How a search runs, from keyword to next question
Five stages, in the order they happen. Everything below is illustrated with one worked example so you can see the same search change shape as it moves through each one.
sleep apnea
Does treating it actually improve daytime function in adults?
Type the keyword, then say what you actually want to know
A keyword is a handle, not a question. The first screen asks for both, because everything downstream is scored against the question rather than the string.
Enter one keyword or a small cluster of them. Underneath, a single optional line asks what you are trying to find out. You can leave it blank and still get a ranked list, but filling it in is what separates a source that mentions your keyword from a source that answers you.
Three optional filters sit beside it: a date floor, a set of source classes to prefer or exclude, and a language. None of them are required, and none of them silently exclude anything without telling you. If a filter removed sources, the results header says how many and offers to put them back.
Nothing about your phrasing needs to change. There is no query syntax to learn, no operators to memorise and no field codes. If you already search well, the tool starts from what you typed and shows you where it could go further.
What this stage gives you
- A saved search object with your keyword, question, filters and timestamp
- A plain-language echo of what the tool understood you to be asking
- The count of sources retrieved before any ranking is applied
Read a list ordered by two scores, never one
Every source comes back with a relevance score and a trust score, side by side. They are never averaged, because a compelling blog post and a rigorous paper about the wrong question are different problems.
Results arrive as a ranked list. Each row carries the title, where it came from, what kind of source it is, and the two scores as segmented bars you can read at a glance from across a desk.
You choose the sort. Rank by relevance when you are still working out what your question is. Rank by trust when you know the question and need something you can cite. Rank by the lower of the two when you want to see, quickly, whether anything in the list is strong on both counts.
Sources that score high on one axis and low on the other are the interesting ones, so they are marked rather than buried. A row that reads 92 relevance and 31 trust is telling you something useful: this is exactly your question, and the answer on offer is not one you can lean on yet.
What this stage gives you
- A ranked list with two independent scores per source
- Sort by relevance, by trust, or by the weaker of the two
- Split markers on every source that is strong on one axis and weak on the other
Open any score and see what produced it
A score you cannot interrogate is a number you should not trust. Click either bar and the six trust signals open up, each with the evidence that moved it.
The panel shows every signal separately: who is behind the source, what it is built on, whether anyone checked it, how old the claim is against how fast the field moves, who funded it, and whether independent sources agree.
Each signal carries its receipts. Where the score says a source cites primary research, the citations are listed and linked. Where it says the claim is contradicted, the contradicting source is named. Where a signal could not be assessed, it says so and is excluded from the total rather than being quietly scored as average.
You can disagree. Any signal can be overridden for your workspace, with a note explaining why, and the override travels with the search when you export or share it. Your judgement is the last step, not the tool.
What this stage gives you
- Six trust signals and four relevance signals, each scored separately
- Named evidence behind every signal, linked where a link exists
- A per-workspace override with a reason, carried through exports
See what your results never answer
The gap map is the part that changes how people work. It reads your question against the whole result set and marks the sub-questions nothing in it addresses.
A good result list feels complete, which is precisely the risk. The gap map breaks your question into its parts and checks each one against every source you were given, then reports the parts that came back empty or thin.
Gaps are typed, not lumped together, because the fix is different for each. A coverage gap means nothing addresses it and you need a new query. An evidence gap means the topic is covered but only by opinion. A currency gap means your sources predate the last change in the field. A population gap means the evidence exists but is about somebody else. A disagreement means two credible sources contradict each other and nothing resolves it.
Every gap is a claim about your result set, so every gap is checkable. Open one and it lists which sources it looked at and why each fell short.
What this stage gives you
- Your question broken into its component sub-questions
- Each sub-question marked covered, thin, contested or absent
- The reasoning behind each gap, with the sources it examined
Take a rewritten query, or ignore it
Every gap produces a query that would close it. Each one arrives with the change it makes and the reason it makes it, so you can judge the suggestion instead of just running it.
Suggestions are written as queries you could have typed yourself. No proprietary syntax, no black box: a rewritten search string, the move it represents, and the sentence explaining why this search rather than yours.
Run one and it opens as a new search alongside the original, not on top of it. You keep both, you can see what changed between them, and you can go back. Research is not linear, and losing the previous result set to try an idea is a good way to stop trying ideas.
Refinements you run are kept with the search, so six weeks later you can see not just what you found but what you asked, in what order, and what made you change direction.
What this stage gives you
- A queue of rewritten queries, each tied to the gap it closes
- The move and the reasoning stated in plain language for every one
- Side-by-side comparison against the original result set
All suggestions for this search
9 refinements- R01Split the condition
obstructive sleep apneaYour results mix obstructive and central sleep apnea. They have different causes, different treatments and largely separate literatures. Naming the subtype removes the thirty-one sources that are about the other one.
- R02Name the population
obstructive sleep apnea adults over 70 outcomesThe gap map flagged that every trial you were given recruited adults between 30 and 65. If your question includes older adults, this is the query that goes looking for them.
- R03Name the intervention
CPAP adherence obstructive sleep apneaSearching for treatment returns overviews of the options. Searching for adherence returns work with something measurable in it, which is what you need to answer whether treatment helps in practice rather than in principle.
- R04Name the outcome and the design
obstructive sleep apnea daytime sleepiness randomized controlled trialDaytime function and cardiovascular endpoints are two separate bodies of research, and yours is the first. Adding the design term lifts primary trials above the commentary currently sitting between them.
- R05Use the field's own vocabulary
obstructive sleep apnoea OR OSAHS OR "sleep-disordered breathing"British spelling, the older OSAHS acronym and the umbrella term sleep-disordered breathing each unlock indexed literature that the phrase you typed will never reach on its own.
- R06Search the measure, not the condition
"apnea-hypopnea index" severity threshold disagreementMost of the disagreement in your result set traces back to different AHI cut-offs for what counts as moderate. Searching the measure itself surfaces the papers arguing about the threshold rather than the ones quietly assuming one.
- R07Put a floor under the date
obstructive sleep apnea home testing guidelines 2021..2026Guidance on at-home testing changed after the oldest of your top-ranked sources was written. A date floor stops superseded advice from outranking current advice on the strength of being better cited.
- R08Clear the commercial crowd
obstructive sleep apnea screening -"best cpap machine" -mouthpieceNine of your top twenty results are retail comparison pages that score high on the keyword and low on everything else. Two exclusions clear them without touching a single clinical source.
- R09Close the evidence gap
CPAP long-term adherence cohort 5 yearsAdherence is discussed everywhere in your results and measured almost nowhere. Asking for long-run cohort work is how you find out whether the twelve-month figure holds.
The whole loop takes about four minutes
Keyword in, ranked list, open a couple of signals, read the gap map, take one refinement. Do it once on a subject you know well and you will know immediately whether the scoring agrees with you.
Twenty-five ranked searches a month on the free plan. No card.