THE INTERACTIVE LAB SYNTHETIC MODEL

Break the assumption.
Then test the test.

Change one condition. Compare what the implementation does
with what the evaluation claims.

BH is an invented input in conversion units, not speed or flight telemetry. These illustrative barriers are not flight-certified fixes.

This demo requires JavaScript. The complete research, methods, and published results remain available without it.

One conversion opportunity

START WITH A QUESTION

01 Input & lifecycle

BH conversion units

-40,00065,535

Signed 16-bit result: -32,768 to 32,767.
Ties-away overflow starts at ≤ -32,768.5 or ≥ 32,767.5.

Time seconds after lift-off

0 s / lift-off40 s / cutoff80 s

Alignment executes only when time < 40 s, if retained. This is a fresh snapshot on every change, not recovery of a previously halted processor.

02 System design Retain alignment after lift-offKeep the unnecessary conversion path reachable. Guard the conversionReport out-of-range status; do not silently clamp BH. Isolate alignment exceptionsKeep navigation running when alignment raises an error. Reject diagnostic frames as flight dataBlocks the unsafe-command proxy, not navigation loss. Identical SRI replicas

Shared deterministic faults are not independent hardware failures.

03 Test apparatus What is in the test loop? What counts as failure?

A stub changes the test, not the implementation. A diagnostic frame is still a frame.

ACTUAL IMPLEMENTATION / WITH SELECTED BARRIERS

Awaiting model

Failed units: --
Navigation: --
Unsafe command: --

APPARENT EVALUATION / SELECTED HARNESS & ORACLE

Awaiting evaluation

01 / CONVERSION--
02 / EXCEPTION POLICY--
03 / DIAGNOSTIC INTERFACE--

The left-hand controls are interventions on a Boolean mechanism model, not a digital twin. Navigation loss and unsafe commands are modeled outcomes, not a physical breakup calculation. Read the state semantics.

TEST BUDGET EXPLORER / ANALYTIC, NOT A RANDOM RUN

When does more testing help?

Assume independent opportunities, each with probability p of both exercising and detecting failure. This calculator is separate from the snapshot controls above.

P(at least one detection) = 1 - (1 - p)ⁿ

Baseline profiles assume retained alignment, no containment, a faithful implementation, and a safety oracle. The weak oracle and idealized stub have detection p = 0 in this model. No browser random generator is used; these are analytic probabilities, not a replay of Python's seeded counts.

A snapshot is not a trajectory.

The published temporal demonstration latches navigation loss: ending alignment does not restart a halted processor. Its invented linear BH ramp first crosses the sampled threshold at 32.8 seconds, not the historical failure time.

Synthetic linear BH trace with latched navigation loss at 32.8 seconds; not flight telemetry — Published synthetic trace, copied from the research artifact at build time. Read its method and limitations.

Take the BLIND-SPOT teaching exercise →