Australian not-for-profit · ACN 699 651 771Open research · Open software · Public benefitPublic record

Learn / Engineering

Bosun: learn from supervised engineering

Trace a scoped task into a candidate change, validation, and acceptance.

Experimental open-source software · Reviewed 3 October 2026

Explore the mechanism

Trace the processBosun experimental software • repository-dependent behaviour

From a task to a reviewed change.

Select a stage to see who acts, what changes, and what evidence remains.

Human maintainer

Define the requested change and acceptance criteria.

The task describes the authority granted to an executor.

Stage 1 of 4
Stages explain a design and its evidence; a highlighted stage is not proof of real execution. Sources: underlying documentation.

01 / Explanation

A practical feedback path

Bosun explores orchestration for software-engineering agents. Its public source lets contributors inspect coordination, executor routing, review workflows, and recovery mechanisms. It is a narrower applied track, with configuration-dependent behaviour.

02 / Explanation

Completion must be observable

An executor exit code, a completed run, passing checks, and an accepted change are different events. Useful evidence includes the actual diff, meaningful check results, reviewer decisions, and the state of the repository after recovery. Throughput without quality is not a sufficient measure.

03 / Explanation

Feed lessons back without overstating them

Failure recovery, scope preservation, review bypass attempts, and audit trails can inform DSEMA research questions. Record the version, environment, failure mode, and reproduction procedure. A working coding workflow does not demonstrate a self-evolving constitutional agent economy.

Check your understanding

What is stronger evidence of a successful engineering task?

Next: How DET.io’s research fits together →All guides