LEARN · for anyone who acts on AI output

Can you catch the errorwhen the AI sounds certain?

The fastest way to over-rely on AI is to trust a confident answer. This Check seeds real AI text with a fabricated statistic, false claims, and biased framing, then scores how many you catch — and flags when you were sure and wrong.

Get the Check — $69one-time · instant download · yours to keep

Scope. For individual development and training planning. It does not certify, rank people against each other, or make employment decisions, and is not a hiring, promotion, performance-review, discipline, or termination tool. Not legal advice.

Five deliverables · runnable
Runnable Python engine
engine
Self-scoring workbook (.xlsx)
workbook
Facilitator's Guide
playbook
The Over-Reliance Field Guide
playbook
Seeded sample + 6 learners
sample
Works alongside
AI Fluency Diagnostic · Prompt Practice Lab · AI Data Analysis Starter Kit
01.The Problem

Confident AI output is the easiest thing to trust and the easiest to get wrong.

A made-up number

AI invents precise-sounding statistics with no source. The more specific, the more convincing — and the more likely to ship unchecked.

A plausible lie

Wrong dates, misattributed quotes, false cause-and-effect. They pass because they sound right, not because they are.

Sure, and wrong

The dangerous combination is confidence without verification. People wave errors through precisely when they feel most certain.

02.See It Work

Review a seeded AI draft. Your verdict comes from your own marks.

The full scoring model runs right here: six seeded-error items with weights that sum to 100, a hard gate on the fabricated statistic, and an over-confidence flag. Toggle a catch or change a confidence rating and watch the verdict move.

Try it: review the seeded AI output

Each line below is from an AI draft with a planted error. Mark whether you caught it, and how sure you were. Your verdict updates live from your own marks — nothing is invented.

Fabricated statisticgate itemweight 30

“Studies show 73% of teams that adopt this workflow cut their review time in half within the first month.”

Confidence
Plausible-but-false factual claimweight 18

“The technique was first standardized in the ISO 9001:2008 revision, which made it mandatory for all vendors.”

Confidence
Plausible-but-false causal claimweight 17

“Because response times dropped, customer satisfaction rose — the two always move together.”

Confidence
Biased / one-sided framingweight 13

“This is clearly the only sensible option; no serious team would consider the alternatives.”

Confidence
Loaded-language framingweight 12

“Obviously, anyone who still does this manually is simply wasting everyone’s time.”

Confidence
Outdated-as-current claimweight 10

“The current limit is 4,000 tokens, so keep every document under that ceiling.”

Confidence
70%ASSISTED

ASSISTED — you catch some errors but lean on the AI for others. Verify before you act.

Gate: you missed the fabricated statistic, so the verdict is capped at ASSISTED — your catch-rate would otherwise read INDEPENDENT. Catching a confident fake number is the load-bearing skill.

Over-confidence flag: you rated yourself highly confident on 1 item you missed (S1) — the over-reliance signal. Being sure while wrong is the habit to break.

For individual development only · does not rank people or make employment decisions · not legal advice.

03.Runs Offline, Same Math

A Python engine and a workbook reproduce the on-page verdict exactly.

Score one learner or a whole cohort from a CSV. The cohort headline is the weakest individual tier — not the average — so one untrained reviewer can't be hidden by a strong team mean.

$ python3 oar_engine.py --input sample_learners.csv

Dana Reyes               88%   INDEPENDENT
Marcus Hale              70%   ASSISTED  [OVERCONFIDENT]
  (gate: missed the fabricated stat -> capped to ASSISTED)
  high confidence on missed item(s): S1  (over-reliance signal)
Priya Nair               73%   INDEPENDENT
Sam Okafor               27%   RELIANT
Robin Cho               100%   INDEPENDENT
Lee Donovan              59%   ASSISTED  [OVERCONFIDENT]

COHORT: TEAM RELIANT  (MIN learner tier = RELIANT)
Most-missed error type: false_claim
Mean catch-score (context only): 70%  |  learners over-confident: 2/6

Mean reads 70%, but the cohort is TEAM RELIANT — the floor is the honest signal. Marcus caught everything except the fake number, and that alone keeps him out of INDEPENDENT.

04.The Standard

Three operating principles keep the verdict honest.

The fake number is the gate

Miss the fabricated statistic and you cannot score INDEPENDENT — capped at ASSISTED regardless of catch-rate. The cap only lowers; it never lifts.

Confidence is calibrated, not rewarded

High confidence on a missed item trips the over-reliance flag. It annotates the verdict for reflection; it never changes the tier.

Scored from your own marks

No AI, no invented value, no baked-in multiplier. The same marks always produce the same verdict, in the engine, the workbook, and on this page.

05.What This Is — And Isn't

A development tool that measures a skill, not a verdict on a person.

It is
  • A deterministic check of whether you catch planted AI errors.
  • A calibration tool that surfaces confident wrong answers.
  • A team training input — find the weakest skill, train it, re-run.
  • A deep-dive on the verification skill the AI Fluency Diagnostic flags.
It is not
  • A way to rank people against each other.
  • A hiring, promotion, performance-review, or termination tool.
  • A certification of anyone's competence.
  • Legal advice.

Scope. For individual development and training planning. It does not certify, rank people against each other, or make employment decisions, and is not a hiring, promotion, performance-review, discipline, or termination tool. Not legal advice.

06.Who It's For

Anyone who acts on AI output, and the people training them.

Individuals

Knowledge workers, analysts, marketers, founders — anyone who pastes AI output into something that ships. Find out whether you catch the errors you can't afford to miss.

L&D and team leads

Run it across a team, read the weakest-link cohort verdict, train the most-missed error type, and re-run on a cadence to build the habit.

07.Pairs Well With

Built to slot into the LEARN line and the verification toolkit.

AI Data Analysis Starter Kit

The honesty rules that keep AI output checkable — the verification skill this Check tests, made operational.

View
Anti-Slop Content System

Spot the AI tells and over-reliance patterns in your own published work, not just in a test.

View
AI Governance & Acceptable Use Starter Kit

The org-level safe-use policy that sits behind individual error-catching habits.

View

New to the LEARN line? Run the AI Fluency Diagnostic first to find your weakest skill; if it's verification, this Check is the deep-dive. The Prompt Practice Lab builds the writing side of the same fluency.

08.Common Questions

Direct answers on the gate, the flag, and the scope.

It measures whether you catch errors in AI output instead of trusting it. You review a piece of AI-generated text that has been deliberately seeded with six planted problems — a fabricated statistic, several plausible-but-false claims, and biased framing — and mark which ones you caught. It scores your weighted catch-rate and returns a tier: INDEPENDENT (≥70), ASSISTED (40–69), or RELIANT (<40), plus a cohort roll-up for a team. It's a deterministic, offline check (no AI grades you — your own marks drive the score), built for individual development and training, not for hiring or performance decisions.

Stop trusting the confident answer.
Start catching it.

One purchase, lifetime access, 12 months of updates. $69, once.

Get the Check — $69

Scope. For individual development and training planning. It does not certify, rank people against each other, or make employment decisions, and is not a hiring, promotion, performance-review, discipline, or termination tool. Not legal advice.

Sold by RedHub AI LLC · Secured by Stripe · redhub.ai