Read a step result

What you'll do

Read a synthetic producer result and decide what it establishes before ledger recording. Identify the operation, status, check identity, actual observations and retained evidence. You need F1-F2; this exercise uses a local worksheet rather than a live source or Hosted account.

Start with the producer's operation

A producer executes a bounded operation and emits a structured step result. It may generate a model, observe a query or compare a particular projection. The result describes that invocation. Only the ledger records receipts; an author's "done" message is neither a step result nor a receipt.

Read the operation before its status. A successful semantic-generation operation can coexist with a failed qualification because no deployed query has been observed. A successful test process can also contain an intentionally failing gate. The level at which success is reported matters.

Read actuals against expectations

Imagine a synthetic query returns revenue of 149 while an independently frozen expectation is 150. The actual is a valid observation even though the comparison fails. Do not change the expected number to make the result green. Investigate population, grain, precision and the exact query state.

Teaching worksheet - not a runtime JSON schema
Operation: observe the declared semantic query
Actual: 149
Frozen expectation: 150
Comparison: mismatch
Next action: investigate the bound population and definition

The gate recomputes numeric comparisons and exact grouped-key coverage rather than trusting a producer's pass flag. A handwritten "pass" in a result cannot replace required observations or override contradictory evidence. A matching scalar total can still fail a grouped check when categories are missing or misplaced.

Preserve the observed evidence

Before recording, identify the actual output and supporting files. Keep them in approved local evidence custody. Raw results and local receipts may contain business values, query text and other private material. They do not belong in a public academy submission or a Hosted qualification envelope.

Do not overwrite an earlier result to disguise a retry. Retained evidence should remain the evidence the recorded claim actually used. A later observation is a new invocation with its own bindings and result. This makes a failure explainable rather than erasing the path that produced it.

Distinguish a negative test from live proof

The brief's offline regression example intentionally evaluates missing evidence and expects failures with reasons. The outer test can pass because refusal is correct. It does not follow that the inner qualification passed, nor that a live source-to-report run took place.

Use that distinction in your worksheet. Write one sentence for the operation result and another for the qualification claim. For example: "The negative test observed the expected refusal; no live Fabric evidence was produced." That is more precise than reporting "all tests green" to a reviewer asking whether sales is reconciled.

Your expected output identifies the operation, scope, observations, evidence and next owning action. Add the failure reason in plain language. If the required artifact is missing, keep the claim blocked or unknown as appropriate; do not manufacture a result to complete the exercise.

Check your understanding

1. Why is a successful generation step insufficient for semantic qualification?
2. Who records a receipt, and what does the producer supply first?
3. How can an intentionally failing gate belong to a successful regression test?