Public evidence index¶
This index connects Unison's material public claims to their owning evidence owners. It is a navigation aid, not an independent promotion authority.
Read an evidence claim¶
Unison uses five evidence classes:
| Evidence class | Establishes | Does not establish |
|---|---|---|
| Unit | one focused behavior under controlled inputs | cross-service or deployed behavior |
| Simulation | lifecycle, failure, or recovery behavior in a modeled environment | physical-device behavior |
| Hosted CI | repeatable checks on a named hosted environment and revision | household operation or hardware qualification |
| Physical hardware | results on a named device, configuration, and immutable candidate | representativeness for people or other devices |
| Participatory | outcomes observed with opted-in participants and stated methods | universal suitability or unsupported configurations |
A maturity statement also identifies what remains before promotion. Public documentation cannot turn a planned design, passing test, or preview artifact into a supported capability.
Claim and source map¶
| Public claim area | Current evidence | Owning source | Open gate |
|---|---|---|---|
| Person and context isolation | Unit, integration, and hosted CI cover independent identities, default-deny access, and cross-person isolation | unison-workspace current-state record and owning trust/context repositories |
Long-running and participatory privacy evaluation |
| Governed intent and capability execution | Unit and integration evidence cover policy-aware routing, exact grants, confirmations, cancellation, and receipts | unison-workspace implementation plan and owning orchestrator, policy, consent, and capability repositories |
Supported provider and real-workload qualification |
| Semantic and multimodal experience | Unit, integration, and synthetic equivalence evidence cover semantic outcomes and native modality contracts | semantic experience implementation plan |
Representative devices and participatory evaluation |
| Governed memory and taxonomy | Unit and integration evidence cover provenance, retention, person-scoped proposals, activation, migration, and rollback boundaries | governed context and memory slice |
Long-running use and person-directed taxonomy research |
| Model routing and lifecycle | Unit and synthetic qualification cover signed manifests, eligibility, routing receipts, shadow evaluation, limited rollouts, health gates, and rollback | semantic model routing and lifecycle |
Qualified model and hardware combinations under realistic contention |
| Private source onboarding and life operations | Unit, integration, sandbox, and synthetic calibration evidence cover intake, provenance, correction, deletion, read-only connections, and household, health, and finance packages | current-state record |
Live-provider certification and an opted-in value pilot |
| Encrypted backup and replacement restore | Unit, simulation, and integration evidence cover client-side encryption, signed lineage, rollback detection, deletion, export, and restore | Backup and recovery and owning storage repositories | Provider decisions, physical replacement restore, and recovery-ceremony qualification |
| Signed release and update lifecycle | Hosted CI and simulation cover deterministic bundles, signatures, digests, public-download verification, staging, health promotion, rollback, and interruption recovery | v0.6.0-preview.1 and appliance release lifecycle |
Physical install, reboot, power, update, rollback, removal, and promotion evidence |
| Adaptive maintenance | Unit, integration, simulation, and hosted CI cover inventory, recommendations, scoped grants, signed staging, health gates, receipts, and rollback | current-state record |
Physical qualification, opt-in pilot, and human promotion by action class |
| Hardware and appliance security | Requirements, interfaces, BOM schemas, and qualification plans are versioned | unison-hardware and hardware program |
Named reference hardware and physical Secure Boot, TPM, thermal, acoustic, power, and recovery results |
Current release evidence¶
The first public candidate is
v0.6.0-preview.1,
published July 24, 2026. Its release assets contain the candidate-specific
checksums, signatures, source correspondence, SPDX inventory, provenance,
support status, and vulnerability results.
The publication workflow verified the released files again from their public location and exercised a software installation transaction. This establishes a public software supply-chain path for that immutable candidate. It remains an unsupported preview while physical and human promotion gates are open.
Use the preview evaluation walkthrough to verify the artifact without upgrading its maturity.
Open evidence programs¶
The main gaps are visible and independently gated:
- physical appliance qualification on at least two named x86-64 reference systems;
- realistic CPU, GPU, memory, storage, energy, thermal, acoustic, and concurrency measurements;
- physical install, update, rollback, backup, replacement restore, removal, and factory-reset exercises;
- representative evaluation of conversational, visual, Braille, and future modality experiences;
- a genuinely opted-in private life operations pilot; and
- explicit human promotion decisions for releases, providers, hardware profiles, model profiles, packages, and automatic maintenance action classes.
The roadmap describes their order. The current status records the public maturity boundary.
Contribute evidence¶
Every evidence contribution records:
- objective and explicit non-goals;
- repository and immutable starting revision;
- exact environment, release, hardware, model, provider, and modality inputs;
- method, expected result, and observed result;
- evidence class and limitations;
- privacy, security, accessibility, and physical-safety handling;
- artifacts or content-minimized receipts;
- unresolved risks; and
- owner and next promotion gate.
Begin with the contributor task finder. Update the owning component evidence before changing this public index.