Component review

Memory

One component. Three subjects, six memory types, three stakeholder lenses. Switch the subject to An agent to see the honest-refusal state, and the lenses to see the same memory answering a different reader’s question.

The rows below are examples. The live viewer belongs at /u/[slug]/memory, behind the workspace authority walk. Two panels are deliberately empty — those types return a note rather than data, and the panel says which.

Four actors, four shapes

The same component, the same six types. What changes is who the memory is about — and that changes which panels have anything in them.

A customer of two years. Mostly conclusions — the system has watched long enough to generalise, and the raw history matters less than what it now expects.

Ada Lovelace

Is this subject behaving sensibly, and what is it acting on?

What we concluded

Beliefs held about this subject, and where each one came from.

  • prefers invoices on the last Friday of the month

    91% confidentfrom observed7 observations
  • asks about delivery windows before price

    78% confidentfrom observed12 observations
  • escalates when a reply takes longer than a day

    83% confidentfrom observed5 observations

What leads to what

Which routes have worked, and which have stopped working.

  • billing question → store credit answer41 used+62 / −3highway
  • delivery question → tracking link27 used+38 / −2open
  • refund request → policy link12 used+8.5 / −11blocked

Who it knows

Groups this subject belongs to, and its role in each.

  • acme · member
  • acme-billing · owner

The operator view

The same data with the machinery showing: the subject picker, every type including the ones that are not wired, each caveat and its file:line. Switch to An agent to see the honest refusal — the read path coerces agent ids to humans, so a live panel there would be a false blank.

Ada Lovelace

Is this subject behaving sensibly, and what is it acting on?

A lens changes what is shown, never what may be read — memory is returned whole or refused.

What we concluded

Beliefs held about this subject, and where each one came from.

  • prefers invoices on the last Friday of the month

    91% confidentfrom observed7 observations
  • escalates when a reply takes longer than a day

    44% confidentfrom asserted2 observationspending

What leads to what

Which routes have worked, and which have stopped working.

  • billing question → store credit answer41 used+62 / −3highway
  • refund request → policy link12 used+8.5 / −11blocked

What the shape implies

computed for this view

The band a route falls in — open, closing, blocked, toxic, highway.

  • billing question → store credit answer41 used+62 / −3highway
  • refund request → policy link12 used+8.5 / −11blocked

These bands are computed in TypeScript when this view is built, not derived by the graph. Nothing writes them back, so nothing else in the system can query them.

resolvers/memory.ts:144-150 — under a comment crediting schema/sui.tql, which has zero rules

What happened

not read here

Every message, in order, with who said it and when.

No conversations recorded for this subject yet.

The history is real, but it lives in the channels D1 store and this door does not read it — reveal returns a note here, not rows.

resolvers/memory.ts — episodic: { note: "conversation history in channels D1" }

What it knows how to do

not read here

The capabilities available to this subject.

No capabilities listed.

Skills are scoped to a workspace, not to an actor, so this door has nothing actor-shaped to return. Discovery answers this question instead.

resolvers/memory.ts — procedural: { note: "skills are workspace-scoped" }

Who it knows

Groups this subject belongs to, and its role in each.

  • acme · member
  • acme-billing · owner