See It Work — Reward the AI for staying in bounds — not for raw output | S2 Vol 5 · Yield Organism & Generational Prosperity

Every agentic system rewards something, even if you never said so out loud — usually throughput: more tasks, more output, more speed. This replaces that with alignment scoring: each action is scored on four simple questions — was it in scope, did it respect your sensitivity rules, did it leave checkable evidence, and did it strengthen or erode your constitution over time.

This is the single highest-leverage choice in the whole volume. Reward speed and the AI accelerates toward whatever proxy you picked, eroding your rules over thousands of tiny steps. Reward fidelity and staying within your rules becomes the path of least resistance.

What this means for you

The AI is rewarded for staying within your rules, not for raw output — so it compounds your constitution instead of eroding it. What this means for you: your AI's incentives point at the thing you actually care about (staying trustworthy), so it gets safer over time on its own, rather than drifting toward whatever metric you happened to measure.

Each action gets a four-part alignment score — none of it about speed or volume:

Alignment Score

in scope?+1 / 0 / -1

sensitivity rules respected?+1 / -1

evidence left to check?+1 / 0 / -1

strengthened the constitution?+1 / 0 / -1

Reward fidelity, not throughput — and the agents compound your constitution instead of your extraction.

For the technical reader — the command, and how to verify it yourself

# one line · you do not need to run this
see walkthrough

./bl-verify
# -> the AI scored on rulebook fidelity, not output

Full step-by-step is in Appendix RX: Hands-On Demonstrations in the book.