Reading path
When can we trust a model?
From a convincing fit to a useful guarantee: ask what the observations, evaluation and bounds actually establish.
Three core readings: about 53 min · With the optional extension: about 80 min
Understand the problem, then examine the evidence.
The first three articles form the core. The fourth asks for more mathematical background and is optional. Reading times are estimates; there is no need to finish in one sitting.
Core 1 of 3About 8 min
When the Fit Is Not the Model
See why several parameter explanations can fit the same observations, then ask where a new measurement could separate them.
- Useful starting points
- Graphs and the idea of fitting parameters. Derivatives help with the detailed uncertainty discussion.
- Evidence and limits
- A finite-grid demonstration with deterministic pseudo-noise is not a confidence region or a proof of structural non-identifiability.
Core 2 of 3About 25 min
When Solar Power Changes Fast: Why Better Intervals Are Still Not Reliable Enough
Distinguish improvement over a baseline from meeting a reliability requirement in the hardest eligible groups.
- Useful starting points
- Percentages and prediction intervals. The article introduces calibration and coverage.
- Evidence and limits
- No deployable candidate clears the absolute criterion in the five confirmation stations; origins without current power are excluded.
Core 3 of 3About 20 min
Random Stress Tests Are Not Certificates
Separate the best evaluated value from a valid upper bound over the entire declared set.
- Useful starting points
- Maxima, derivative signs and the meaning of a numerical bound.
- Evidence and limits
- The certificate concerns one synthetic dimensionless function. Fixed zero-hit designs establish no general miss probability or reactor-safety claim.
Optional advanced readingAbout 27 min
Two Networks, One Spectrum, Different Alarms
Apply the distinction to a specified sensor, deadline and threshold, and see why a valid bound can still leave a decision unresolved.
- Useful starting points
- Network diagrams and exponential decay; the detailed derivations use linear algebra.
- Evidence and limits
- The compact exact formula assumes an acyclic equal-holding-rate network with uniform removal. The alarm thresholds are research choices, not safety limits.