Methodology

How Advennt builds and maintains gambling jurisdiction intelligence.

Advennt provides structured, comparable intelligence on gambling regulation across every jurisdiction and regional bloc in our coverage register. This page explains what we do and why — our source hierarchy, confidence framework, scoring model, editorial process, and the trust tiers that distinguish lawyer-reviewed content from AI-generated content.

Source Hierarchy

Every fact in Advennt traces back to a source. We grade every source on the four-tier scale the whole Asymmetric Intelligence estate uses, so a tier means the same thing on every site:

Tier	What it is	Examples
Tier 1 — Authoritative	Authoritative source: government, regulator, central bank, official journal or court	Gambling Act 2005, UKGC licence conditions, MGA regulatory decisions, official gazettes, court judgments
Tier 2 — Reporting on a primary source	Press or NGO reporting on a primary source	Trade and general press reporting a regulator's decision or a new statute
Tier 3 — Commentary	Vendor, analytics, law-firm or consultancy commentary	Law-firm client alerts, compliance-vendor briefings, industry association reports
Tier 4 — Aggregator	Aggregator or syndicator carrying someone else's reporting	News aggregators and syndicated copies of another outlet's story

We always prefer Tier 1 where available. A finding with no Tier 1 source behind it is flagged as such on the page rather than hidden, and lower tiers are shown with their tier so you can weigh them yourself. We never rely on reporting when the primary source is publicly accessible.

Confidence Framework

Every rating, score, and factual claim carries a confidence level reflecting the strength of the underlying evidence:

Level	Meaning	Evidence Required
Confirmed	Verified against primary sources	Tier 1 source + at least one corroborating source
Probable	Strong evidence, no contradictions	Multiple Tier 2 sources without Tier 1 contradiction
Uncertain	Single or mixed evidence	Single source or conflicting sources
Speculative	Forward-looking inference	Analytical inference from current trajectory

Auto-downgrade rule: If content is marked "Confirmed" but the underlying sources lack a Tier 1 reference, the system automatically downgrades the confidence to "Probable." This prevents unsubstantiated ratings from persisting in the data.

The 11-Category Model

Each jurisdiction is assessed across 11 structured categories. Every category combines narrative analysis with structured, comparable data fields and a traffic light (TL) rating:

#	Category	What It Covers
1	Overview	Market description, regulation status, size, digital maturity, key attractions and headwinds
2	Licensing & Regulation	Licensing authority, licence types, local presence, product coverage, B2B licensing
3	Marketing	Advertising channels, bonus rules, sponsorship, affiliate regime
4	Enforcement	Enforcement style, targeting, tools, recent activity, enforcement events database
5	Fees	Application fees, annual fees, capitalisation requirements
6	Taxes	Tax basis, headline rate, VAT treatment, special levies
7	Outlook	Regulatory direction, reform pipeline, expected triggers, asymmetric signals
8	Payments & Banking	PSP availability, banking risk, deposit/payout methods, payment blocking
9	Technical Compliance	Testing standards, RNG, geolocation, hosting, responsible gambling tech
10	Distribution & Platform Rules	App store policies, ad platform restrictions, geo-gating, affiliate constraints
11	Market Entry Practicality	Time-to-launch, local presence, bottlenecks, adviser stack, sequencing

Traffic Lights

Each category carries a traffic light (TL) rating that provides an at-a-glance assessment:

Green — Favourable, open, or low burden
Amber — Mixed, restricted, or average burden
Red — Unfavourable, closed, or high burden

Ratings are comparative across jurisdictions and may be adjusted as more markets are scored.

Product Classes and Legal Status

Every jurisdiction is assessed against the same fixed list of product classes, and each class carries one legal status drawn from a fixed vocabulary. Both lists are closed: a jurisdiction cannot invent a class, and an assessment cannot invent a status. That is what makes one jurisdiction comparable with another.

The definitions below are rendered directly from the contract the pipeline itself uses. They are not a summary of it, so this page cannot drift away from what the data actually means.

Two different questions. Legal status asks whether a product can lawfully be offered in a jurisdiction. Legal accessibility asks whether a commercial operator can realistically get in. These are not the same, and they can diverge: a state lottery monopoly is perfectly lawful and completely closed to you. Read them together.

Legal status

What the law says about offering this product here. Each status shows the definition we publish to readers and, beneath it, the test our analysts and expert reviewers apply when assigning it — because a legal status is only auditable if the rule that produced it is visible.

Loading the status vocabulary…

Product classes

The classes a jurisdiction is assessed against. Player products are things a customer plays or bets on; supply roles are business-to-business positions; settlement rails are payment and settlement functions. Classes from different families are never compared as peers.

Loading the class vocabulary…

Derived Scores

Two headline scores are computed deterministically from the structured data — no AI is involved in scoring:

Legal Accessibility

Classifies each product in each jurisdiction as open, restricted, or closed according to whether a commercial operator can realistically enter that market, and rolls those up to a headline band per jurisdiction.

This sits alongside legal status rather than being computed from it, because the two questions genuinely come apart: a state lottery monopoly is lawful and closed, and supplying an incumbent in a prohibited market may be open to you as a vendor. Where a product is assessed as accessible on terms its legal status would not obviously predict, that is a judgement about the route in, and the licensing detail carries the reason. Both vocabularies are published in full.

Commercial Attractiveness

A weighted score derived from the traffic lights of six categories — with heavier weight on licensing framework and tax environment, lighter weight on marketing rules, enforcement intensity, fee burden, and forward outlook. Classified as high (score ≥ 0.70), medium (0.40–0.69), or low (< 0.40). Modifier rules adjust for edge cases such as legally closed markets or punitive enforcement regimes.

AI-content disclosure

Advennt's jurisdiction intelligence is assembled by an AI research pipeline from cited public sources, and it passes evidence gates before it publishes. Where an expert reviewer has reviewed a jurisdiction, its page shows the Expert reviewed badge naming the reviewer, what kind of reviewer they are, and the as-at date of the content they reviewed (see Trust Tiers below). Expert reviewers are law firms and specialist advisers, such as compliance or market-research firms. Anything added after the review date is marked pending review, and everything else is AI-generated. The ai-generated meta tag in a page's code is its machine-readable AI-content mark: the analysis on that page was assembled by an AI research pipeline, and it is not legal advice. Facts are machine-assembled from cited public sources. Unless marked Expert reviewed, individual claims have not each been independently verified by a human reviewer. The pipeline roles are Research, Interpreter, Reviewer, Composer, Applier, Curator and Publisher. Where the pipeline has produced nothing, the page says so rather than substituting a generic disclaimer or a plausible estimate.

Silence when nothing material. A category or jurisdiction with no findings is shown as an explicit empty state, never filled with invented content.
No fabricated counts. A count or date is shown only where pipeline data backs it. Figures still typed into a page by hand are listed by the fleet site-facts census and are being moved behind data bindings.
Cited provenance. Structural values such as module lists, jurisdiction lists and pipeline role names come from this site's own configuration and the fleet doctrine, not invented for the page.

Trust Tiers

Advennt covers jurisdictions at different levels of editorial verification. Every jurisdiction page displays a badge indicating its trust tier:

EXPERT REVIEWED

Content reviewed and approved by a named expert reviewer for the jurisdiction: a law firm or a specialist adviser. The reviewer, their type and the as-at date of the reviewed content are displayed. A review older than its review cycle is shown as refresh due.

EXPERT REVIEW PENDING

AI-assembled, evidence-gated content in a jurisdiction where an expert reviewer is engaged and the review has not yet been returned.

AI-ASSEMBLED

Content assembled by the Advennt intelligence system from cited sources, using the methodology described on this page, and passed through its evidence gates. Not yet expert-reviewed.

The AI-assembled tier is not a disclaimer — it reflects a structured, systematic intelligence process with tiered sources, confidence scoring, and version-controlled data. The methodology is published here precisely so that users can assess the rigour for themselves.

What the AI system does

Regularly polls Tier 1 and Tier 2 sources on a scheduled basis
Classifies developments by jurisdiction, product, category, and impact
Generates draft update proposals with sources, confidence levels, and proposed changes
Computes derived scores deterministically from structured fields

What the AI system does not do

It does not update the baseline directly — all changes are proposals
It does not overwrite reviewed text — changes after an expert review are marked pending until the reviewer returns them
It does not compute scores using AI — scoring is pure deterministic logic

Coverage Tiers

Jurisdictions are classified by depth of coverage:

Tier	Count	Depth
Tier A — Core	15–30	Full 11-category baseline, rich updates, enforcement events, all extended fields
Tier B — Strategic	20–30	Full baseline with lighter ancillary fields (categories 8–11 may have partial data)
Tier C — Watchlist	30–40	Skeleton baseline with strong monitoring for triggers and outlook

Promotion and demotion between tiers follows documented criteria: draft legislation, consultations, enforcement campaigns, or sustained demand trigger promotion. Demotion requires editorial review and recorded rationale. No jurisdiction is ever fully removed.

Version History and Transparency

Every baseline change is recorded with a version history entry: prior value, new value, date, reason, and source. This creates an auditable trail of how each jurisdiction's profile has evolved over time. Users can see exactly what changed, when, and why.

Persistent vs. transient: Core regulatory positions (licensing requirements, tax rates, enforcement style) are persistent — they change only when there is material new information. Individual events (consultations, enforcement actions, policy announcements) are transient — they populate timelines and update feeds without altering the baseline unless they represent a structural change.

Research benchmark — gambling

Run window: 16 to 19 October 2026 · Caveats: run by us, AI-judged, not independently validated.

Each cell compares stored pipeline research for Advennt (gambling) against a single-prompt Google Gemini Deep Research Max report on the same jurisdiction and sector. Figures are ours versus the report.

Cell	Verified points	Depth	Accuracy	Result
UK · gambling	217 vs 68	72% vs 29%	85.2% vs 89.5%	Pass
EU/EEA · gambling	54 vs 43	77% vs 34%	79.8% vs 65.4%	Pass, under review

TRACE run-1 dataset audit: 152 published records, 9,295 citations (98.4% provenance publisher match; 100% non-empty record integrity).

Full per-consumer tab and methodology on Asymmetric Intelligence: https://a-i.gi/benchmark/#tab-advennt · summary.json.

Editorial Standards

Restrictive baseline: When sources conflict, the more restrictive interpretation is recorded as the baseline. This protects users from underestimating compliance risk.
No silent overwrites: All baseline changes are recorded in version history.
Materiality thresholds: Traffic light (TL) changes are always material. Narrative updates without structured field changes are timeline-only.
Sampling checks: Periodic human spot-checks of AI proposals against Tier 1 sources.