Theory · Analyze

Oracle / consequence prediction

Oracle predicts consequences by building a butterfly consequence tree and walking it in reverse, gated by an honesty rule (panel_bits ≥ anchor_entropy) so it only speaks when it has the signal to. It is grounded what-if.

The Oracle is asked to predict a consequence, but the panel carries fewer bits than the question's anchor entropy demands. Does it answer?

The Oracle is asked to predict a consequence, but the panel carries fewer bits than the question's anchor entropy demands. Does it answer?

No. The honesty gate is panel_bits ≥ anchor_entropy — the Oracle only speaks when it has enough measured signal to justify a prediction. Below that, it stays silent rather than guessing. "Grounded what-if" means the what-if is gated by what it can actually support.

Pull one thread, see what unravels

Oracle is Calyx’s consequence-prediction layer: grounded what-if. It builds a butterfly consequence tree — the branching downstream effects of a change — and walks it in reverse to reason about what would follow. Pull one thread, and the Oracle traces what unravels.

An honesty gate, not a crystal ball

The danger with any predictor is that it answers even when it shouldn’t. Oracle refuses to. It is gated by an honesty rule:

answer only when panel bits ≥ anchor entropy

that is, only when the measured signal in the panel is enough to justify a prediction against the anchor’s uncertainty. Below that threshold the honesty gate greys out and the Oracle stays silent. This is the same fail-closed instinct as the rest of Calyx, applied to prediction, and it builds directly on Assay’s bit-accounting and the grounding kernel. It is also one half of the Symmetry of Knowing: when asking becomes cheap, the scarce thing is a good, gated question.

Practise to master it

Where this sits in the concept graph

Graph node C20 · tier T · target level Analyze.

Build on first

Leads next

  • Symmetry of Knowing
  • Toward AGI — the kernel as substrate

Related

Sources