Proxy Hunt — test the measure, not the story

Or try:

Syntax reference

Hunt

title: Lantern invite pressure and date: 2026-08-13: the authored context and date for this hunt
outcome: Groups retain after week one: the desired outcome — it is not interchangeable with the proxy
proxy: Invitation rate: the operating measure under pressure, or monitored as a guardrail
action: Prompt every active member: the action whose consequences you are testing
mode: optimise or monitor: monitor also needs optimisation-pressure:
verdict: Keep the measure paired with the guardrail: an author-stated hunt-level note for a shared artefact; it never replaces the tool-derived review state or causal limit. Use verdict: off to suppress only this note.
palette: ocean: use ocean, slate, ember or plum
accent: #C05621: override the named palette with a 6-digit hex accent

Theories

intended-theory:: one authored mechanism:; the route is action → mechanism → desired outcome
protects:: one or more indented - Protected outcome lines
failure-theory fatigue:: up to three named challenges, each with mechanism:, harmed-outcome:, guardrail:, basis: and weaken-with:
basis: reasoned-mechanism: or speculative-concern; neither is evidence certification

Reported pattern

reported-pattern:: author-entered proxy-reading: and outcome-reading:, plus the named outcome, population, horizon, comparator and source. A pattern never proves the mechanism.
trade-off: and decision-rule:: name an explicit conflict; an empty rule leaves the hunt unresolved

Select a failure theory to inspect its scoped receipt.

Start typing — or load an example.

Export

Authored theories and reported patterns remain separate · full exports omit the live selection · the URL carries the hunt and selected receipt

Why hunt the proxy?

Operating measures are useful because they move sooner than the outcomes we care about. That same convenience makes them dangerous targets: a team can improve the number by changing behaviour in ways that do not improve the goal — or that quietly damage an outcome the product needs to protect.

Proxy Hunt makes the challenge inspectable. The intended route and each failure route contain actions, mechanisms and outcomes only. The proxy and any reported readings sit outside those routes, so a visual connection cannot smuggle in a causal claim. A co-moving pattern can justify investigation; it cannot establish why the outcomes moved.

The result is a guardrail decision, not an approval stamp. An unfinished hunt stays visibly unfinished. A speculative concern stays speculative. A reasoned mechanism still says what evidence would weaken it. Use Gauge to elicit a room’s judgement, Fermi to test a conditional calculation, and Paths when new knowledge changes what work should happen.