A prose summary is useful to read, but a workflow needs fields it can inspect. Ask the assistant for a small JSON record: run_id, status and the cited log line. JSON syntax only establishes that a parser can read the text. A schema establishes the types and required keys. Neither proves that the cited line actually supports the diagnosis.

Three different checks

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

import json

def validate(text, known_runs):
    try:
        rec = json.loads(text)
    except (ValueError, TypeError):
        return False
    if type(rec) is not dict or set(rec) != {"run_id", "status", "line"}:
        return False
    if type(rec["run_id"]) is not str or rec["run_id"] not in known_runs:
        return False
    if type(rec["status"]) is not str or rec["status"] not in {"PASS", "FAIL"}:
        return False
    if type(rec["line"]) is not int or rec["line"] < 1:
        return False
    return True

known = {"run-014"}
assert validate('{"run_id":"run-014","status":"FAIL","line":12}', known)
assert not validate('{"run_id":"run-014","status":"FAIL","line":true}', known)
assert not validate('{"run_id":"unknown","status":"FAIL","line":12}', known)
print("PASS: synthetic record shape checks; source meaning must still be reviewed")

This standard-library example validates a deliberately tiny contract. It accepts only three keys and does not inspect a real source file. The strict Python type check matters because bool is a subclass of int. A production workflow should use its established schema library and define the permitted fields explicitly; do not silently coerce a surprising value and then execute it.

Exercise

A reply has the required keys and cites line 12. The actual error is on line 21, while line 12 contains a routine startup message. Explain why this record must not be accepted as an engineering diagnosis. Record an unresolved question in the human review note, outside this example's strict three-key JSON record. A richer schema could explicitly add an unresolved_questions field, but both producer and validator must agree before changing the format.

Related topics

• Tool schemas and validation

Related topics

• Up: Practical foundations

• Next: Repositories, Diffs and Isolated Changes