Quadrant Shifting #

Stage 5Compound

The Verification Quadrant is not destiny. Capital investments can move tasks to better positions. These are the five moves that change where a task sits on the grid.

Start with a scenario

Verification trap

Easy calc, hard check - build a verifier

Avoid zone

Hard both ways - decompose the task

Hard calculation

Expensive compute - enrich inputs

Fuzzy verification

Subjective checks - constrain outputs

Download JPEGCopy ImageCopy URLEmbed

Drag the before-dot to reposition the task. Pick a shift below to see where it moves.

Customize shiftStart: (8, 8) | Move: none▾

The Grid #

Select an investment type below to see how it shifts tasks on the quadrant.

Verification Difficulty →

Verification Trap

Do Not Automate

Automate Now

AI Sweet Spot

Calculation Difficulty →

The Five Moves #

↓ Build a VerifierShifts tasks downward

Do Not Automate / Verification Trap → AI Sweet Spot / Automate Now

↙ Decompose the TaskShifts tasks left and down

Do Not Automate → Multiple smaller tasks in better quadrants

← Enrich InputsShifts tasks left

High calculation difficulty → Lower calculation difficulty

↓ Constrain OutputsShifts tasks down

High verification difficulty → Lower verification difficulty

↓ Build a RubricShifts tasks down

Subjective "is this good?" verification → Structured scoring against criteria

The Capital Allocation Frame #

Each of these five moves is a capital investment. The same questions you would ask about buying a machine apply here:

NPVWhat is the present value of all future labor savings minus the investment cost? Example: a verifier that saves 40 hours/week at $50/hour is worth $104k/year before discounting.

DEPRECIATIONModels depreciate (distribution shift, competitive catch-up). But verifiers and data moats can appreciate - each failure case makes them smarter. The net rate determines whether this is a wasting asset or a compounding one.

RISK-ADJUSTEDThe Dollarized Confusion Matrix gives you the variance. High cost asymmetry means high variance in outcomes. Risk-adjusted return = E[NPV] - λ × Var[NPV].

MOATWhat does it cost a competitor to replicate your position? Models: low moat (commodity). Data from operations: high moat. Custom verifiers encoding institutional knowledge: highest moat.

DUAL CURVEPhysical assets depreciate. Knowledge assets have a dual curve: the model depreciates (like a truck) but the data and verifiers appreciate (like land under the depot). Invest in the appreciating side.

The Decision Sequence #

These tools form a workflow:

1.DIAGNOSE - Plot the task on the Verification Quadrant. Calculate the Ablaza Ratio.

2.CALIBRATE - Use the Dollarized Confusion Matrix to set the right threshold from error costs.

3.INVEST - Choose which shift moves the task to a better quadrant. Evaluate the investment using capital allocation math.

4.VALUE - Calculate the NPV of the automation to compare against physical alternatives. Same math, different asset class.

When to Use This #

Use when #

+A task sits in a hard quadrant and you are willing to spend capital to move it
+You can estimate the cost of reducing calculation or verification difficulty
+Task volume is large enough to pay back the shift investment
+You are choosing between making the task easier vs trying harder at it

Skip when #

-The task is already in the sweet spot - no shift is needed
-Shift cost exceeds the NPV of the shifted task (run Automation NPV first)
-The difficulty is intrinsic (adversarial input, ambiguous ground truth)
-You have not yet diagnosed the quadrant (use Verification Quadrant first)

Rosetta Stone #

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

[Allocator

CapEx for OpEx leverage. Spend once on a verifier or a constraint; the task moves to a cell where automation has positive NPV forever after.](/positions/allocator/)[Operator

The move where you invest in tooling so next quarter's version of the same decision is cheaper. Makes the work easier to scale without growing headcount.](/positions/operator/)[Builder

Building the test infrastructure, the constraint language, the decomposition, the rubric. Each move shifts a stubborn task into a cell you can actually automate.](/positions/builder/)[Scientist

A structural intervention on the cost function. Instead of optimizing over the existing quadrant, you change which quadrant the task occupies - a change of measure in cost-space.](/positions/scientist/)