Tensions in Practice: Search speed in tension with finding every match¶
A tiny archive with two relevant records
Imagine three archive records inspected in order: A is relevant to the question, B is not, and C is relevant. A quick search stops after finding A. A full scan also reaches C, but does more inspection before finishing. Both use the same relevance test; what changes is the part of the archive they reach. The labels disclose the answer for this invented example so the missing result can be seen.
Return a useful result promptly
Avoid inspecting the whole archive when one suitable record will answer the current question.
Find the relevant set
Reach relevant records that appear beyond the first match.
Why these aims pull against each other
Stopping early saves inspection work by leaving some candidates unexamined. That also leaves later relevant records outside the returned set, even when the test correctly classifies every record it sees.
Choose an arrangement to see what changes and what remains difficult.
An invented three-record search. Disconnected records are not inspected in that arrangement; distances do not measure time.
What this choice protects
What it costs
When it fits
Compare the arrangements
Stop after one match
Inspect A, return it, and end this search before inspecting B or C.
- What it protects
- One relevant result is returned with a short search path.
- What it costs
- C remains unfound in this run; a user needing all matches receives an incomplete set.
- When it fits
- Fits a need satisfied by one acceptable record. It is unsuitable when completeness is part of the requested result.
Illustration note: Three records and their relevance labels are invented. Early termination is the supported mechanism; no real search latency is estimated.
Inspect all three
Test A, B and C, then return the two matching records.
- What it protects
- Both relevant records in this stated three-record archive are found.
- What it costs
- The search spends effort on B as well as the matches and waits for the final inspection.
- When it fits
- Fits this finite archive when completeness matters and the inspection budget permits it; the relevance test and archive coverage still need to be correct.
Illustration note: The ordered scan is one editorial search procedure, not a claim that every index must read every record to be complete.
What this illustration does—and does not—establish
Search and Retrieval: Precision vs Recall vs Query Latency supplies the early-termination versus coverage tradeoff. The finite sequence makes an omitted match visible while holding the relevance criterion fixed.
- The example changes search reach, not the correctness or strictness of the relevance test. A wider search does not repair a bad test.
- Finding every match here means every match in the declared three-record archive, not every relevant fact in the world.
- No numeric latency, precision, recall benchmark, or guarantee about an actual search engine is claimed.
Source entries
Search and Retrieval
This source passage supplies the contextual tension. The concrete arrangements and schematic examples are editorial illustrations, not measured findings.
Precision vs Recall vs Query Latency
- T1: Precision vs Recall vs Query Latency. Exhaustive search retrieves all relevant items (high recall) but is slow. Approximate or early-termination retrieval is fast (low latency) but misses some results (lower recall). Tightening relevance criteria improves precision (fewer false positives) but may reduce recall. No single point is optimal across all use cases.
The source operation
Every search-and-retrieval system faces trade-offs between precision (false positives excluded), recall (false negatives excluded), and query latency, and must determine both what is relevant and how efficiently to locate it.