Compare named vehicles across the standard sims with opt-trade

_4_OptSim/StandardSens/configs/trade_study.yaml

@@ -0,0 +1,49 @@
+# A trade study: named vehicles compared across the standard sims.
+#
+#   make opt-trade
+#   make opt-trade STUDY=path/to/another_study.yaml
+#
+# See docs/doe-reverse-engineering.md. Copy this file for each study; results land
+# in _4_OptSim/results/trade/<name>.md and .csv.
+name: rear_roll_stiffness
+
+# Each candidate is the baseline vehicle with these variables changed. Any
+# variable declared in vehicle_architecture.yaml may be used, mass and CG
+# included, because every candidate is compiled rather than overridden. The
+# baseline is always included and needs no entry.
+#
+# `both` makes exactly the changes of the other two, so the report also says
+# whether those two changes stack or interact.
+candidates:
+  stiff_rear_bar:
+    rear.stabar.rate_n_m_per_rad: 961.495352
+  soft_front_spring:
+    front.actuation.spring_rate_n_per_m: 21015.2202
+  both:
+    rear.stabar.rate_n_m_per_rad: 961.495352
+    front.actuation.spring_rate_n_per_m: 21015.2202
+
+# What to compare, per standard: SteadyStateEval, RampSteerEval or TransientEval.
+# All three run against the one executable compiled for each candidate.
+#
+# `resolution` is the smallest change worth acting on. A smaller delta is still
+# reported, marked `~`. Leave it off and the delta is reported unjudged. There is
+# no score and no ranking: what a tenth of a degree of understeer is worth in
+# settling time is your call, not the tool's.
+metrics:
+  SteadyStateEval:
+    understeer_gradient_deg_per_g: {resolution: 0.02}
+    roll_gradient_deg_per_g: {resolution: 0.02}
+    peak_handwheel_torque_Nm: {resolution: 0.5}
+  TransientEval:
+    yaw_rise_time_s: {resolution: 0.005}
+    yaw_overshoot_pct: {resolution: 1.0}
+    roll_overshoot_pct: {resolution: 1.0}
+    settling_time_s: {resolution: 0.05}
+
+# Also render each standard's PDF report per vehicle. Worth having when the study
+# goes in front of a design review; costs roughly ten seconds per run.
+render_reports: true
+
+# CPUs available to simulations (the Docker container's count, not the host's).
+cpus: 12

_4_OptSim/StandardSens/pipeline/batch.py

@@ -1,4 +1,4 @@
-"""batch.py — Run the SteadyStateEval report wrapper for all variants.
+"""batch.py — Run each configured standard for all variants.
 
 For each variant_XXXX/build/<standard>/ that has a compiled executable:
   1. Skip if metrics.csv already exists and is valid (correct row count)
@@ -20,7 +20,7 @@
 import pandas as pd
 import yaml
 
-from StandardSens.pipeline.steady_state_eval_report import run_report
+from StandardSens.pipeline.standards import get_standard, run_standard
 
 # ---------------------------------------------------------------------------
 # Config
@@ -83,7 +83,10 @@ def run_variant(
     results_dir.mkdir(parents=True, exist_ok=True)
 
     try:
-        metrics_csv = run_report(
+        # Dispatch on the standard's name: running one standard's report against
+        # another's executable fails late and unhelpfully.
+        metrics_csv = run_standard(
+            get_standard(standard),
             variant_dir=variant_dir,
             build_dir=build_dir,
             exec_name=standard_cfg["model"],

_4_OptSim/StandardSens/pipeline/compiler.py

@@ -17,6 +17,7 @@
 
 from __future__ import annotations
 
+from collections.abc import Collection
 import platform
 import subprocess
 import sys
@@ -55,6 +56,7 @@
 # list would keep caches its neighbour had already declared stale.
 PIPELINE_TOOLING_INPUTS = (
     DEFAULT_REPORT_WRAPPER,
+    STANDARD_DIR / "pipeline/standards.py",
     DEFAULT_STEADY_STATE_SIM,
     DEFAULT_STEADY_STATE_CONFIG,
     DEFAULT_MODELICA_RUNNER,
@@ -269,16 +271,22 @@ def compile_all(
         template_path: Path = DEFAULT_MOS_TEMPLATE,
         doe_config_path: Path = DEFAULT_DOE_CONFIG,
         architecture_config_path: Path = DEFAULT_ARCHITECTURE_CONFIG,
+        only_standards: Collection[str] | None = None,
 ) -> dict[str, list[Path]]:
     """Compile all variants in population_dir for all standards in config.
 
+    `only_standards` narrows that to the named ones, for callers that run several
+    standards against one executable and must not pay for a build per standard.
+
     Skips variants that are already compiled and whose inputs haven't changed.
     Runs variants in parallel using ProcessPoolExecutor.
     Returns dict mapping standard -> list of successful exe paths.
     Failed variants are logged and skipped.
     """
     cfg = load_compiler_config(compiler_config_path)
     standards: dict[str, dict] = cfg["standards"]
+    if only_standards is not None:
+        standards = {name: standards[name] for name in only_standards}
     max_workers: int = cfg.get("max_workers", 2)
 
     # Resolve boblib_path relative to the config file