CarrierAn experiment in expression

Chromatic studies / 05

Chromatic
Inheritance.

What a reply carries forward.

A colour can remain, pass to another cell,
or change with the field.
Trace the choice before carrying it on.

01 / The encounter

Preparing the local model…

Your field
Colour · position · relationSelect a cell to begin
Compare requested replies This input, this tab

Choose a model and request its reply. Switching back recalls the same saved run; it does not run again. Editing the input or protection clears this comparison. The comparison covers colours, cell channels and typed relations. These are numerical counts, not quality scores.

The words can wait.

Select a colour. Its numerical identity is already part of the field.
Read every cell, relation and membrane Accessible text view
Inside the calculation Raw output & repairs

Inheritance learns source choices and numerical edits from the input field, including colour, position, layer, emphasis, membranes and birth. The preceding Continuity and original Field models remain available on the same input. Raw decisions, grammar repairs and optional software protection are recorded separately.

No response has been requested.

The complete run includes the exact input and reply JSON. The reply’s parent hash verifies the matching exported input bytes; it records lineage, not identity, approval or meaning.

Show numerical decisions
No numerical output yet.

Exact colour cores preserve allocated RGB values. Halos, paper grain and the outer frame are display marks. Cell size shows emphasis. Solid rings show positive layers and dashed rings negative layers; ring count gives the distance from layer zero. Arrowheads preserve direction. Membrane openings show permeability.

02 / Traces of the exchange

Fields left
along the way.

Complete fields retain their colours, openings, relations and layers. Open a recorded work; reveal its reading when you are ready.

03 / Keep the source close

A small model.
Visible seams.

A reply inherits a field through particular choices. Inheritance uses fitted numeric operators and source pointers to retain or copy colours, edit the other channels, and propose a new cell from the predicted geometry. It is a new trained numerical model, not a new neural backbone. The source choices, edits and repairs are visible. Earlier models remain available for comparison. You can alter the field and decide whether its reply is worth carrying forward.

The benchmark concerns authored numerical exercises. Precision on those exercises is distinct from an understanding of what a relationship means. A small geometric change can cross a learned decision boundary. Optional protection is visible software policy. Readings of the resulting fields are interpretations, not evidence of model understanding, subjective experience or care.

Method, measurements & limits ↗Reproducible record ZIP ↓