---
title: "How AcuiQ is built"
url: "https://acuiq.com/wiki/method/how-acuiq-is-built"
updated: 2026-10-15
last_verified: 2026-10-15
license: CC-BY-4.0
license_url: "https://creativecommons.org/licenses/by/4.0/"
section: "method"
basis: "Modern research"
audience: "Anyone"
published: "2026-08-30"
reading_minutes: 4
sources: 3
---

# How AcuiQ is built

_What the index holds, where each entry came from, and how a search is ranked._

AcuiQ indexes point selections that other people published, and lets you [search them by symptom](/symptoms). It is a reference tool. It does not diagnose, it does not differentiate patterns, and it does not decide what to treat – those are the parts of acupuncture it deliberately leaves alone, and they are set out in [what AcuiQ does not do](/wiki/method/what-acuiq-does-not-do).

## What the database holds

6,965 entries linking symptoms to points, drawn from 773 catalogued points across 14,457 indexed symptom terms. Every entry is one of the kinds below, and each result card says which kind it is.

| Kind | Entries | What it is |
| --- | --- | --- |
| Study-reported | 5,777 | Point selections extracted from published papers. Each carries a DOI or article link you can open and check. This is the bulk of the database, and the kind ranked highest. |
| Text or case report | 855 | Prescriptions recorded in clinical texts, compiled point-prescription sets, and individual case reports. Cited by name; many are Chinese-language sources indexed in CNKI without an open link. |
| Website | 329 | Point selections taken from acupuncture websites rather than from the literature. Linked, but not peer-reviewed and not always attributed at source. This is not a spread of sites: sacredlotus.com alone supplies 317 of the 329. They carry the heaviest ranking discount of any sourced tier. A few come from flashcard sites, which are study aids rather than clinical references. |
| Unsourced | 4 | Entries whose source was lost in extraction. Shown with no source claim. |

Only human evidence reaches the index. An earlier version of this database held protocols derived from animal research. They are gone, and every extraction path now runs a species gate before a paper reaches the pipeline. The July 2026 audit that found them is [published in full](/wiki/research/nothing-ever-fails#what-we-found-in-our-own-database), with the count and what else it turned up.

Each entry also records what its source reported for the points: a benefit, no effect, or no results yet, for a trial that registered its protocol and has not published. Only entries whose source reported a benefit reach the search, the counts and the point pages. A symptom page lists the other two kinds apart, under their own headings, as [Reading a protocol](/wiki/method/reading-a-protocol) describes.

## How results are ranked

Your search terms are resolved to symptom concepts, expanded over synonyms and narrower terms, and matched by text similarity against each entry’s symptom list. Coverage is the fraction of your query an entry addresses; its score is that coverage multiplied by a weight for what kind of source it is – study-reported 1.0, text or case report 0.9, website 0.7, unsourced 0.6. Because that is a multiplier and not a tiebreak, a website entry matching your query exactly still scores below a study matching three quarters of it. The scorer runs fuzzy string matching over a curated index, discounted by source. There is no model and nothing is learned from your searches. The percentage on a card is that discounted score, so two entries with the same text similarity show different percentages. It is not clinical confidence and it is not a claim that the treatment works.

Score does not set the order on its own. Every entry naming more than one point is listed above every entry naming one, whatever the two score, because a prescription is the unit this index holds. A card lower down the page can therefore show a higher percentage than a card above it.

One line on a card is computed across the result set rather than from the entry alone. Every entry in a set answers the same search, so the points on one card can be compared with the points the study-reported entries chose, and each card states both halves of that comparison: which of its points a study also chose, and what the overlap does not establish. A study covering the same complaint did not test this prescription, and a second publication is not yet a second research group. An empty overlap is stated as plainly as a full one.

698 entries name a single point, because that is what their source reported. They are labelled “single-point entry” rather than “protocol”. Where no entry matches at all, the search falls back to showing points that merely list your symptom among their indications – labelled as an index lookup, with no score attached.

## Corrections

If an entry misattributes a source, misreads a prescription, or pairs a point with a symptom its source never claimed, that is a bug and it will be fixed. Write to [info@acuiq.com](mailto:info@acuiq.com). What extraction at this scale reveals about the literature itself is written up under [Research](/wiki/research).

### Sources

1. AcuiQ protocol corpus, counted live at render time — /symptoms
2. What AcuiQ does not do — /wiki/method/what-acuiq-does-not-do
3. Nothing in This Literature Ever Fails – the audit behind these tiers — /wiki/research/nothing-ever-fails