Evidence ceiling E3Proof boundary
Public capability status
Inspect the scope. Keep the evidence boundary.
Historical evidence from the tested V132 APEX component checkpoint. Its measurements describe that checkpoint, not the current website version. Website publication and customer production qualification are separate.
Current engineering checkpoint
V145 evaluation and V146 research integration.
V145: 1,593 Node and 33 Python tests passed. V146: 1,624 Node and 33 Python tests passed. These dated seller engineering results do not establish browser acceptance, model qualification or customer-live operation.
Actual V146 APEX-0 diagnostic: 0/20 correct with one candidate and 0/20 with three; no measured reasoning gain. The historical records below retain their original scope.
Inspect build fingerprint and native hosting observationAPEX-0Research lifecycle tested
Qualified intelligence models0
APEX production admissionHOLD
Customer-liveNot established
Release and evidence
Seller service live is separate from customer live.
- Historical hosting baseline
- V126
- Historical component checkpoint
- V132 · source-scoped tests
- Original checkpoint release assessment
- HOLD
- Protocol
- FRP-2.0.0-draft.1
- Core
- FSK-1.0.0
- Evidence ceiling
- E3 · seller controlled
- Independent validation
- Not established
- Outside witness operators
- 0
- Customer-live
- No
- Production qualification
- HOLD
Hosting version, public product version and frozen protocol/core versions are separate identities. A saved source checkpoint is not a deployment.
Verified checkpoint tests
This historical component evidence is source-scoped.
V132: 1486 Node tests and 18 Python tests pass, with zero failures or skips. 1504 distinct primary executions; repeat runs are not added again.
Build, typecheck and lint: PASS. npm and Python dependency audits: zero known vulnerabilities in the audited inventory. These results do not establish whole-estate security qualification.
Browser interaction and accessibility review: NOT RUN. Static and route checks do not become browser evidence.
Capability matrix
A status is a classification, not an upgrade path.
| Capability | Classification | Established scope | Boundary |
|---|---|---|---|
| Finality APEX | RESEARCH | Finality Group's sovereign intelligence program and architecture. | The full intelligence program is incomplete; no frontier capability is claimed. |
| Finality Sovereign Intelligence | BLOCKED | Typed proposal, planning, state and research interfaces are implemented. | No qualified capable FSI model is admitted. |
| Finality Industry Intelligence | BLOCKED | Eight domain packs and bounded reference simulators are tested. | No qualified specialist learned model, expert approval or customer deployment. |
| Finality Solutions Intelligence | BLOCKED | Problem mandates, lifecycle records and solution admission boundaries are implemented. | The installed runtime has not built and qualified a working solution; isolated execution is blocked. |
| Finality Intelligence Fabric | TESTED | Local memory, rights, records, tools, model admission, jobs and budgets pass component tests. | Full model-backed coordination and production service integration remain incomplete. |
| Finality Core | TESTED | Deterministic verification and bounded hostile-condition tests pass. | Protected high-assurance admission remains HOLD; bounded tests are not whole-estate assurance. |
| Finality Control | TESTED | Authority and consequence checks refuse unadmitted production actions. | Models cannot create authority or override STOP. |
| Reality Memory | TESTED | Persistent records, correction, rights revocation and dependency invalidation are tested. | Learned memory quality and complete backup purge are not established. |
| Outcome Corpus | IMPLEMENTED | Rights-bound outcome and experiment records form a local research substrate. | Customer data, broad capability corpora and independent benchmarks are not established. |
| World Models | TESTED | Bounded ledger, flow, energy, quality and dependency reference computations execute. | Reference mathematics does not establish a learned predictive world model. |
| Discovery Engine | IMPLEMENTED | Versioned problem, hypothesis, investigation and evidence records are supported. | Substantive model-led discovery and verified inventions are not established. |
| Model Foundry | TESTED | APEX-0 training, signed checkpoints, local reload and inference execute. | This is lifecycle validation, not capable general or frontier intelligence. |
| Domain Forge | RESEARCH | Domain proposals and reviewed applicability boundaries are represented. | New-domain qualification and measured cross-domain transfer remain open. |
| AIOS | SELLER CONTROLLED EVIDENCE | The existing governed planning and tool architecture is retained. | It does not establish model authority, autonomous production operation or external assurance. |
| CAFIN | SELLER CONTROLLED EVIDENCE | Existing federated-learning and governed-promotion architecture is retained. | No independent participant network or qualified learned capability is established. |
| Uncertainty Engine | TESTED | Bounded uncertainty flags and rejection of stale or unauthorized measurements execute. | Learned calibration, unfamiliarity detection and acquisition value remain unqualified. |
| Improvement Governor | TESTED | Rights, immutable plans, checkpoint and final-evaluation locks constrain local improvement. | Successful self-improvement and qualified release promotion are not established. |
| Versioned Intelligence Registry | TESTED | Artifact admission, byte integrity, rights and revocation checks execute. | A registered artifact is not automatically capable or production-qualified. |
| Rollback Engine | TESTED | Local checkpoint resume and committed-state restart/restore are exercised. | Protected global trust restore, distributed recovery and second-host portability remain open. |
| Learning Explorer | RESEARCH | Versioned learning proposals and controlled exploration are part of the architecture. | Retained capability and useful autonomous exploration have not been qualified. |
| Counterfactual Reasoning | RESEARCH | Declared alternatives and assumption boundaries can be recorded and challenged. | Measured learned counterfactual or causal reasoning is not established. |
| Simulation | TESTED | All eight bounded industry reference simulators pass. | Simulation results cannot authorize a real institutional or physical outcome. |
| Evidence Discovery | IMPLEMENTED | Evidence requests and source applicability can be represented within rights and budgets. | The system does not manufacture authoritative observations. |
| Contradiction Discovery | TESTED | Declared omissions, conflicts and dependency-triggered invalidation are exercised. | Bounded declared-input checks do not establish universal discovery completeness. |
| Adversarial Systems | TESTED | The preserved verifier's hostile-condition and mutation checks run. | Seller-generated cases do not become buyer-authored or independent evidence. |
| Research Preregistration | TESTED | Immutable plans, negative results and distinct replication references persist. | A replication reference does not self-award independent reproduction or a causal result. |
| Tool Permissions | TESTED | Host-bound actor, tenant, scope, rights and budget checks constrain tool use. | Protected external effects require real authority and deployment admission. |
| Shared Budgets | TESTED | Atomic reservations and settlement prevent tested double-spend races. | Useful-result energy and production-scale economics have not been measured. |
| Local Identity & Gateway | TESTED | Local authenticated sessions, recovery, tenant isolation and restart are tested. | This owner-local evaluation profile is not production identity migration or complete MFA/SSO. |
| Isolated Solution Execution | BLOCKED | The runtime denies generated-code execution when the required isolation is unavailable. | No successful sandbox execution or unsafe fallback is claimed. |
| Distributed / GPU Training | BLOCKED | CPU checkpoint formats and transport-denial behavior are tested. | Successful distributed or GPU training is not established on this host. |
| Existing Seller Service | DEPLOYED | The historical V126 hosting observation is recorded as succeeded. | This historical component record does not identify the current website release. Seller service live is separate from customer live. |
| Protected Production Qualification | EXTERNALLY GATED | Hardware, protected trust, outside evidence and target-environment release gates remain binding. | The current release assessment remains HOLD. |
| Institution Customer-live | NOT ESTABLISHED | No authentic qualifying institution activation is recorded. | All institution, workflow and evidence gates must pass; no manual promotion. |
| Independent Validation & Witness | NOT ESTABLISHED | No qualifying outside evaluator, implementation or witness evidence was obtained. | Seller-controlled verification remains within E1–E3. |
Status vocabulary
Every capability has one explicit classification.
- IMPLEMENTED
- A defined software component exists; broader validation is not implied.
- TESTED
- The disclosed bounded component or reference behavior passed seller-controlled tests.
- DEPLOYED
- A specified hosting or operating release has a deployment receipt.
- MEASURED
- A specified numeric result was recorded under stated conditions.
- SELLER_CONTROLLED_EVIDENCE
- Attributable seller evidence; no independent conclusion is implied.
- EXTERNALLY_GATED
- Acceptance needs a real outside authority, source or environment.
- RESEARCH
- An active architecture or objective; capability is not established.
- BLOCKED
- A required model, execution resource or admission is unavailable.
- NOT_ESTABLISHED
- The required qualifying evidence has not been obtained.
G01–G23
All full-program gaps remain open.
| Gap | Classification | Component result | Remaining scope |
|---|---|---|---|
| G01 · Trainable learning machine | RESEARCH | Actual APEX-0 weights and fresh NumPy/CPU checkpoint training execute. | APEX-1+ capability, qualified models and scaling are not established. |
| G02 · Curriculum | RESEARCH | Easy-to-hard curriculum, family-disjoint splits and one-use final seal are executed. | Substantive curriculum quality and externally sealed populations are unqualified. |
| G03 · Memory and forgetting | RESEARCH | Four memory types persist; correction, transitive interruption, revocation and content tombstones pass. | Learned-memory evaluation, model-weight unlearning and backup purge are not established. |
| G04 · Operational uncertainty | RESEARCH | Bounded structural uncertainty, stale-data rejection and unauthorized measurement HOLD pass. | Learned calibration, unfamiliarity and measured acquisition value remain unqualified. |
| G05 · Coordination | RESEARCH | Leases, cycle rejection, bounded planning and aggregate reservations pass. | Complete model-backed agent coordination, backpressure and cost comparison remain incomplete. |
| G06 · Experiment and replication discipline | RESEARCH | Immutable preregistration, failed/negative results and distinct execution-reference replication pass. | No causal discovery, physical experiment or independent replication was established. |
| G07 · Research portfolios | RESEARCH | Durable project/hypothesis/plan/results governance is tested. | Six expanded research programs and their measured discoveries remain incomplete. |
| G08 · Typed precision | RESEARCH | Typed record boundaries, exact reference finance and bounded units/time components pass. | Complete precision/currency/unit/entity/time semantics across the program remain incomplete. |
| G09 · Claim boundaries | RESEARCH | Claim dimensions and proposal/observation/authority boundaries pass. | Causal and economic assertions require substantive evidence and qualification. |
| G10 · Dependency invalidation | RESEARCH | Restart-safe memory invalidation and saved FINAL-to-REOPEN/STOP tests pass. | Production propagation, protected trust and complete offboarding remain unvalidated. |
| G11 · Verification boundary | RESEARCH | Deterministic verifier, false-FINAL campaign and denial without protected release admission pass. | Inherited high-assurance admission remains HOLD. |
| G12 · Shared resource accounting | RESEARCH | Atomic aggregate reservations and forced stale-read settlement race pass. | Production global serialization, multi-party measurements and capacity sources are unverified. |
| G13 · Durable execution | RESEARCH | Actual local worker death/reconciliation and fresh-process CPU restore pass. | Production orchestration and successful multi-rank restore are unvalidated. |
| G14 · Security | RESEARCH | Parser/signature/cache/auth/tenant failure tests pass; npm and Python advisory audits are zero. | Whole-estate security qualification and OS sandbox execution are not established. |
| G15 · Controlled improvement | RESEARCH | Externally bound budgets, immutable plans, checkpoint locks and post-final training denial pass. | Complete learning/release promotion and model improvement acceptance remain unqualified. |
| G16 · HOLD / STOP recovery | RESEARCH | Permanent STOP, REOPEN, bounded retries and reconciliation after local effects pass. | Domain-approved fallbacks/escalation and useful production liveness remain unestablished. |
| G17 · Industry qualification | RESEARCH | All eight bounded reference simulators execute while preserving unqualified state. | All eight learned FII systems remain BLOCKED_MODEL and unqualified. |
| G18 · Physical experimentation | RESEARCH | Host permissions and fail-closed generated-code sandbox boundary are exercised. | No physical facility, observed physical result or successful OS-isolated experiment is available. |
| G19 · Solution ownership | RESEARCH | Lifecycle records and rejection of ineligible builders/packages pass. | Actual first-party solution construction, deployment, maintenance and retirement are incomplete. |
| G20 · Problem mandate | RESEARCH | Mandate/proposal schemas and fit/approval boundary tests pass. | Legitimate buyer mandates and measured non-build alternatives are not established. |
| G21 · Resource measurement | RESEARCH | Budget reserve/settle tests and bounded CPU execution measurements exist. | Energy and cost per established useful result are not measured. |
| G22 · Rights lifecycle | RESEARCH | Rights admission, revocation after restart and first-party research provenance checks pass. | Commercial title, contributor assignments, legal exclusivity and invention review are unresolved. |
| G23 · Independent profiles | RESEARCH | Seller-only evaluator state and cross-language verification remain distinct from independent assurance. | No external evaluator, witness or buyer organization has supplied qualifying evidence. |
Sovereignty boundary
First-party local research. Full sovereignty remains unqualified.
The tested owner-local APEX-0 lifecycle used no external AI model, key, teacher, judge or embedding service. This narrower result does not establish a complete sovereign platform or production independence.
S01–S24 remain the original acceptance requirements. A scoped test pass is not full-platform qualification.
| Gate | Classification | Original result | Scope |
|---|---|---|---|
| S01 · Canonical architecture naming | TESTED | PASS_SCOPED | Canonical naming/legacy compatibility tests pass. |
| S02 · Entity and ownership | NOT ESTABLISHED | NOT_ESTABLISHED | Requested ownership is recorded; actual entity/title/assignments are unresolved. |
| S03 · Asset and dependency inventory | NOT ESTABLISHED | PARTIAL | Node SBOM, Python wheel licenses/hashes and private artifact inventory exist; title/contributor review remains open. |
| S04 · First-party lineage | NOT ESTABLISHED | PARTIAL | APEX-0 random-initialization and signed learned lineage exist; mandatory qualified FSI/FII/FSIx lineage does not. |
| S05 · Specialist model artifacts | NOT ESTABLISHED | PARTIAL | Actual APEX-0 weights/tokenizer/recipe/data delivered; needed capable specialist artifacts are absent. |
| S06 · Train and resume lifecycle | TESTED | PASS_SCOPED | APEX-0 train/resume/export/reload and further update on a fresh run execute exactly. Multi-rank restore is blocked. |
| S07 · Capability measurement | NOT ESTABLISHED | PARTIAL | One-use held-out next-token loss was measured; substantive intelligence capability is unqualified. |
| S08 · Stack integration | NOT ESTABLISHED | PARTIAL | Local records, registry, inference gateway, auth, jobs and host mediation pass separately. Qualified full-stack routing is incomplete. |
| S09 · External AI dependency | TESTED | PASS_SCOPED | Owner-local APEX-0 requires no external AI key/model/teacher/judge/embedding or remote fallback. |
| S10 · Local operating profile | NOT ESTABLISHED | PARTIAL | Declared local components use local identity/SQLite; current production still uses hosted infrastructure and identity. |
| S11 · Offline distribution | NOT ESTABLISHED | PARTIAL | Fourteen hashed CPU wheels are packaged; complete platform distribution and all rights review are not established. |
| S12 · Cold-start verification | NOT ESTABLISHED | PARTIAL | Fresh venv installation used --no-index and hashed supplied wheels, then all 18 tests passed. Complete platform cold start is unverified. |
| S13 · Network sovereignty | NOT ESTABLISHED | PARTIAL | Provider credentials are absent and selected training runs deny Python socket calls. OS-wide network denial is blocked. |
| S14 · Industry qualification | NOT ESTABLISHED | PARTIAL | Eight bounded reference workloads pass; learned FII capability, domain experts and actual institution workloads remain unqualified. |
| S15 · Solutions execution | BLOCKED | BLOCKED_MODEL_SECURITY | No eligible first-party solution builder; OS namespace execution is also denied. |
| S16 · Fail-closed capability boundaries | TESTED | PASS_SCOPED | Missing/corrupted/revoked/unpinned models and forbidden fallbacks fail closed, including cached inference. |
| S17 · Identity and administration | NOT ESTABLISHED | PARTIAL | Owner-local auth/recovery and authenticated cross-tenant HTTP persistence pass. Full MFA/SSO/service administration is incomplete. |
| S18 · Protected trust | NOT ESTABLISHED | PARTIAL | Offline signatures/freshness boundaries remain enforced; protected hardware/trust admission stays HOLD. |
| S19 · Restore and portability | NOT ESTABLISHED | PARTIAL | Fresh process, copied APEX-0 artifact and fresh venv restore pass on this host. Second clean host/protected trust restore is unverified. |
| S20 · Source and operator recovery | NOT ESTABLISHED | PARTIAL | Native source control and local operator recovery components are tested. Actual human handoff and a qualified hardware envelope are unverified. |
| S21 · Confidentiality | TESTED | PASS_SCOPED | Confidentiality/build gates pass, no private source/model/data published and candidate remains unpublished. |
| S22 · Production migration | NOT ESTABLISHED | NOT_RUN | No production identity/storage/runtime migration was attempted. Current V126 is retained. |
| S23 · Commercial and exclusive title | NOT ESTABLISHED | NOT_ESTABLISHED | Licensed dependencies remain licensed; commercial/exclusive title and contributor assignments are unresolved. |
| S24 · Complete evidence report | TESTED | PASS_SCOPED | The 28-section report separates software, capability, title, sovereignty, publication, external assurance and customer activation. |
Remaining requirements
Publication and activation still require authentic evidence.
- Qualified first-party FSI, FII and FSIx models and approved substantive data.
- Complete coordination, service integration, precision and customer administration.
- Successful isolated solution building, multi-rank / GPU training and portability.
- Protected hardware and trust, staging, multi-region safety and recovery.
- Legal title, data rights, institutional mandates and independent assurance.
- Full browser and accessibility review, release admission and customer activation.
The preserved release gate records 18 blockers. No public-content edit, admin switch or seller fixture satisfies them.
Public-safe projection
Inspect the status without transferring protected material.
Public results omit weights, proprietary tokenizer artifacts, private corpora, sealed cases, source, credentials, keys and customer records.