← AI Operations Tools

The Verification Quadrant

Based on the Templeton Ratio

Before automating a task with AI, ask two questions: How hard is it to do? and How hard is it to check? The ratio between these two numbers - the Templeton Ratio - determines whether AI will save you time or double it.

Start with a scenario

Drag the dot in the chart, or open Customize below for sliders.

Calculation Difficulty →
EasyHard
Hard to verify
Easy to verify

The Templeton Ratio

AI value isn't determined by how impressive the output looks. It is determined by one number: how much cheaper it is to check than to produce.

# The Templeton Ratio
T = time_to_do / time_to_check
# T = 1 → no value (you are doing the work twice)
# T = 10 → 9x leverage per unit
# T = 100 → transformative (check 100 outputs in the time of 1)

The Verification Quadrant maps the two components of this ratio. Tasks in the AI Sweet Spot have high T - hard to calculate, easy to verify. This is the P vs NP intuition applied to operations. Generating a solution is expensive but checking it is cheap, and that asymmetry is where automation creates leverage.

The Verification Trap kills more AI pilots than technical failure. The task looks automatable because generation is easy. But "is this output correct?" requires human judgment that scales linearly with volume. T approaches 1. You end up paying for generation and review instead of just doing the work once.

How to Score a Task

Calculation Difficulty

How hard is it for a human to produce correct output?

Low (0.0 - 0.3)
Lookup, extraction, classification against a known taxonomy. Anyone trained on the format can do it reliably.
High (0.7 - 1.0)
Requires domain expertise, synthesis across sources, or creative judgment. Takes a skilled human meaningful time.

Verification Difficulty

How hard is it to determine whether the output is correct?

Low (0.0 - 0.3)
Binary pass/fail, compiles-or-not, diff against ground truth, spot-check values. A machine or a quick glance can verify.
High (0.7 - 1.0)
Requires reading carefully, expert judgment, or "is this good?" assessment. Verification takes nearly as long as doing the work.

Decision Rules

LOW / LOWAutomate for volume - value comes from throughput, not skill replacement, and full autonomy is safe.
HIGH / LOWAutomate aggressively. This is where AI creates transformative ROI. Build quality gates on the cheap verification and scale.
LOW / HIGHDon't automate until you can make verification cheaper. Invest in rubrics, deterministic checks, or decomposition before deploying AI.
HIGH / HIGHDo not automate yet. No cheap feedback signal means no quality gates - unless you build one. See below.
Framework →Stage 5: Compound

The Capital Value of Verifiers

The quadrant above describes the current state of a task; it is not permanent. Build a verifier and you permanently move the task across the diagram - red into blue, amber into green. And unlike almost every other capital asset, the verifier appreciates through operating use: every failure caught is encoded, and the next run inherits the catch.

Read the framework: investment math, what counts as a verifier, fail conditions, and why this is the highest-leverage move in an AI operating budget →

When to Use This

Use when

  • +Triaging a backlog of AI candidates and you need to sort them fast
  • +Deciding whether the capital investment is the generator or the verifier
  • +Engineering and business need a shared vocabulary for what is hard
  • +You want a quick visual before doing the heavier Automation NPV work

Skip when

  • -You need a single number - compute the Templeton Ratio directly
  • -The task is fully deterministic and no AI is involved
  • -Both axes are already well understood and the team just needs to ship
  • -You are past categorization and into execution (use the Promotion Protocol)

Rosetta Stone

Four circles, four readings of the same object. Each role reads the artifact through its own lens.

See also