Papers / Preprints
Two papers,
one mechanism.
Both come out of the same question: when maintainers take a part from one aircraft to fly another, what does that do to the readiness of the fleet, and does the optimization problem sitting on top of it keep its structure. The long paper carries the model, the simulator and the operational estimates. The companion carries the proofs.
3 preprints / Not peer reviewed / Code and data public
The papers
Updated 19 October 2026Primary paper
Version 10 / 19 October 2026
Readiness at the Critical Point
One sign decides it: a curvature criterion for cannibalization feedback, measured on real fleets, and why greedy scheduling survives either answer
Readiness under cannibalization collapses as a saddle-node bifurcation rather than declining smoothly, the sign of the feedback is set by part-removal time against repair time, and on three independent public panels cannibalization intensity rises with supply-driven unavailability in all nine specifications they admit. The sign is what the theory needs; no magnitude is established.
- Kind
- Theory, simulation, and three estimates from operational data
- Figures
- 12, each labelled by kind
- References
- 39
- Reading time
- about 45 minutes
Submodular optimizationSaddle-node bifurcationPanel regressionDiscrete-event simulationSustainment schedulingCompanion note
Version 3 / 19 October 2026
A Curvature Criterion for Cannibalization Feedback
One inequality decides whether scarcity feedback destroys diminishing returns, and for a propensity derived from mechanism it reduces to a condition on logistics timing
Expected readiness is a scalar response composed with a monotone submodular coverage function, so every structural claim reduces to the curvature of that response. Concavity holds exactly when 2b'(x) + x b''(x) clears a threshold set by the balance law, and for a propensity derived from mechanism the condition collapses to a single inequality on logistics timing, decided at the origin.
- Kind
- Theory, with one measured section
- Figures
- 5, each labelled by kind
- References
- 19
- Reading time
- about 25 minutes
Submodularity ratioApproximation guaranteesCurvature criteriaGreedy algorithmsMatroid constraintsEmpirical note
Version 1 / 19 October 2026
The Sign of Cannibalization Feedback
Does cannibalization intensity rise with supply-driven unavailability? Three panels of published fleet statistics, nine specifications, one sign and no magnitude
Within unit, cannibalization intensity rises with supply-driven unavailability on three independent public panels and in all nine specifications they admit, six of them clearing five per cent under independent errors. No magnitude is established: no two panels measure the same pair of quantities, so none may be pooled and none identifies a direction of causation.
- Kind
- Three estimates from published operational statistics
- Figures
- 0, each labelled by kind
- References
- 11
- Reading time
- about 30 minutes
Panel regressionRandomization inferenceChart digitisationReliability dataReproducible provenance
The programme behind them
Where this
work sits.
Both papers come out of one line of work on complex systems and computational research. The programme overview sets out the questions behind them and what is and is not public.
Every numbered result in both papers is asserted by a test in one public repository, which also carries the four operational datasets as CSV with Parquet mirrors and per-file checksums, and the scripts that produced the two that were not transcribed by hand.
Status
What these are, and are not.
- Both are preprints. Neither has been peer reviewed, and neither is submitted to a venue.
- No experiment was run on an actual fleet. The operational sections estimate from published statistics, and each paper states which of its results is proved, computed, simulated, estimated, or not yet executed.
- Each version lists what the previous one withdrew. Version 9 of the long paper withdraws three p-values and corrects a count; those entries are kept rather than quietly fixed.