Evidence tour

Five stops · about eight minutes

For reviewers who will not browse blindly. Each stop links to a live page. Read it as a vendor benchmark we ran on ourselves — internally generated, not independently verified — and finish at stop 4, where the measurement rules are defined.

1. 01

   Where SolvSRK loses: Arenstorf & Lotka–Volterra

   On the Arenstorf orbit SolvSRK drops to 0% survival under noise while SciPy holds 100%; on Lotka–Volterra it survives just 30% vs SciPy's 100%. These losses lead the dead-zones page — not the wins.

   Key point: We publish competitor wins

   Open stop →
2. 02

   Total failure: the E5 reaction

   E5's stiffness ratio near 10^11 defeats every arm — SolvSRK and all 13 baselines fail. We show the cells where nobody finishes.

   Key point: We publish dead zones

   Open stop →
3. 03

   Where it wins: Robertson under noise

   Clean Robertson is parity for implicit methods. The real edge is noisy survival — every SciPy arm hits 0% at σ=0.1 while SolvSRK holds 100%.

   Key point: The headline is scoped to noise

   Open stop →
4. 04

   Methodology § accuracy

   Survival and SCD are separate questions, and the SCD reference is an independent high-fidelity solver — SolvSRK is never its own reference.

   Key point: Accuracy is defined, not hand-waved

   Open stop →
5. 05

   Four objectives, one free tool

   Every problem is ranked on balanced, survival, precision, and efficiency — the same objective set SolvScout gives away free. When another arm wins an objective, you see it.

   Key point: We show the losing objective too

   Open stop →