Part 1 · Operator’s Guide

Quality

The six-rung QA ladder and ten failure modes.

Public teaching edition · 1 October 2026

Agents checking agents

Self-checking is the first pass. It catches obvious mistakes but can repeat the assumptions that caused them. A reviewer starts fresh, reads the actual artifact and standard, and tests the claims.

Give the reviewer the request, acceptance criteria, relevant authorized source and exact revision. Do not instruct it to agree with the builder. A different model is useful only if it performs an independent review; its name alone proves nothing.

Keep publisher, executor, auditor and human decision owner distinct when consequence warrants it. Prevent concurrent reviewers from silently changing the production artifact while the builder is still working. Record disagreements in the task and resolve them with evidence.

For one low-risk chat exercise, inspect the draft yourself and label the assistant's pass a self-check. Before a consequential action, obtain the required independent or human review. Do not hide the gap behind the word audited.

The QA ladder

Rung Who Evidence
L0 self-check The doer Result opened; acceptance test checked; observation recorded
L1 machine checks Deterministic tools Exact commands, revision and results for tests, links, rendering or lint
L2 fresh-context review A context that did not build it Artifact and source opened; findings and fixes recorded
L3 different-model review A different model reviewing independently Claim, evidence, tone and omissions checked; disagreements reconciled
L4 human owner sign-off The required human decision owner The particular approval or review actually given
L5 scheduled audit A later checker Dated evidence that the result still holds

A human authorization record proves permission for the stated action. It does not prove the human reviewed quality. Record both separately if both happened. The packet tool records L0–L4; L5 is an external scheduled audit.

The source protocol uses L0 for an internal note, L1 for a repository change, L3 for a shared rule, and L2 before external delivery plus L3 before closing publish/partner/client work. Money, credentials and deletion classes require the human authority applicable to the action. These method labels do not override a tool's or organization's stricter approval requirements.

Required rungs compound rather than substitute. A waiver must be explicitly authorized, named and visible; it cannot bypass an applicable safety requirement. An unavailable check stays missing. A failed check reopens the work with a specific next action.

Ten failure modes

ID Rule Notice this before acting
F1 Wrong authority Use the authoritative source You compare totals without naming the master
F2 Unread context Open the relevant record You state a person's circumstances from memory
F3 Human as workaround Check supported capabilities first You ask a person to paste or click before checking the available route
F4 Unverified write Check the intended match and destination You changed text but have not read it back
F5 Stale read Verify the current result A cached page is your only publication evidence
F6 Unlooked-at output Render and inspect the artifact You exported a document without viewing its pages
F7 Stale cross-reference Recheck derived pointers An upstream name, list or version changed
F8 Wrong surface Confirm the actual environment You assume a path exists on another computer
F9 No abort rule Stop repeated blind attempts Two interactions failed to land where expected
F10 Silence as status Own the next safe step A waiting row has no recheck, fallback or decision owner

Some checks can be automated; others require judgment. Document the real enforcement and its limits. A printed preflight reminder is not proof that a later action obeyed it. Never treat silence as permission. When required approval is missing, continue independent authorized preparation and keep the dependent action held.