Results

Experiments

Questions, registered hypotheses and results from the reviewed experiment releases. Every chart and table reads the same verified summary CSV offered for download.

What these results describe

Network outcomes are MODELLED; comparison statistics and size accounting are DERIVED. C5 orbital inputs are REAL public data. C2 orbits and positions are MODELLED. Starlink’s operational routing is UNKNOWN; Starlink does not use T-CRH.

Intervals are taken from the result files. Exact correctness and byte-accounting checks have no sampling interval. These experiments do not establish a universal winner; each verdict below says whether the registered hypothesis is supported, not supported or mixed, and each experiment page lists the caveats behind it.

Registered experiments

Reviewed releases; no pilot or failed attempt is selected automatically.
Experiment	Question	Verdict
E0: Correctness gate	Do geometry, physics and all wire implementations agree?	Hypothesis supported (Pass): all 34 recorded checks have 0 violations. Only the cross-implementation and fixed-value rows are external references; the rest are self-consistency or invariant checks.
E1: Detection of up-but-slow links	Can deadlines detect an up-but-slow ISL without excessive false alarms, and where do they fail relative to local and conventional monitors?	Mixed, mostly not supported: at the frozen pilot rule (clean-packet false-positive rate 0.0041 to 0.00587 across cells, within the registered 0.01 limit), the hypothesis is not supported for the nominal rate cut (factor 0.3) and not supported for the added-delay family, supported for the strong rate cut (factor 0.2) and supported for ARQ loss (significantly more detections than both LS-H and CRH-16 in 10 of 24 nominal and strong cells; neither comparator has a slowness detector). LQ beats ChronoRoute significantly in 13 cells, and the LS-D comparison is not valid evidence.
E2: Header overhead	What wire cost buys per-hop time checking?	Hypothesis supported: T-CRH adds more bytes than CRH-16 and fewer than SRv6 SRH at every registered entry count (1–40). Against CRH-32 there is a crossover: CRH-32 is smaller at 1–3, 5, 7 entries, equal at 4, 6, 8–11, 13, 15, and larger at 12, 14, 16–40.
E4: Hard failures	Does local fallback preserve delivery under repairable link and satellite outages?	Mixed: against continue and CRH-16 the greedy-with-visited fallback has higher PDR in every outage cell (Holm-significant in 2) and lower registered blackhole time in every outage cell (Holm-significant in 7), but that metric excludes the fallback’s own hop_limit and loop losses, and the no-tail-penalty hypothesis is not supported against LS-H (tail-failure indicator above zero in 3 cells).
E5: Fallback policies	Does loop avoidance improve fallback completion, and does a Grid detour justify its cost?	Mixed: greedy-with-visited reduces the stuck/loop loss rate against pure greedy in 2 of 2 shells; the Grid detour shows no extra benefit, and continue wins triggered completion in C5.
E16: Python and ns-3 agreement	Are the Python and ns-3 implementations interchangeable within the registered scope?	Hypothesis supported within the replayed scope: every endpoint is equivalent, but by a common-random-numbers design that makes the test degenerate (all paired differences exactly 0), and no run lost a packet, so loss paths are not tested.