Counted
Counted: proof of which real quantum computer each team used.
An organiser's own AI agent can build the same evidence page from their IBM credentials.
Judges at quantum hackathons usually cannot verify whether teams really ran their projects on quantum hardware.
Simulators can imitate hardware, so a claim of real quantum hardware is easy to fake or to miss in the fine print. Organisers also hand out scarce quantum computer time with no per-team record, so honest teams cannot prove their work either.
Counted gives every team a checkable evidence page built from the real hardware job records.
- 1Organiser exports job records locally; the API key never leaves their machine.
- 2Each team gets an evidence card showing backend, shots, time and fingerprint.
- 3Judges check claims against the job records, with each flag linked to its source.
Why it failed
Check 2 · AI can do itThe trust comes from the organiser holding the provider credentials, not from Counted. A quantum club could ask Claude Code or ChatGPT for a script that pulls job IDs, backends, shots and counts hashes from its own IBM instance, hashes them, compares circuits across teams, tracks per-team QPU minutes and publishes a static page. Judges already trust the organiser, so no separate third-party trust layer is needed, and the idea itself notes there is no verified case of organisers paying for verification.
- 01
Get three organisers to pay before building anything beyond a script.
- 02
Prove judges need third-party trust that organiser-run scripts cannot give.
- 03
Interview ten organisers about verification pain and what they already do manually.
- PhaseScope
A Fall Fest winner that traces a result on ibm_fez back to the raw counts. PhaseScope is a team's own self-reported trace for one experiment. Counted is the organiser-side layer that gives every team in an event the same provider-sourced evidence card, so judges compare like with like.
- GhostQPU
A hackathon MVP that detects quantum cloud jobs forged by a calibration-based simulator. GhostQPU infers fakery statistically from the data. Counted avoids the arms race by pulling records directly from the organiser's instance, and it adds the event workflow, judge view and Devpost linking.
- quantum-verifier
A verification gate that checks an AI-generated circuit's claim against simulation, device noise and a control experiment before it runs. It is a pre-run safety check for one developer. Counted is a post-run, multi-team audit trail for events. It checks what a team says it ran against what the provider logged, and it does not simulate.
- A Fall Fest winner built a reproducibility bundle by hand: per-job raw counts, versions, timing, seeds and file hashes. Most teams will not do this, and judges have no standard way to check it.github.com ↗
- Another 2026 Fall Fest project puts a claim-versus-evidence table in its README, with a hardware job ID and a link to the raw counts. Teams are inventing this format themselves.github.com ↗
- The standard hackathon requirement is a live demo showing outputs and counts. That is a demo, not a verification of provenance.atomiq-hackathon-qff-26-iem.devpost.com ↗
- A 2026 hackathon repo discloses that all of its scored results are simulated and that its one hardware run did not reproduce the simulated features. Hardware and simulator claims are mixed in practice.github.com ↗
- A public hackathon MVP shows a simulator built from public calibration data can forge hardware-like jobs. Judges cannot assume pasted counts are genuine, which gives a reason to anchor to provider records.github.com ↗
- A 2026 arXiv paper says quantum computing lacks provenance infrastructure and that provenance is fragmented across providers.arxiv.org ↗