processors and drop column

docs/canvas-lab-architecture.md

@@ -202,13 +202,21 @@ File:
     model_providers: [...],
     model_configs: [...],
     columns: [...],
-    processors: [],
+    processors: [...],
   },
   run: { rows: 5, preview: true, output_formats: ["jsonl"] },
   ui: { nodes: [...], edges: [...] }
 }
 ```
 
+Current processor UI surface:
+- `schema_transform` (sheet -> `Processors` -> `Schema Transform`)
+- mapped to recipe processor with `build_stage: "post_batch"` and JSON `template`.
+
+Current drop policy:
+- column dialogs (`sampler` / `llm` / `expression`) expose `drop` toggle.
+- payload writes column `drop` directly (preferred over drop-columns processor in v1).
+
 How relation is enforced:
 - collect `model_alias` values used by LLM columns
 - ensure each alias exists in `recipe.model_configs`
@@ -253,6 +261,8 @@ Model dialogs:
 `/Volumes/Expansion/projects/new-ui-prototype/studio/frontend/src/features/canvas-lab/dialogs/models/model-provider-dialog.tsx`
 - model config:
 `/Volumes/Expansion/projects/new-ui-prototype/studio/frontend/src/features/canvas-lab/dialogs/models/model-config-dialog.tsx`
+- processors:
+`/Volumes/Expansion/projects/new-ui-prototype/studio/frontend/src/features/canvas-lab/dialogs/processors-dialog.tsx`
 
 `ModelConfigDialog` and `LlmDialog` use shadcn `Combobox` fed from store configs:
 - model config `provider` suggests model-provider node names
@@ -291,3 +301,36 @@ Minimal path for a new block type:
 - payload/import utils = external contract boundary.
 
 If one piece changes, keep all six in sync.
+
+## 14) Processors roadmap (decision)
+
+Current decision: **Option 3 (hybrid)**.
+
+v1 scope:
+- add `drop` toggle on column blocks (sampler/llm/expression).
+- add processor config surface for `schema_transform`.
+- keep payload builder as single mapper to `recipe.processors`.
+- keep processor state separate from node graph for now.
+
+Reason:
+- fastest ship path.
+- matches Data Designer column-level `drop`.
+- avoids duplicate/complex processor edge logic in v1.
+
+Future option noted: **Option 2 (processor chain in graph)**.
+
+Option 2 structure:
+- add virtual node `Dataset Output`.
+- processors become graph nodes: `Schema Transform`, `Drop Columns`, future processors.
+- processor order derived from chain edges:
+  `Dataset Output -> P1 -> P2 -> ...`
+- enforce chain rules:
+  - no cycles
+  - one incoming max per processor
+  - one outgoing max per processor
+  - chain must start at `Dataset Output`
+
+Migration from option 3 -> 2:
+- keep same processor schema/payload contracts.
+- move order source from list/order field to edge traversal.
+- UI changes mostly in canvas rendering + validation; payload adapter stays mostly same.

studio/frontend/src/features/canvas-lab/canvas-lab-page.tsx

@@ -22,11 +22,13 @@ import { CanvasEdge } from "./components/canvas-edge";
 import { CanvasNode } from "./components/canvas-node";
 import { ConfigDialog } from "./dialogs/config-dialog";
 import { ImportDialog } from "./dialogs/import-dialog";
+import { ProcessorsDialog } from "./dialogs/processors-dialog";
 import { useCanvasLabStore } from "./stores/canvas-lab";
 import type { CanvasNodeData, SamplerConfig } from "./types";
 import { isCategoryConfig } from "./utils";
 import { importCanvasPayload } from "./utils/import";
 import { buildCanvasPayload } from "./utils/payload";
+import { buildDefaultSchemaTransform } from "./utils/processors";
 
 const NODE_TYPES: NodeTypes = { builder: CanvasNode };
 const EDGE_TYPES: EdgeTypes = { canvas: CanvasEdge, semantic: CanvasEdge };
@@ -68,6 +70,7 @@ export function CanvasLabPage(): ReactElement {
     nodes,
     edges,
     configs,
+    processors,
     sheetView,
     activeConfigId,
     dialogOpen,
@@ -84,6 +87,7 @@ export function CanvasLabPage(): ReactElement {
     updateConfig,
     isValidConnection,
     setSheetView,
+    setProcessors,
     setDialogOpen,
     loadCanvas,
     setLayoutDirection,
@@ -93,6 +97,7 @@ export function CanvasLabPage(): ReactElement {
       nodes: state.nodes,
       edges: state.edges,
       configs: state.configs,
+      processors: state.processors,
       sheetView: state.sheetView,
       activeConfigId: state.activeConfigId,
       dialogOpen: state.dialogOpen,
@@ -109,6 +114,7 @@ export function CanvasLabPage(): ReactElement {
       updateConfig: state.updateConfig,
       isValidConnection: state.isValidConnection,
       setSheetView: state.setSheetView,
+      setProcessors: state.setProcessors,
       setDialogOpen: state.setDialogOpen,
       loadCanvas: state.loadCanvas,
       setLayoutDirection: state.setLayoutDirection,
@@ -120,6 +126,7 @@ export function CanvasLabPage(): ReactElement {
   );
   const [previewLoading, setPreviewLoading] = useState(false);
   const [importOpen, setImportOpen] = useState(false);
+  const [processorsOpen, setProcessorsOpen] = useState(false);
   const [statusMessage, setStatusMessage] = useState<{
     tone: "success" | "error";
     text: string;
@@ -146,7 +153,12 @@ export function CanvasLabPage(): ReactElement {
     setPreviewLoading(true);
     setStatusMessage(null);
     try {
-      const { payload, errors } = buildCanvasPayload(configs, nodes, edges);
+      const { payload, errors } = buildCanvasPayload(
+        configs,
+        nodes,
+        edges,
+        processors,
+      );
       if (errors.length > 0) {
         setStatusMessage({
           tone: "error",
@@ -172,7 +184,12 @@ export function CanvasLabPage(): ReactElement {
 
   const handleCopyRecipe = async (): Promise<void> => {
     setStatusMessage(null);
-    const { payload, errors } = buildCanvasPayload(configs, nodes, edges);
+    const { payload, errors } = buildCanvasPayload(
+      configs,
+      nodes,
+      edges,
+      processors,
+    );
     if (errors.length > 0) {
       setStatusMessage({
         tone: "error",
@@ -211,6 +228,17 @@ export function CanvasLabPage(): ReactElement {
     return null;
   };
 
+  const handleOpenProcessorsFromSheet = useCallback(() => {
+    if (
+      !processors.some(
+        (processor) => processor.processor_type === "schema_transform",
+      )
+    ) {
+      setProcessors([...processors, buildDefaultSchemaTransform()]);
+    }
+    setProcessorsOpen(true);
+  }, [processors, setProcessors]);
+
   return (
     <div className="min-h-screen bg-background">
       <main className="w-full px-6 py-8">
@@ -313,6 +341,7 @@ export function CanvasLabPage(): ReactElement {
                 onAddModelProvider={addModelProviderNode}
                 onAddModelConfig={addModelConfigNode}
                 onAddExpression={addExpressionNode}
+                onOpenProcessors={handleOpenProcessorsFromSheet}
               />
             </Panel>
             <Controls position="bottom-left" />
@@ -333,6 +362,13 @@ export function CanvasLabPage(): ReactElement {
         onImport={handleImport}
         container={sheetContainer}
       />
+      <ProcessorsDialog
+        open={processorsOpen}
+        onOpenChange={setProcessorsOpen}
+        processors={processors}
+        onProcessorsChange={setProcessors}
+        container={sheetContainer}
+      />
     </div>
   );
 }

studio/frontend/src/features/canvas-lab/components/block-sheet.tsx

@@ -8,7 +8,9 @@ import {
 } from "@/components/ui/sheet";
 import {
   ArrowLeft02Icon,
-  ArrowRight01Icon, type Database02Icon,
+  ArrowRight01Icon,
+  CodeIcon,
+  type Database02Icon,
   PlusSignIcon,
 } from "@hugeicons/core-free-icons";
 import { HugeiconsIcon } from "@hugeicons/react";
@@ -16,8 +18,15 @@ import type { ReactElement } from "react";
 import type { LlmType, SamplerType } from "../types";
 import { BLOCK_GROUPS, getBlocksForKind } from "../blocks/registry";
 
-type SheetView = "root" | "sampler" | "llm" | "expression";
+type SheetView = "root" | "sampler" | "llm" | "expression" | "processor";
 type SheetKind = "sampler" | "llm" | "expression";
+type RootSheetView = Exclude<SheetView, "root">;
+type RootGroup = {
+  kind: RootSheetView;
+  title: string;
+  description: string;
+  icon: typeof Database02Icon;
+};
 
 type BlockSheetProps = {
   container: HTMLDivElement | null;
@@ -28,6 +37,7 @@ type BlockSheetProps = {
   onAddModelProvider: () => void;
   onAddModelConfig: () => void;
   onAddExpression: () => void;
+  onOpenProcessors: () => void;
 };
 
 function getSheetTitle(view: SheetView): string {
@@ -40,6 +50,9 @@ function getSheetTitle(view: SheetView): string {
   if (view === "expression") {
     return "Expression blocks";
   }
+  if (view === "processor") {
+    return "Processor blocks";
+  }
   return "LLM blocks";
 }
 
@@ -48,8 +61,19 @@ const VIEW_KIND: Record<SheetView, SheetKind | null> = {
   sampler: "sampler",
   llm: "llm",
   expression: "expression",
+  processor: null,
 };
 
+const ROOT_GROUPS: RootGroup[] = [
+  ...BLOCK_GROUPS,
+  {
+    kind: "processor",
+    title: "Processors",
+    description: "Output schema + post batch.",
+    icon: CodeIcon,
+  },
+];
+
 function BlockSheetButton({
   icon,
   title,
@@ -97,6 +121,7 @@ export function BlockSheet({
   onAddModelProvider,
   onAddModelConfig,
   onAddExpression,
+  onOpenProcessors,
 }: BlockSheetProps): ReactElement {
   const title = getSheetTitle(view);
   return (
@@ -138,7 +163,7 @@ export function BlockSheet({
         <div className=" py-4">
           <div className="mt-4 flex flex-col gap-2">
             {view === "root" &&
-              BLOCK_GROUPS.map((item, index) => (
+              ROOT_GROUPS.map((item, index) => (
                 <BlockSheetButton
                   key={item.kind}
                   icon={item.icon}
@@ -148,7 +173,17 @@ export function BlockSheet({
                   onClick={() => onViewChange(item.kind)}
                 />
               ))}
+            {view === "processor" && (
+              <BlockSheetButton
+                icon={CodeIcon}
+                title="Schema Transform"
+                description="Transform final dataset schema."
+                isActive={true}
+                onClick={onOpenProcessors}
+              />
+            )}
             {view !== "root" &&
+              view !== "processor" &&
               getBlocksForKind(VIEW_KIND[view] ?? "sampler").map(
                 (item, index) => (
                 <BlockSheetButton

studio/frontend/src/features/canvas-lab/dialogs/config-dialog.tsx

@@ -1,5 +1,6 @@
 import { Button } from "@/components/ui/button";
 import { Dialog, DialogContent, DialogFooter } from "@/components/ui/dialog";
+import { Switch } from "@/components/ui/switch";
 import type { ReactElement } from "react";
 import type { NodeConfig, SamplerConfig } from "../types";
 import { renderBlockDialog } from "../blocks/registry";
@@ -41,6 +42,22 @@ export function ConfigDialog({
         {config && (
           <div className="space-y-4">
             <ValidationBanner config={config} />
+            {(config.kind === "sampler" ||
+              config.kind === "llm" ||
+              config.kind === "expression") && (
+              <div className="flex items-center justify-between gap-3 rounded-2xl border border-border/60 px-3 py-2">
+                <div>
+                  <p className="text-sm font-semibold">Drop from final dataset</p>
+                  <p className="text-xs text-muted-foreground">
+                    Keep for generation but omit from exported rows.
+                  </p>
+                </div>
+                <Switch
+                  checked={config.drop ?? false}
+                  onCheckedChange={(value) => onUpdate(config.id, { drop: value })}
+                />
+              </div>
+            )}
             {renderBlockDialog(config, categoryOptions, onUpdate)}
           </div>
         )}

studio/frontend/src/features/canvas-lab/dialogs/processors-dialog.tsx

@@ -0,0 +1,134 @@
+import { Button } from "@/components/ui/button";
+import { Dialog, DialogContent, DialogFooter } from "@/components/ui/dialog";
+import { Input } from "@/components/ui/input";
+import { Switch } from "@/components/ui/switch";
+import { Textarea } from "@/components/ui/textarea";
+import { type ReactElement, useMemo } from "react";
+import type { CanvasProcessorConfig } from "../types";
+import { buildDefaultSchemaTransform } from "../utils/processors";
+type ProcessorsDialogProps = {
+  open: boolean;
+  onOpenChange: (open: boolean) => void;
+  processors: CanvasProcessorConfig[];
+  onProcessorsChange: (processors: CanvasProcessorConfig[]) => void;
+  container?: HTMLDivElement | null;
+};
+
+export function ProcessorsDialog({
+  open,
+  onOpenChange,
+  processors,
+  onProcessorsChange,
+  container,
+}: ProcessorsDialogProps): ReactElement {
+  const schemaIndex = useMemo(
+    () =>
+      processors.findIndex(
+        (processor) => processor.processor_type === "schema_transform",
+      ),
+    [processors],
+  );
+  const schemaProcessor = schemaIndex >= 0 ? processors[schemaIndex] : null;
+  const nameId = schemaProcessor ? `${schemaProcessor.id}-name` : "schema-transform-name";
+  const templateId = schemaProcessor
+    ? `${schemaProcessor.id}-template`
+    : "schema-transform-template";
+
+  const setSchemaEnabled = (enabled: boolean) => {
+    if (enabled) {
+      if (schemaProcessor) {
+        return;
+      }
+      onProcessorsChange([...processors, buildDefaultSchemaTransform()]);
+      return;
+    }
+    onProcessorsChange(
+      processors.filter(
+        (processor) => processor.processor_type !== "schema_transform",
+      ),
+    );
+  };
+
+  const updateSchema = (patch: Partial<CanvasProcessorConfig>) => {
+    if (!schemaProcessor) {
+      return;
+    }
+    const next = [...processors];
+    next[schemaIndex] = { ...schemaProcessor, ...patch };
+    onProcessorsChange(next);
+  };
+
+  return (
+    <Dialog open={open} onOpenChange={onOpenChange}>
+      <DialogContent
+        container={container}
+        position="absolute"
+        overlayPosition="absolute"
+        overlayClassName="bg-transparent"
+        className="max-h-[85vh] overflow-auto sm:max-w-2xl"
+      >
+        <div className="space-y-4">
+          <div className="flex items-center justify-between gap-3 rounded-2xl border border-border/60 px-3 py-2">
+            <div>
+              <p className="text-sm font-semibold">Schema transform</p>
+              <p className="text-xs text-muted-foreground">
+                Transform final rows to target schema (post-batch).
+              </p>
+            </div>
+            <Switch
+              checked={Boolean(schemaProcessor)}
+              onCheckedChange={setSchemaEnabled}
+            />
+          </div>
+
+          {schemaProcessor && (
+            <div className="space-y-3">
+              <div className="grid gap-2">
+                <label
+                  className="text-xs font-semibold uppercase text-muted-foreground"
+                  htmlFor={nameId}
+                >
+                  Name
+                </label>
+                <Input
+                  id={nameId}
+                  className="nodrag"
+                  value={schemaProcessor.name}
+                  onChange={(event) => updateSchema({ name: event.target.value })}
+                />
+              </div>
+              <div className="grid gap-2">
+                <label
+                  className="text-xs font-semibold uppercase text-muted-foreground"
+                  htmlFor={templateId}
+                >
+                  Template (JSON)
+                </label>
+                <Textarea
+                  id={templateId}
+                  className="nodrag min-h-[220px]"
+                  value={schemaProcessor.template}
+                  onChange={(event) =>
+                    updateSchema({ template: event.target.value })
+                  }
+                />
+                <p className="text-xs text-muted-foreground">
+                  Use Jinja refs like {"{{ customer_review }}"} in values.
+                </p>
+              </div>
+            </div>
+          )}
+        </div>
+        <DialogFooter>
+          <Button
+            type="button"
+            variant="outline"
+            onClick={() => onOpenChange(false)}
+          >
+            Done
+          </Button>
+        </DialogFooter>
+      </DialogContent>
+    </Dialog>
+  );
+}

studio/frontend/src/features/canvas-lab/stores/canvas-lab.ts

@@ -11,6 +11,7 @@ import {
 import { create } from "zustand";
 import type {
   CanvasNode,
+  CanvasProcessorConfig,
   LayoutDirection,
   LlmType,
   ModelConfig,
@@ -34,12 +35,13 @@ import {
   updateNodeData,
 } from "./canvas-lab-helpers";
 
-type SheetView = "root" | "sampler" | "llm" | "expression";
+type SheetView = "root" | "sampler" | "llm" | "expression" | "processor";
 
 type CanvasLabState = {
   nodes: CanvasNode[];
   edges: Edge[];
   configs: Record<string, NodeConfig>;
+  processors: CanvasProcessorConfig[];
   sheetView: SheetView;
   activeConfigId: string | null;
   dialogOpen: boolean;
@@ -47,6 +49,7 @@ type CanvasLabState = {
   nextId: number;
   nextY: number;
   setSheetView: (view: SheetView) => void;
+  setProcessors: (processors: CanvasProcessorConfig[]) => void;
   setDialogOpen: (open: boolean) => void;
   openConfig: (id: string) => void;
   setLayoutDirection: (direction: LayoutDirection) => void;
@@ -68,6 +71,7 @@ export const useCanvasLabStore = create<CanvasLabState>((set, get) => ({
   nodes: [],
   edges: [],
   configs: {},
+  processors: [],
   sheetView: "root",
   activeConfigId: null,
   dialogOpen: false,
@@ -75,6 +79,7 @@ export const useCanvasLabStore = create<CanvasLabState>((set, get) => ({
   nextId: 3,
   nextY: 280,
   setSheetView: (view) => set({ sheetView: view }),
+  setProcessors: (processors) => set({ processors }),
   setDialogOpen: (open) => set({ dialogOpen: open }),
   openConfig: (id) => set({ activeConfigId: id, dialogOpen: true }),
   setLayoutDirection: (direction) =>
@@ -168,6 +173,7 @@ export const useCanvasLabStore = create<CanvasLabState>((set, get) => ({
         state.layoutDirection,
       ),
       edges: snapshot.edges,
+      processors: snapshot.processors,
       nextId: snapshot.nextId,
       nextY: snapshot.nextY,
       activeConfigId: null,

studio/frontend/src/features/canvas-lab/types/index.ts

@@ -47,6 +47,7 @@ export type SamplerConfig = {
   // biome-ignore lint/style/useNamingConvention: api schema
   sampler_type: SamplerType;
   name: string;
+  drop?: boolean;
   // biome-ignore lint/style/useNamingConvention: api schema
   convert_to?: "float" | "int" | "str";
   values?: string[];
@@ -107,6 +108,7 @@ export type LlmConfig = {
   // biome-ignore lint/style/useNamingConvention: api schema
   llm_type: LlmType;
   name: string;
+  drop?: boolean;
   // biome-ignore lint/style/useNamingConvention: api schema
   model_alias: string;
   prompt: string;
@@ -156,10 +158,21 @@ export type ExpressionConfig = {
   id: string;
   kind: "expression";
   name: string;
+  drop?: boolean;
   expr: string;
   dtype: ExpressionDtype;
 };
 
+export type SchemaTransformProcessorConfig = {
+  id: string;
+  // biome-ignore lint/style/useNamingConvention: api schema
+  processor_type: "schema_transform";
+  name: string;
+  template: string;
+};
+
+export type CanvasProcessorConfig = SchemaTransformProcessorConfig;
+
 export type NodeConfig =
   | SamplerConfig
   | LlmConfig

studio/frontend/src/features/canvas-lab/utils/import/importer.ts

@@ -1,4 +1,4 @@
-import type { NodeConfig } from "../../types";
+import type { CanvasProcessorConfig, NodeConfig } from "../../types";
 import { buildEdges } from "./edges";
 import { isRecord, parseJson, readString } from "./helpers";
 import {
@@ -21,6 +21,40 @@ type UiInput = {
   edges?: unknown;
 };
 
+function parseProcessors(input: unknown): CanvasProcessorConfig[] {
+  if (!Array.isArray(input)) {
+    return [];
+  }
+  const processors: CanvasProcessorConfig[] = [];
+  input.forEach((item, index) => {
+    if (!isRecord(item)) {
+      return;
+    }
+    const type = readString(item.processor_type);
+    const templateRaw = item.template;
+    const isSchemaTransform =
+      type === "schema_transform" || isRecord(templateRaw);
+    if (!isSchemaTransform) {
+      return;
+    }
+    const name = readString(item.name) ?? `schema_transform_${index + 1}`;
+    const template =
+      typeof templateRaw === "string"
+        ? templateRaw
+        : isRecord(templateRaw)
+          ? JSON.stringify(templateRaw, null, 2)
+          : "{\n  \"text\": \"{{ column_name }}\"\n}";
+    processors.push({
+      id: `p${index + 1}`,
+      // biome-ignore lint/style/useNamingConvention: api schema
+      processor_type: "schema_transform",
+      name,
+      template,
+    });
+  });
+  return processors;
+}
+
 export function importCanvasPayload(input: string): ImportResult {
   const parsed = parseJson(input);
   if (!parsed.data || !isRecord(parsed.data)) {
@@ -41,6 +75,7 @@ export function importCanvasPayload(input: string): ImportResult {
 
   const errors: string[] = [];
   const configs: NodeConfig[] = [];
+  const processors = parseProcessors(recipe.processors);
   const nameToId = new Map<string, string>();
 
   let nextId = 1;
@@ -129,6 +164,7 @@ export function importCanvasPayload(input: string): ImportResult {
       configs: Object.fromEntries(configs.map((config) => [config.id, config])),
       nodes,
       edges,
+      processors,
       nextId,
       nextY: maxY + 140,
     },

studio/frontend/src/features/canvas-lab/utils/import/parsers.ts

@@ -72,6 +72,7 @@ function parseSampler(
   id: string,
   errors: string[],
 ): SamplerConfig | null {
+  const drop = column.drop === true;
   const samplerType = readString(column.sampler_type);
   if (!samplerType || !SAMPLER_TYPES.includes(samplerType as SamplerType)) {
     errors.push(`Sampler ${name}: unsupported sampler_type.`);
@@ -99,6 +100,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "category",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       values,
@@ -122,6 +124,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "subcategory",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -137,6 +140,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "uniform",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       low: readNumberString(params.low),
@@ -150,6 +154,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "gaussian",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       mean: readNumberString(params.mean),
@@ -163,6 +168,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "bernoulli",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       p: readNumberString(params.p),
@@ -175,6 +181,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "datetime",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -197,6 +204,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "timedelta",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -216,6 +224,7 @@ function parseSampler(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "uuid",
       name,
+      drop,
       // biome-ignore lint/style/useNamingConvention: api schema
       convert_to: normalizedConvertTo,
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -232,6 +241,7 @@ function parseSampler(
     id,
     kind: "sampler",
     name,
+    drop,
     // biome-ignore lint/style/useNamingConvention: api schema
     sampler_type: samplerType as SamplerType,
     // biome-ignore lint/style/useNamingConvention: api schema
@@ -297,6 +307,7 @@ function parseLlm(
     // biome-ignore lint/style/useNamingConvention: api schema
     llm_type: llmType,
     name,
+    drop: column.drop === true,
     // biome-ignore lint/style/useNamingConvention: api schema
     model_alias: readString(column.model_alias) ?? "",
     prompt: readString(column.prompt) ?? "",
@@ -378,6 +389,7 @@ function parseExpression(
     id,
     kind: "expression",
     name,
+    drop: column.drop === true,
     expr: readString(column.expr) ?? "",
     dtype: normalized,
   };

studio/frontend/src/features/canvas-lab/utils/import/types.ts

@@ -1,10 +1,15 @@
 import type { Edge } from "@xyflow/react";
-import type { CanvasNode, NodeConfig } from "../../types";
+import type {
+  CanvasNode,
+  CanvasProcessorConfig,
+  NodeConfig,
+} from "../../types";
 
 export type CanvasSnapshot = {
   configs: Record<string, NodeConfig>;
   nodes: CanvasNode[];
   edges: Edge[];
+  processors: CanvasProcessorConfig[];
   nextId: number;
   nextY: number;
 };

studio/frontend/src/features/canvas-lab/utils/index.ts

@@ -68,6 +68,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "category",
       name,
+      drop: false,
       values: ["A", "B", "C"],
       weights: [null, null, null],
     };
@@ -79,6 +80,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "subcategory",
       name,
+      drop: false,
       // biome-ignore lint/style/useNamingConvention: api schema
       subcategory_parent: "",
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -97,6 +99,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "uniform",
       name,
+      drop: false,
       low: "0",
       high: "1",
     };
@@ -108,6 +111,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "gaussian",
       name,
+      drop: false,
       mean: "0",
       std: "1",
     };
@@ -119,6 +123,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "bernoulli",
       name,
+      drop: false,
       p: "0.5",
     };
   }
@@ -129,6 +134,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "datetime",
       name,
+      drop: false,
       // biome-ignore lint/style/useNamingConvention: api schema
       datetime_start: "",
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -144,6 +150,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "timedelta",
       name,
+      drop: false,
       // biome-ignore lint/style/useNamingConvention: api schema
       dt_min: "0",
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -161,6 +168,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "uuid",
       name,
+      drop: false,
       // biome-ignore lint/style/useNamingConvention: api schema
       uuid_format: "",
     };
@@ -172,6 +180,7 @@ export function makeSamplerConfig(
       // biome-ignore lint/style/useNamingConvention: api schema
       sampler_type: "person_from_faker",
       name,
+      drop: false,
       // biome-ignore lint/style/useNamingConvention: api schema
       person_locale: "",
       // biome-ignore lint/style/useNamingConvention: api schema
@@ -188,6 +197,7 @@ export function makeSamplerConfig(
     // biome-ignore lint/style/useNamingConvention: api schema
     sampler_type: "person",
     name,
+    drop: false,
     // biome-ignore lint/style/useNamingConvention: api schema
     person_locale: "",
     // biome-ignore lint/style/useNamingConvention: api schema
@@ -221,6 +231,7 @@ export function makeLlmConfig(
     // biome-ignore lint/style/useNamingConvention: api schema
     llm_type: llmType,
     name,
+    drop: false,
     // biome-ignore lint/style/useNamingConvention: api schema
     model_alias: "allenai/olmo-3.1-32b-instruct",
     prompt:
@@ -302,6 +313,7 @@ export function makeExpressionConfig(
     id,
     kind: "expression",
     name: nextName(existing, "expr"),
+    drop: false,
     expr: "",
     dtype: "str",
   };

studio/frontend/src/features/canvas-lab/utils/payload.ts

@@ -1,5 +1,6 @@
 import type { Edge } from "@xyflow/react";
 import type {
+  CanvasProcessorConfig,
   CategoryConditionalParams,
   CanvasNode,
   ExpressionConfig,
@@ -316,6 +317,7 @@ function buildLlmColumn(
 ): Record<string, unknown> {
   const base = {
     name: config.name,
+    drop: config.drop ?? false,
     // biome-ignore lint/style/useNamingConvention: api schema
     model_alias: config.model_alias,
     prompt: config.prompt,
@@ -401,16 +403,52 @@ function buildExpressionColumn(
     // biome-ignore lint/style/useNamingConvention: api schema
     column_type: "expression",
     name: config.name,
+    drop: config.drop ?? false,
     expr: config.expr,
     dtype: config.dtype,
   };
 }
 
+function buildProcessors(
+  processors: CanvasProcessorConfig[],
+  errors: string[],
+): Record<string, unknown>[] {
+  const output: Record<string, unknown>[] = [];
+  for (const processor of processors) {
+    if (processor.processor_type !== "schema_transform") {
+      continue;
+    }
+    const name = processor.name.trim();
+    if (!name) {
+      errors.push("Schema transform: name is required.");
+      continue;
+    }
+    const template = parseJsonObject(
+      processor.template,
+      `Schema transform ${name} template`,
+      errors,
+    );
+    if (!template) {
+      continue;
+    }
+    output.push({
+      // biome-ignore lint/style/useNamingConvention: api schema
+      processor_type: "schema_transform",
+      name,
+      // biome-ignore lint/style/useNamingConvention: api schema
+      build_stage: "post_batch",
+      template,
+    });
+  }
+  return output;
+}
+
 // biome-ignore lint/complexity/noExcessiveCognitiveComplexity: payload build
 export function buildCanvasPayload(
   configs: Record<string, NodeConfig>,
   nodes: CanvasNode[],
   edges: Edge[],
+  processors: CanvasProcessorConfig[] = [],
 ): CanvasPayloadResult {
   const errors: string[] = [];
   const columns: Record<string, unknown>[] = [];
@@ -442,6 +480,7 @@ export function buildCanvasPayload(
         // biome-ignore lint/style/useNamingConvention: api schema
         column_type: "sampler",
         name: config.name,
+        drop: config.drop ?? false,
         // biome-ignore lint/style/useNamingConvention: api schema
         sampler_type: config.sampler_type,
         params: buildSamplerParams(config, errors),
@@ -585,6 +624,7 @@ export function buildCanvasPayload(
       },
     ];
   });
+  const recipeProcessors = buildProcessors(processors, errors);
 
   return {
     errors,
@@ -595,7 +635,7 @@ export function buildCanvasPayload(
         // biome-ignore lint/style/useNamingConvention: api schema
         model_configs: modelConfigs,
         columns,
-        processors: [],
+        processors: recipeProcessors,
       },
       run: {
         rows: 5,

studio/frontend/src/features/canvas-lab/utils/processors.ts

@@ -0,0 +1,11 @@
+import type { CanvasProcessorConfig } from "../types";
+
+export function buildDefaultSchemaTransform(): CanvasProcessorConfig {
+  return {
+    id: "schema-transform-1",
+    // biome-ignore lint/style/useNamingConvention: api schema
+    processor_type: "schema_transform",
+    name: "schema_transform",
+    template: '{\n  "text": "{{ column_name }}"\n}',
+  };
+}