Learning

Watch a year pass.

Move along the timeline. What changes isn't how clever Grace is — it's how much of what actually happened in your world she can reason from.

Day 1Month 1Month 6Year 1

What has actually happened by day 1

A first question is asked

A handful of files are given

No outcomes have happened yet

Every assumption still has to be explained

Day one is honest work: Grace understands the information you give her, and nothing more than that.

The loop

Experience → outcome → evidence → a better next decision.

Remembering more isn't learning. Learning is what happens when a real outcome is compared against what was expected, the evidence is kept, and the next similar situation begins from that instead of from nothing.

A failed attempt is as useful as a successful one — it removes a route that would otherwise be tried again with confidence.

The learning loop

The loop closes on itself: the last step feeds the first one the next time round.

The architecture underneath

Walk the architecture, step by step.

Twelve stops between the container your world lives in and the deeper domain it becomes. Open any step — and where the domain itself is involved, you can move around inside it.

Where does learning live?

FDI container

Everything Grace knows about your world sits inside one bounded container — your Focus Domain. Learning happens in there, not in a general pool shared with strangers.

Grace insight · FDI
“The folder is not just where information is stored. It defines part of the world Grace can reason about.”

1 / 12

Focus Domain Intelligence

Sarah / Dog walking

Interactive · illustrative

Current scope

Sarah / Dog Walking

Grace is operating inside this part of the domain while keeping the permitted relationships to the wider world. Narrowing the scope changes what she works on — not what she is allowed to know.

Open a folder to move the scope.
Create one to watch the domain expand.
Nothing here reaches a real customer record.

The same structure, six different worlds.

The folder names change completely. The architecture doesn't.

Sarah

Business → Clients → Dogs → Finance

Mia

Education → Biology → Research → Sources

Emma

Executive → People → Meetings → Commitments

Daniel

Repository → Components → Tests → Failures

Sofia

Finance → Reconciliation → Evidence → Exceptions

Priya

Company → Strategy → Markets → Decisions

The same loop, three levels of complexity.

A week of dog walks, a company's strategy, and a fault in a codebase. Different worlds, one architecture.

AprilAction
Sarah introduces 30-minute puppy visits at £12 and starts offering them across the week.
April – JuneOutcome
Demand, cancellations, travel time between visits, customer response and margin per slot are all observed as they happen.
JuneEvidence
Tuesday and Thursday afternoon puppy slots fill consistently. Late-Friday slots create travel time that eats the margin.
JuneFailure kept
The late-Friday experiment is recorded as a weak result rather than quietly forgotten.
SeptemberDecision
Sarah asks whether she should expand puppy visits.
SeptemberOutcome
Grace answers from Sarah's own history: increase Tuesday and Thursday capacity first, where demand is demonstrated. Friday expansion has weaker evidence.

Nothing here comes from generic dog-walking advice. The recommendation exists because Sarah's own months produced it.

Focus Domain Intelligence

Deep in your world, not shallow in every world.

Focus Domain Intelligence is the consequence of this loop running for a while. Connected, domain-specific evidence accumulates in one place, so Grace becomes genuinely deeper in your world rather than pretending to know everything equally well.

It isn't a bigger model. It's a better-evidenced starting point — and it only ever grows inside the boundaries you granted.

Outcomes

What actually happened — not what was predicted.

Evidence

The proof behind each outcome, kept attached to it.

Approaches that worked

Ways of working that delivered, reachable again.

Failures

What didn't work, and why — so it isn't repeated confidently.

Unknowns

Open questions stay open until evidence resolves them.

History

The full path that led here, available to every later decision.

Grace insight · learningHuman meaning

Grace retains failures as well as successes.

day 1year 1

Where this honestly stands.

These labels describe Grace's capabilities. The eleven worlds on this site are illustrative use cases, not external customer trials.

Proven

Self-healing repairs with independent verification and rollback, measured on real software with recorded results.

Tested

Retaining context, provenance, outcomes, failures and evidence where it has been implemented and exercised.

In development

Playbooks: converting successfully verified experience into reusable, governed ways of working.

Research direction

Bounded autonomous learning — Grace improving behaviour without being allowed to silently rewrite her own rules.

Day 1

Grace understands the information you give her.

Over time

Grace understands what actually happened.

Then

Grace can use verified experience when the next situation arrives.

Grace learning is not simply remembering more. It is preserving what happened, why it happened and the evidence behind it, so future behaviour can improve without abandoning governance.

Case studies