# About OCME

OCME (Open Chinese Mathematical Encyclopedia) publishes knowledge, teaching derivations, practice and evidence in Chinese and English. The name preserves its origins; the reading experience is bilingual.

Bilingual question derivations36,000

Knowledge objects302

Concept guides299

Learning paths59



## Understanding verification states



## Question derivations

36,000 have second-context AI mathematical and bilingual semantic review; 0 remain display/symbol-checked candidates.





## Formal evidence

5 Lean evidence objects are linked to 4 knowledge objects; 298 objects have no linked formal evidence.

[Read exact claims and Lean sources](/en/formal/)



## Source and review are separate

Source questions retain candidate and English retains ai_candidate. Second-context AI review is neither human review nor a whole-bank Lean proof. Finite checks support only their stated scope.









## Public data & reading formats

HTML, Markdown and JSON use the same public sources. Search, filters, pagination and language changes retain URL state. Existing question numbers and derivation links remain valid.

Original mathematical data, formulas, ASTs, source snapshots and revision history are preserved. ISQL-M/SREG remain internal research and are not published in this website update.

[AI / plain-text reading entry](/llms.txt) · [Public site catalogue JSON](/data/site/catalogue.json)

Collaboration & publication principlesContent generation, bilingual checks, mathematical review, source adoption, formal proof and publication are separate steps. Evidence supports its exact stated claim; a website update does not promote Canonical truth.

CC BY-SA 4.0 · MIT · EveMissLab



Technical governance: R1–R7 & rolesTrust comes from visible scope, replayable procedures, preserved disagreement, and human retention of decision authority.



## R1 · State and events

Preserve objects, claims, and transitions as verifiable events.



## R2 · Materialization queue

Derive dependency-safe work batches from the Atlas.



## R3 · Design / build / verify

Role outputs follow explicit contracts and do not modify Canonical state directly.



## R4 · Mechanical trust

Bind test results to an exact candidate revision.



## R5 · Risk and disagreement

Do not flatten high risk, correlated bias, or unresolved objections.



## R6 · Real providers

Connect external models through credential boundaries and retain exact model versions.



## R7 · High-order adjudication

Reserve for cross-domain integration and strict mathematical review.



## Three layers of work without authority collapse



## Daily lead architecture

Site, topic selection, research packets, MKOs, user-triggered batches, publication decisions, and cross-role integration.

High frequency · accountable lead



## High-order mathematical review

Frontier models may perform low-frequency whole-library rigor and architecture integration only after explicit discussion.

Low frequency · high-risk review



## Batch question production

An approved fast model produces 100-question candidate batches only after the user starts a governed brief.

Manual start · candidate production