Your scraper healed.
Your data didn't.

Bright Data's AI self-healing rewrote our scraper and reported success. Vanished caught 28 silently corrupted records that no validation would ever surface.

See the experiment
28
records silently corrupted
0
errors thrown
11
findings surfaced
14s
after self-heal completed

Self-healing hides failures.

Your scraper reports success. Your HTTP client returns 200. Your JSON parser passes. But the data is wrong — and no downstream system will ever know.

01

Scraper breaks

Open-Meteo changed their docs page. The scraper's selectors stopped matching. Bright Data's AI self-healing kicks in.

02

AI claims success

The self-healer rewrote selectors, added navigation guards, and reported the scraper as HEALTHY. 69 records returned.

03

Data is wrong

28 records lost forecast_days. Record 0 is now WHO radiation data. No error was thrown. No validation failed.

Three steps. One truth.

01

Snapshot

Capture what your scraper actually returned. Not what it was supposed to return.

02

Compare

Detect field-level semantic changes across snapshots — even when record order is scrambled.

03

Expose

Surface meaningful drift, disappearance, and contradictions. Silence structural noise.

Can you prove your scraper healed correctly?

Vanished compares snapshots before and after self-healing and surfaces exactly what changed — silently.