Crossover instrument · v0.7 · suitability + economics

Who should hold this task?

describe a task (or set the dials) · one atomic, repeatable sub-task at a time
Describe a task
🔑 Claude API key

Stored only in this browser (localStorage) — never in this page or anywhere but Anthropic. The classify call runs straight from your browser (anthropic-dangerous-direct-browser-access) and spends your key. Bring your own key — other visitors can't see yours and must paste their own; the sliders and presets work without one.

What the task demands · Claude sets these on Classify placeholder defaults
Compare against · reference human (role & level)
Intrinsic suitability · before economics (structural)
Cost verdict · at today's dial
Zone · verifiability × cost-of-miss
The why — four lenses on the same task
Lens 1 · cost vs cost-of-miss (how much a miss hurts)consequence
AI + envelope human expert cyan: AI cheaper · amber: pay the expert
Lens 2 · cost vs required quality bar (how good it must be)curves blow up at each ceiling
AI human expert your bar q*
Lens 3 · role × level (human staircase vs uniform AI line)where AI crosses the ladder
AI displaces human cheaper AI can't reach
Lens 4 · role portfolio (what's left of the role)role + token dial + AI capability
Lens 5 · zone map (verifiability × cost-of-miss)the Zone tile, drawn · dial-invariant
automate (cyan) cheap augment danger — human-led ● = this task
v0 scaffold · constants are choices, not measurements. The token dial is the only thing forecast forward; capability ceilings are frozen inputs. The "Classify with Claude" box calls the Anthropic API directly from your browser with your own key (kept in localStorage) — it is the only feature that needs a key. Ported from the AI-in-PDLC Task Rubric project.