| Type | README |
|---|---|
| Updated | 2026-08-07 11:47:00 +0900 |
| Status | discussion-draft |
| Tags | licium, digital-identity, authentication, repository, evaluation, sqlite |
| Description | A minimal substrate and executable reference model for composing digital identities from values and relations. |
Start here/はじめに: Concept Guide | コンセプトガイド
Identity emerges from values and relations.
Licium explores a deliberately small substrate for composing Digital Identity: multivalued ID–Value associations below directories, protocols, and application-specific identity objects.
ID -> { Value, Value, Value, ... }
A Digital Identity is not an Entity or a permanent user entry. It is a purpose-specific set of values and relations derived from explicitly pinned Source data.
Source data at a pinned knowledge cut
-> complete Repository state / Root
-> Definition / Profile / Context evaluation
-> derived Identity values and relations
-> optional materialized projection
Directory, LDAP, GraphQL, SCIM, OIDC, SAML, authorization systems, and credentials may consume or project this data. They are not mandatory primitives in the minimal substrate.
New to Licium? Start with the concept guide or its Japanese version. It explains the model through one Alice example before the contract and evidence documents introduce their precise boundaries.
Licium is a design and model-validation project, not a stable specification or production-ready IdP.
What exists:
- public-facing design summaries;
- a disposable SQLite and POSIX-shell reference slice;
- fixed expected outputs and negative mutations for that slice.
- a finite runnable Protocol Integration Proof sample.
What does not exist:
- Rust source or a Cargo workspace;
- a production service;
- a real Spanner adapter or conformance result;
- production OIDC, SAML, LDAP, or SCIM servers;
- a completed authentication or authorization product.
- A disposable finite Protocol Integration Proof is implemented and included in this repository tree. Its technical presence does not record repository-specific review or authorization state.
- The sample runs an Authorization Code + S256 PKCE loopback flow with pinned
oidc-provider9.11.1, a replaceable adapter, and exact SQLite and flat-file/POSIX providers. - The engine owns protocol flow, sessions, tokens, and signing. The test RP owns its verifier, state, nonce, and token validation. Licium supplies a protocol-neutral authentication request, pinned evaluation, subject decision, and selected Identity projection.
- The aggregate runner binds 63 suites: 43 positive/boundary verifiers and 20 negative self-tests. Static closure covers 71 shell files and 15 project-authored ESM files.
- Dependencies are not vendored. Reproduction requires exact Node v22.22.2
and retrieves the pinned npm dependency graph from
package-lock.json. - The runtime uses loopback
127.0.0.1:56100;127.0.0.1:56101is only a redirect identifier and needs no listener. - This remains a synthetic, disposable, non-production slice. It does not make Licium a production IdP, put OIDC into Licium Core, certify OIDC or security, establish arbitrary-backend portability, or implement Spanner or Rust.
- Read the proof summary or run the sample.
- Artifact review, local preparation, commit review, and push authorization are external operational records. Inclusion of these bytes grants no standing authorization for a later repository mutation or push.
- Treat this finite integration as the pre-Rust toy IdP vertical slice for learning the boundary, not as source code to translate line by line.
- Keep the pinned generic OIDC engine and protocol-facing adapter outside Licium Core. Licium continues to expose a protocol-neutral backend contract.
- In the first Rust milestone, replace exactly one protocol-neutral provider while leaving the OIDC engine and protocol-facing adapter unchanged.
- Keep the SQLite and flat-file/POSIX providers as comparison oracles and add the Rust provider as a third implementation of the same bounded contract.
- This Roadmap does not select a stable Core API, arbitrary-backend support, Spanner, an async runtime, an ORM, a transport, or a production IdP design.
The exact test-only sqlite-reference-v0 tuple later recorded 83/83 PASS in
two fresh isolated sealed sessions. BC01--BC12 and the overall report were
PASS; FAIL, UNTESTED, UNAVAILABLE, and INVALID were all zero. This
result is limited to the recorded implementation, adapter, profile, SQLite
runtime, configuration, and tested SUT revision
630adfd92cee8ce19249f401b6238e5261a4b33d.
The accepted result also passed current-source closure, session binding,
evidence sealing, 19 run self-test controls, 6 session self-test controls, and
an independent exact-report review with disposition ACCEPTED. This
609-file reviewer package includes the complete 472-file SQLite replay source
used to generate a fresh full session.
Publication still has distinct states: technical acceptance, project-owner artifact countersign, sibling local preparation, local-commit review, and public-push authorization. Before sibling local preparation, the project owner must countersign the exact manifest, all 609 rights/data rows, the material-claim inventory, and the safety exceptions. That countersign permits only the bound local preparation; it does not authorize a public push. Public push requires a later, separate project-owner authorization after local-commit review.
Read SQLite Backend Conformance v0 Result. It separates the quick static integrity check of a compact 47-file evidence reference from multi-hour fresh runtime and negative self-tests; the static reference is not a standalone sealed session or a substitute for a fresh full gate.
Reviewers are invited to inspect the exact tuple, evidence boundary, and
reproduction instructions, and to report reproducible counterexamples or
claims that exceed the stated boundary. This is not SQLite product or
production-service certification, and Spanner remains UNTESTED.
Technical review does not authorize a public push. Public GitHub publication
requires separate explicit project-owner authorization.
The Repository/Evaluation lifecycle boundary and its backend-independent conformance questions are now concrete enough for external review.
Feedback is especially useful on:
- whether Root, Publication, Head, Result, Effect, View, and Observation remain cleanly separated;
- whether the candidate claims stay within the included evidence;
- whether the Backend Conformance observations are independent of SQLite;
- which Identity Definition, Context, delegation, and lineage semantics are still missing.
The shortest review path is:
README
-> Candidate Core Contract
-> Candidate Backend Conformance Contract
-> Evidence Map / executable reference slice
The candidate contracts are proposed requirements. The included executable evidence remains E67–E72; predecessor evidence and E73R are not bundled.
The included E67–E72 reference slice connects one small lifecycle:
Pair ingest
-> complete Root
-> accepted Publication / Head
-> pinned Evaluation
-> Result
-> optional Repository Effect + Decision Observation
-> View
-> restart / historical replay / explanation
Requirements are a POSIX shell, common Unix utilities, and sqlite3.
sh model/run-reference-slice-tests.sh
The successful aggregate covers six targeted experiments, 30 targeted relations, and 25 negative identities. See the reproduction guide and evidence map.
The reference slice does not establish production durability, distributed consensus, external-effect atomicity, performance, cryptographic trust, authorization correctness, real Spanner conformance, a final Rust API, or a final storage schema. SQLite tables are a disposable realization, not the Licium Core or a wire format.
- Concept guide / 日本語
- Current design / 日本語
- Model boundary / 日本語
- Evidence and non-conclusions / 日本語
- Candidate Repository / Evaluation Lifecycle Core Contract / 日本語
- Candidate Backend Conformance Contract / 日本語
- Public Sample Policy / 日本語
- Reference slice in English
- Reference slice in Japanese
- Comparisons with Zanzibar, Macaroons, and related systems / 日本語
- Reproduction guide / 日本語
The two superseded historical records are already written primarily in Japanese and remain at their original paths: Identity Generation Hypothesis and Licium Minimal Data Model.
- the exact Pair/Tuple boundary between physical storage and semantic data;
- which logical objects require stable lineage identifiers;
- final Definition, Profile, Context, delegation, and Grant semantics;
- distributed publication and convergence;
- archive capacity, legal erasure, and an explicit destructive protocol;
- Rust type boundaries and backend capability contracts;
- production SQLite deployment behavior and real Spanner adapter conformance and performance.
MIT License. See LICENSE.
アイデンティティは、値と関係から立ち現れる。
Liciumは、デジタルアイデンティティを構成するための、意図的に小さい基盤を 探究しています。Directory、protocol、application固有のidentity objectより下に 位置する、多値のID–Value associationが出発点です。
ID -> { Value, Value, Value, ... }
デジタルアイデンティティはEntityそのものでも、永続的なuser entryでも ありません。明示的に固定されたSource dataから導出される、目的別の値と関係の 集合です。
固定されたknowledge cut上のSource data
-> 完全なRepository state / Root
-> Definition / Profile / Context evaluation
-> 導出されたIdentityの値と関係
-> 任意のmaterialized projection
Directory、LDAP、GraphQL、SCIM、OIDC、SAML、authorization system、credentialは、 このdataを利用またはprojectionできます。これらはminimal substrateの必須 primitiveではありません。
Liciumを初めて読む場合は、まず コンセプトガイドまたは English versionを参照してください。一つのAliceの例から modelを説明し、その後にcontractとevidence文書が正確な境界を示します。
Liciumは設計とmodel-validationのprojectであり、stable specificationでも production-ready IdPでもありません。
存在するもの:
- 公開用の設計summary;
- 使い捨てる前提のSQLite/POSIX shell reference slice;
- reference sliceに対する固定expected outputとnegative mutation。
- finite runnable Protocol Integration Proof sample。
存在しないもの:
- Rust sourceまたはCargo workspace;
- production service;
- 実際のSpanner adapterまたはconformance result;
- production OIDC、SAML、LDAP、SCIM server;
- 完成したauthentication/authorization product。
- 使い捨てのfinite Protocol Integration Proofを実装し、このrepository treeへ 含めた。技術的な存在はrepository固有のreview/authorization stateを記録しない。
- sampleは、固定した
oidc-provider9.11.1、交換可能なadapter、exact SQLite/ flat-file/POSIX providersを使い、Authorization Code + S256 PKCEのloopback flowを実行する。 - engineがprotocol flow、session、token、署名を所有する。test RPがverifier、 state、nonce、token validationを所有する。Liciumはprotocol-neutralな authentication request、pinned evaluation、subject decision、selected Identity projectionを供給する。
- aggregate runnerは63 suitesを固定する。43のpositive/boundary verifierと20の negative self-testである。static closureは71 shell filesと15 project-authored ESM filesをcoverする。
- dependencyはvendorしない。再現にはexact Node v22.22.2が必要で、
package-lock.jsonから固定npm dependency graphを取得する。 - runtimeはloopback
127.0.0.1:56100を使う。127.0.0.1:56101はredirect identifierだけで、listenerは不要である。 - これはsynthetic、disposable、non-productionのsliceである。Liciumをproduction IdPにせず、OIDCをLicium Coreへ入れず、OIDC/security certification、任意backend portability、Spanner/Rust implementationを確立しない。
- proof summaryを読む、 またはsampleを実行する。
- artifact review、local preparation、commit review、push authorizationはartifact 外のoperational recordである。このbytesの存在は、後続repository mutation/push へのstanding authorizationを与えない。
- このfinite integrationを、source codeの逐語移植対象ではなく、境界を学習する Rust前のtoy IdP vertical sliceとして扱う。
- 固定した汎用OIDC engineとprotocol-facing adapterをLicium Core外に保つ。Liciumは 引き続きprotocol-neutral backend contractを公開する。
- 最初のRust milestoneではOIDC engineとprotocol-facing adapterを変更せず、 protocol-neutral provider一個だけを置換する。
- SQLite/flat-file/POSIX providersをcomparison oraclesとして保持し、同じbounded contractの第三implementationとしてRust providerを追加する。
- このRoadmapはstable Core API、任意backend support、Spanner、async runtime、 ORM、transport、production IdP designを選択しない。
Backend Conformance v0 requirementsは、その後exact test-only
sqlite-reference-v0 tupleで実行された。二つのfresh isolated sealed sessionで
83/83、BC01--BC12、overall reportはPASS、FAIL、UNTESTED、UNAVAILABLE、
INVALIDはすべて0だった。このresultは記録されたimplementation、adapter、profile、
SQLite runtime、configuration、tested SUT revision
630adfd92cee8ce19249f401b6238e5261a4b33dに限られる。
accepted resultはcurrent-source closure、session binding、evidence sealing、
19 run self-test controls、6 session self-test controls、independent exact-report
reviewのACCEPTEDもPASSした。この609-file reviewer packageはfresh full sessionを
生成するための完全な472-file SQLite replay sourceを含む。
publicationにはtechnical acceptance、project-owner artifact countersign、 sibling local preparation、local-commit review、public-push authorizationという 別々のstateがある。sibling local preparationの前に、project ownerはexact manifest、 609件すべてのrights/data row、material-claim inventory、safety exceptionを countersignしなければならない。このcountersignが許可するのはbindingされたlocal preparationだけであり、public pushはauthorizeしない。public pushにはlocal-commit review後の、project ownerによる別の明示的なauthorizationが必要である。
SQLite Backend Conformance v0 Resultは、compactな 47-file evidence referenceのquick static integrity checkとmulti-hour fresh runtime/ negative self-testを分けて説明する。static referenceはstandalone sealed sessionでも fresh full gateの代用でもない。
reviewerはexact tuple、evidence boundary、reproduction instructionsを検査し、再現可能な
counterexampleまたは明記したboundaryを越えるclaimを報告してほしい。これはSQLite
product一般またはproduction serviceのcertificationではなく、SpannerはUNTESTEDのままである。
technical reviewはpublic pushをauthorizeしない。public GitHubへの公開には、project
ownerによる別の明示的な許可が必要である。
Repository/Evaluation lifecycleの境界とbackend-independentなconformanceの問いが、 external review可能な具体性を持つ段階になりました。
特に次のfeedbackを求めています。
- Root、Publication、Head、Result、Effect、View、Observationの分離が明確か;
- candidate claimsが同梱evidenceの範囲を越えていないか;
- Backend Conformanceの観測がSQLiteから独立しているか;
- Identity Definition、Context、delegation、lineage semanticsに何が不足するか。
最短のreview path:
README
-> Candidate Core Contract
-> Candidate Backend Conformance Contract
-> Evidence Map / executable reference slice
candidate contractsは提案中のrequirementsです。同梱するexecutable evidenceは E67–E72のままであり、predecessor evidenceとE73Rは含みません。
収録されたE67–E72 reference sliceは、一つの小さなlifecycleを接続します。
Pair ingest
-> complete Root
-> accepted Publication / Head
-> pinned Evaluation
-> Result
-> optional Repository Effect + Decision Observation
-> View
-> restart / historical replay / explanation
必要なのはPOSIX shell、一般的なUnix utility、sqlite3です。
sh model/run-reference-slice-tests.sh
aggregateの成功は、6個のtargeted experiment、30個のtargeted relation、25個の negative identityを対象とします。詳細は 再現手順と evidence mapを参照してください。
reference sliceは、production durability、distributed consensus、external-effect atomicity、performance、cryptographic trust、authorization correctness、実際の Spanner conformance、最終的なRust API、最終的なstorage schemaを確立しません。 SQLite tableは使い捨てのrealizationであり、Licium Coreでもwire formatでも ありません。
- コンセプトガイド / English
- 現在の設計 / English
- Model boundary / English
- Evidenceと非結論 / English
- Repository/Evaluation Lifecycle Core Contract候補 / English
- Backend Conformance Contract候補 / English
- Public Sample Policy / English
- Reference slice日本語版
- Reference slice English version
- Zanzibar、Macaroons、関連systemとの比較 / English
- 再現手順 / English
supersededとなった次の二つのhistorical recordは、もともと主に日本語で書かれており、 元のpathに保存されています: Identity Generation Hypothesis、 Licium Minimal Data Model。
- physical storageとsemantic dataの間の正確なPair/Tuple境界;
- stable lineage identifierを必要とするlogical object;
- 最終的なDefinition、Profile、Context、delegation、Grant semantics;
- distributed publicationとconvergence;
- archive capacity、legal erasure、明示的なdestructive protocol;
- Rust type boundaryとbackend capability contract;
- production SQLite deploymentのbehavior、real Spanner adapterのconformanceと performance。
MIT Licenseです。LICENSEを参照してください。