Curator Value Function V_curator · Ch 3(Signal selected ÷ noise available) × context depth. Sell the taste layer, not the generator.
Worked example — A security analyst's Friday email
She reads roughly five hundred vendor advisories a week and forwards the four that actually matter to her estate, each with one line on why it matters here.
It came out at 0.06 on this calculator's scale — and that is strong curation, not weak.
Read the direction, not the digit. Triple the advisories to 1,500 and her four picks stay four: the ratio falls, and what she sells — the sifting — becomes more valuable, not less. Drop her context to zero and it collapses to nothing however good the ratio looks, because that's a Top-10 listicle. Sell the taste layer, never the generator.
Your turn — change one number
Result 0.06
Yellow. Thin. Deepen the context or select harder.
Book's benchmark: 🟢 ≥ 5 · 🟡 1.5–5 · 🔴 < 1.5. Explodes as noise → ∞, **only if** context is real; context_depth ≈ 0 = a Top-10 listicle.
Heads up: Appendix A bands this at ≥ 5 for green, but the formula divides by the noise — so any realistic amount of sifting scores far below 1, and the appendix's own note that it 'explodes as noise → ∞' points the other way. The light here is scaled to this calculator's inputs, not to the printed benchmark. Read the direction of travel, not the digit.
⌁ ask what this number means for me
Authenticity Token T_authentic · Ch 3Human trust × 1/AI-substitutability. Prove you have context AI can't fake — that's the whole moat.
Worked example — Eleven years inside pharma quality systems
A consultant the client trusts almost completely. A model can draft the SOP text perfectly well — but it doesn't know which regulator reads this one, or how that regulator reads.
It came out at 30.0. Green, exactly on the line.
Hand the same person a task a model does at 0.8 substitutability and the token falls to 11 — yellow, AI-washable. Her trust didn't change. The task did. The moat is which work you accept, not how good you are.
Your turn — change one number
Result 30.00
Green. A sommelier's pick. High trust, low substitutability.
Book's benchmark: 🟢 ≥ 30 (high trust / low substitutability — a sommelier's pick) · 🟡 10–30 · 🔴 < 10 (AI-washable).
⌁ ask what this number means for me
Risk Tax RT · Ch 4(P_fail × cost of disaster) + (panic × career cost). The number running in their amygdala.
Worked example — A $500K pilot in front of a VP of Ops
He privately puts the odds of failure at 30%. He's fairly rattled about it — call the panic 0.5 — and he reckons owning a failure like this costs him his next role, about $800K of career.
It came out at 1.1× the deal. Green — he can say yes without it being a career decision.
Raise his panic to 0.9 and the same deal reads 1.7× — yellow. Nothing about your product changed. Your entire job in that room is the second term: who else has done this, what the rollback looks like, and whose name is on it if it fails.
Your turn — change one number
Result 550
Green. Low tax. They can say yes without fear.
Book's benchmark: 🟢 < 1.5× · 🟡 1.5–3× · 🔴 > 3× (career risk > any upside; walk).
⌁ ask what this number means for me
Decision Fatigue DF · Ch 4Decisions made ÷ glucose remaining. Why the 4pm 'no' is automatic.
Worked example — The 4pm meeting you asked for
The director you need a yes from has made about forty real decisions today and is running on coffee.
It came out at 13.3. Red — after 2pm the default answer is no.
Same pitch at 9:30am: twelve decisions in, rested — 1.5, green. This is the cheapest variable on the entire list and it costs you nothing but a different calendar invite. No costs them zero calories. Stop making them spend calories on yes.
Your turn — change one number
Result 13.33
Red. After-2-PM depletion. 'No' costs zero calories. Come back tomorrow morning — or bring food.
Book's benchmark: 🟢 < 5 (before 11 AM) · 🟡 5–<10 (post-lunch) · 🔴 ≥ 10 (after 2 PM — default answer is "no").
⌁ ask what this number means for me