close guardrail gaps and fix the scoring pipeline top to bottom

Constraint resolver: aggregate mass/footprint across a combo instead of
pairwise-only checks, treat medium/atmosphere as agreement not supply/demand,
reduce multi-provider checks by best/sum instead of AND-ing every provider,
fail closed on unrecognized mutex values, add a propulsion-viability
(thrust-to-weight) rule. Seed data updated to match (nuclear/solar-sail
footprint floors, water-medium exclusions, explicit ground/gravity providers).

Domain metric units were stored globally per metric name instead of
per-domain, silently corrupting cost_efficiency for every domain but the
first one seeded — fixed with a schema migration.

Stub estimator's cost_efficiency/safety/availability/reliability were a
backwards formula and flat constants; replaced with heuristics grounded in
each entity's thrust_profile/energy_form/infrastructure.

LLM estimate_physics() now receives each metric's unit and expected range
instead of a bare name, fixing wildly miscalibrated estimates traced back to
the prompt's own hardcoded example anchoring the model to the wrong order of
magnitude. Sharpened the safety-estimation and plausibility-review prompts.
Deduped provider parsing logic into llm/parsing.py.

Web pipeline form can now pick an LLM provider per run instead of only via
server env var.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

src/physcom_web/routes/pipeline.py

@@ -21,6 +21,9 @@ def _run_pipeline_in_background(
     passes: list[int],
     threshold: float,
     run_id: int,
+    llm_provider: str | None = None,
+    llm_model: str | None = None,
+    llm_host: str | None = None,
 ) -> None:
     """Run the pipeline in a background thread with its own DB connection."""
     from physcom.db.schema import init_db
@@ -45,7 +48,8 @@ def _run_pipeline_in_background(
         from physcom.llm.registry import build_llm_provider
         resolver = ConstraintResolver()
         scorer = Scorer(domain)
-        pipeline = Pipeline(repo, resolver, scorer, llm=build_llm_provider())
+        llm = build_llm_provider(provider=llm_provider, model=llm_model, host=llm_host)
+        pipeline = Pipeline(repo, resolver, scorer, llm=llm)
 
         pipeline.run(
             domain, dim_list,
@@ -108,11 +112,17 @@ def pipeline_run():
         flash("Select at least one dimension.", "error")
         return redirect(url_for("pipeline.pipeline_form"))
 
+    llm_provider = request.form.get("llm_provider", "").strip() or None
+    llm_model = request.form.get("llm_model", "").strip() or None
+    llm_host = request.form.get("llm_host", "").strip() or None
+
     # Create pipeline_run record
     config = {
         "passes": passes,
         "threshold": threshold,
         "dimensions": dim_list,
+        "llm_provider": llm_provider,
+        "llm_model": llm_model,
     }
     run_id = repo.create_pipeline_run(domain.id, config)
 
@@ -123,7 +133,8 @@ def pipeline_run():
     # Start background thread
     t = threading.Thread(
         target=_run_pipeline_in_background,
-        args=(db_path, domain_name, dim_list, passes, threshold, run_id),
+        args=(db_path, domain_name, dim_list, passes, threshold, run_id,
+              llm_provider, llm_model, llm_host),
         daemon=True,
     )
     t.start()

src/physcom_web/templates/pipeline/run.html

@@ -49,6 +49,29 @@
       </div>
     </fieldset>
 
+    <fieldset>
+      <legend>LLM Provider</legend>
+      <p class="form-hint">Used for Pass 2 estimation and Pass 4 review. Leave on "server default" to use whatever LLM_PROVIDER is configured in the server environment (or the physics stub if none).</p>
+      <div class="form-group">
+        <select name="llm_provider" id="llm_provider">
+          <option value="">— server default —</option>
+          <option value="stub">Stub (fast, no LLM)</option>
+          <option value="ollama">Ollama (local)</option>
+          <option value="gemini">Gemini (cloud, requires server-side GEMINI_API_KEY)</option>
+        </select>
+      </div>
+      <div class="form-group">
+        <label for="llm_model">Model</label>
+        <p class="form-hint">Leave blank to use the provider's default model.</p>
+        <input type="text" name="llm_model" id="llm_model" placeholder="e.g. qwen2.5:7b or gemini-2.0-flash">
+      </div>
+      <div class="form-group">
+        <label for="llm_host">Ollama host</label>
+        <p class="form-hint">Only used when Ollama is selected. Leave blank for http://localhost:11434.</p>
+        <input type="text" name="llm_host" id="llm_host" placeholder="http://localhost:11434">
+      </div>
+    </fieldset>
+
     <div class="form-group">
       <label for="threshold">Score Threshold</label>
       <p class="form-hint">Minimum composite score (0–1) for a combination to pass scoring. Lower values keep more results; higher values are more selective.</p>