candidates processed
SCAMNET / HORTRAME RESEARCH
Audited scam-network statistics
Analysis is currently paused.
ScamNet studies fraudulent storefront infrastructure. This public beta pairs audited aggregate network-membership precision with a sanitized table of provisional matches.
HORTRAME has gathered and preserved substantially more research data. The table publishes only provisional domains and observed-title identity hypotheses; raw evidence, authoritative attribution claims, and detection details remain private. Researchers may contact us to discuss methods or collaboration.
ScamNet is an automated, continuously updated research system. Its public records are versioned observations—not a final or exhaustive blacklist.
Beta. Collection, validation, and analysis operate as one continuous pipeline. Results evolve through audited 1,000-domain batches, with timestamps, lens versions, and uncertainty intervals. No deferred bulk-cleanup stage; raw evidence is never published automatically.
This website is currently being updated while the first round of analysis is completed. Data, victim domains, countries, and regions may still be missing.
LATEST AUDITED QUALITY CHECKPOINT
Precision evidence for the provisional population
——
PASS applies only to aggregate membership-precision evidence. The 95% lower bound is a conservative population-level floor under the stated random-sampling model—not the chance that any particular row is correct. Released rows remain provisional.
provisional population—
fresh random sample—
conservative point precision—
exact one-sided 95% interval—
confirmed network failures—
future audited cadence1,000
Complete calculation and limitations
What is measured
This estimate measures only precision of provisional positive network-membership classifications at a frozen checkpoint. It does not measure recall, total network size, entity-attribution accuracy, individual-domain correctness, future performance, or final replay status.
Population and sample
The frozen provisional positive population is N = 11,656. Fifty development observations were excluded from the fresh draw to avoid recycling known information, leaving Nf = 11,606. A precommitted sample of n = 500 was selected without replacement by a deterministic procedure fixed before outcomes were known. Identities, exact order, and seed remain confidential.
Conservative accounting retained 496 successes and treated all 4 alerts as errors, although later review confirmed zero network-membership failures. Point precision is 496 / 500 = 99.20%.
Exact one-sided calculation
P_F(X ≤ x) = Σ [C(F,k) × C(Nf−F,n−k)] / C(Nf,n)
x = 4, Nf = 11,606, n = 500
F_fresh = 208
F_total = 208 + 50 = 258
p_min = (11,656 − 258) / 11,656 = 97.7865477%Exact one-sided 95% confidence interval: [97.79%, 100%]. Across comparable repetitions, at least 95% of resulting lower bounds would not exceed true precision. This is not a 95% probability assigned to each domain.
Under the stated model, the lower bound corresponds to at least 11,398 of 11,656 classifications being correct and at most 258 being misclassified. It is unknown which classifications those could be. This guarantees no individual domain.
Presumed victim — unverified may later describe an apparent copied identity in a released row. That field is outside the 99.20% estimate and its confidence interval. Ambiguous identity is shown as Unresolved and requires a separate future audit.
Limits and disclosure boundary
Random-sampling uncertainty is quantified; shared systematic bias may remain. Full replay and final review are separate safeguards. ScamNet’s methodology is versioned and evolves as evidence and conditions change. Operational detection, discovery, and adaptation details are deliberately withheld to reduce evasion risk.
Citation: Guillaume Bolduc (2026). ScamNet provisional network-membership precision checkpoint, public method version 1. HORTRAME. Statistical resolution seal SHA-256: 9dda7c0814098d56d847d30c0eb5d42750e5926b00a40ef42c1c0bf174bd9786.
Ce qui est mesuré
Cette estimation mesure seulement la précision des classifications positives provisoires d’appartenance au réseau à un checkpoint gelé. Elle ne mesure ni rappel, ni taille totale du réseau, ni exactitude de l’attribution d’entités, ni domaine individuel, ni performance future, ni état final du replay.
Population et échantillon
La population positive provisoire gelée est N = 11 656. Cinquante observations de développement furent exclues du tirage frais pour éviter de recycler de l’information connue, laissant Nf = 11 606. Un échantillon pré-engagé de n = 500 fut choisi sans remplacement par une procédure déterministe fixée avant de connaître les résultats. Identités, ordre exact et graine restent confidentiels.
Le calcul conservateur retient 496 succès et traite les 4 alertes comme erreurs, même si l’examen ultérieur n’a confirmé aucun échec d’appartenance. Précision ponctuelle: 496 / 500 = 99,20 %.
Calcul unilatéral exact
P_F(X ≤ x) = Σ [C(F,k) × C(Nf−F,n−k)] / C(Nf,n)
x = 4, Nf = 11 606, n = 500
F_frais = 208
F_total = 208 + 50 = 258
p_min = (11 656 − 258) / 11 656 = 97,7865477 %Intervalle de confiance unilatéral exact à 95 %: [97,79 %, 100 %]. Dans des répétitions comparables, au moins 95 % des bornes inférieures ne dépasseraient pas la vraie précision. Ce n’est pas une probabilité de 95 % attachée à chaque domaine.
Selon le modèle déclaré, la borne inférieure correspond à au moins 11 398 classifications correctes sur 11 656 et au plus 258 classifications erronées. On ignore lesquelles. Cela ne garantit donc aucun domaine individuel.
Victime présumée — non vérifiée pourra décrire plus tard une identité apparemment copiée dans une ligne diffusée. Ce champ reste hors de l’estimation 99,20 % et de son intervalle. Une identité ambiguë est affichée comme Non résolue et exige un futur audit séparé.
Limites et frontière de divulgation
L’incertitude du tirage aléatoire est quantifiée; un biais systématique partagé peut subsister. Replay complet et examen final restent des barrières distinctes. La méthode ScamNet est versionnée et évolue avec les preuves et conditions. Les détails opérationnels de détection, découverte et adaptation sont retenus pour réduire le risque d’évasion.
Citation: Guillaume Bolduc (2026). Checkpoint de précision provisoire d’appartenance au réseau ScamNet, méthode publique version 1. HORTRAME. SHA-256 du sceau de résolution statistique: 9dda7c0814098d56d847d30c0eb5d42750e5926b00a40ef42c1c0bf174bd9786.
provisional strict membership decisions
fresh random audit sample
public provisional rows
SEPARATELY AUDITED SUBSET
Where victim organizations are based
—
—
Country shading shows each country’s share of the mapped victim organizations with separately reviewed legal-country evidence. It does not locate scam operators, servers, customers, or people.
Coverage is currently too small to infer a global distribution. The 11,656 provisional network matches are excluded until victim identity and location pass their own evidence gates.
Map uses only its separately audited victim geography aggregate. Table country and state/province values are not yet analysed and do not color the map.
Map geometry: Natural Earth 5.1.1
PHASE 1 · PROVISIONAL CHECKPOINT
Validation is still in progress
14,000 candidates were processed; 11,656 received provisional positive membership classifications. 11,652 provisional rows are currently displayed. No row is individually guaranteed and identity hypotheses remain unverified.
—candidates processed
—provisional strict matches
0public provisional rows
—
PROVISIONAL PUBLIC TABLE
Provisional network matches
Read every row as provisional.
Population-level membership precision does not guarantee any individual domain. “Presumed victim — unverified” is an observed page-title hypothesis, not a verified attribution. Presumed victim domain, country, and state/province are marked “Not yet analysed” until separate evidence qualifies.
Warning: domain links open observed live sites that may be malicious, deceptive, changed, or unavailable. Use an isolated research environment.
Columns
Select a column heading to sort. Shift-select adds a secondary sort.
Filter records
Combine conditions with nested AND / OR groups.
No records match these filters.
PUBLIC EVIDENCE BOUNDARY
Bounded provisional disclosure
This page publishes aggregate confidence evidence and a sanitized provisional table. It excludes authoritative entity-attribution claims, raw captures, infrastructure fingerprints, operational methods, and historical or quarantined records.