Reviewed 2026-09-29
We’re working on model quality, execution costs, supported conversions and tools people can run themselves. Each item lists the test needed for its next step.
The retained programme below was reviewed on 2026-09-29. The current engineering record contains newer implementation checks and measurements.
Next decisive tests
The next research question is whether richer cognition and greater stored capability improve useful quality while the active path stays constrained.
Complete qualified fixed-holdout, stored-capacity-matched and active-path-matched comparisons with repeated independent evaluation.
Fracture is the migration path for existing neural capability; the decisive question is whether useful task behaviour survives conversion under policy-governed evaluation.
Pass bounded quality, safety and resource gates on qualified source models before any production conversion commitment.
Move from founder-run evidence to a bounded partner environment with named owners, reproducible acceptance criteria and an explicit handoff path.
Qualify one partner problem with evidence access, security boundaries, a decision owner and an operational handoff owner.
Further work required
These items remain under development. Each record identifies the remaining evidence and operating requirements.
This hold describes the July release candidate. It does not date or suppress the newer current systems receipts.
Regenerate a new evidence release from one authoritative validated handoff before treating it as current publication authority.
This hold applies to /chat/ and /api/chat only. The public /v1 native text preview and Unified World follow their own live availability and capability contracts.
Qualify the protected workspace backend, identity, retention, security, monitoring, rollback and capacity controls.
Research execution and bounded professional engagements are not equivalent to a production ERAIS product.
Establish repeatable quality, security, observability, recovery and operating-cost envelopes in an independently run environment.
Mechanism evidence and source-pinned receipts remain public with their scope; promotion waits for fair, repeated, matched-quality comparisons.
Reconcile qualified dense controls, fixed holdouts and full resource accounting under an approved claim review.
How a result moves
Record the exact source revision, input artifacts, hashes and verification result.
State what was observed, what it supports, what it does not support, and what evidence would change the decision.
A technical result does not become a product. Operations, rights, security, recovery, and delivery must pass separately.
Go deeper