A worked example of how a single message moves through Honcho’s reasoning pipeline — from ingestion to extracted facts to a dreamed-over hypothesis. Good onboarding read for understanding the deriver and dreamer in concrete terms.

The example message

A user (Alex) sends:

“I’ve been feeling really stuck at work lately. My manager isn’t supportive and I’ve been looking into product management roles. Actually took a UX course last month.”

Step 1 — Message arrives

The message is passed into Honcho’s ingestion path. This enqueues background reasoning work for the deriver.

Step 2 — Extraction (synchronous)

The extractor breaks the message into atomic facts (premises) — single, independent claims. Each is stored immediately and is available to downstream dreaming tasks.

PremiseContent
p1Alex has been feeling stuck at work
p2The feeling is recent (“lately”)
p3Alex’s manager is not supportive
p4Alex has been looking into product management roles
p5Alex took a UX course
p6The UX course was last month

Step 3 — Dreaming process

The background dreaming process kicks off, running reasoning over the extracted observations.

3.1 — Abduction phase

The abducer agent receives the extracted premises (plus any existing patterns and hypotheses — none yet here) and asks:

“What single explanation would make all of these observations expected?”

It generates the hypothesis: “Alex is considering a career transition to product management”, supported by p1, p3, p4, and p5.

Content
HypothesisAlex is considering a career change to product management
p1Feeling stuck → motivation to leave
p3Unsupportive manager → push factor
p4Looking into PM roles → exploring the target
p5Taking a UX course → building relevant skills

3.2 — Prediction phase

The prediction agent receives only the hypothesis (not the supporting context) and generates predictions about the user from it.

Note: the source document ends here. The prediction phase and any subsequent steps were not yet written in the original. Extend this page as the worked example is completed.