MDH: dataset-driven type inference for aggregation variables

Whole-quoted "{var}" placeholders now default to the matched field's data
type (detected via $type and cached) instead of the input-text shape — so a
numeric-looking value typed against a string field no longer silently fails
to match. The resolved type is folded into an "Auto (Type)" selector per
variable and can be overridden (String/Number/Boolean/Null); Boolean/Null are
offered only when the value makes them meaningful, and the "Auto (X)" label
always reflects the true inferred type even under a manual override.

Substitution stays textual-then-JSON5.parse (byte-identical when no type
resolves); bulk/download and persistence (last pipeline, per-collection
state, query history) all carry the chosen types. Field types reuse the
Stats $type facet machinery. New modules: placeholderSyntax, placeholderFields,
fieldTypes. Full unit coverage; suite green.

Author: mrtnzlml

src/console/console.css

@@ -419,15 +419,37 @@ body {
   font-size: 11px;
   color: var(--accent);
   flex-shrink: 0;
-  min-width: 80px;
 }
 
 .placeholder-input {
   flex: 1;
-  padding: 4px 8px;
+  min-width: 0;
+  height: 24px;
+  padding: 2px 8px;
   font-size: 12px;
 }
 
+.placeholder-type-select {
+  flex-shrink: 0;
+  height: 24px;
+  padding: 0 6px;
+  font-family: var(--font-mono);
+  font-size: 11px;
+  border: 1px solid var(--border);
+  border-radius: var(--radius);
+  background: var(--bg-input);
+  color: var(--text-primary);
+  cursor: pointer;
+}
+
+.placeholder-warn {
+  flex-shrink: 0;
+  font-size: 13px;
+  line-height: 1;
+  color: var(--warning-fg);
+  cursor: help;
+}
+
 /* ── Pipeline debug ───────────────────────────── */
 
 .pipeline-debug {

src/mdh/components/DataPanel.jsx

@@ -48,7 +48,7 @@ export default function DataPanel() {
   // that drives the Variables inputs and the Pipeline Debug — see
   // useEditorSnapshot. recomputeEditorState() is called on every editor edit and
   // on placeholder changes.
-  const [editorState, recomputeEditorState] = useEditorSnapshot(editorRef, pipeline.computeEditorState);
+  const [editorState, recomputeEditorState] = useEditorSnapshot(editorRef, pipeline.computeEditorStateWithTypes);
 
   const collection = selectedCollection.value;
 
@@ -95,9 +95,10 @@ export default function DataPanel() {
   async function runQuery() {
     if (!collection || !editorRef.current) return;
     const rawText = editorRef.current.getValue();
-    const result = await query.runQuery(collection, rawText, pipeline.substitutePlaceholders);
+    await pipeline.ensureFieldTypes(collection, pipeline.referencedFields(rawText));
+    const result = await query.runQuery(collection, rawText, pipeline.substituteWithTypes);
     if (result) {
-      addToHistory(collection, rawText, { ...pipeline.placeholderValues.value });
+      addToHistory(collection, rawText, { ...pipeline.placeholderValues.value }, { ...pipeline.placeholderTypes.value });
     }
   }
 
@@ -119,6 +120,7 @@ export default function DataPanel() {
     if (external && external.collection === collection) {
       pendingPipelineLoad.value = null;
       if (external.variables) pipeline.placeholderValues.value = { ...external.variables };
+      if (external.placeholderTypes) pipeline.placeholderTypes.value = { ...external.placeholderTypes };
       setTimeout(() => {
         if (!editorRef.current) return;
         pipeline.suppressSync.value = true;
@@ -133,6 +135,7 @@ export default function DataPanel() {
     if (pending) {
       pendingLoadRef.current = null;
       if (pending.variables) pipeline.placeholderValues.value = { ...pending.variables };
+      if (pending.placeholderTypes) pipeline.placeholderTypes.value = { ...pending.placeholderTypes };
       setTimeout(() => {
         if (!editorRef.current) return;
         pipeline.suppressSync.value = true;
@@ -148,6 +151,7 @@ export default function DataPanel() {
     if (saved) {
       skip.value = saved.skip || 0;
       if (saved.variables) pipeline.placeholderValues.value = { ...saved.variables };
+      if (saved.placeholderTypes) pipeline.placeholderTypes.value = { ...saved.placeholderTypes };
       setTimeout(() => {
         if (!editorRef.current) return;
         pipeline.suppressSync.value = true;
@@ -172,11 +176,18 @@ export default function DataPanel() {
     return () => { saveStateForCleanup(collection); };
   }, [collection]);
 
+  useEffect(() => {
+    if (!collection || !editorRef.current) return;
+    pipeline.ensureFieldTypes(collection, pipeline.referencedFields(editorState.text))
+      .then((changed) => { if (changed) recomputeEditorState(); });
+  }, [editorState.text, collection]);
+
   function saveStateForCleanup(col) {
     if (!editorRef.current) return;
     savePipelineState(col, {
       pipelineText: editorRef.current.getValue(),
       variables: { ...pipeline.placeholderValues.value },
+      placeholderTypes: { ...pipeline.placeholderTypes.value },
       skip: skip.value,
     });
   }
@@ -197,7 +208,7 @@ export default function DataPanel() {
     // Falls back to {} when the pipeline has no $match or is unparseable.
     if (!editorRef.current) return {};
     try {
-      const text = pipeline.substitutePlaceholders(editorRef.current.getValue());
+      const text = pipeline.substituteWithTypes(editorRef.current.getValue());
       const parsed = JSON5.parse(text);
       if (Array.isArray(parsed)) {
         const match = parsed.find((s) => s && typeof s === 'object' && s.$match);
@@ -318,7 +329,7 @@ export default function DataPanel() {
     clearTimeout(persistTimerRef.current);
     persistTimerRef.current = setTimeout(() => {
       if (!editorRef.current) return;
-      saveLastPipeline(editorRef.current.getValue(), pipeline.placeholderValues.value);
+      saveLastPipeline(editorRef.current.getValue(), pipeline.placeholderValues.value, pipeline.placeholderTypes.value);
     }, 400);
   }
 
@@ -338,16 +349,17 @@ export default function DataPanel() {
     }
   }
 
-  function handleLoadPipeline(pipelineText, col, variables) {
+  function handleLoadPipeline(pipelineText, col, variables, placeholderTypes) {
     if (col && col !== collection) {
       // Defer to the [collection] effect — it will apply the pipeline and variables
       // after reset() instead of racing the default path.
-      pendingLoadRef.current = { pipelineText, variables };
+      pendingLoadRef.current = { pipelineText, variables, placeholderTypes };
       selectedCollection.value = col;
       return;
     }
     if (selectionMode.value) selectionPipelineDirty.value = true;
     if (variables) pipeline.placeholderValues.value = { ...variables };
+    if (placeholderTypes) pipeline.placeholderTypes.value = { ...placeholderTypes };
     if (editorRef.current) {
       pipeline.suppressSync.value = true;
       editorRef.current.setValue(pipelineText);
@@ -427,6 +439,14 @@ export default function DataPanel() {
     handleSetPlaceholder._timer = setTimeout(runQuery, 400);
   }
 
+  function handleSetPlaceholderType(name, type) {
+    pipeline.setPlaceholderType(name, type);
+    persistLastPipeline();
+    recomputeEditorState();
+    clearTimeout(handleSetPlaceholder._timer);
+    handleSetPlaceholder._timer = setTimeout(runQuery, 400);
+  }
+
   async function downloadAll() {
     const tc = pagination.totalCount.value;
     if (tc !== null && tc > 10_000) {
@@ -455,7 +475,7 @@ export default function DataPanel() {
 
     let pipelineStages;
     try {
-      const text = pipeline.substitutePlaceholders(editorRef.current.getValue());
+      const text = pipeline.substituteWithTypes(editorRef.current.getValue());
       const parsed = JSON5.parse(text);
       if (!Array.isArray(parsed)) throw new Error('pipeline must be a JSON array');
       pipelineStages = stripPaginationStages(parsed);
@@ -533,7 +553,7 @@ export default function DataPanel() {
     if (!editorRef.current) return;
     let pipelineStages;
     try {
-      const text = pipeline.substitutePlaceholders(editorRef.current.getValue());
+      const text = pipeline.substituteWithTypes(editorRef.current.getValue());
       const parsed = JSON5.parse(text);
       if (!Array.isArray(parsed)) throw new Error('pipeline must be a JSON array');
       pipelineStages = stripPaginationStages(parsed);
@@ -614,7 +634,7 @@ export default function DataPanel() {
     if (!editorRef.current) return;
     let pipelineStages;
     try {
-      const text = pipeline.substitutePlaceholders(editorRef.current.getValue());
+      const text = pipeline.substituteWithTypes(editorRef.current.getValue());
       const parsed = JSON5.parse(text);
       if (!Array.isArray(parsed)) throw new Error('pipeline must be a JSON array');
       pipelineStages = stripPaginationStages(parsed);
@@ -706,7 +726,7 @@ export default function DataPanel() {
     if (!editorRef.current) return;
     let pipelineStages;
     try {
-      const text = pipeline.substitutePlaceholders(editorRef.current.getValue());
+      const text = pipeline.substituteWithTypes(editorRef.current.getValue());
       const parsed = JSON5.parse(text);
       if (!Array.isArray(parsed)) throw new Error('pipeline must be a JSON array');
       pipelineStages = stripPaginationStages(parsed);
@@ -847,8 +867,11 @@ export default function DataPanel() {
         <PlaceholderInputs
           names={placeholderNames}
           values={pipeline.placeholderValues.value}
+          types={pipeline.placeholderTypes.value}
           onSetValue={handleSetPlaceholder}
+          onSetType={handleSetPlaceholderType}
           onRunQuery={runQuery}
+          resolvedTypeFor={(name) => pipeline.resolvedTypeForName(name, editorState.fieldMap || {}, editorState.parsed != null)}
         />
         <PipelineDebug pipeline={editorState.parsed} />
       </div>

src/mdh/components/PipelineEditor.jsx

@@ -81,9 +81,9 @@ export default function PipelineEditor({ editorRef, initialValue, onChange, onVa
     updateSaveBtn();
   }
 
-  function loadFromPanel(pipeline, collection, variables) {
+  function loadFromPanel(pipeline, collection, variables, placeholderTypes) {
     setLibraryOpen(false);
-    onLoadPipeline(pipeline, collection, variables);
+    onLoadPipeline(pipeline, collection, variables, placeholderTypes);
   }
 
   return (

src/mdh/components/PlaceholderInputs.jsx

@@ -1,6 +1,42 @@
 import { h } from 'preact';
 import { useState } from 'preact/hooks';
 import { domain, token } from '../store.js';
+import { isJson5NumberLiteral } from '../hooks/usePipeline.js';
+
+// True when the typed value can be matched as the given type (drives the
+// "won't match" hint). string / null / auto / undefined accept anything.
+export function isCompatibleWithType(val, type) {
+  if (type === 'number') return isJson5NumberLiteral(val);
+  if (type === 'boolean') return val === 'true' || val === 'false';
+  return true;
+}
+
+const CAP = { string: 'String', number: 'Number', boolean: 'Boolean', null: 'Null' };
+
+// Static tooltip describing what the control is — not a repeat of the resolved
+// type, which is already shown in the "Auto (X)" option text.
+const TYPE_SELECT_TITLE = 'Data type for this variable in the query — Auto infers it from the matched field';
+
+// What the value-based (Auto, no dataset type) path coerces a value to — mirrors
+// renderWholeToken's default branch order. Used to label "Auto (X)".
+export function valueBasedType(val) {
+  if (val === 'true' || val === 'false') return 'boolean';
+  if (val === 'null') return 'null';
+  if (isJson5NumberLiteral(val)) return 'number';
+  return 'string';
+}
+
+// Type options to offer for a variable. Auto/String/Number are always available;
+// Boolean only when the value is true/false; Null only when the value is empty or
+// 'null' (matching JSON null is pointless otherwise). The current override is
+// always included so a previously-saved Boolean/Null choice is never dropped.
+export function typeOptionsFor(value, override) {
+  const v = value || '';
+  const opts = ['auto', 'string', 'number'];
+  if (v === 'true' || v === 'false' || override === 'boolean') opts.push('boolean');
+  if (v === '' || v === 'null' || override === 'null') opts.push('null');
+  return opts;
+}
 
 export function parseAnnotationId(input) {
   if (/^\d+$/.test(input)) return input;
@@ -31,7 +67,7 @@ function extractDatapoints(nodes, fields) {
   }
 }
 
-export default function PlaceholderInputs({ names, values, onSetValue, onRunQuery }) {
+export default function PlaceholderInputs({ names, values, types, onSetValue, onSetType, onRunQuery, resolvedTypeFor }) {
   const [annotRow, setAnnotRow] = useState(false);
   const [annotStatus, setAnnotStatus] = useState('');
 
@@ -72,19 +108,38 @@ export default function PlaceholderInputs({ names, values, onSetValue, onRunQuer
           <span class="placeholder-annotation-status">{annotStatus}</span>
         </div>
       )}
-      {names.map((name) => (
-        <div class="placeholder-row" key={name}>
-          <span class="placeholder-name">{`{${name}}`}</span>
-          <input
-            class="input placeholder-input"
-            value={values[name] || ''}
-            onInput={(e) => {
-              onSetValue(name, e.target.value);
-            }}
-            onKeyDown={(e) => { if (e.key === 'Enter') onRunQuery(); }}
-          />
-        </div>
-      ))}
+      {names.map((name) => {
+        const rt = resolvedTypeFor ? resolvedTypeFor(name) : { type: undefined, autoType: undefined };
+        const value = values[name] || '';
+        const override = (types && types[name]) || '';
+        const autoLabelType = rt.autoType || valueBasedType(value); // what Auto yields, ignoring override
+        const effective = override || rt.type; // override-first effective type, for the compat check
+        const incompatible = value !== '' && !isCompatibleWithType(value, effective);
+        return (
+          <div class="placeholder-row" key={name}>
+            <span class="placeholder-name">{`{${name}}`}</span>
+            <input
+              class="input placeholder-input"
+              value={value}
+              onInput={(e) => { onSetValue(name, e.target.value); }}
+              onKeyDown={(e) => { if (e.key === 'Enter') onRunQuery(); }}
+            />
+            <select
+              class="placeholder-type-select"
+              value={override || 'auto'}
+              title={TYPE_SELECT_TITLE}
+              onChange={(e) => onSetType(name, e.target.value)}
+            >
+              {typeOptionsFor(value, override).map((opt) => (
+                <option value={opt} key={opt}>{opt === 'auto' ? `Auto (${CAP[autoLabelType]})` : CAP[opt]}</option>
+              ))}
+            </select>
+            {incompatible && (
+              <span class="placeholder-warn" title={`This value won't match as ${CAP[effective]}`}>{'⚠'}</span>
+            )}
+          </div>
+        );
+      })}
     </div>
   );
 }

src/mdh/components/QueryHistory.jsx

@@ -26,20 +26,22 @@ function dedupKey(collection, pipeline) {
   return collection + '::' + normalized;
 }
 
-export async function addToHistory(collection, pipeline, variables) {
+export async function addToHistory(collection, pipeline, variables, placeholderTypes) {
   const queryHistory = await readList('queryHistory');
   const key = dedupKey(collection, pipeline);
   const filtered = queryHistory.filter((e) => dedupKey(e.collection, e.pipeline) !== key);
   const entry = { collection, pipeline, ts: Date.now() };
   if (variables && Object.keys(variables).length > 0) entry.variables = variables;
+  if (placeholderTypes && Object.keys(placeholderTypes).length > 0) entry.placeholderTypes = placeholderTypes;
   filtered.unshift(entry);
   await writeList('queryHistory', filtered.slice(0, MAX_HISTORY));
 }
 
-export async function saveQuery(collection, pipeline, name, variables) {
+export async function saveQuery(collection, pipeline, name, variables, placeholderTypes) {
   const savedQueries = await readList('savedQueries');
   const entry = { collection, pipeline, name, ts: Date.now() };
   if (variables && Object.keys(variables).length > 0) entry.variables = variables;
+  if (placeholderTypes && Object.keys(placeholderTypes).length > 0) entry.placeholderTypes = placeholderTypes;
   savedQueries.push(entry);
   await writeList('savedQueries', savedQueries);
 }
@@ -67,7 +69,7 @@ function formatTime(ts) {
 function QueryRow({ item, currentCollection, savedName, onLoad, onDismiss, showUnsave, onUnsave }) {
   return (
     <div class={'query-history-item' + (item.collection === currentCollection ? ' query-history-item-current' : '')}>
-      <div class="query-history-item-info" onClick={() => { onLoad(item.pipeline, item.collection, item.variables); onDismiss(); }}>
+      <div class="query-history-item-info" onClick={() => { onLoad(item.pipeline, item.collection, item.variables, item.placeholderTypes); onDismiss(); }}>
         <span class="query-history-collection">{item.collection}</span>
         {savedName && <span class="query-history-name">{savedName}</span>}
         <span class="query-history-time">{formatTime(item.ts)}</span>