Research Guides
Keep Conditional Source Checks Out of the Failure Count
A research monitor checks a source again and receives HTTP 304. If its export labels every non-200 response as a broken page, the resulting failure count can be misleading. If it labels the response as a newly downloaded article, the evidence trail is misleading in a different way.
This worksheet separates a conditional check from a new content capture. It is a manual review method for an authorized research workflow, not a description of a monitoring feature offered by Colquhoun.shop.
1. Classify the response before counting it
MDN explains that a 304 Not Modified response to a conditional GET or HEAD means the resource does not need to be retransmitted. The response has no body; the client can use its valid cached representation. Conditional headers include If-None-Match and If-Modified-Since. MDN: 304 Not Modified
For a research ledger, create separate outcomes for a newly retrieved representation, a validated existing representation, an access restriction and an unsuccessful request. These are editorial record categories, not a replacement for the actual HTTP status. Keep both fields.
Do not turn a validation result into a claim that the source's facts remain correct. A price, deadline or policy can need independent verification even when the retrieved page has not changed.
2. Pair the check with the content it refers to
Assign each saved capture a stable internal identifier. Record which capture a later conditional request is validating, along with the requested address, request time and validator used. Keep credentials and private header values out of shared worksheets.
If the associated capture is missing, record that gap. A status-only row does not supply text that another reviewer can inspect. Obtain an authorized content capture before quoting the page, rather than reconstructing a missing body from memory.
3. Work through an invented monitoring export
A fictional editor reviews 20 source checks. Twelve returned new bodies, six returned 304 responses and two could not be completed. A report that says eight sources failed combines two different outcomes. A report that says 18 new articles were downloaded is also inaccurate.
The editor instead reports 12 new body retrievals, six conditional validations and two unresolved checks. Those figures describe requests, not necessarily 20 unique publishers or 20 independent pieces of evidence.
During review, one of the six validation rows points to a deleted local capture. The editor retains the observed 304 but marks the evidence attachment as missing. The network result and the research-record defect are separate facts, so repairing one must not erase the other.
4. Copy this validation ledger
- Source: requested URL and the claim being checked.
- Request: method, check time and relevant non-secret conditional header.
- Response: actual status and selected response metadata.
- Capture: identifier and original retrieval time of the body used for review.
- Evidence status: attached, missing or not yet inspected.
- Editorial action: retain citation, inspect again or investigate a discrepancy.
Keep “last checked” separate from “body captured.” Otherwise, a routine validation can make an old saved document look like a newly collected one. Neither timestamp should automatically replace the publisher's stated publication or revision date.
5. Test the handoff, not just the counter
Ask a colleague to open one newly retrieved record and one conditionally validated record. Can they locate the actual source text, understand when it was captured and see which later check relied on it? If not, improve the record before circulating a research summary.
Finish with a bounded conclusion: which requests completed, which content remains inspectable and which claims still need review. Correct classification makes a monitoring report more useful, but it does not establish source accuracy, successful indexing or search performance.