The short answer
Store the claim and its supporting passage, with enough context to check the interpretation later.
Capture sources, distinguish quotations from interpretations, and keep claim-level evidence so your research library supports useful answers instead of plausible summaries.Keep the evidence behind the bookmark
A bookmark preserves a location. It may lose the passage you relied on, the version you read or the conditions attached to a result. A summary can lose the same evidence when it drops the original qualifications.
Build the library around questions you expect to revisit. For each consequential claim, retain source identity and a precise location. Preserve only material you are authorized to store, and keep attribution attached to excerpts.
Separate three kinds of note
| Note type | Example from a fictional study | Review question |
|---|---|---|
| Source statement | The method was evaluated on 40 short documents. | Does the passage say this exactly? |
| Interpretation | The evidence covers a narrow document set. | Does that follow from the method and limitations? |
| Hypothesis | The method may struggle on our scanned archive. | What test would support or reject it? |
Label the inference so the next reader can distinguish your reasoning from the study's results. Otherwise, a later summary can turn a hypothesis into something the study supposedly found.
Create a claim card with a usable source trail
Question: What does this source establish?
Claim: One bounded statement
Type: Source statement / interpretation / hypothesis
Source: Author, title and direct URL or approved file
Location: Section, page, table or passage
Version and access date: Record what was actually read
Evidence: Short excerpt or faithful paraphrase
Limitations: Population, method, scope and uncertainty
Counterevidence: Related sources or unresolved disagreement
Next action: Apply, test, revisit or leave unresolvedThis is a proposed research record, not a required Gigabrain schema. W3C's provenance overview provides a formal reference for tracing information and its derivation. A small working library can begin with plain, source-linked records.
Follow a claim back through secondary coverage
If several articles repeat a research claim, find the original paper, dataset or official documentation where possible. Three reports of the same experiment are not three independent experiments. Preserve the original method and distinguish commentary from evidence.
When the source is unavailable, mark the claim as secondhand. Do not invent a quotation or report that you inspected a method you could not access. Say which source you could not check, so the next reader knows where verification stopped.
Keep changes and disagreements visible
A provider can update documentation, and a later study can narrow an earlier result. Record the version you relied on and add the new source as a change, rather than silently rewriting the old note. Explain which claim is affected.
To test the library, ask an agent for a short answer with one supporting source, one limitation and one unresolved question. Open the passages yourself. Check whether the answer distinguishes the study's observation from your proposed application.
Turn the library into a working system
Start with five sources around a real project question and use the research-source template. Five sources chosen for one question are easier to check than a folder of unrelated bookmarks.
Gigabrain can connect source-linked knowledge across later tasks. See the architecture guide and product evidence. Compare packages if you want the underlying system and supported workflows for maintaining that collection.
