The mouse that rolled forever
The report was one sentence: “holding the mouse still at a certain heading, the jet seems to weave back and forth, never actually achieving the desired heading — prop planes aren't affected.” All three clauses turned out to be measurably, embarrassingly true.
The feeling, as numbers
The house method: before touching a constant, make the weave a number. A headless probe flew the real desktop chain — virtual chase-camera aim ray, the mouse instructor, the jet path hold, the auto-trim fold, the pointer-lock reticle bleed — and held a quarter-deflection of cursor to the right, the plainest turn a mouse can ask for. The Sparrowhawk settled into a rock-steady carve: bank 78°, never wandering outside a 67–89° band. Every jet in the era rolled full circle: bank sweeping −180° to +180°, forever, heading corkscrewing past the mark and back. Same hand, same reticle, same instructor — one input method flies a turn, the other flies a drill bit.
Then the live client made it lethal: a real headless-Chromium pilot joined the jets front, settled hands-off, dragged the mouse one firm sweep right and held it still. The Starfang banked through 97° while the drag was still in progress, kept rolling to 137° on its own, put the nose 40° below the horizon, and flew into the Brick Sea from 478 m in six and a half seconds. At altitude you live through that — it reads as the jet weaving drunkenly through the heading you asked for. Near the deck it reads as a crash you didn't order.
Why pursuit worked until it didn't
The mouse instructor has always flown pursuit: roll until the target sits in the lift plane, then pull it onto the nose. For a prop that's a fine plan, because a prop's pull can actually cash the check — at 100 m/s the wing turns the nose fast enough that the target lands in the lift plane and the roll command dies out at a steady bank. A jet's pull cannot. At 250 m/s the 6 g ceiling is a turn rate of about 12°/s, a modest reticle offset asks for more, so the target never centers — and the pursuit answers the only way it knows: keep rolling “toward” it. Past 90° of bank the pull stops turning the jet and starts slicing the nose into the ground, the aim ray rides the nose down, and the loop closes on a corkscrew. The faster the plane, the tighter the trap: the exact failure the touch stick shipped with in its first cut (“wound the wings past vertical, spiralled 550 m into the dirt”) — and the touch stick's cure is the cure here too.
The carve capture
For jets, the horizon game changes from pursuit to capture: heading error commands a bank angle, capped at the 80° a 6 g carve can actually use, and the roll rate-damps onto it with the same gains the touch instructor has already proven on these airframes at these speeds. Aim well above or below the horizon and the carve yields back to pure pursuit, because a loop or a split-S really is “roll it into the lift plane and haul.” Props fly the original pursuit bit-for-bit — the WW2 mouse feel players have tuned their hands to is its own spec.
And one more thing the capture's clean endgame exposed: the mouse assist's “coordinated” fine-aim rudder has been feeding rudder away from the target since mouse aim began — a sign slip (positive yaw is nose-left in this airframe basis, the touch instructor even documents it), buried the whole time under the pursuit's saturated roll on props. On the new capture it surfaced as a standing nine-degree miss: the rudder yawing off exactly what the shallow final bank turned in. Jets now get the rudder the right way around, fading as the bank steepens, with the same Dutch-roll damper the autopilot uses.
Before and after
Capture of a heading 40° off the nose, held frozen (the freelook steer-to case — the report's exact words): before, 25 to 33 bank reversals and heading swings past ±160°, never settling. After: one smooth bank in, one smooth rollout, converging monotonically to under half a degree — zero overshoot, altitude held the whole way.
The maximum-effort sustained mouse turn (target pinned 90° abeam, fourteen seconds): before, the pursuit “won” 21.6°/s on the Balalaika — by riding 93° of bank, nose below the horizon, trading 626 m of altitude for it. A spiral with good press. After: 15.8°/s at 80° of bank, gaining 144 m — and identical to the touch stick's maximum to the decimal, on every jet. The two input methods now agree on what the instructor's ceiling is; nobody out-turns the other by borrowing gravity.
And the drag that killed the Starfang: same gesture, same hold. It banks to 54°, carves 20° of heading as the reticle bleeds home, and rolls out wings-level on the new course — altitude pinned at 500 m, bank reversals zero, tail heading wander 0.9°. A hard mouse pull still climbs at 25 m/s, a diagonal aim still banks and climbs, the porpoise-meter is untouched, and the WW2 fleet's rows in the feel probe are byte-identical.
Confession
The weave shipped with the jets era itself, and two full balance passes flew straight through it — because every probe we aimed at the jets measured held maximum-effort inputs: rim slams, full pulls, targets pinned abeam. Nobody ever measured the plainest thing a mouse does — point at a heading and wait. The pursuit's corkscrew even flattered the sustained-turn numbers we bragged about in the last post; 21.6°/s was partly gravity's money. The capture test — frozen heading, hands still, does it settle — now lives in the probes, and the two-year-old backwards rudder it flushed out is the reminder of why: a bug can hide indefinitely under a louder one.