v8.1.0
v8.1.0
We measured whether a leader can defend position by blocking in a touge-style race — start and finish in different places, several cars at once (Stage AX, tabletop × precise v2 engine, measured on v8.0.0. Note the tabletop default is the dynamics model, so this is not what you get by leaving the defaults alone). The answer is no — not one of the four leader tactics raised the win rate significantly above simply taking the fastest line, in any condition (0 of 66, full sweep of 10,368 races; still 0 when the tactic's wins are restricted to races with no contact and no close approach). What defends position is the road width, not the tactic: simply racing wins 85% on the touge courses as shipped, 85% at 4.5 car widths (flat), 100% at 3 car widths and 100% at 2 car widths — narrower than 3 car widths and it reaches 100%. Blocking the inside does not work because in a hairpin the reachable corridor narrows to a single point, so racing and inside-blocking go through the corner in the same place (the only difference is hugging one side on the straight, which leaves the other side open), and 23 of the 60 points in the inner band are unreachable for the blocker too. We also added 18 narrowed variant courses ([4.5/3/2-Car Width]) at the end of the course list, with the road width squeezed down to the ratio of a real mountain pass. The built-in sample programs struggle to finish on the narrow ones, and the course descriptions say so. The physics notes ("11. When Drift Does / Doesn't Hold") and the in-app design Q&A were updated too. Not one byte of the driving physics changed (record reproducibility is preserved).
日本語
峠のような「スタートとゴールが別・多車」のレースで、先頭がブロックすれば順位を守れるのかを測りました (Stage AX・卓上×精密v2エンジン v8.0.0 で測定。卓上の既定は動力学モデルなので、既定のまま走らせたときの話ではありません)。答えは守れないです — 先頭の戦術4種のどれも、素直に最速線を走る場合より1位率を有意に上げる条件は0/66でした (全掃引10,368レース。戦術側の勝ちを「接触も接近も無い」ものに限っても0)。順位を守るのは戦術ではなく道幅で、素直に走ったときの1位率は現行の峠85%→4.5台分85%(横ばい)→3台分100%→2台分100%と、3台分より狭くすると100%になります。内側を塞いでも効かないのは、ヘアピンでは回廊が一点に絞られて素直な線と内側占有がコーナーでは同じところを通るため (違いは直線での寄りだけで、寄れば反対側が空いて抜かれる) で、そもそも内側の帯60点のうち23点は塞ぐ側も到達できません。あわせて、道幅を実車の峠の比率まで詰めた派生コース18本 (〔道幅 4.5/3/2 台分〕) をコース一覧の末尾に追加しました。狭路版は出荷の既定サンプルでは完走しにくいので、その旨をコース説明に明記しています。物理解説の「11. ドリフトが成り立つ/成り立たない条件」と、アプリ内の設計Q&Aにも追記しました。走行物理は1バイトも変えていません (記録の再現性は維持)。
Verifying this release yourself
git clone https://github.com/RumiCar-group/RumiCar-Simulator.git
cd RumiCar-Simulator
node wf_run_all.mjs # the full assertion-gate suiteThe suite runs from a fresh clone with no dependencies beyond Node.js.
The real-browser checks under browser/ additionally need npm install and a display.