mirror of
https://github.com/hanzo-apps/world.git
synced 2026-10-30 14:53:09 +00:00
zooqueenandHanzo Dev
97834743ae
Two tiles on world.hanzo.ai named a routing strategy the flywheel never built. "learned-engine share" (router-stats) and "engine-routed" (the enso ledger) both counted decisions whose source was an external /route engine, so both read 0 permanently — the dashboard said the learned router loses every request while cost-saving ran at 77%. The router-stats tile follows the gateway to learned_share: traffic served by the tuned prefer table rather than the static seed. The ledger tile now folds a bySource histogram and shows the LEADING source with its share; the old two-way engine/heuristic switch also dropped explore, override, explicit and family rows on the floor, under-reporting the total it split. Co-authored-by: Hanzo Dev <[email protected]>
62 lines
2.9 KiB
TypeScript
62 lines
2.9 KiB
TypeScript
import { expect, test } from '@playwright/test';
|
|
|
|
// Enso Live Training must reflect REAL models (the served catalog), never the opaque
|
|
// upstream "arm-N" routing mix that read as random/meaningless amounts. Both feeds are
|
|
// stubbed with real-shaped payloads; the router-stats stub deliberately includes an
|
|
// arm-N by_model map to prove it is NOT rendered anymore.
|
|
|
|
const ENSO = '.panel[data-panel="enso-training"]';
|
|
|
|
const ROUTER_STATS = {
|
|
scope: 'platform',
|
|
window: { since: '2026-07-17T00:00:00Z', until: '2026-07-18T00:00:00Z', events: 12345 },
|
|
cost: { saved_pct: 0.23, cumulative_saved_index: 98765, baseline_model: 'x', priced_events: 100 },
|
|
quality: { reward_rate: 0.5, rewarded_events: 10, learned_share: 0.62, avg_confidence: 0.8, shadow_agreement: 0.91 },
|
|
by_task: {},
|
|
by_model: { 'arm-1': 100, 'arm-2': 50, 'arm-3': 25 }, // opaque arms — must NOT render
|
|
throughput: { per_hour: [1, 2, 3, 4, 5, 6], total_window: 21 },
|
|
retrain: {
|
|
version: 'v9', trained_time: '2026-07-17T12:00:00Z', events: 100, gate_passed: true,
|
|
published: true, gate_kind: 'reward', gate_metric: 'rate', gate_value: 0.5, gate_base: 0.4, note: '',
|
|
},
|
|
};
|
|
|
|
const CLOUD_MODELS = {
|
|
updatedAt: '2026-07-18T00:00:00Z',
|
|
totalModels: 42,
|
|
zenModels: 8,
|
|
cloudRegions: 5,
|
|
cloudPlans: 3,
|
|
families: ['Zen', 'Fable'],
|
|
models: [
|
|
{ id: 'zen5', name: 'Zen 5', provider: 'hanzo', tier: 'flagship', context: 200000, inPrice: 3, outPrice: 15 },
|
|
{ id: 'zen5-flash', name: 'Zen 5 Flash', provider: 'hanzo', tier: 'fast', context: 128000, inPrice: 0.5, outPrice: 1.5 },
|
|
{ id: 'fable5', name: 'Fable 5', provider: 'hanzo', tier: 'premium', context: 400000, inPrice: 5, outPrice: 20 },
|
|
],
|
|
};
|
|
|
|
test.describe('Enso Live Training — real models only', () => {
|
|
test.use({ viewport: { width: 1440, height: 900 } });
|
|
|
|
test('renders the real served-model catalog, never opaque arm-N amounts', async ({ page }) => {
|
|
await page.route('**/v1/world/cloud/router-stats*', (r) => r.fulfill({ json: ROUTER_STATS }));
|
|
await page.route('**/v1/world/cloud/models', (r) => r.fulfill({ json: CLOUD_MODELS }));
|
|
|
|
// Enso Live Training is a Cloud-flagship panel (added only in the cloud
|
|
// variant); ?variant=cloud wins over the build-time VITE_VARIANT at runtime.
|
|
await page.goto('/?variant=cloud');
|
|
await page.waitForSelector(ENSO, { timeout: 45000 });
|
|
const panel = page.locator(ENSO);
|
|
|
|
// Real model names from the catalog appear.
|
|
await expect(panel).toContainText('Zen 5 Flash', { timeout: 15000 });
|
|
await expect(panel).toContainText('Fable 5');
|
|
await expect(panel).toContainText('models served');
|
|
// Real aggregates still render (unchanged, real telemetry).
|
|
await expect(panel).toContainText('cost saved');
|
|
// The opaque per-arm amounts are GONE — no meaningless "Enso arm N" rows.
|
|
await expect(panel).not.toContainText('Enso arm');
|
|
await expect(panel).not.toContainText('arm-1');
|
|
});
|
|
});
|