Research Guides
Keep Transformed Web Captures Separate From Publisher Quotes
A website research capture can succeed without delivering exactly what the publisher's origin server sent. When a response explicitly identifies itself as transformed, keep that information beside the quotation, table or screenshot before another editor treats the capture as the publisher's exact wording.
This is a provenance worksheet for public web research. It does not diagnose a particular intermediary, recommend bypassing access controls or describe a feature of Colquhoun.shop.
1. Read the signal without overstating it
RFC 9110's definition of HTTP 203 describes a successful request whose response content was changed by a transforming proxy from the origin's HTTP 200 response. The status is evidence about how the delivered response was produced. It is not a statement that the publisher is untrustworthy or that every visible sentence changed.
The status also does not tell you what changed. Another response returning 200 is not automatically proof that its content is the origin's unmodified text. Record what each path actually establishes before comparing the two.
2. Mark the claim that needs a second look
Imagine a fictional public research report listing an inventory of 120 pages. A reader-friendly service returns a 203 response containing 120, but the transformed capture does not show the scope note. A separately verified publisher copy states “120 English-language pages in the sampled directory.”
Without the note, an editor might write “120 pages across the entire site.” The number did not need to change for the claim to become misleading. A defensible comparison records the same number in both captures and the missing scope note in the first. It does not speculate about how or why the service omitted it.
Use the verified publisher copy for the exact quotation and keep the transformed copy as evidence of what that route presented. If no publisher copy is available, report that the quotation or scope could not be independently confirmed. Do not imply direct verification from an unconfirmed route.
3. Keep the response path with the evidence
- Question: identify the exact claim, value or passage being checked.
- Capture: record time, public address, method and status.
- Route: note the capture service or intermediary when known, without guessing from the status alone.
- Comparison: identify the independently verified publisher copy, or label it unavailable.
- Difference: record wording, unit, date, notes or links that affect interpretation.
- Decision: quote, paraphrase with limits, or hold the claim.
Retain only evidence you are permitted to store. Use a narrow excerpt or a structural diff when the entire response is not needed. Remove credentials and private identifiers from any research handoff.
4. Keep other explanations in the comparison
Try to match time, language, selected version and other relevant request conditions. If the original capture was made on Monday and the direct publisher copy on Thursday, a legitimate edit is another explanation for a text difference. Avoid attributing all differences to the intermediary unless the evidence distinguishes that cause.
Recheck the consequential passage rather than requiring every byte to match. A formatting change may be immaterial to a summary; a dropped exception, unit or date is not. The accuracy threshold depends on what your article will claim.
5. Close with a claim-level finding
One useful note is: “The transformed capture showed 120 but omitted the scope note. The verified publisher copy limits the figure to English-language pages in the sampled directory, and the published sentence now preserves that limit.” Keep the route and timestamp beside the note so another reviewer can repeat the check.
That demonstrates a verified difference for one piece of evidence. It does not establish how all clients see the page, what Google indexed, or that the intermediary behaves the same way on other pages.