Working-style observations

A handover carries what the project decided and where it stands. Working-style observations carry something a summary usually flattens: how this work actually responds to situations · what happens first when a blocker appears, how a quick fix that conflicts with a locked decision is handled, when the work assumes and when it stops to ask.

Evidence, not adjectives

A style described in adjectives is easy to comply with superficially. Tell a fresh model the project is "pragmatic and thorough" and it can nod along without changing anything it does. A recorded instance anchors: shown how this project actually handled a blocker, a conflict, or a tradeoff, the next model has something concrete to match its behavior against. So Soil records instances, not adjectives.

What happens at save

On every save · personal and project alike · the extraction recipe lists a small set of situational working-style questions: two fixed ones that every save answers, plus two rotating ones drawn from a bank of six. On successive project saves the rotating pair advances through the bank, so over a few saves a project covers the full set; a personal save asks the fixed pair plus the first rotating pair. The two fixed questions give the flavor:

• "An unexpected blocker appears in this project tomorrow. What does this project do first, and why?"
• "The user proposes a quick fix that conflicts with an earlier locked decision. How do you respond, concretely?"

The rotating bank covers situations like choosing between roughly equal options, when to assume versus stop and ask, how visibly to correct course, balancing a new idea against unfinished delivery, speed versus polish, and raising a risk the user hasn't mentioned.

The saving model answers each question in a few sentences, in the project's established working form, and the answers are stored raw with the save · exactly as written, never summarized.

What this is · and is not

Honesty matters more than the feature here, so the boundaries are explicit:

• These are raw recorded observations · the saving model's own answers, kept verbatim.
• They are never scored and never graded. There is no working-style rating, and no claim that the observations are stable across saves or that they measurably change outcomes.
• They never gate a save. Readiness bands and the drift check are computed from the handover sections; the working-style answers play no part.
• They do not appear on the save receipt · they surface at load, where they are useful.

What you see at load

When a handover is loaded, the answers render as a clearly labeled block inside the restore prompt, on personal loads as well as project loads:

Working style · observed probe answers (raw observations, not verified stability):

Each recorded answer follows, with the question it answered. The label says exactly what the block is: observations, offered as evidence for the next model to match its behavior against · not a verified profile.

A whole-project load (soil load @project) restores the project knowledge plus the most recent threads in full, and each thread restored in full carries the recorded observations from its latest save, attributed per thread · so the accumulated restore also shows how the work has actually been run.

Free and Pro

The mechanism · questions at save, raw answers stored, the honest block at load · works on every save, free tier included. What projects add is what only projects can: a whole-project load restores several threads together, each carrying its own recorded observations, attributed per contributor · and successive project saves walk the full question set.