Integrity policy
Ranking integrity
Last updated 24 August 2026. Current draft: TVS-IG-1.0.0. Compatible with scoring draft TVS-1.2.0. Ranking editions will record both version identifiers. Detectors are versioned separately from the score.
A measurable ranking creates an incentive to optimise the paperwork instead of the wedding. This policy is how we treat suspected manipulation. It is not a secret penalty table and it is not a paid appeals desk.
Integrity is separate from the Top100 Venue Score. If a signal looks gamed, we hold or drop that evidence until it is verified. We do not subtract hidden points. Unrelated metrics keep their own facts. Historic editions stay frozen. Scoring stays TVS-1.2.0. A later integrity version is not a new wedding-venue scoring methodology.
Exact detector thresholds stay in versioned configuration. Named calibration findings stay internal. This page is the public doctrine, the statuses we use, the mechanisms, the audit rates, and the kinds of games we test for.
Doctrine
- Integrity work does not change category percentages in the scoring model.
- There are no hidden negative Top100 Venue Score points.
- The default is to neutralise or hold the affected signal, then verify.
- Unrelated metrics are left alone.
- Once the issue is resolved with stronger evidence, the signal can be used again.
- Published ranking editions are not rewritten when a later integrity version appears.
Integrity statuses
CLEAR
No integrity anomaly detected.
WATCH
Monitor or refresh that source. WATCH alone does not change a score.
REVIEW
The affected evidence is not accepted until a human has verified it.
HOLD
Ranking eligibility, or the affected category, is blocked until the issue is resolved.
Mechanisms
01
Neutralise, do not penalise
If a signal is suspected of manipulation, we hold or drop only that evidence. The fact may become unknown. Reputation may become null. Ranking eligibility may pause. Other metrics are not contaminated.
02
One evidence origin
The same claim copied across a homepage, FAQ, brochure, sister site, or syndicated directory is one origin. It is not independent corroboration. We keep the strongest current source.
03
One venue, one ranking entity
One physical wedding venue has one canonical ranking record. Rooms, packages, microsites, and near-duplicate listings do not get extra ranks unless they are operationally and legally distinct.
04
Physical space identity
Marketing aliases for the same room count once, unless the spaces can operate independently at the same time.
05
Controlled taxonomy
Synonyms do not create extra features. Extraction maps claims to canonical concepts before scoring. One concept is one fact.
06
Review integrity
We watch unusual review velocity, sudden count jumps, rating shifts with abnormal volume, source saturation, and source identity changes. Reputation can be held or set to null pending verification. There is no automatic negative score.
07
High-leverage fact changes
New accepted facts are compared with earlier snapshots, especially capacity, bedrooms, accessibility, and exclusivity. The previous accepted fact stays until stronger evidence replaces it.
08
Freshness is content, not a timestamp
A new page date does not reset freshness if the relevant content is materially unchanged.
09
Visible evidence, not markup alone
Schema or meta tags are not enough for high-impact facilities unless visible official content or a trusted structured source corroborates them. Otherwise the fact stays unknown.
10
Official sources override marketing
Heritage, licensing, and distances come from official registers or our own geospatial work. If marketing conflicts, the official source wins and the conflict is logged.
11
Transparency has to be usable
Teaser copy, third-party prices, a from-price without material terms, or a stale PDF cannot manufacture full transparency. We score only the facts actually disclosed after a defined search.
12
Correction churn
Corrections stay welcome. Repeated contradictory submissions on high-leverage facts go to human verification. There is no punitive score.
13
Operator portfolios
A confirmed pattern at one operator can trigger an audit of sibling venues. Siblings are never automatically penalised. Each venue is still judged on its own evidence.
14
Top-rank audit
Before an official edition we manually verify high-leverage and required-category evidence for every provisional Top 120, and for every REVIEW or HOLD flag.
15
Random audit
We randomly audit at least 10% of remaining eligible venues so gaming cannot focus only on known review targets. The sample rate is public. The sample list stays hidden until the audit is complete.
16
Versioned evolution
Integrity detectors change in their own versions. New abuse patterns can be addressed going forward without silently rewriting historic editions.
17
Clean re-entry
There is no permanent secret blacklist. Once the affected evidence is independently verified, it is usable again under the normal rules. The audit log is kept.
Edition audits
120
Provisional top set, full high-leverage audit
10%
Random sample of remaining eligible venues
All
REVIEW and HOLD flags audited before eligibility
Edition quality control. Every provisional Top 120 gets a full high-leverage audit. At least 10% of the remaining eligible set is sampled at random. Every REVIEW or HOLD is audited.
What we test for
Illustrative games we test for. These are attack patterns, not accusations about named venues.
| If someone tries this | What should happen |
|---|---|
| The same lounge is named four different ways on one page. | One space and one facility fact. No score multiplication. |
| The same barn is marketed under three package names. | Count one space unless simultaneous independent operation is verified. |
| An operator launches separate microsites for the same address and packages. | One canonical ranking record. |
| A venue gains 200 reviews in 30 days after a stable low-volume history. | Reputation HOLD for quality control. No automatic negative score. |
| An old pricing PDF gets a new updated page date with the same content. | Freshness does not reset. |
| Machine-readable markup claims wheelchair access but visible pages have no access evidence. | The accessibility fact stays unknown. |
| Capacity jumps from 120 to 220 just before an edition cutoff. | Reverification is required before the new capacity is accepted. |
| A venue claims nearby sister-hotel rooms as on-site bedrooms. | Those rooms are excluded unless they sit inside the canonical venue boundary. |
| A venue adds ten paid commercial award badges. | No Top100 Venue Score effect. Awards are not scoring inputs. |
| The homepage says weddings from £2,000 with no scope, inclusions, or terms. | That does not automatically satisfy full transparency metrics. |
| Two sibling venues share identical facility copy. | No cross-venue corroboration. Each venue needs entity-specific evidence. |
| A high-leverage fact flips repeatedly through venue submissions. | The fact goes to REVIEW or HOLD until a human adjudicates the evidence. |
Calibration queues that name specific venues stay in versioned integrity configuration. They are not published on this page.