WHAT WOULD CHANGE YOUR MIND?
an inspectable hinge field · no probabilities

Try the worked example: on “Learn before committing”, move Permanent room’s amber bead from +1.6 to about +2.8. The lead crosses the hinge.

The field currently leans

—

Reading the threads…

The closest hinge

Finding the smallest plausible reversal…

now
hinge test

Test the closest hinge to compare the current field with one minimally changed assumption.

Which assumptions carry the result?

These are contributions to the gap between the current leader and runner-up. A large thread can still be uncertain; “load-bearing” means it changes this comparison, not that it is true.

    Keep, reload, or share your reasoning

    JSON preserves every editable assumption and round-trips back into this field. The Markdown note is a human-readable record of the current lead, ranges, closest hinge, and stated limits.

    Nothing is uploaded. Loading replaces the current field only after validation succeeds.

    Keep one copy in this browser

    This uses local storage on this device. Export JSON if the decision matters; browser storage can be cleared.

    No saved copy found.

    How the hinge is calculated — and where it stops
    1 · Weigh each threadEach assessment from −4 to +4 is multiplied by its consideration’s importance from 1 to 5. Weighted values are added, then divided by total importance.
    2 · Admit uncertainty“Honest uncertainty ±” is a score-point bound, not a probability. Each option’s low and high outcome move every assessment within that stated bound, clipped to the scale.
    3 · Find one reversalFor each challenger and thread, the engine divides the current raw gap by that thread’s importance. It keeps the smallest relative leader-to-challenger shift that fits within the stated uncertainty.
    • The hinge assumes one consideration changes while all others stay fixed. Correlated assumptions, feedback, thresholds, and interactions between considerations are not modeled.
    • Ranges are stress bounds, not confidence intervals. A wide range means you admitted room to be wrong; it does not say outcomes inside are equally likely.
    • The tool cannot discover a missing option or consideration, judge whether your evidence is good, or tell you what ought to matter.
    • Weights and scores are prompts for explicit disagreement, not measurements with objective decimal precision. The interface rounds readings even though the deterministic engine retains exact inputs.
    • A lead is not a recommendation to act. Irreversible, ethical, legal, health, or safety decisions need domain evidence and accountable human judgment.