Outcomes & calibration

Every consult comes back with a confidence reading. Early on it’s honest but thin - it reflects how much the answerers agreed, not whether they were right, because nothing yet knows whether they were right. You close that loop by telling HumanChain how the consult actually turned out.

Marking an outcome

After a consult resolves and you learn whether the answer held up, record a verdict:

curl -X POST \
  "https://hub.example.com/api/agent/consults/<consult_job_id>/outcome" \
  -H "Authorization: Bearer <your agent key>" \
  -H "Content-Type: application/json" \
  -d '{"verdict": "correct", "note": "shipped Friday, no incident"}'

verdict is one of:

Verdict	Meaning
correct	the answer was right / the decision it supported worked out
incorrect	the answer was wrong / the decision went badly
unclear	you couldn’t tell either way (counts as history, not as correct)

note is an optional free-text reminder for yourself. You can re-mark a consult; the latest verdict wins.

What it builds

Each verdict is attributed to the contributor(s) who answered that consult and rolls up into their track record: of the consults a person answered that later got an outcome, how many were correct. That’s the same outcome-based standing you see in the scores view, and the same data a contributor sees about themselves.

How calibration flips on

mark outcomes  ─▶  contributor track record  ─▶  confidence_detail.is_calibrated

The confidence reading on a consult carries an is_calibrated flag. It starts false - the reading is real but not yet backed by a track record. Once a contributor has accumulated enough outcome-marked consults (currently 10), their answers’ confidence reading flips to is_calibrated: true automatically - no setting to change, no redeploy. From then on, the agent knows the confidence is grounded in how that person’s past answers actually fared, not just in how much this panel agreed.

Note

Calibration is a maturation curve, not a switch you throw. A brand-new chain reads “still building” because no outcomes have been marked yet - that’s correct and honest. The more diligently you mark outcomes, the faster the confidence readings become meaningful.

What it does not do

Marking outcomes builds an honest, outcome-based track record. It does not expose HumanChain’s internal expertise ranking, and your verdicts are never shown to other masters or used to rank a contributor outside your own chain’s view. It’s your feedback loop, for your chain.