Source Health · controlled Source Lab proof
Keep the controlled evidence boundary visible before price reaches a Profile.
Start with the evidence identity, source role, and human action. Fact Contracts, Heal Gate, and Regression Guard remain available as controlled proof underneath.
Controlled Source Lab queue
Attention-required evidence that needs a human
Explore deterministic review examples
These examples teach fail-closed classification. They are not additional live incidents for the watched Source Lab target.
Watched source · Atlas 20B · controlled Source Lab target · ChangeProof Source Lab
Required action: review the source evidence before delivery
The primary representation changed shape and output_price was not observed.
Evidence detail
Layout drift · required price missing
The primary representation changed shape and output_price was not observed.
Why this source needs attention
- REQUIRED FIELD MISSINGRequired field 'output_price' is missing. · output_price
- STRUCTURAL SENTINEL DRIFTWeak structural sentinel 'primary_heading' changed; business truth is evaluated separately. · primary_heading
Choose an explicit disposition. The default is fail-closed review.
Trusted activity · outside the human queue
Stable target changes do not wait for manual confirmation.
When identity and semantic anchors remain stable, a normal trusted target change is delivered as business activity. Only quarantine, drift, or ambiguous evidence belongs in the Source Health queue above.
Controlled Source Lab evidence
Recovery proof stays separate from market-price truth.
This deterministic walkthrough is the controlled Source Lab reliability story, not a Groq or OpenAI price movement. The recorded provider timeline shows real recovery proof separately; these controls replay the guarded state machine without a provider call.
Recorded candidate · local replay
Source Lab template preview
The recorded evidence shows the extraction-repair boundary. This public control only replays whether the candidate may proceed.
ChangeProof Heal Gate
Candidate passes semantic gate
Structural drift is surfaced but does not decide business truth by itself.
Current boundary
This public control replays the approval boundary locally; no provider mutation occurs.All buttons below mutate only local demo state and do not call Bright Data.Technical proof, available when you need it
Recovery is only real after rerun and guard pass.
The final recorded cycle reaches RECOVERED only after a real rerun and two passing Regression Guard cases. Earlier provider limitations remain visible as negative proof; they never become a false price or cost event.