[fix.physics.rho.D8-] the remedy the cost gate prescribes erases the answer, so I withdrew it

I shipped a basis selector two commits ago because the refusal said "use a smaller
basis (sto-3g is ~31 s here)". Then I measured what that basis does to the question
this panel exists to answer, on molecules whose answer is not in doubt:

                    def2-SVP        6-31G        STO-3G
  bromobenzene   44.02  σ-hole 1   refused     18.85  σ-hole 0
  chlorobenzene  31.60  σ-hole 1   38.42 σ1    13.94  σ-hole 0
  fluorobenzene  18.01  σ-hole 0   25.81 σ0    12.52  σ-hole 0

STO-3G finds ZERO σ-holes on bromobenzene and chlorobenzene. It does not make the
number less precise — it flips the classification. Every halogen comes back looking
like fluorine, which is the one element that genuinely has no σ-hole and the whole
reason this panel and the halogen audit exist. A chemist picking the fast option to
get an answer sooner gets the OPPOSITE answer, rendered identically.

So STO-3G is withdrawn from this panel, with the measurement printed next to the
control rather than in a commit nobody reads. The list stops at 6-31G.

What replaces it is the remedy that does not cost the conclusion: a budget selector
(90 / 300 / 900 s) that is actually SENT. It never was — post()'s third argument only
set the client-side abort, so the server always used its own default and the refusal's
second suggestion, "raise max_seconds deliberately", could not be taken from the
interface at all.

The def2-SVP controls double as a positive control for the endpoint and it passes
cleanly: Br 44.0 > Cl 31.6 > F 18.0 with σ-hole counts 1, 1 and 0. Fluorobenzene's
positive maximum sits on the ring hydrogens, not on the C–F axis, and the classifier
says so — the exact distinction PLIP's geometry-only test cannot make.

NOT RESOLVED, and newly surfaced: def2-SVP on lapatinib with a 900 s budget does not
time out, it FAILS — "RYS_build_jk kernel for (dp|dd) failed" — a GPU kernel error on
the d-function combinations that basis brings in for S and Cl. A 6-31G retry returned
"cudaErrorDevicesUnavailable: device busy", which is another session warming caches on
the same GPU, so that one says nothing about the molecule. Both need isolating on a
quiet GPU and on CPU before anyone concludes anything about lapatinib; I am not
claiming the budget fix makes this molecule work.

Author: ivanicu

src/app.frontend.facets.molstar-rdkit.editable/facets/ligand-physics/index.ts

@@ -328,16 +328,35 @@ function selectedBasis(): string {
     return el?.value || 'def2-svp';
 }
 
+/**
+ * The remedy that actually works on a large ligand.
+ *
+ * The cost gate's first suggestion is a smaller basis, and for THIS question that
+ * suggestion is wrong: measured on this backend, STO-3G reports zero σ-holes on both
+ * bromobenzene and chlorobenzene, which carry one at def2-SVP (44.0 and 31.6 kcal/mol).
+ * It does not make the number less precise, it flips the classification the panel exists
+ * to produce. So the basis list stops at 6-31G and the budget is what the user raises —
+ * and max_seconds is now actually SENT, which it never was: post()'s third argument only
+ * ever set the client-side abort, so the server always used its own default and the
+ * refusal's "raise max_seconds deliberately" was unreachable from the interface.
+ */
+function selectedBudget(): number {
+    const el = document.getElementById('phys-budget') as HTMLSelectElement | null;
+    const v = parseInt(el?.value || '', 10);
+    return Number.isFinite(v) && v > 0 ? v : BUDGET_SECONDS;
+}
+
 async function runSurface() {
     if (!molfile || busy) return;
     setBusy(true);
     const basis = selectedBasis();
+    const budget = selectedBudget();
     setStatus('phys-surface-status',
         `Solving surface electrostatics on ${ligandLabel ?? 'ligand'} in ${basis}… `
-        + `(gives up at ${BUDGET_SECONDS} s · Cancel to stop now)`, 'busy');
+        + `(gives up at ${budget} s · Cancel to stop now)`, 'busy');
     const generation = ligandGeneration;
     try {
-        const out = await post('/surface/mep', { molfile, basis }, BUDGET_SECONDS);
+        const out = await post('/surface/mep', { molfile, basis, max_seconds: budget }, budget);
         if (!isCurrentLigandGeneration(generation)) {
             setStatus('phys-surface-status',
                 'Ligand changed while solving — result discarded rather than '

src/app.frontend.facets.molstar-rdkit.editable/index.html

@@ -793,6 +793,7 @@
         .bond-atlas-block { margin-top: var(--sp-4, 16px); padding-top: var(--sp-3, 12px); border-top: 1px solid var(--border); }
         .phys-basis-label { display: inline-flex; align-items: center; gap: var(--sp-1, 4px); color: var(--text-3); font-size: var(--fs-2xs, 9px); letter-spacing: var(--track-wide, .08em); text-transform: uppercase; }
         .phys-basis-label select { text-transform: none; letter-spacing: normal; }
+        .phys-why-warn { border-left: 2px solid var(--warn); padding-left: var(--sp-2, 8px); }
         .halogen-block { margin-top: var(--sp-3, 12px); padding-top: var(--sp-3, 12px); border-top: 1px dashed var(--border); }
         .halogen-note-inline { color: var(--text-3); font-size: var(--fs-2xs, 9px); letter-spacing: var(--track-wide, .08em); text-transform: uppercase; }
         .halogen-row { border-left: 2px solid var(--border-strong); padding: var(--sp-2, 8px) 0 var(--sp-2, 8px) var(--sp-2, 8px); margin-bottom: var(--sp-2, 8px); }
@@ -1476,8 +1477,14 @@ <h3>σ-hole · surface electrostatics</h3>
                                 <label class="phys-basis-label">Basis
                                     <select id="phys-basis">
                                         <option value="def2-svp" selected>def2-SVP · has iodine</option>
-                                        <option value="6-31g">6-31G · faster</option>
-                                        <option value="sto-3g">STO-3G · fastest</option>
+                                        <option value="6-31g">6-31G · faster, no Br/I</option>
+                                    </select>
+                                </label>
+                                <label class="phys-basis-label">Budget
+                                    <select id="phys-budget">
+                                        <option value="90" selected>90 s</option>
+                                        <option value="300">300 s</option>
+                                        <option value="900">900 s · large ligand</option>
                                     </select>
                                 </label>
                                 <button type="button" class="phys-btn" id="phys-run-surface">Compute</button>
@@ -1487,6 +1494,12 @@ <h3>σ-hole · surface electrostatics</h3>
                         <p class="phys-why">A halogen with no σ-hole cannot make a halogen
                            bond however good the geometry looks. The angle is the evidence:
                            below 150° from the C–X axis a positive cap is not a σ-hole.</p>
+                        <p class="phys-why phys-why-warn">STO-3G is <b>not offered here</b>, although it is
+                           three times faster and the cost gate recommends it. Measured on this backend:
+                           it finds <b>zero</b> σ-holes on bromobenzene and chlorobenzene, both of which
+                           have one at def2-SVP (44.0 and 31.6 kcal/mol). A minimal basis does not shrink
+                           the answer, it erases it — every halogen comes back looking like fluorine. For
+                           a large ligand raise the budget instead; that costs time, not the conclusion.</p>
                         <div id="phys-surface-status" class="field-status" data-tone="idle">Load a ligand first.</div>
                         <div id="phys-surface-body"></div>