Concepts
The ideas that make Soil work. Handovers, the extraction and verification engine, how Soil verifies on your own models, working-style observations, provenance, safe projections, and projects.
These ideas carry the whole product. Read them in order and the rest of the docs will make sense.
Handovers
The unit Soil produces: everything a fresh AI needs to continue a project without drift · and nothing it doesn't.
Extraction & verification
The engine: the active model fills a recipe, then Soil checks it over multiple iterations and runs a drift check before saving.
How Soil verifies a handover
The load-time measurement: the model you save from writes the answer key, the one you load into answers cold, Soil scores it · live on project loads.
Working-style observations
Every save records how the work actually runs · situational answers stored raw and shown at load. Evidence, not adjectives · never scored.
Provenance
Every load-bearing claim is labelled Stated, Inferred, or Unknown · so you never trust a confidence Soil didn't earn.
Safe projections
Soil Cloud stores meaning, never secrets or raw prompts. Carry-meaning-not-verbatim is built into extraction, not bolted on after.
Projects & saving over time
A personal save is a standalone snapshot; a project is where your work accumulates · save threads into it and restore the whole thing at once. The paid capability.