Search, by meaning.
Sourcipedia's public search is semantic. Type what you mean — a term, a phrase, or a plain-language question — and the system returns the cards that best match the meaning, not just the words.
§ 01How it works
The public search surface uses vector retrieval. Your query is embedded into the same space as the cards in the corpus; the closest cards by cosine similarity are returned, ranked. There is no keyword-only mode toggle and no separate boolean operator language on the public surface.
That choice keeps the experience simple: anyone can type any phrasing and get a sensible answer, without having to learn a query syntax.
§ 02What you can type
Single terms
"photosynthesis", "GDP", "cassini-huygens". A bare term resolves to the closest card about that topic.
Noun phrases
"grid-scale battery storage", "deep ocean carbon flux", "post-quantum key exchange". A multi-word phrase pins the query to a more specific intent than a single word usually can.
Plain-language questions
"What is the difference between mass and weight?", "How do reverse repos work?", "Why is the sky blue?". Questions are embedded the same way as topics; you do not need to drop the question mark or reformat.
§ 03Reading the results
Search results land at /search?q=…. Each row is a candidate card and shows:
- The card title.
- A short snippet of the summary.
- A relevance signal — how closely the card matched your query.
Results are ordered by similarity, top to bottom. The closest match is first; relevance falls off down the list. You can click any result to open the article at its permanent URL.
§ 04What ranking actually does
Vector ranking measures semantic distance between your query and each card. A card that uses different words but describes the same topic can still rank highly — that is the point. A card that shares vocabulary with your query but talks about something else will rank lower, even if it would have scored well on a pure keyword match.
This is also why search behaves well across paraphrases. "How does an MRI work" and "principle of magnetic resonance imaging" land near the same card; you do not need to discover the canonical title to find it.
§ 05When nothing matches
If no card in the corpus is close enough to your query, the result set is empty. Two things to try, in order:
- Rephrase. The search is forgiving across paraphrases, but it cannot guess a topic it has no signal for. A different way of asking often surfaces an existing card you would not otherwise have found.
- Come back. The auto-generator observes demand. Topics that get queried and have no card tend to be picked up for generation on a later cycle.
If you need the answer right now, run the same prompt through ResearchAnything.ai. A research run will go out to the open web and build a defensible report — and the sources it admits will feed back into the corpus that Sourcipedia draws from.
§ 06Public vs admin surfaces
Everything described on this page is the public, no-account surface at sourcipedia.com. Administrators of a Sourcipedia deployment have additional tools — corpus introspection, audit logs, the ability to inspect why a card landed where it did — but those are not exposed to readers. The reader experience is intentionally a single input box and a ranked list of results.
§ 07Where to go next
- Look up something now → sourcipedia.com
- Understand the structure behind a result → Articles & sources
- Run a long-form query instead → ResearchAnything quickstart