Choose a claim
Paste one factual claim or provide supporting source URLs. Keep the date and scope clear, and consent to external research before submission.
MOCHI
AI Jury
MOCHI © 2026
Who We Are:
Check the claim before you share it. Evidence, context and model disagreement for crypto and AI.
1:
Our Ethos
Description
Reliable decisions require independent review and traceable evidence. MOCHI compares findings from multiple models and records where they differ. Agreement informs the assessment; source quality remains essential.
2:
About Us
● What We Do:
MOCHI is building claim research for crypto and AI posts, articles and announcements. Inspect retrieved sources, compare three model findings and choose whether to share the result. The research pilot is separate from the attested document protocol below.
Protocol
Description
5
Independent model classes in the default jury configuration
7
Document schemas specified for the initial release
3–9
Configurable jury size for
different review requirements
80%
Default field-level agreement threshold for five jurors
100%
Required fields must cite evidence; unresolved fields stay HUNG
Independent Review Across the Stack
The Document Protocol
● Layer 00
Documents & versioned task schemas
● Layer 01
Independent, attested model juries
● Layer 02
Evidence, dissent & receipts
Structured findings.
Traceable evidence.
4:
Our Principles
Independent review
01•Model jury
The default jury contains two large generalists from different lineages, a document specialist, a small fast model and a dissenter. Each receives the same source and versioned task schema. The protocol records structured fields, confidence and evidence hashes before comparing results.
MOCHI is in development. The dashboard demonstrates the intended review flow with clearly labelled sample data.
Evidence first
02•Cited fields
A field-level verdict contains the modal answer, the agreement share, evidence references and minority answers. If a required field remains below its configured threshold, the overall status is HUNG. Model consensus is not proof that the source or answer is correct.
Inspect a sample Decision Packet in the dashboard. No demonstration result is a live model run or an on-chain attestation.
Private by design
03•Controlled access
The proposed live protocol encrypts documents to enclave keys and provides payer-controlled evidence access. This frontend does not upload or encrypt documents to live enclaves. Its local previews help developers implement and test those states before production integration.
Production launch requires enclave attestation validation, authentication, secure key management and audited access controls.
Fail closed
04•Honest outcomes
A HUNG result remains unresolved until the requester chooses a larger jury, a human panel or acceptance of the split. Feeds must reject unresolved or cross-check-failed updates and expose the timestamp of the last accepted verdict.
The demo includes a conflicting reserve statement
to show how the unresolved field remains visible.
5:
Build With MOCHI
Bring claim research into your product
MOCHI is designed for readers, community moderators, researchers and applications checking factual claims. The standalone claim research experience returns sources, individual findings and unresolved disagreements. Anonyma is one planned integration.
The claim research pilot requires configured providers and an access token. Paid reviews and production attestations remain pending. Document samples stay available in the original workspace.
6:
How It Works
Description
Review independently
Inspect the evidence bundle first. Three different configured models then assess the claim independently against retrieved sources.
Assess the findings
See whether evidence supports, contradicts or adds context to the claim. Missing evidence and jury disagreement stay visible; agreement does not guarantee correctness.
Retain the evidence
Download the research result or explicitly publish a share link. The pilot record includes sources and model findings; it is not an enclave attestation or chain receipt.
7:
Evidence In Practice
Description
ILLUSTRATIVE CASE STUDY · SAMPLE DATA
One earnings release.
Five independent readings.
In this illustrative review, a fictional issuer reports EPS of 2.11. Four models return 2.11 and one returns 2.10. The decision packet retains both readings and the cited source.
4/5Agreement on the EPS field
One dissenting reading retained
One dissenting reading retained
The EPS field meets the configured 80% threshold. A required field below its threshold would return HUNG and prevent a feed update.
Read the case study →8:
The Road Ahead
Description
PLANNED SEQUENCE · NO RELEASE DATES COMMITTED
Core jury
Versioned schemas, juror selection, attestation verification and field-level consensus.
Verifiable records
Verdict contracts, evidence commitments, signed receipts and independent verification.
Standing feeds
Corporate actions, earnings and attestation feeds with freshness and cross-check controls.
Builder access
Developer SDK, agent access, payment integration and requester-controlled escalation.
Work With Evidence
Investigate the claim. Inspect the sources. Share the evidence when you choose.
On The
Record.
MOCHI / PROPOSED TOKENOMICS
Useful reviews.
Accountable revenue.
The proposal: eligible service surplus funds MOCHI purchases and manual burns. Developer fees stay separate. Paid access and the purchase program are not live.
Explore the fee flow and tokenomics →