Every piece of information on Yerleske has a source, a date and a confidence level. This page explains where data comes from, how it is verified and how the Yerleske Score will be calculated.
How a record goes live
- Draft: internal only, not published.
- Discovered: created by an import or discovery run. Shown with a “Pending verification” notice and kept out of search engines (noindex).
- Verified: checked against an official source, ready to publish.
- Published: passed the quality gate; public and indexable.
- Archived: removed, with history and old links preserved.
To be indexed, a profile needs at least a name, type, city, official website, a short description, a source and an address or campus. Pages that do not meet this bar remain visible but are not indexed.
Confidence score
Each value receives a 0–100 confidence score based on its source type. Appearing on the institution’s official domain or being corroborated by independent sources raises the score within the range for that type. Editors can adjust the ranges.
| Source type | Range |
|---|---|
| Official source Official public source | 95–100 |
| Official source University website | 90–100 |
| Official source Unit website | 88–98 |
| Official source Official API | 90–100 |
| Verified institution Verified institution | 90–98 |
| Official source RSS / Atom feed | 85–95 |
| Public source Structured data | 70–92 |
| Public source Web search (discovery) | 50–90 |
| Public source Public web | 40–80 |
| Yerleske editorial Yerleske editor | 60–90 |
| Community Student contribution | 10–60 |
| AI-assisted AI extraction | 20–85 |
| Public source Open data (Wikidata / OpenStreetMap) | 55–85 |
Discovery and crawling
Web search (Serper) is only a discovery tool for Yerleske: it helps locate official websites, but search results are never stored as facts. Official sources are crawled after an editor confirms them. Our crawler respects robots.txt, rate-limits itself per domain and identifies as YerleskeBot.
Ratings, reviews and photos from third-party services such as Google Maps are not turned into Yerleske data.
How AI is used
AI never invents information. It only helps extract structured facts from a source text we hold; every extracted fact links to its source, is labelled “AI-assisted” and passes validation before publication. Without a source, the answer is “This could not be verified.”
Status: AI-assisted extraction is not enabled yet (pending integration).
Conflicting information
When two sources disagree, existing data is never silently overwritten. The conflict is recorded and an editor decides with both sources side by side. Official public sources and the institution’s own website take precedence.
Freshness
- pages.methodology.freshness.university.description: 30 days
- pages.methodology.freshness.university.contact: 90 days
- pages.methodology.freshness.campus.address: 180 days
- Campus facilities: 90 days
- News: 1 hours
- Events: 1 hours
Yerleske Score
The Yerleske Score is our own independent measure. It is not a Google rating or any other platform’s rating and is never presented as one. It will be a weighted average of category ratings from students:
- Academic environment (weight 1.5)
- Campus life (weight 1.2)
- Transportation (weight 1)
- Library (weight 0.8)
- Food (weight 0.8)
- Sports (weight 0.6)
- Social life (weight 0.8)
- Technology (weight 0.7)
- Safety (weight 1)
- Accessibility (weight 0.8)
- Dormitory (weight 0.7)
- Value (weight 0.9)
Verified students carry more weight; rate limits, duplicate detection and moderation guard against manipulation. Paid plans never influence scores, rankings or student reviews. Any sponsored content is clearly labelled “Sponsored”.
To write a review, a student is verified with a code sent to an address on the university’s own domain (the address is not stored; only the domain and a one-way hash). Every review is moderated before publication; e-mails, phone numbers and links in the text are masked; the same text from another account is rejected.
The score is a Bayesian average: a handful of reviews cannot push it to an extreme; it starts at the site-wide mean and moves towards the institution’s own mean as reviews accumulate. No score is shown with fewer than three published reviews.
