Journey 06 — Limits and non-goals

What SolvScout deliberately does not do, where SolvSRK fits, and how to interpret low confidence and dead zones.

Prerequisites: Journeys 01–05 complete. Scenario: What SolvScout / SolvTune is NOT

Key APIs

solvscout.recommend()
solvsrk.run_profile()
solvsrk.run()

Concepts to internalize

Advisor, not integrator — compare() rows are corpus medians, not live nfev from your machine.
Validate with a real run — low confidence or dead_zone means extrapolation; prove finalists with SolvSRK.
IVP ODE scope — not PDE analysis, not symbolic math, not a substitute for domain expertise.
Garbage in → garbage profile — RHS quality matters.

Important code

The guided tour script wraps this pattern. Copy the production shape into your pipeline:

# SolvScout advises — it never steps your ODE:
#   classify() / recommend() / compare() / recommend_product()
#
# Execution layer:
token = scout.recommend_product(profile).profile
solvsrk.run_profile(token, ndim, t0, t_end, y0, rhs_fn=rhs)
 
# Or map recommend().best.config into solvsrk.run(cfg, ...)

Act-by-act walkthrough

Act 1 — SolvScout advises; it does not integrate

classify() and recommend() analyse and rank. They never step your ODE. compare() reads benchmark medians from the corpus - it does not run a solver either. If a table row says nfev=..., that is a looked-up median, not this call's cost.

Act 2 — Where execution lives

To actually integrate, use SolvSRK: pass recommend_product()'s profile token to solvsrk.run_profile(), or map recommend().best.config into solvsrk.run() with your RHS. SolvScout is the 'which solver + which knobs' front end; SolvSRK runs it.

Pattern: token = recommend_product(profile).profile -> solvsrk.run_profile(...)

Act 3 — Recommendations are corpus-conditioned

Confidence reflects how close your problem is to benchmarked ones. A 'low' confidence or a dead_zone flag means SolvScout is extrapolating - treat it as a hint, and validate with a real run.

Field	Value	Notes
`problem`	rossler
`best_arm`	solvsrk
`confidence`	high
`n_dead_zones`	0

Act 4 — Not a general-purpose classifier

SolvScout characterizes initial-value ODE systems for solver routing. It is not a PDE analyzer, not a symbolic solver, and not a substitute for domain knowledge about your model. Garbage RHS in -> garbage profile out.

Guided tour output

Verified console capture from 06_limits.py:

What SolvScout / SolvTune is NOT

-- Act 1 - SolvScout advises; it does not integrate --
  -> classify() and recommend() analyse and rank. They never step your ODE. compare() reads benchmark medians from the corpus - it does not run a solver either. If a table row says nfev=..., that is a looked-up median, not this call's cost.

-- Act 2 - Where execution lives --
  -> To actually integrate, use SolvSRK: pass recommend_product()'s profile token to solvsrk.run_profile(), or map recommend().best.config into solvsrk.run() with your RHS. SolvScout is the 'which solver + which knobs' front end; SolvSRK runs it.
  log: Pattern: token = recommend_product(profile).profile -> solvsrk.run_profile(...)

-- Act 3 - Recommendations are corpus-conditioned --
  -> Confidence reflects how close your problem is to benchmarked ones. A 'low' confidence or a dead_zone flag means SolvScout is extrapolating - treat it as a hint, and validate with a real run.
  problem: rossler
  best_arm: solvsrk
  confidence: high
  n_dead_zones: 0

-- Act 4 - Not a general-purpose classifier --
  -> SolvScout characterizes initial-value ODE systems for solver routing. It is not a PDE analyzer, not a symbolic solver, and not a substitute for domain knowledge about your model. Garbage RHS in -> garbage profile out.

-- Run complete --
  summary: You have completed the SolvScout / SolvTune journeys.

Run it

cd python
# Guided tour — narrated scenario walkthrough (recommended first run):
python examples/journeys/06_limits.py
 
# Production one-liner — same API call you ship:
python -c "import solvscout as s; print('compare() is corpus lookup, not a live run')"

What comes next

Continue to SolvSRK examples for integration, diagnostics, and UQ.

Navigation

← Journey 05 — Product profile token

05 Product profile token Integration