Your total drift is small, but your map is noisy

Why a clean net score can hide a broken map

If you over-claim on three tasks and under-claim on three tasks, the net drift is zero. The score looks like an accurate read. But what it actually shows is a map with errors on both ends — not a map that is right everywhere in the middle.

A person with zero net drift who got there by being accurate is in a completely different position from a person with zero net drift who got there by making two types of opposite mistakes. The first person has a reliable map and can act on it. The second person has a map with hidden noise, and every action based on it carries more risk than the score suggests.

This result is saying: the total looks close, but the individual readings are unreliable. The fix is not to change one number. It is to sort the whole calendar again with clearer criteria.

What errors in both directions actually look like

The over-claim side: some tasks you are calling your edge are execution work that AI handles competently. These are usually tasks that still feel skilled because they used to require concentrated effort, or because they sit close to important outputs. A polished summary. A well-researched brief. A formatted report. The quality of the output can be real while the production layer is no longer scarce.

The under-claim side: some tasks you are giving to AI are judgment work where context, stakes, or relationships change the correct answer. These are usually tasks where AI produces a coherent response quickly, which makes them look like execution work. An escalation decision. A read on whether a customer is genuinely satisfied. A call on whether a situation is an exception worth making. The fluency of the AI output disguises the fact that the task required someone to hold the context, not just produce a response.

Both errors are happening simultaneously. The execution work is getting your attention when AI should have it. The judgment work is getting AI's attention when it should have yours.

Why this combination is harder to fix than a single-direction error

If you only over-claim, the fix is clear: move execution work to AI and reclaim the time. If you only under-claim, the fix is also clear: reclaim the judgment tasks from AI. When both errors are present, fixing one without the other leaves the calendar just as noisy, only tilted differently.

The deeper issue is that both errors share a cause: the classification criteria are not sharp enough. You are not distinguishing reliably between tasks where AI should own the production and tasks where you should own the decision. When the criteria are fuzzy, errors go in both directions.

The sharp criterion is simple to state and hard to apply consistently: execution tasks are ones where the correct answer can be produced from the inputs alone, and context only improves the output at the margin. Judgment tasks are ones where the correct answer changes with the specific context, and getting it wrong has consequences that require someone accountable.

The two fixes, applied in sequence

First: name the specific tasks, not the categories. Go through your result and identify the individual tasks flagged by the model. On the over-claim side: which specific task are you protecting that AI handles well? On the under-claim side: which specific task are you giving to AI that requires you to hold the context? One task on each side, named specifically, is more actionable than a general reorientation.

Second: make a swap, not a policy. Move the identified execution task to AI on its next occurrence. Reclaim the identified judgment task on its next occurrence. Do not announce a new system. Do not try to reclassify everything at once. Make the two specific changes and observe what happens to the quality of both. The execution task should be fine with AI. The judgment task should feel different when you are the one making the call.

Use the contrast to sharpen the criteria. After making the swap, you will have fresh evidence about what the difference between execution and judgment feels like in practice. Use that evidence to review the rest of the flagged tasks. The criterion gets sharper with real examples attached to it.

What a clean map actually produces

When the map is accurate and the calendar reflects it, two things happen that do not happen when the map is noisy. Execution work gets done faster because AI handles the production. Judgment work gets done better because your full attention is available for the decisions that need it.

The combination is not just more efficient. It changes the kind of skill you are building over time. Every hour spent on judgment work in real situations builds judgment. Every hour spent on execution that AI handles builds nothing new. A noisy map distributes your attention across both categories randomly. A clean map concentrates your attention where it compounds.