This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Two layouts give a spread, not an estimate, and the interesting claims rested on counts in single digits.
What was tested
Every combined-condition case behind a claim was rerun on four more layouts: 26 cases and 1,380 runs, with 95 % confidence ranges.
What was found then
Five of ten claims changed. Withdrawn: the storm-plus-slow-link interaction. For the eVTOL, storm with the slow link gave 0.42 collisions per 1,000 km, barely above the slow link alone (0.33). Also withdrawn: 'crowding multiplies most'. The slow-responding craft's calm rate halved (0.16 to 0.08), so earlier multipliers were only about half the six-layout values. Turbulence came in at 3 times calm, though its range touches calm's. Impact speeds along the line between craft were typically 7.3 m/s, lower than the relative speeds quoted before. Survived: crowding and the link multiply losses (about 8 times at the moderate level), speed raises severity rather than frequency, and agility matters most.
What we learned later
Losing messages separately at each receiver (2026-09-11) moved no link result outside its range (all link cases together 109 → 91 collisions); the ideal craft's contacts at 0.5 messages a second went from 4 to 1.
research/envelope.md (The link pillar under independent loss)
The eVTOL results were redone after two layouts run on another Mac were replaced: storm with the slow link 0.36, the slow link alone 0.31. No conclusion changed.
research/envelope.md (conjunction map, evtol note)
Crowding and speed results were redone with a new way of placing craft at the start (slow-responding craft at 4× crowding 1.65, eVTOL 0.72); none moved outside its range.
research/envelope.md (conjunction map † note)
Later speed work found that losses do rise with cruise speed for craft with response lag; at the phase-1 close, speed became a condition that matters for safety in some cases.
research/design/phase1_close.md:37
What changed
Nothing (the safety map was rewritten).
Numbers and source references
- 26
- cells re-runresearch/reviews/brief-c5.md:57
- 1,380
- new runsresearch/envelope.md:5787
- 1.02 million
- craft-kmresearch/envelope.md:5787
- 95 %
- Poisson intervalsresearch/envelope.md:5787
- Five of ten
- claims that changedresearch/envelope.md (Six seeds instead of two)
- 0.16 → 0.08
- lagged calm rate, two vs six seedsresearch/envelope.md (Six seeds instead of two)
- 3×
- turbulence 3 against calmresearch/envelope.md (What multiplies the vehicle penalty)
- 7.3 m/s
- median radial closing speedresearch/envelope.md (What the collisions are; What cycle 3 found)
- 50.4
- max radial closing speed, m/sresearch/envelope.md (What the collisions are)
- 8×
- multiplier at the moderate levelresearch/envelope.md (Six seeds instead of two)
- 1 Hz
- link rateresearch/envelope.md (Six seeds instead of two)
- 0.42
- eVTOL storm 10 + 1 Hz, per 1,000 kmresearch/envelope.md@e735ce4:393
- 0.33
- eVTOL 1 Hz link aloneresearch/envelope.md@e735ce4:391
- 0.56
- lagged storm 10 + 1 Hzresearch/envelope.md@e735ce4:393
- 0.62
- lagged 1 Hz link aloneresearch/envelope.md@e735ce4:391
- 16 vs 30 m/s
- speed comparisonresearch/envelope.md@e735ce4