.
- A README: what it does, where memory is load-bearing, which partner stacks and where, a short “how memory made this possible” note, and a Prior Work declaration.
- Two public posts: the demo video and at least one build-log, tagging @sibylcap and each claimed partner.
09Terms & eligibility
You keep all IP. By submitting, you grant Sibyl Labs and the named partners a non-exclusive, royalty-free license to feature your video, repo, screenshots, and description in event coverage, always with attribution.
Sibyl Labs reference builds and staff are showcase-only and prize-ineligible. Entrants must be 18+, excluding sanctioned jurisdictions. Prizes pay in USDC on Base; to receive one, winners provide payout details, and top-five teams join a short case-study interview. The Network School residency is non-transferable; if a winner cannot take it, it passes to the next rank.
10Utility over slop
This event rewards useful products that solve a real problem. Memory used as a trivial notepad passes the gate but scores at the floor and will not place. Build something people would actually use.
Know the rules? Finish the build.
Registration closed Aug 31. Registered teams: add your repo, demo, and posts to your private team link and mark it ready before Sep 10, 23:59 UTC.
Submit your build →