Evaluate policy impacts

Evaluate policy impacts — real work, not an imagined feature: named inside 5 evidenced career tasks. Below are four ready AI prompts for it, one per height of help: do it, make it easier to accept, decide when you are stuck, and change the pattern for good.

5career tasks name it
3prompt heights

The four heights

The same task, four distances: today's deadline, the next reviewer, the stuck moment, the pattern.

Improve — make it easier to accept

+
Before I send this to the ministry, make the evaluation easier to use for policy decisions: put the…
Before I send this to the ministry, make the evaluation easier to use for policy decisions: put the headline recommendation and expected fiscal impact on the first page, make the counterfactual assumptions explicit, flag where data are weak, and add a one-sentence implementation implication for each recommendation.

Decide — diagnose the stuck moment

+
Our draft says the tax credit increased investment by 8 percent. The treasury analyst emailed…

The treasury asks whether behavioral responses invalidate our estimates

Our draft says the tax credit increased investment by 8 percent. The treasury analyst emailed asking if increased tax avoidance or shifting investment timelines might be biasing our estimate. I can't fully test those channels with available data. What is the most defensible answer to give now about the robustness of our estimate and the right next empirical step?

Become — change the pattern

+
On three evaluations we understated the fiscal cost of incentives because off-budget credits and…

Repeatedly underestimating fiscal costs of incentives erodes trust

On three evaluations we understated the fiscal cost of incentives because off-budget credits and carryforwards weren't tracked consistently. That led to heated meetings and policy reversals. Which habit should I change to avoid this pattern: expand fiscal data requests, add conservative sensitivity bounds, or require treasury sign-off before publication? Recommend one habit and how to implement it across reports.

Where the evidence lives

Who was seen doing this, and what people really ask.

Software tasks in the LLOS Work Atlas come from evidence, never a feature list: careers attested to do the work, real job descriptions, and the questions people actually ask (with their view counts). Facets — feature, workflow, troubleshoot, administer, deploy, scale — are open metadata: the work decides, not a taxonomy.
Copyright © LLOS.ai · 2026 — original pedagogy, voice, and design — all rights reserved.

The rest of the map

Same library, five ways in.