feat(tests): add Playwright e2e suite + debug probe for a-immersive

- Add window.__nisps debug probe (gated on ?debug=1) exposing iml state,
  getOutputs/getLoss/getWeights/getExampleCount, and action triggers
  (thumbsUp/thumbsDown/train/randomise/clearExamples/saveState)
- Fix WasmIML bug: this.dataset was a plain object; import Dataset and
  use new Dataset(100) so computeWeights() is available for training
- Fix WasmIML.addExample/clearDataset to use Dataset API methods
- 44 Playwright e2e tests across 4 spec files:
  - ml-engine.spec.js: WASM inference bounds, training loss, thumbs
    up/down behavior, async training, example capture semantics
  - ui-interactions.spec.js: drawer open/close, mode switching,
    heatmap bar counts, preset chips, keyboard shortcuts (1/2/Z)
  - input-pipeline.spec.js: input→output variation, clamping, joystick
    drag, post-training output bounds across the full input space
  - persistence.spec.js: URL params (?preset, ?spread), localStorage
    round-trip, saveState probe

.gitignore

@@ -41,3 +41,9 @@ playground/osc-bridge/node_modules/
 # Dolt database files (added by bd init)
 .dolt/
 *.db
+
+# Node / Playwright
+node_modules/
+bun.lock
+playwright-report/
+test-results/

package.json

@@ -0,0 +1,13 @@
+{
+  "name": "memlnaut-nisps-tests",
+  "version": "1.0.0",
+  "private": true,
+  "scripts": {
+    "test": "playwright test",
+    "test:ui": "playwright test --ui",
+    "test:headed": "playwright test --headed"
+  },
+  "devDependencies": {
+    "@playwright/test": "^1.59.1"
+  }
+}

playground/js/a-app.js

@@ -869,6 +869,54 @@ async function init() {
   // Auto-save every 10 seconds
   setInterval(saveState, 10000);
 
+  // Debug probe — exposed on window when ?debug=1 is in the URL.
+  // Used by Playwright e2e tests. Zero footprint in production.
+  if (new URLSearchParams(window.location.search).has('debug')) {
+    window.__nisps = {
+      get iml()    { return iml; },
+      get imlJoy() { return imlJoy; },
+      get imlHand(){ return imlHand; },
+      getOutputs:      () => [...iml.getOutputs()],
+      getLoss:         () => iml.lastLoss,
+      getWeights:      () => iml._getFlatWeights(),
+      getExampleCount: () => iml.exampleCount,
+      setInputs: (x, y) => {
+        iml.setInput(0, x);
+        iml.setInput(1, y);
+        iml.process();
+        const outputs = iml.getOutputs();
+        routeOutputs(outputs);
+        updateHeatmap(outputs);
+      },
+      thumbsUp:   () => onThumbsUp(),
+      thumbsDown: () => onThumbsDown(),
+      train: () => {
+        const loss = trainModel();
+        const outputs = iml.getOutputs();
+        routeOutputs(outputs);
+        updateHeatmap(outputs);
+        syncRawParamsFromOutputs(outputs);
+        updateStatus();
+        drawLossPlot();
+        return loss;
+      },
+      trainAsync: () => new Promise(resolve => trainModelAsync(resolve)),
+      randomise:  () => {
+        iml.randomiseWeights(spreadLevel);
+        const outputs = iml.getOutputs();
+        routeOutputs(outputs);
+        updateHeatmap(outputs);
+        syncRawParamsFromOutputs(outputs);
+        updateStatus();
+      },
+      clearExamples: () => {
+        iml.clearDataset();
+        updateStatus();
+      },
+      saveState: () => saveState(),
+    };
+  }
+
   // Start animation
   requestAnimationFrame(animate);
 }

playground/js/nisps/nisps-wasm.js

@@ -2,6 +2,8 @@
 // Uses nisps-core compiled to WASM for inference, training, and weight ops.
 // Training runs in a Web Worker for non-blocking operation.
 
+import { Dataset } from './dataset.js';
+
 // Activation function IDs matching C++ nisps::ACTIVATION_FUNCTIONS enum
 const ACTIVATION = { SIGMOID: 0, TANH: 1, LINEAR: 2, RELU: 3 };
 
@@ -116,8 +118,8 @@ export class WasmIML {
     this.totalTrainingIterations = 0;
     this.logFn = null;
 
-    // Dataset (JS-side for persistence/visualization access)
+    // Dataset (JS-side for persistence/visualization access and sample weighting)
-    this.dataset = { features: [], labels: [], maxExamples: 100 };
+    this.dataset = new Dataset(100);
 
     // Persistent WASM buffers for inference (avoid alloc/free per frame)
     const inputDim = nInputs + BIAS;
@@ -200,18 +202,11 @@ export class WasmIML {
     while (inVec.length < this.nInputs) inVec.push(0);
     const outVec = outputs.slice(0, this.nOutputs);
     while (outVec.length < this.nOutputs) outVec.push(0);
-
+    this.dataset.add(inVec, outVec);
-    if (this.dataset.features.length >= this.dataset.maxExamples) {
-      this.dataset.features.shift();
-      this.dataset.labels.shift();
-    }
-    this.dataset.features.push([...inVec]);
-    this.dataset.labels.push([...outVec]);
   }
 
   clearDataset() {
-    this.dataset.features = [];
+    this.dataset.clear();
-    this.dataset.labels = [];
     this.log('Dataset cleared.');
   }
 

playwright.config.js

@@ -0,0 +1,27 @@
+const { defineConfig } = require('@playwright/test');
+
+module.exports = defineConfig({
+  testDir: './tests/e2e',
+  timeout: 30_000,
+  expect: { timeout: 10_000 },
+  use: {
+    baseURL: 'http://localhost:7331',
+    headless: true,
+    // WASM + AudioContext require a stable origin
+    ignoreHTTPSErrors: true,
+  },
+  // Start a static file server against the playground/ dir before tests run.
+  // python3 -m http.server serves directory listings and static assets fine.
+  // WASM files need correct MIME type — Python's server handles .wasm correctly.
+  webServer: {
+    command: 'python3 -m http.server 7331',
+    cwd: './playground',
+    url: 'http://localhost:7331',
+    reuseExistingServer: true,
+    timeout: 10_000,
+  },
+  projects: [
+    { name: 'chromium', use: { browserName: 'chromium' } },
+  ],
+  reporter: [['list'], ['html', { open: 'never' }]],
+});

tests/e2e/helpers.js

@@ -0,0 +1,32 @@
+/**
+ * Shared helpers for e2e tests.
+ */
+const { expect } = require('@playwright/test');
+
+/**
+ * Navigate to a-immersive with ?debug=1 and wait for the WASM engine to
+ * initialise and expose window.__nisps.
+ *
+ * @param {import('@playwright/test').Page} page
+ * @param {string} extraParams  - additional query string, e.g. '&preset=beginner-1'
+ */
+async function loadApp(page, extraParams = '') {
+  // Clear app state but mark help as seen so the overlay doesn't block clicks.
+  await page.addInitScript(() => {
+    localStorage.removeItem('nisps-a-immersive');
+    localStorage.setItem('nisps-help-seen', '1');
+  });
+  await page.goto(`/a-immersive.html?debug=1${extraParams}`);
+  // Wait until the debug probe is ready (WASM init is async).
+  await page.waitForFunction(() => window.__nisps !== undefined, { timeout: 20_000 });
+}
+
+/**
+ * Returns the current text content of the status line.
+ * @param {import('@playwright/test').Page} page
+ */
+async function statusText(page) {
+  return page.locator('#status-text').textContent();
+}
+
+module.exports = { loadApp, statusText };

tests/e2e/input-pipeline.spec.js

@@ -0,0 +1,110 @@
+/**
+ * Input pipeline tests — joystick → MLP inputs → outputs → heatmap.
+ */
+const { test, expect } = require('@playwright/test');
+const { loadApp } = require('./helpers');
+
+test.describe('Joystick → output pipeline', () => {
+  test('different input positions produce different outputs', async ({ page }) => {
+    await loadApp(page);
+
+    await page.evaluate(() => window.__nisps.setInputs(0.1, 0.1));
+    const out1 = await page.evaluate(() => window.__nisps.getOutputs());
+
+    await page.evaluate(() => window.__nisps.setInputs(0.9, 0.9));
+    const out2 = await page.evaluate(() => window.__nisps.getOutputs());
+
+    const anyChanged = out1.some((v, i) => Math.abs(v - out2[i]) > 0.0001);
+    expect(anyChanged).toBe(true);
+  });
+
+  test('all outputs remain in [0, 1] across input positions', async ({ page }) => {
+    await loadApp(page);
+    const corners = [[0, 0], [0, 1], [1, 0], [1, 1], [0.5, 0.5]];
+    for (const [x, y] of corners) {
+      await page.evaluate(([x, y]) => window.__nisps.setInputs(x, y), [x, y]);
+      const outputs = await page.evaluate(() => window.__nisps.getOutputs());
+      for (const v of outputs) {
+        expect(v).toBeGreaterThanOrEqual(0);
+        expect(v).toBeLessThanOrEqual(1);
+      }
+    }
+  });
+
+  test('heatmap bar widths change when inputs change', async ({ page }) => {
+    await loadApp(page);
+
+    const widths1 = await page.evaluate(() =>
+      Array.from(document.querySelectorAll('.heatmap-cell-bar')).map(el => el.style.width)
+    );
+
+    // Move to far corner
+    await page.evaluate(() => window.__nisps.setInputs(0.95, 0.05));
+
+    const widths2 = await page.evaluate(() =>
+      Array.from(document.querySelectorAll('.heatmap-cell-bar')).map(el => el.style.width)
+    );
+
+    const anyChanged = widths1.some((w, i) => w !== widths2[i]);
+    expect(anyChanged).toBe(true);
+  });
+
+  test('mouse drag on joystick container updates MLP inputs', async ({ page }) => {
+    await loadApp(page);
+
+    const box = await page.locator('#joystick-container').boundingBox();
+    const cx = box.x + box.width / 2;
+    const cy = box.y + box.height / 2;
+
+    // Drag from center towards bottom-right
+    await page.mouse.move(cx, cy);
+    await page.mouse.down();
+    await page.mouse.move(cx + box.width * 0.3, cy + box.height * 0.3, { steps: 10 });
+    await page.mouse.up();
+
+    // After drag, at least one input axis should have moved from 0.5
+    const [x, y] = await page.evaluate(() => [
+      window.__nisps.iml.inputState[0],
+      window.__nisps.iml.inputState[1],
+    ]);
+    const moved = Math.abs(x - 0.5) > 0.01 || Math.abs(y - 0.5) > 0.01;
+    expect(moved).toBe(true);
+  });
+
+  test('setInputs clamps values to [0, 1]', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(() => window.__nisps.setInputs(-5, 99));
+    const [x, y] = await page.evaluate(() => [
+      window.__nisps.iml.inputState[0],
+      window.__nisps.iml.inputState[1],
+    ]);
+    expect(x).toBeGreaterThanOrEqual(0);
+    expect(x).toBeLessThanOrEqual(1);
+    expect(y).toBeGreaterThanOrEqual(0);
+    expect(y).toBeLessThanOrEqual(1);
+  });
+
+  test('after training, moving inputs produces smoothly varying outputs', async ({ page }) => {
+    await loadApp(page);
+
+    // Load calm-to-chaotic preset (3 examples spanning the input space)
+    await page.click('[data-drawer="training"]');
+    await page.click('[data-preset="calm-to-chaotic"]');
+    await page.waitForFunction(
+      () => document.getElementById('status-text').textContent.includes('loss'),
+      { timeout: 15_000 }
+    );
+
+    // Sample 5 positions and verify all outputs are bounded
+    const positions = [0.0, 0.25, 0.5, 0.75, 1.0];
+    for (const t of positions) {
+      await page.evaluate((t) => window.__nisps.setInputs(t, 1 - t), t);
+      const outputs = await page.evaluate(() => window.__nisps.getOutputs());
+      expect(outputs).toHaveLength(126);
+      for (const v of outputs) {
+        expect(v).toBeGreaterThanOrEqual(0);
+        expect(v).toBeLessThanOrEqual(1);
+      }
+    }
+  });
+});

tests/e2e/ml-engine.spec.js

@@ -0,0 +1,170 @@
+/**
+ * ML engine sanity tests — verify the WASM IML behaves correctly:
+ *  - outputs are always bounded [0, 1]
+ *  - randomize produces different outputs
+ *  - thumbs-up captures the current rawParamValues as the training label
+ *  - training completes and produces a finite loss
+ *  - thumbs-down moves weights and changes outputs
+ *  - async training (triggered by thumbs-up) updates the status line
+ */
+const { test, expect } = require('@playwright/test');
+const { loadApp, statusText } = require('./helpers');
+
+// Two contrasting examples with known inputs and all-low / all-high targets.
+const EXAMPLE_LOW  = { input: [0.1, 0.9], output: new Array(126).fill(0.1) };
+const EXAMPLE_HIGH = { input: [0.9, 0.1], output: new Array(126).fill(0.9) };
+
+test.describe('ML engine (WASM IML)', () => {
+  test('probe is exposed after WASM init', async ({ page }) => {
+    await loadApp(page);
+    const probe = await page.evaluate(() => typeof window.__nisps);
+    expect(probe).toBe('object');
+  });
+
+  test('initial outputs are all in [0, 1]', async ({ page }) => {
+    await loadApp(page);
+    const outputs = await page.evaluate(() => window.__nisps.getOutputs());
+    expect(outputs).toHaveLength(126);
+    for (const v of outputs) {
+      expect(v).toBeGreaterThanOrEqual(0);
+      expect(v).toBeLessThanOrEqual(1);
+    }
+  });
+
+  test('initial state is 0 examples, untrained', async ({ page }) => {
+    await loadApp(page);
+    const count = await page.evaluate(() => window.__nisps.getExampleCount());
+    expect(count).toBe(0);
+    const loss = await page.evaluate(() => window.__nisps.getLoss());
+    expect(loss).toBeNull();
+  });
+
+  test('randomize changes outputs', async ({ page }) => {
+    await loadApp(page);
+    const before = await page.evaluate(() => window.__nisps.getOutputs());
+    await page.evaluate(() => window.__nisps.randomise());
+    const after = await page.evaluate(() => window.__nisps.getOutputs());
+    const anyChanged = before.some((v, i) => Math.abs(v - after[i]) > 0.001);
+    expect(anyChanged).toBe(true);
+  });
+
+  test('thumbs-up increments example count by 1', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(() => window.__nisps.thumbsUp());
+    // Give async training a moment to start but we only need to check example count
+    await page.waitForTimeout(100);
+    const count = await page.evaluate(() => window.__nisps.getExampleCount());
+    expect(count).toBe(1);
+  });
+
+  test('thumbs-up captures current input position and all 126 output values', async ({ page }) => {
+    await loadApp(page);
+    // Set a known joystick position via the probe
+    await page.evaluate(() => window.__nisps.setInputs(0.25, 0.75));
+    await page.evaluate(() => window.__nisps.thumbsUp());
+    await page.waitForTimeout(100);
+
+    const { features, labels } = await page.evaluate(() => ({
+      features: window.__nisps.iml.dataset.features,
+      labels:   window.__nisps.iml.dataset.labels,
+    }));
+
+    expect(features).toHaveLength(1);
+    expect(labels).toHaveLength(1);
+
+    // Input dimension = 2 (joystick x/y, pipeline-processed)
+    expect(features[0]).toHaveLength(2);
+    // The input pipeline may transform values; inputs must stay in [0, 1]
+    expect(features[0][0]).toBeGreaterThanOrEqual(0);
+    expect(features[0][0]).toBeLessThanOrEqual(1);
+    expect(features[0][1]).toBeGreaterThanOrEqual(0);
+    expect(features[0][1]).toBeLessThanOrEqual(1);
+
+    // Labels = all 126 output values, captured from rawParamValues at click time
+    expect(labels[0]).toHaveLength(126);
+    for (const v of labels[0]) {
+      expect(v).toBeGreaterThanOrEqual(0);
+      expect(v).toBeLessThanOrEqual(1);
+    }
+  });
+
+  test('sync train() returns a finite non-negative loss', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(([low, high]) => {
+      window.__nisps.iml.addExample(low.input, low.output);
+      window.__nisps.iml.addExample(high.input, high.output);
+    }, [EXAMPLE_LOW, EXAMPLE_HIGH]);
+
+    const loss = await page.evaluate(() => window.__nisps.train());
+    expect(typeof loss).toBe('number');
+    expect(isFinite(loss)).toBe(true);
+    expect(loss).toBeGreaterThanOrEqual(0);
+  });
+
+  test('training with contrasting examples produces a lower loss than initial', async ({ page }) => {
+    await loadApp(page);
+    // Initial inference — loss is null (never trained), so randomise to get a baseline
+    await page.evaluate(() => window.__nisps.randomise());
+
+    await page.evaluate(([low, high]) => {
+      window.__nisps.iml.addExample(low.input, low.output);
+      window.__nisps.iml.addExample(high.input, high.output);
+    }, [EXAMPLE_LOW, EXAMPLE_HIGH]);
+
+    const loss1 = await page.evaluate(() => window.__nisps.train());
+    const loss2 = await page.evaluate(() => window.__nisps.train());
+
+    // Second training run on same data should converge further (loss2 <= loss1)
+    expect(loss2).toBeLessThanOrEqual(loss1 + 1e-6);
+  });
+
+  test('status line reflects example count and loss after training', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(([low, high]) => {
+      window.__nisps.iml.addExample(low.input, low.output);
+      window.__nisps.iml.addExample(high.input, high.output);
+      window.__nisps.train();
+    }, [EXAMPLE_LOW, EXAMPLE_HIGH]);
+
+    // updateStatus() is called inside trainModel()
+    const text = await page.locator('#status-text').textContent();
+    expect(text).toContain('2 examples');
+    expect(text).toContain('loss');
+  });
+
+  test('thumbs-down changes outputs (weight noise)', async ({ page }) => {
+    await loadApp(page);
+    const before = await page.evaluate(() => window.__nisps.getOutputs());
+    await page.evaluate(() => window.__nisps.thumbsDown());
+    const after = await page.evaluate(() => window.__nisps.getOutputs());
+    const anyChanged = before.some((v, i) => Math.abs(v - after[i]) > 0.0001);
+    expect(anyChanged).toBe(true);
+  });
+
+  test('async training via thumbs-up button updates status with loss', async ({ page }) => {
+    await loadApp(page);
+    await page.click('#btn-thumbsup');
+    // Wait for the async training to complete and status to update
+    await page.waitForFunction(
+      () => document.getElementById('status-text').textContent.includes('loss'),
+      { timeout: 15_000 }
+    );
+    const text = await page.locator('#status-text').textContent();
+    expect(text).toContain('1 example');
+    expect(text).toContain('loss');
+  });
+
+  test('clear examples resets to 0 and marks untrained', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(([low]) => {
+      window.__nisps.iml.addExample(low.input, low.output);
+      window.__nisps.train();
+    }, [EXAMPLE_LOW]);
+    expect(await page.evaluate(() => window.__nisps.getExampleCount())).toBe(1);
+
+    await page.evaluate(() => window.__nisps.clearExamples());
+    expect(await page.evaluate(() => window.__nisps.getExampleCount())).toBe(0);
+    const text = await page.locator('#status-text').textContent();
+    expect(text).toContain('0 examples');
+  });
+});

tests/e2e/persistence.spec.js

@@ -0,0 +1,118 @@
+/**
+ * State persistence tests — localStorage round-trip and URL param application.
+ */
+const { test, expect } = require('@playwright/test');
+const { loadApp, statusText } = require('./helpers');
+
+const STORAGE_KEY = 'nisps-a-immersive';
+
+test.describe('State persistence', () => {
+  test('fresh load with no localStorage starts untrained', async ({ page }) => {
+    await loadApp(page); // helpers.js clears localStorage before load
+    const text = await statusText(page);
+    expect(text).toContain('0 examples');
+    expect(text).toContain('untrained');
+  });
+
+  test('URL ?preset=beginner-1 applies preset on load', async ({ page }) => {
+    await loadApp(page, '&preset=beginner-1');
+    const selected = await page.locator('#synth-preset-select').inputValue();
+    expect(selected).toBe('beginner-1');
+  });
+
+  test('URL ?preset=advanced-2 applies advanced preset', async ({ page }) => {
+    await loadApp(page, '&preset=advanced-2');
+    const selected = await page.locator('#synth-preset-select').inputValue();
+    expect(selected).toBe('advanced-2');
+  });
+
+  test('URL ?spread=0 is accepted without crash', async ({ page }) => {
+    await loadApp(page, '&spread=0');
+    // App should be functional — probe must still exist
+    const probe = await page.evaluate(() => typeof window.__nisps);
+    expect(probe).toBe('object');
+  });
+
+  test('URL ?spread=1 is accepted without crash', async ({ page }) => {
+    await loadApp(page, '&spread=1');
+    const probe = await page.evaluate(() => typeof window.__nisps);
+    expect(probe).toBe('object');
+  });
+
+  test('examples + weights persist across reload via localStorage', async ({ page }) => {
+    // Navigate fresh — no init script so localStorage is untouched between loads.
+    // Suppress the help overlay by pre-setting nisps-help-seen via evaluate.
+    await page.goto(`/a-immersive.html?debug=1`);
+    await page.evaluate(() => localStorage.removeItem('nisps-a-immersive'));
+    await page.evaluate(() => localStorage.setItem('nisps-help-seen', '1'));
+    await page.waitForFunction(() => window.__nisps !== undefined, { timeout: 20_000 });
+
+    // Write 2 examples directly to localStorage in the app's save format.
+    await page.evaluate(([key]) => {
+      const state = {
+        features: [[0.1, 0.9], [0.9, 0.1]],
+        labels:   [new Array(126).fill(0.2), new Array(126).fill(0.8)],
+        handFeatures: [],
+        handLabels:   [],
+        noiseLevel:   0.05,
+        outputMode:   'visual',
+        inputMode:    'joystick',
+        joyX: 0.5, joyY: 0.5,
+        groupOverrides:    null,
+        visualOverrides:   null,
+        midiCCOverrides:   null,
+        audioCanvasState:  null,
+        synthPresetId:     null,
+      };
+      localStorage.setItem(key, JSON.stringify(state));
+    }, [STORAGE_KEY]);
+
+    // Reload — no addInitScript registered, so localStorage is preserved.
+    await page.reload();
+    await page.waitForFunction(() => window.__nisps !== undefined, { timeout: 20_000 });
+
+    // loadState() runs sync training; wait for status to show a loss value.
+    await page.waitForFunction(
+      () => document.getElementById('status-text').textContent.includes('loss'),
+      { timeout: 15_000 }
+    );
+
+    const count = await page.evaluate(() => window.__nisps.getExampleCount());
+    expect(count).toBe(2);
+    const text = await statusText(page);
+    expect(text).toContain('2 examples');
+    expect(text).toContain('loss');
+  });
+
+  test('saveState probe writes valid JSON to localStorage', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(() => window.__nisps.saveState());
+
+    const raw = await page.evaluate(([key]) => localStorage.getItem(key), [STORAGE_KEY]);
+    expect(raw).not.toBeNull();
+
+    const state = JSON.parse(raw);
+    expect(Array.isArray(state.features)).toBe(true);
+    expect(Array.isArray(state.labels)).toBe(true);
+    expect(typeof state.outputMode).toBe('string');
+    expect(typeof state.noiseLevel).toBe('number');
+  });
+
+  test('localStorage state is overwritten on next saveState call', async ({ page }) => {
+    await loadApp(page);
+    await page.evaluate(() => window.__nisps.saveState());
+
+    const before = await page.evaluate(([key]) => localStorage.getItem(key), [STORAGE_KEY]);
+
+    // Add an example and save again
+    await page.evaluate(() => {
+      window.__nisps.iml.addExample([0.3, 0.7], new Array(126).fill(0.5));
+      window.__nisps.saveState();
+    });
+
+    const after = await page.evaluate(([key]) => localStorage.getItem(key), [STORAGE_KEY]);
+    // State changed — the feature array should now include our example
+    const parsed = JSON.parse(after);
+    expect(parsed.features.length).toBeGreaterThan(0);
+  });
+});

tests/e2e/ui-interactions.spec.js

@@ -0,0 +1,176 @@
+/**
+ * UI state machine tests — drawers, mode switching, preset chips, keyboard shortcuts.
+ */
+const { test, expect } = require('@playwright/test');
+const { loadApp, statusText } = require('./helpers');
+
+test.describe('UI interactions', () => {
+  test.describe('Dock drawers', () => {
+    test('Train dock button opens training drawer', async ({ page }) => {
+      await loadApp(page);
+      await expect(page.locator('#drawer-training')).toHaveClass(/hidden/);
+      await page.click('[data-drawer="training"]');
+      await expect(page.locator('#drawer-training')).not.toHaveClass(/hidden/);
+    });
+
+    test('Drawer close button hides the drawer', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="training"]');
+      await expect(page.locator('#drawer-training')).not.toHaveClass(/hidden/);
+      await page.click('#drawer-training .drawer-close');
+      await expect(page.locator('#drawer-training')).toHaveClass(/hidden/);
+    });
+
+    test('Mode dock button opens mode drawer', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="mode"]');
+      await expect(page.locator('#drawer-mode')).not.toHaveClass(/hidden/);
+    });
+
+    test('Engine dock button opens engine drawer', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="params"]');
+      await expect(page.locator('#drawer-params')).not.toHaveClass(/hidden/);
+    });
+
+    test('multiple drawers can be open simultaneously', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="training"]');
+      await page.click('[data-drawer="mode"]');
+      // The app doesn't auto-close drawers — both can be open
+      await expect(page.locator('#drawer-training')).not.toHaveClass(/hidden/);
+      await expect(page.locator('#drawer-mode')).not.toHaveClass(/hidden/);
+    });
+  });
+
+  test.describe('Output mode switching', () => {
+    test('default mode is visual — synth quick controls hidden', async ({ page }) => {
+      await loadApp(page);
+      await expect(page.locator('#synth-quick-controls')).toHaveClass(/hidden/);
+      await expect(page.locator('#midi-cc-quick-controls')).toHaveClass(/hidden/);
+    });
+
+    test('switching to Synth shows synth quick controls', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="mode"]');
+      await page.click('[data-mode="synth"]');
+      await expect(page.locator('#synth-quick-controls')).not.toHaveClass(/hidden/);
+    });
+
+    test('switching to MIDI CC shows midi-cc quick controls', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="mode"]');
+      await page.click('[data-mode="midi-cc"]');
+      await expect(page.locator('#midi-cc-quick-controls')).not.toHaveClass(/hidden/);
+    });
+
+    test('switching back to Visual hides synth controls', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="mode"]');
+      await page.click('[data-mode="synth"]');
+      await page.click('[data-mode="visual"]');
+      await expect(page.locator('#synth-quick-controls')).toHaveClass(/hidden/);
+    });
+
+    test('heatmap shows 20 bars in visual mode', async ({ page }) => {
+      await loadApp(page);
+      const count = await page.locator('.heatmap-cell').count();
+      expect(count).toBe(20);
+    });
+
+    test('heatmap shows 126 bars in synth mode', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="mode"]');
+      await page.click('[data-mode="synth"]');
+      const count = await page.locator('.heatmap-cell').count();
+      expect(count).toBe(126);
+    });
+  });
+
+  test.describe('Status line', () => {
+    test('shows 0 examples · untrained on fresh load', async ({ page }) => {
+      await loadApp(page);
+      const text = await statusText(page);
+      expect(text).toContain('0 examples');
+      expect(text).toContain('untrained');
+    });
+  });
+
+  test.describe('Training controls', () => {
+    test('Randomize button changes heatmap bar widths', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="training"]');
+
+      const widthsBefore = await page.evaluate(() =>
+        Array.from(document.querySelectorAll('.heatmap-cell-bar')).map(el => el.style.width)
+      );
+      await page.click('#btn-randomize');
+      const widthsAfter = await page.evaluate(() =>
+        Array.from(document.querySelectorAll('.heatmap-cell-bar')).map(el => el.style.width)
+      );
+
+      const anyChanged = widthsBefore.some((w, i) => w !== widthsAfter[i]);
+      expect(anyChanged).toBe(true);
+    });
+
+    test('preset chip loads examples and trains', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="training"]');
+      await page.click('[data-preset="calm-to-chaotic"]');
+
+      // calm-to-chaotic has 3 examples; training is async
+      await page.waitForFunction(
+        () => document.getElementById('status-text').textContent.includes('loss'),
+        { timeout: 15_000 }
+      );
+      const text = await statusText(page);
+      expect(text).toContain('3 examples');
+      expect(text).toContain('loss');
+    });
+
+    test('Clear Ex resets example count to 0', async ({ page }) => {
+      await loadApp(page);
+      await page.click('[data-drawer="training"]');
+      await page.click('[data-preset="calm-to-chaotic"]');
+      await page.waitForFunction(
+        () => document.getElementById('status-text').textContent.includes('loss'),
+        { timeout: 15_000 }
+      );
+
+      await page.click('#btn-clear-examples');
+      const text = await statusText(page);
+      expect(text).toContain('0 examples');
+    });
+  });
+
+  test.describe('Keyboard shortcuts', () => {
+    test('key 2 triggers thumbs-up (adds example)', async ({ page }) => {
+      await loadApp(page);
+      const before = await page.evaluate(() => window.__nisps.getExampleCount());
+      await page.keyboard.press('2');
+      await page.waitForTimeout(200);
+      const after = await page.evaluate(() => window.__nisps.getExampleCount());
+      expect(after).toBe(before + 1);
+    });
+
+    test('key 1 triggers thumbs-down (changes outputs)', async ({ page }) => {
+      await loadApp(page);
+      const before = await page.evaluate(() => window.__nisps.getOutputs());
+      await page.keyboard.press('1');
+      const after = await page.evaluate(() => window.__nisps.getOutputs());
+      const anyChanged = before.some((v, i) => Math.abs(v - after[i]) > 0.0001);
+      expect(anyChanged).toBe(true);
+    });
+
+    test('key Z triggers undo after thumbs-down', async ({ page }) => {
+      await loadApp(page);
+      const before = await page.evaluate(() => window.__nisps.getWeights());
+      await page.keyboard.press('1'); // thumbs-down
+      await page.keyboard.press('z'); // undo
+      const after = await page.evaluate(() => window.__nisps.getWeights());
+      // Weights should be restored (approximately)
+      const maxDiff = before.reduce((m, v, i) => Math.max(m, Math.abs(v - after[i])), 0);
+      expect(maxDiff).toBeLessThan(1e-4);
+    });
+  });
+});