Status: Public extension v0.4.2
Parent: CIEA Core
Type: Cross-cutting extension for AI automation and LLM productivity claims
Scope: AI-generated answers, drafts, summaries, code, customer responses, operational recommendations, and other outputs that are presented as productivity or automation gains.
AI automation should not be evaluated by generation speed alone. It must be evaluated by total quality cost: verification, rework, rollback, responsibility recovery, and downstream harm.
Shorter:
Automation has not removed work if it has merely moved quality assurance.
Many AI products are evaluated through visible surfaces:
Those surfaces can be real.
But they are not enough.
A system can generate faster while moving the remaining work into:
CIEA therefore asks:
Did automation reduce the real work?
Or did it move verification, repair, and responsibility outside the performance table?
Definition
Output QA Displacement occurs when an automated system reduces generation or processing time while shifting quality assurance to another actor.
Common receiving actors:
Diagnostic sentence:
If output quality is not warranted and verification capacity is not funded, automation has not removed work; it has relocated quality assurance.
Definition
Warranty Gap occurs when a system is sold or adopted as operational capability, while the accuracy, completeness, or fitness of its outputs remains weakly warranted, unwarranted, or shifted to the customer’s verification duty.
The audit does not need to claim bad faith.
It asks a narrower question:
Is the output treated as operationally useful in the sales or adoption story,
while treated as unguaranteed or customer-verified in the responsibility layer?
Diagnostic sentence:
Capability is sold upstream; verification is assigned downstream.
Definition
Rollback Surface is the counter-surface that asks whether an automated result or change can be reversed, how much recovery costs, and who owns the recovery pathway.
A fast automated action is not necessarily efficient if rollback is unclear or expensive.
Required questions:
What was changed?
Can the change be reversed?
Who can reverse it?
How long does reversal take?
What is lost during recovery?
Was recovery cost included in the performance claim?
Diagnostic sentence:
A faster automated change is not interpretable until rollback cost, recovery ownership, and post-action audit burden are checked.
Definition
Accountability Collateral refers to the structure that makes an actor answerable because they have both meaningful control and actual exposure to loss if the system fails.
The relevant question is not only whether a name appears on a process chart.
It is:
Can this actor stop the system?
Can this actor be made answerable?
Does this actor actually lose something if failure occurs?
Diagnostic sentence:
If no one can meaningfully lose from automation failure, no one can meaningfully be trusted with automation authority.
Definition
Reliance–Warranty Split occurs when an interface or product presentation encourages practical reliance, while the responsibility layer withdraws or weakens warranty for the output.
Common pattern:
interface creates reliance
→ output looks actionable
→ terms or responsibility layer withdraw warranty
→ user/customer inherits risk classification and verification burden
Diagnostic sentence:
The interface creates trust; the responsibility layer withdraws warranty.
| Improved surface | Required counter-surfaces |
|---|---|
| answer generation speed | factual error, verification time, reliance risk |
| document drafting speed | review burden, revision cycles, expert correction cost |
| coding speed | test failure, security review, rollback cost |
| customer-service automation | resolution rate, escalation, recontact, abandonment, frustration |
| call deflection | unresolved demand, human escalation, quiet exit |
| productivity gain | hidden review time, rework, downstream error cost |
| lower staffing need | remaining-case complexity, worker stress, service degradation |
| faster operational change | rollback clarity, recovery time, blast radius |
| lower complaint volume | complaint access difficulty, suppressed reporting, external complaints |
[1. Performance claim]
- What improved?
- Who claims it improved?
- Which metric is used?
[2. Core]
- What was actually supposed to improve?
- Is the core output quality, user resolution, cost, speed, trust, or accountability?
[3. Output quality]
- Is the output factual, advisory, customer-facing, operational, legal, financial, medical, or administrative?
- Is output quality measured?
- Is output quality warranted?
- What kinds of error are known or foreseeable?
[4. Verification]
- Who verifies the output?
- Is verification mandatory or informal?
- Is verification time counted?
- Is expert review required?
- Is verification capacity funded?
[5. Warranty gap]
- Is the system sold or adopted as operational capability?
- Does the responsibility layer require customer/user validation?
- Does the claimed productivity gain subtract validation cost?
[6. Rollback]
- Can wrong output or automated change be reversed?
- Who owns rollback?
- What is time-to-recovery?
- What is lost during recovery?
[7. Accountability collateral]
- Who has stopping authority?
- Who has final responsibility?
- Who bears loss if failure occurs?
- Is there a compensation, correction, or appeal path?
[8. Judgment]
- Real improvement
- Surface optimization
- Output QA displacement
- Warranty gap risk
- Rollback insufficiency
- Responsibility recoverability failure
- Insufficient evidence
Use only when:
Use when:
Use when:
Use when:
Use when:
Claim:
An AI system increased document drafting productivity by 40%.
CIEA does not accept this claim directly.
It asks:
Did final usable documents increase?
Did expert review time increase?
Did factual correction increase?
Did legal/policy review increase?
Were rejected drafts counted?
Were downstream errors counted?
Was rollback or correction cost counted?
Who is answerable if a draft causes damage?
Possible judgment:
The system may have improved draft generation speed, but real productivity cannot be accepted unless verification time, correction cycles, and responsibility paths are included. Without those counter-surfaces, the claim remains vulnerable to Output QA Displacement and Warranty Gap risk.
This CIEA extension is not a replacement for a delegation cutoff protocol.
| Tool | Function |
|---|---|
| Stop–Go | Pre-delegation authority freeze or expansion decision. |
| CIEA | Performance-claim audit after or during adoption. |
Short version:
Stop–Go asks whether the system may act. CIEA asks whether the claimed improvement survived the hidden-cost audit.
This extension does not claim:
It claims only:
AI automation performance claims must include verification, rollback, and responsibility costs before they can be treated as real efficiency.
Generation speed is not productivity until verification cost is counted.
A model output is not operational value unless someone can verify, correct, and answer for it.
Capability without output warranty creates customer-side quality assurance.
A cost is not gone merely because it became a review burden.
Automation has not removed work if it has relocated quality assurance, rollback, and responsibility.