Known Limits
What We Won't Claim.
The permanent edge of the record: what we deliberately don't automate, what's still unresolved, what a local model honestly can't do, and where our capacity ends.
01Why This Page Exists
The Most Valuable Section.
Internally, we call the known-issues section the most valuable section of any document — because a record that only contains wins isn't a record, it's a brochure. This page is the permanent, public edge of ours. Nothing anywhere on this site is allowed to be softer than this page.
02The Current Edge
Known Limits — Today.
Stated plainly, next to the things we sell
- The human stays at the actuation gate — by design. The capability pipeline runs live: candidates are nominated, shadowed, examined, and promoted on evidence, and the governed decision gateway adjudicates every consequential action — human GO/NO-GO, hash-chained and externally anchored ledgers, fail-closed, an instant kill switch. What we deliberately do not enable is fully unsupervised actuation: an agent touching a live system with no human in the loop. That gate is a discipline we keep, not a capability we lack. Agents propose; humans decide.
- The formal research claims are pre-registered — not yet independently proven. The mechanisms themselves run in production; what stays pre-registered is the generalizable, falsifiable science about how they behave at scale — published with its falsification paths, before results, and unresolved until an independent replay closes it. The label is the point.
- Local models are not frontier models. For the deepest reasoning, the big labs' newest models are better — full stop. A sovereign stack gives you honest routing: local for the private and the routine (most of a working day), frontier by your choice under your data rules when a task warrants it. Anyone selling a local box that "beats the frontier at everything" is selling you something.
- Capacity is one steward deep. Stewarded seats are capped at what one person can verifiably serve. The cap is public before launch, and we don't sign response-time promises we can't keep. The promise must stay smaller than the capacity — that's our own reach doctrine applied to ourselves.
- Some cells stay empty on purpose. Our rule is that no claim prints above its actual maturity. When something can't be honestly stated, its cell prints empty — you may see a "—" where a number usually lives. The empty cell is the policy working, not an oversight.
03The Promise
How This Page Changes.
Entries leave this page one way: by being resolved in the record, with the resolution linked. Entries arrive the moment we know about them. If you ever find this page stale, that staleness is itself a finding — tell us, and it goes in the ledger.