Journey 04 — Configuration and SolvScout hand-off

import solvscout as scout
import solvsrk
 
# Production path:
profile = scout.classify(rhs, y0, t_span).profile
token = scout.recommend_product(profile).profile
result = solvsrk.run_profile(token, 3, t0, t_end, y0, rhs_fn=rhs, jac_fn=jac)
 
# When you need wrappers before run():
cfg, mode = solvsrk.profile_to_config(token)
cfg.handoff_conservation_sum = 1.0
solvsrk.run(cfg, 3, t0, t_end, y0, rhs, jac_fn=jac)

SolvScout tunes; SolvSRK integrates

-- Act 1 - Tuning is the product --
  -> config_defaults() is a blank API baseline — safe for learning the struct, not what you ship on stiff chemistry. Resonix tuning lives in SolvScout: classify your rhs, call recommend_product(), and integrate with run_profile(token, ...) or profile_to_config(token) when you need to layer problem-specific wrappers.
  untuned: solvsrk.config_defaults() — not production-ready on stiff problems
  production: solvscout.classify -> recommend_product -> solvsrk.run_profile(token, ...)

-- Act 2 - Untuned vs tuned on Robertson --
  -> Same Robertson problem, same analytic Jacobian — only the knobs change. Untuned is config_defaults() with no overrides. Tuned is a SolvScout profile token (or the bundled tuned preset when Scout is offline). Compare finished_ok, wall_s, nfev, and y_final.
  _section: Untuned (config_defaults)
  finished_ok: False
  failure_mode: blowup
  wall_s: 8.941e-05
  nfev: 9
  y_final: [-992.191, -2.95518e+06, 2.95618e+06]
  _section: Tuned (SolvScout tuned (solvsrk, high))
  finished_ok: True
  failure_mode: ok
  wall_s: 0.002542
  nfev: 795
  y_final: [0.715827, 9.18553e-06, 0.284164]
  -> Untuned blew up (blowup) at t=0.02; tuned knobs reached t_end. That gap is why you do not ship config_defaults() on stiff plants.

-- Act 3 - Production API — run_profile --
  -> run_profile decodes the opaque token and integrates in one call. Use profile_to_config when you need to add physical wrappers before run().
  solvscout_solver: solvsrk
  confidence: high
  token_prefix: rx-UlgBAQAAAgkAcm9iZ...
  _section: run_profile(token, ...)
  finished_ok: True
  failure_mode: ok
  wall_s: 0.002516
  nfev: 795
  y_final: [0.715827, 9.18553e-06, 0.284164]

-- Act 4 - Fine-tune on top of tuned knobs --
  -> Once Scout gives you a good baseline, tighten rtol/atol only if you need more accuracy — start from robertson_config() or a decoded token, not from config_defaults().
  _section: Tuned baseline, loose rtol/atol
  finished_ok: True
  failure_mode: ok
  wall_s: 0.002517
  nfev: 871
  y_final: [0.715834, 9.18581e-06, 0.284157]
  _section: Tuned baseline, tight rtol/atol
  finished_ok: True
  failure_mode: ok
  wall_s: 0.003887
  nfev: 1244
  y_final: [0.715827, 9.18553e-06, 0.284164]

-- Act 5 - Step-size cap --
  -> cvode_max_step bounds the largest internal step CVODE may take — useful when a forcing term or event must not be skipped over. 0.0 means 'no explicit cap' (the solver chooses). Applied on the tuned Robertson preset.
  cvode_max_step: 0.05
  cvode_steps: 828

-- Run complete --
  next_journey: 05_diagnostics.py

cd python
# Guided tour — narrated scenario walkthrough (recommended first run):
python examples/journeys/04_configuration.py
 
# Production one-liner — same API call you ship:
python examples/journeys/04_configuration.py