Skip to content

The register

The evidence register: status and sources.

The claims covered by Proof, and the rule we judge them by. The proof page carries the capabilities that run and the ones being proven; the designed ones are listed here.

Open the recorded run Back to proof

22 capabilities, by status

  • Implemented 3
  • In production validation 5
  • Proposed 14

Each status describes the evidence behind the capability.

01 The table

22 claims, strongest evidence first.

Filter by evidence status, or read the whole table.

  • Implemented Runs in Luminids, with tests and saved character state.
  • In production validation Runs in the game and is undergoing broader validation.
  • Proposed The capabilities and tools we’re working towards.
Show

All 22 claims on this page.

The 3 claims marked Implemented, out of 22 on this page.

The 5 claims marked In production validation, out of 22 on this page.

The 14 claims marked Proposed, out of 22 on this page.

Every claim on this page, with its status, its name, and what you can open.
Status Name The claim What you can open
Implemented event and autobiographical record

Each character keeps a bounded, curated record of what it has lived through, and that record changes how it develops.

 Test suite, character layer
Implemented persistent individual identity

Each character keeps a stable identity, and that identity survives saving and loading.

 Test suite, character layer
Implemented grounded authored expression

What a character says is assembled from what it has actually lived, and it stays silent rather than claim something it cannot support.

 Test suite, character layer
In production validation shared pair history

Two individuals share one bounded record of the history between them, written during ordinary play.

 Test suite, relationship tests
In production validation evidence held per character

Evidence is held per individual, and it gates what that individual is allowed to claim.

 Test suite, character layer
In production validation belief-gated response

A character that has come to believe something about itself reacts when the world contradicts it, remembers the day, and lets the belief go; a character that never formed the belief does neither.

 The recorded run
In production validation appraisal and expectation revision

A character can hold an expectation, discover it was wrong, and be changed by that.

 Test suite, character layer
In production validation belief-gated action

A character would act on what it believes about someone else, then find the world has moved on and adjust rather than pretend.
Proposed one-way view of another character

A durable personal view of another individual, held separately from the history the two of them share.
Proposed culture and testimony

Shared colony culture, and testimony: what one character is told by another becoming something the hearer holds, with a record of how it was learned.
Proposed population runtime target

LuminidsAI is being designed towards persistent populations under a real game performance budget.
Proposed population performance budget

The production programme is designed around hundreds of persistent beings and a strict real-time budget.
Proposed CRI Engine

A runtime that would carry contextual relational intelligence beyond the game it was built in.
Proposed CRI Studio

Author the situations, actions and project goals that shape a character's life, then inspect the choices they produce.
Proposed CRI Bench

Change a memory, belief or observation and test what happens to the character's decisions.
Proposed public SDK

A standalone interface that would let developers outside the studio build on the architecture.
Proposed learning by watching

Characters learn skills by watching and practising, then share what they know with others.
Proposed personal projects

Characters pursue personal projects over time, carrying their reasons and progress across interruptions.
Proposed places that matter

Characters form attachments to places through shared experiences, and adapt when those places change.
Proposed helpful initiative

Characters take helpful action from what they have learned about someone, while leaving room to be mistaken.
Proposed remembered preferences

A favourite meal or familiar song carries a remembered reason that can shape later choices.
Proposed repairing misunderstandings

Characters discover misunderstandings, revise their view of what happened and try to put things right.

02 The method

One rule decides whether a capability is cognition or presentation.

We check what caused the behaviour, how widely it has been tested, and whether a player can understand the result.

  1. The rule

    We judge every claim by one rule: withhold the claimed cause, hold the world constant, and see whether the behaviour changes. If the action is unchanged when its claimed cause is withheld, the feature is presentation, not cognition. Presentation is allowed; misnaming it is not.

  2. The rungs

    A passing test is not the whole verdict. Code, causal necessity, live behaviour, player readability, persistence and cost are separate questions, and a capability is only as strong as the highest rung it has reached.

    1. schema and ownership
    2. pure contract
    3. intervention proof
    4. live-world record
    5. player-visible film
    6. population and save validation

  3. The human tests

    The tests the programme is judged against are human before they are technical. None is presented as passed, and no threshold is published for any of them. Most of the test matrix behind them is unrun.

    1. tell apart
    2. predict
    3. be surprised by
    4. be recognised by
    5. society with a memory
    6. survive a save
    7. fit the frame