Feedback

Quality reports and feature requests use the same authenticated feedback service, but carry different information.

Every successful submission returns an opaque report ID. The CLI and MCP confirmations print it as Receipt:, while Python exposes it as FeedbackResponse.report_id. The ID names the submission for later triage; it is not a URL or a report-reading endpoint.

Quality reports

Report what you observed, not what you think caused it. Choose the earliest visible symptom; repeat --reason when more than one label independently describes the result:

analog feedback quality https://example.com/page \
  --reason under_rendered \
  --reason poor_field_naming \
  --note "the final products are missing; prices landed under text_3"

A quality report requires the page’s full HTTP(S) URL and at least one reason label. The note is optional; use it for a short “expected X, got Y” explanation. There is no saved-result handle field.

Page access and rendering

undetected_block

The page was a bot challenge or block that Analog treated as content. A block Analog already identifies is explicit and needs no report.

under_rendered

Content visible on the page is missing from the fetched markdown.

garbled_markdown

The markdown splits characters apart, runs words together, or puts content in the wrong order.

Missing, extra, or misplaced records

no_records_found

The page plainly shows records, but none came back.

under_extraction

Some records visible on the page are missing.

over_extraction

The result contains more records than the page does.

wrong_content_extracted

The records came from the wrong part of the page.

page_chrome_in_records

Navigation, footer, menu, or other page controls appeared as data.

Fields and values

poor_field_naming

Fields have generic names such as text_2.

misfielded_values

Values landed under the wrong fields.

Anything else

other

None of the labels above fit. Include a note explaining what you observed.

Feature requests

Feature requests are free text. They carry no URL and no reason labels:

analog feedback request "diff two handles as a table"

MCP and Python

The MCP server exposes the same two report shapes through analog_feedback. Its input schema enumerates every accepted label:

{
  "kind": "extraction_quality",
  "url": "https://example.com/page",
  "labels": ["under_rendered", "poor_field_naming"],
  "note": "the final products are missing; prices landed under text_3"
}

From Python, submit a validated FeedbackRequest:

from analog import Client, FeedbackRequest

report = FeedbackRequest(
    kind="extraction_quality",
    url="https://example.com/page",
    labels=["under_rendered", "poor_field_naming"],
    note="the final products are missing; prices landed under text_3",
)
receipt = Client().submit_feedback(report)
print(receipt.report_id)

For a feature request, use FeedbackRequest(kind="feature_request", note="…").

Privacy

Extraction URLs are never logged. A quality report includes the URL you explicitly submit; it never includes page content. A quality-report payload contains only the report kind, URL, labels, optional note, and label-set version. It has no fields for page HTML, markdown, records, or a saved-result handle.