When the page breaks, the product doesn’t.

ThreadScout treats scraper health as product data. Every run has a state, every field has a contract, and every repair leaves evidence.

Trace the healing route

Production network0 live collectors

Contract: 13 fields
Delivery: API + JSON
Schedule: Daily · 03:00
Safety: Fail closed

Custom Scraper Studio endpoints for company-owned public career pages. invideo’s original Collector ID survived the documented repair.

Collector control / no dashboard detour

Pull the newest jobs into the product.

This button drives the real Collector ID from ThreadScout, waits for structured JSON, validates every record and applies one source snapshot atomically. Invalid or empty payloads never replace healthy jobs.

No active collector is available yet.

Same ID / repaired contract

A failure you can follow.

Real create, baseline, validation, heal, approval and re-run artifacts are sanitized under evidence/bright-data/examples.

01
BASELINE · 2 ROLES
Baseline run degraded
2 public roles returned, but description and work-mode keys violated the contract.
02
0 JOBS APPLIED
Safety gate held
Strict validation accepted zero jobs, so healthy downstream data stayed untouched.
03
SAME COLLECTOR · APPROVED
Repair reviewed and saved
The agent proposed a 2-step fix; its preview passed validation before approval.
04
2 ROLES · CONTRACT PASSED
Output restored
The full re-run restored both roles at 83.33% optional-field completeness.

Judge evidence / ready slots

Every claim maps to a file.

CreateCLI prompt + collector envelope
Before / afterSanitized structured output
Safety gateInvalid runs update zero jobs

Agent commandbdata scraper heal c_mt38…

See what the output powers

When the page breaks, the product doesn’t.

Pull the newest jobs into the product.

A failure you can follow.

Baseline run degraded

Safety gate held

Repair reviewed and saved

Output restored