Draft for testing. The method has been tried on a small number of real listings and is still changing. Text in [square brackets] is a placeholder to fill before launch.
Version 0.2 (draft) · 26 October 2026
What this is
A published procedure for reading a room-rental listing, and the messages around it, for patterns that often appear in rental fraud — before any money moves. It is built for students looking for a room in Padova.
It reports risk signals. It cannot tell you that a listing is genuine, and it cannot tell you that a person is a fraudster.
What this is not
- Not advice. Nothing here is legal, financial, tax or immigration advice.
- Not an ownership, identity or credit check. v0.1 does not look up who owns the property or who a person is.
- Not an agent or an intermediary. It does not list rooms, introduce you to landlords, take commissions, or hold or move money.
- Not official. It is an unofficial, student-run project. It is not affiliated with the University of Padova or any of its services.
What it reads
Five categories are graded independently. A sixth (photographs) is not run in this version.
1. Language and messages
Reads: the text of the listing and any messages or screenshots you paste. Looks for: recurring fraud scripts — an owner who is abroad and cannot show the place; keys sent by courier or handed over after payment; urgency framed as competition (“first to pay gets the room”); pressure to keep things secret or to move to another channel; generic or copied text; price, photos or address kept for private messages when the post says little else; details that contradict each other, including dates that cannot fit the day the post was written; text that tries to instruct an automated checker. Replying only by private message or on a messaging app is common among genuine private landlords and is not reported on its own. Can establish: that the wording carries catalogued patterns, and how many appear together. Several patterns in one short message is the strongest reading this category can produce. Cannot establish: that the writer is dishonest. Each pattern has an innocent explanation. A genuine landlord who lives abroad writes the same sentence a fraudster writes.
2. Price against the local reference
Reads: the total monthly cost (the asking rent plus any expenses the listing says are extra) and the zone, against a reference range kept by the author. Price is a supporting signal only: many genuine rooms in shared flats are listed cheaply before utilities. Can establish: how far the asking price sits below the reference range. Cannot establish: that a cheap listing is fraudulent. Genuinely cheap rooms exist, and this signal measures a distance, not an intention. The reference covers single rooms; a shared double room or a bed in a room is not compared, and the report says so. Reference data: [source, date and zone table — to be filled from public sources before launch]. Where there is no reference for a zone, the report says “nothing to compare” instead of reporting a clean result.
3. Payment instructions
Reads: any bank account number (IBAN) and any payment method you were asked to use. The IBAN is checked by ordinary program code — structure, checksum and country — not by the language model. Can establish: whether an IBAN is structurally valid; whether it is Italian or foreign; whether the payment method is hard to reverse (gift cards, cryptocurrency, money-transfer services, cash sent by post or courier) or tied to a rechargeable prepaid card. Cannot establish: who holds the account. A valid IBAN belongs to somebody, and this method neither says who nor tries to find out.
4. Address and contract
Reads: whether an address is given and whether it fits the rest of the listing; what kind of contract is mentioned; whether it will be written; how large the deposit is. Can establish: whether the basics of a normal tenancy are present, and whether anything mentioned strays from them. Cannot establish: that the terms are lawful. For contract and legal questions, ask the Sportello Affitti (see “Where to get real help”). Note: many genuine listings hide the exact street number until a viewing. A missing address is reported as “nothing to read”, not as a concern.
A takeover or sublet (subentro) with no mention of the landlord’s consent is reported as a question to ask, not as a concern.
5. Process
Reads: the order of events the advertiser proposes — can you see the place (in person or on a live video call) before paying? do you get a written contract before paying? can you register the address? Can establish: whether the basic steps of a normal rental are offered, and in what order. Cannot establish: that the steps will really be followed.
Not run in v0.1: photographs
Searching the web for copies of the listing photos is not automated in this version. The report says so rather than treating photos as clean. You can do it yourself in a minute with a reverse image search (for example, Google Lens).
Not done: ownership
This method does not check who owns the address. See “Why there is no ownership check”.
Readings
Each category returns one of four readings:
| Reading | Meaning |
|---|---|
| Clear | The category had something to read, and none of its patterns appeared. |
| Informational | The category had little or nothing to read, or cannot be compared. This is not a quiet pass. The report says “nothing either way”. |
| Concern | A pattern appeared that is common in fraud but has innocent explanations. |
| Alert | A pattern appeared that is strongly associated with losing money or having no protection (for example, money asked before you can see the place). |
How readings become a verdict
The categories are graded first and combined second. The combination is arithmetic over the readings, and the readings stay visible so you can disagree with the arithmetic. The language model grades the categories; ordinary program code computes the band and the score, so the same readings always give the same result.
| Band | Rule |
|---|---|
| High risk | At least one alert, or at least three concerns. |
| Proceed with caution | One or two concerns, and no alert. |
| No red flags found | No concern and no alert, and at least three categories read as clear. |
| Couldn’t confirm | No concern and no alert, but fewer than three categories read as clear. There was too little to read. |
There is deliberately no band that means the listing is fine. The best band records that none of the checks fired. That is not the same finding as a genuine landlord, and the wording never lets the two blur.
Score. For convenience the report also shows a number: 100 − 40 × (categories with an alert) − 15 × (categories with a concern), never below 0. These weights are a first guess. They have not been calibrated on real cases, and the number adds no information beyond the band. No score is shown when the band is “Couldn’t confirm”, because a high number there would read as reassurance when the real finding is that there was too little to read.
A thin check is reported as a thin check. When there is little to read, the report names the missing inputs. Less evidence never produces a cleaner-looking answer.
What a verdict establishes — and what it does not
A verdict establishes which catalogued risk patterns are present in the material you supplied, and how strongly they fired.
- It does not establish that a listing is fraudulent. It reports patterns, and the advertiser may have an innocent explanation for every one of them.
- It does not establish that a listing is genuine. No reading this method can produce amounts to that.
- False negatives exist. A well-written scam can read as clean. A scammer can also copy a real owner’s details, or use a real agency’s name. Nothing in v0.1 detects either.
- False positives exist. Honest landlords sometimes live abroad, ask for a reservation, or write in a hurry.
- Accuracy is not measured. v0.1 has not been tested on enough real listings to publish an accuracy rate, and a rate computed from a handful of cases would mislead. Until a tested rate is published, treat the bands as a way of ordering your attention, not as probabilities.
Why there is no ownership check
Checking who owns a property would mean handling third parties’ personal data. In Italy the cadastre is not a register of legal title, and the rules on commercial reuse of property-register data are unclear to the author. For a project run by one student, the risk is not worth it. If the author ever adds such a check, it will be published here first with its limits.
Human checks outrank every signal
Whatever a verdict says:
- See the place — in person, or on a live video call — before you pay anything.
- Get the written contract first. Ask whether it will be registered with the tax authority (Agenzia delle Entrate).
- Pay only by traceable bank transfer, to an account whose holder matches the person who signs the contract. Never use cash, gift cards, cryptocurrency or money-transfer services.
- Do not trust payment links sent in messages. Open the platform’s site yourself by typing its address.
- If something feels wrong, walk away. No room is worth your deposit.
Where to get real help
(Details from the University of Padova pages, October 2026. Check them before relying on them.)
- Sportello Affitti — advice in cases of fraud, discrimination or unfair contracts:
sportelloaffitti.padova@gmail.com. - Cerco Alloggio — the University’s International Housing Office service:
https://cercoalloggio.com/padova·housing@unipd.it. - HousingAnywhere × University of Padova — according to the University’s Erasmus accommodation page (and HousingAnywhere’s own help pages), payments are released to the landlord 48 hours after the move-in date:
https://housinganywhere.com/Padua--Italy/unipd/sign-up. - ESU / SASSA Service — university residences and the Erasmus accommodation form:
https://www.esu.pd.it/it/sassa-service.
What happens to your data
(Design intent for the first release — update this section to match what is actually built.)
- The text you paste is processed to produce the check. Telephone numbers, email addresses and bank account numbers are masked before the text is analysed.
- The text is sent to an AI provider ([provider name and region — confirm before launch]) to be read.
- [Reports are not stored / are stored for N days — decide and write it here.]
- There is no database of landlords, no search by a person’s name, and no list of “bad” advertisers.
Version history
- 0.1 (26 Oct 2026) — first draft. Five categories. No photo search, no ownership check. Weights uncalibrated.
- 0.2 (26 Oct 2026, draft) — one tuning pass after reading 16 real listings from a student group: messaging-app-only contact is no longer a concern on its own; new signal for price, photos or address withheld; dates that cannot fit the post’s date; price compared on total monthly cost and treated as a supporting signal; a registered contract counts as clear even without a street address; a takeover is a question, not a concern.
- Source note (27 Oct 2026) — wording only: the 48-hour payment rule now names its sources, the University’s Erasmus accommodation page and HousingAnywhere’s own help pages. No change to categories, flags or weights.