Same scores, different worlds
The individual scores stay still. The shared failures move.
Cubits11 / Explore / 01
Same scores.
Different worlds.
Two guardrails each miss some cases. Their scores tell you how often. They do not usually tell you whether they miss the same cases.
Change what you know. See what remains possible.
What do the scores leave open?
Interactive construction · no measured systemPossible rate of both missingDerived · marginals alone
0%–10%
A range of compatible answers, not a confidence interval.
0%100%
The amber marker is 1%: what independence assumes.
- Neither misses · plain field
- 85%
- A only · cyan
- 5%
- B only · gold
- 5%
- Both miss · hatched
- 5%
Area represents probability. These four cells sum to one; they are a mathematical witness, not observed cases. Gold locates event B, not an evidential status.
Every point from 0% to 10% is compatible with two 10% marginal miss rates.
This static example works without JavaScript. Enable JavaScript to change the rates and assumptions.
Inspect the calculation
Let a and b be the marginal miss probabilities, and q the probability both miss.
max(0, a + b − 1) ≤ q ≤ min(a, b)
The witness is (1 − a − b + q, a − q, b − q, q), ordered as neither, A only, B only, both. Nonnegative cells and a sum of one give the interval. Independence adds q = a × b.
A supplied interval is intersected with these bounds. This is established two-event Fréchet arithmetic; the interface adds no new theorem. Inspect the calculation source.
What this does and does not establish
Exact input probabilities, a common population, and the same declared miss event. Displayed values are rounded to two decimal places. No sampling uncertainty is modeled. A static stack catches a case if either guard catches it, so its miss event is their intersection.
This is not a deployed route, a vendor comparison, or evidence that a real stack is safe or unsafe. User-entered joint intervals are hypothetical constraints, not measurements.
Inspect the bound’s registered scope · Inspect endpoint witnesses · Run a reproduction
A small cinema of evidence.
Existing films / choose one / sound offThirteen worlds
Follow the possibilities left open by a partial description.
Scope and provenance ↗ · Download film
What this film does not claim (MC-003)
- no new mathematics is claimed; the bounds are Fréchet's and the lower endpoint is Bonferroni's
- not a claim that guardrails in general fail together — this stratum cannot separate shared blind spots from prompt-difficulty heterogeneity
- not a vendor evaluation; the released verdicts are at unstated default configurations, and one supervisor fires exactly once in the 170 released rows
- the identified interval is what the marginals leave open, not a prediction about any deployed stack
- leave-one-out unions identify only exclusive full-stack coverage; they do not identify pairwise or higher-order overlap, Shapley values, or causal contribution
- says nothing about adversarial prompts, which dominate the full evaluation and have no per-item release
What the seal proves
A receipt reaches the edge of what it can establish.
Silent films with on-screen text. Each source manifest describes its thesis, construction, and non-claims. Read about the film method.
Follow the thread.
From intuition to inspectionMove the misses yourself.
Place individual misses on a shared sheet in the original Worldspace experiment.
Enter Worldspace ↗Which column is missing?
Inspect the bounded census of public evaluations, then follow each row to its source.
Open the census ↗What survived correction?
Read the corrections and historical work with their limits still attached.
Read the corrections ↗ · ArchiveAn answer becomes more interesting when you can see what would change it.