Research Guides
Record Request Variants Before Comparing Two Page Captures
Two researchers open the same address and save different page text. One concludes that the publisher changed its claims. The other suspects a broken capture. Before choosing either explanation, check whether the requests could have selected different representations of the resource.
This worksheet records request context for reproducible website research. It does not claim that Colquhoun.AI exposes request-header controls or automatically diagnoses content negotiation.
1. Read Vary as a clue about selection
MDN's Vary reference explains that the response header identifies request fields, beyond the method and URL, that influenced the response. Caches use the named fields when distinguishing responses. The documented guidance calls for consistent Vary values across responses for a URL, including its default and 304 responses.
That is useful context, not a complete explanation of every visible difference. A header observation alone does not prove which version a particular reader saw or establish that a cache served the correct one.
2. Record the capture conditions
Keep the exact requested address, final address after redirects, request time and method. Record relevant non-secret request preferences, the response status and the observed Vary value. Save the body or a permitted excerpt needed to support the specific research claim.
Do not put authentication tokens, session cookies or private account content into a shared worksheet. Describe the access context at a safe level, such as public signed-out view, rather than exporting a browser's entire request history.
3. Compare a fictional language variant
Imagine a fictional documentation page at one public URL. A request expressing an English language preference receives English instructions, while a request expressing a French preference receives French instructions. Both responses declare Vary: Accept-Language.
The reviewer should not describe this observation as proof that the publisher rewrote the page between captures. They first compare like with like: the same URL and language preference, with other recorded conditions held as consistent as practical.
If the two language versions make materially different claims, that becomes a separate editorial question. The capture record helps identify exactly which version supports each quotation instead of merging the texts into one supposedly universal source.
4. Change one condition deliberately
For a permitted diagnostic check, define a small comparison in advance. Use public or otherwise authorized access and vary only the relevant documented preference. Keep the request volume low and respect service limits.
If the tool cannot expose or control that preference, record the limitation. Do not invent a header value from the language shown on screen. Do not use a different account, location or device without noting that the comparison now includes additional differences.
A missing Vary header also does not establish that every visitor receives identical content. Keep the finding narrow: this response did not declare a variation through that header.
5. Copy this representation-comparison card
- Research claim: the passage or fact being checked.
- Address: requested URL and final destination.
- Capture: timestamp, method and safe access context.
- Selection context: relevant observed request preferences.
- Response: status, Vary value and retained evidence.
- Controlled difference: the single intended change between captures.
- Observation: what actually changed in the returned content.
- Uncertainty: settings, intermediaries or timing effects not established.
Use this record when handing evidence to another reviewer. A URL remains essential, but the extra context can explain why opening it later does not immediately reproduce the quoted passage.
6. Separate diagnosis from repair
If a mismatch suggests a configuration defect, pass the evidence to the responsible site owner. Do not prescribe a global cache change from two screenshots. A repair needs its own scoped test, including checks that private responses cannot be mixed between users.
The research outcome is a better account of what was observed, under which conditions, and what still needs verification. It is not a guarantee about all visitors or search crawlers.