This is an earlier experiment. Read the amendments below before treating its findings as current.

The question

The new default changed two things: smaller extra room and early reaction (steering away sooner). In the multirotor storm test it crashes far less than the old controller but spends about a quarter more time close. Which change does which?

What was tested

The multirotor storm test (storms plus a poor radio link) on five sets of layouts, 350 scenario runs per setting, for all four combinations of the two changes; plus eVTOL and multirotor on the mixed test set (145 standard scenarios).

What was found then

Deep time inside the old bubble rose from 0.053 to 0.066 wherever the smaller room was on. Collisions per craft-hour halved wherever early reaction was on (0.0173 with neither, 0.0088 with early reaction alone). So the room brings the closeness and early reaction the crash cut, roughly adding up; in crowded airspace early reaction also adds some closeness. On the mixed set eVTOL crashes fell from 18 to 3 (multirotor too few to read). the researcher then restated the target: each airframe's losses per craft-hour at its own cruise speed (16 m/s), not a flat curve across speeds.

What we learned later

LATER / CYCLE 168

Early reaction at strength 5 with the full extra room was flown on the crossing for the first time next; it later became the default for three airframes (c181, c182, c185).

research/data/oos/mr_cx_2x2.md; research/design/tau_stage_close.md §10

What changed

nothing in the controller; the researcher restated the stage target

Corrections retained in the record

  • Earlier wording: a “charge” that pushes harder when two craft are already inside each other's bubble

    Record correction: The charge reduces the predicted clearance of every closing pair by 5 × closing speed × response delay, so avoidance starts earlier. It is not tied to craft being inside each other's bubble.

    research/data/oos/tau_adopt_j2_record.md (What changed)

  • Earlier wording: So on crowded gusty airspace the adopted setting is exactly the sum of its parts

    Record correction: Additivity holds for the pooled figure only. Within the airspace scenarios the charge also raised closeness (0.079 to 0.105), and reviewers said the split was overstated.

    research/data/oos/mr_bubble_only.md (per family); research/reviews/auditor-c167.md (F-MR-BUBBLE-ATTRIBUTION)

  • Earlier wording: the charge with the full bubble looks better on this suite, but it already failed on the fast crossing scenarios (cycles 137 and 143)

    Record correction: Cycle 137 tested the full bubble with a weaker charge (strength 3), and cycle 143 tested three-quarter and quarter bubbles. The strength-5 charge with the full bubble had not been flown on the crossing.

    research/data/oos/tau_gate2.md; research/data/oos/tau_joint.md

Numbers and source references
350
scenario runs per setting (70 × 5 offsets)research/data/oos/mr_bubble_only.md
0.053 / 0.066 / 0.052 / 0.065
pair-seconds at depth per craft-hour: neither / bubble only / charge only / bothresearch/data/oos/mr_bubble_only.md
0.0173 / 0.0199 / 0.0088 / 0.0099
collisions per craft-hour, same settingsresearch/data/oos/mr_bubble_only.md
18 to 3
eVTOL matrix collisions, old vs adopted, five offsetsresearch/data/oos/mat_seeds.md
5 to 1
multirotor matrix collisions (unreadable)research/data/oos/mat_seeds.md
16 m/s
paired cruise speed per airframe under the restated targetresearch/design/tau_stage_close.md §8.3
Original structured record ↗