Methodology
Data tables · Onward reporting · Definitions · Gaps
This is the data half of the tg.pe transparency report: the full tables behind every summary figure, and how each one is defined. Two parts have pages of their own: the safety pipeline and RDAP bootstrap coverage.
It is written for security researchers, network operators, and internet-governance readers, and for anyone running a small service who wants to copy the parts that work. Where a figure is preliminary or the sample is small, the text says so rather than rounding it into a claim.
Measurement: domain age and RDAP coverage
Resolving the age of every destination's registrable domain, and recording which source answered, turns a routine safety check into a measurement of where RDAP is actually deployed.
That measurement, its per-ccTLD table, and the six limits it cannot be cited without are on RDAP bootstrap coverage. Cache TTLs a reimplementer needs are under definitions; what the instrument still cannot answer is in the gaps list.
The safety pipeline
Every link is checked before it goes live, again in the background once the response is flushed, and again on later clicks.
Each stage's data source, what it does on a hit, its fail mode, its change history, and the reasoning behind the non-obvious choices are on the safety pipeline. What the automation has caught is summarised there; the missing per-stage funnel is in the gaps list.
Where links come from
Most links come from identified Telegram users via @tgpebot; the rest from the anonymous Web form and a few API integrations. The chart breaks the Web share down by submitter country (Cloudflare's CF-IPCountry at submission, so VPS/VPN endpoints in FI/NL/US/RU read as rented infrastructure, not the submitter's real location). Click a legend item to focus it.
Telegram dominates volume every year. Within Web, Taiwan led 2019 and 2022, Russia every other year through 2025 (legitimate users mixed with automated probes), and India so far in 2026. The August 2026 test-link flood is left out of this chart and the one below, because it forged the country tag.
The 2025 Web spike came overwhelmingly from RU/US automated submissions, with sharp single-hour UTC peaks consistent with cron-driven bots. Most appear to have been caught at the safety pipeline, though the per-stage catch rate is not yet quantified (see the gaps list).
By channel, and where removed abuse came from (charts)
Total submissions by channel (Web / Telegram / API), with unique submitters overlaid:
Of the links that were removed, where did the submitter sit? Telegram is deduplicated to distinct users so ban cascades don't inflate it; Web stays per-link.
The asymmetry the charts show is structural. Identified Telegram users face fewer front-door checks, so more slips through at submission.
It gets caught later instead, by external reports, the 7-day click-driven re-check, or manual review, and one ban then clears that user's entire history. Web is filtered hard at the front door, so less of it reaches the point of removal. (The per-channel counts are read off the charts above.)
When links get shortened (submission-rhythm heatmaps)
Submissions over the last 104 weeks, day-of-week × hour-of-day in UTC, Web vs Telegram:
| 00 | 06 | 12 | 18 | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sun | ||||||||||||||||||||||||
| Mon | ||||||||||||||||||||||||
| Tue | ||||||||||||||||||||||||
| Wed | ||||||||||||||||||||||||
| Thu | ||||||||||||||||||||||||
| Fri | ||||||||||||||||||||||||
| Sat |
| 00 | 06 | 12 | 18 | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sun | ||||||||||||||||||||||||
| Mon | ||||||||||||||||||||||||
| Tue | ||||||||||||||||||||||||
| Wed | ||||||||||||||||||||||||
| Thu | ||||||||||||||||||||||||
| Fri | ||||||||||||||||||||||||
| Sat |
Web shows sharp isolated peaks at exact UTC hours, the signature of cron-scheduled bots. Telegram's heat is diffuse across the week, human use across many timezones. That asymmetry is the reason Web gets stricter rate limits and new-domain thresholds than Telegram. The heatmaps confirm that split rather than driving it.
What people shorten
Top destination hosts among active links (self-referential tg.pe links excluded):
| Host | Active links |
|---|---|
| t.me | 2,852 |
| telegra.ph | 1,888 |
| drive.google.com | 1,219 |
| facebook.com | 1,039 |
| hinet.net | 1,016 |
| docs.google.com | 937 |
| youtube.com | 569 |
| elnacional.cat | 557 |
Mostly messaging-app deep links, cloud-file sharing, and Taiwan-local content.
Click distribution (the long tail)
Click counting started in April 2026 (the SQLite→MySQL migration), so earlier clicks aren't counted, and API submissions are excluded from the split. The buckets are a plain group-by with nothing to curate, so they are published live rather than pasted here: Stats › Click distribution. The long tail they show is why click-driven re-check leaves most links unchecked.
Removal timing & re-check coverage
Removal authority comes from the public Terms every submitter accepts. One Telegram ban soft-deletes every link that user ever created. Re-check is click-driven: a click on a link not checked in the last 7 days triggers a fresh Safe Browsing call.
The verified figure is a ~4-hour median from external report to removal. That median describes active response, not all removals: backlog cleanup and ban-cascade sweeps run against much older links and would pull it out of shape.
Time from creation to removal
How old each link was when it came down. The top group is active response to fresh abuse; the bottom group is backlog cleanup and ban-cascade sweeps, whose timings reflect when the operator audited, not when the abuse was first detected.
| URL age at removal | Removals | % |
|---|---|---|
| Active response | ||
| Under 1 hour | 79 | 7.0% |
| 1 – 24 hours | 272 | 24.2% |
| 1 – 7 days | 212 | 18.9% |
| Backlog cleanup & ban cascade | ||
| 7 days – 1 year | 488 | 43.4% |
| Over 1 year | 73 | 6.5% |
1,124 removals, author self-deletes excluded. Two batches are left out: 1,296 links soft-deleted together on 2025-01-06 21:24, a one-time inventory cleanup whose synthetic timestamps would distort the distribution; and the 391 links of the August 2026 test-link flood (what it was), removed within days of creation, which would read as fast active response.
Re-check coverage
Links re-checked against Google Safe Browsing on a later click since the April 2026 integration. Only the latest check of each link is stored, so a link counts once, in the window that holds its most recent re-check; how many calls were made in total is not recorded.
| Most recent re-check | Links |
|---|---|
| Last 7 days | 457 |
| Last 30 days | 1,859 |
| Ever (since integration) | 4,759 |
As of 2026-10-26. A link counts as re-checked when its last_safety_check_at is at least 7 days after its creation, since the check at submission is stamped at creation and a re-check only fires once 7 days have passed.
Of the 30,965 active links, 26,414 have never been re-checked. Most pre-date the Safe Browsing integration; their last_safety_check_at was backfilled to creation date at the April 2026 migration, and their first live check fires on the next click. Most never get one (see the long tail under Click distribution), which is by design: idle URLs shouldn't burn API quota until somebody is about to be exposed.
Abuse reports: reporters & yearly volume
External reports come from national CERTs, anti-fraud vendors, and ordinary recipients. Over the service's life the top three sources (CIRCL, the US IRS, and Netcraft) are ~77% of volume; the long tail confirms tg.pe sits on multiple watchlists. The narrative cases, including the worst-category one, are in the transparency report.
Yearly volume and response time
Reports = external incoming. Links reported = distinct tg.pe/{code} short links named in those reports. Response time = median time from first report to removal.
| Year | Reports | Links reported | Response time | Notes |
|---|---|---|---|---|
| 2026 (YTD) | 6 | 4 | 4.0 h | Includes the parcel-delivery scam reported by an ordinary user (the featured case). One July report named a link already removed the day before, so it has no response time. |
| 2025 | 18 | 3 | 4.6 h | Q4 spike (mail-server-side incident, not tg.pe-shortened content); same-day handling. |
| 2024 | 8 | 3 | 5.2 h | Mostly Netcraft; same-day handling. |
| 2023 | 39 | 7 | 2.9 h | Volume dropped; remaining reports handled within hours. |
| 2022 | 207 | 26 | 5.1 h | IRS impersonation + CIRCL bulk wave, the largest sustained external pressure to date. |
| 2021 | 25 | 37 | 2.5 h | First-line defences built (see timeline). Many reports came as multi-code batches. |
| 2020 | 12 | 2 | 4.8 h | First reports arrived (Cloudflare / Netcraft / DigitalOcean). Tiny volume. |
The 2022 row is the most meaningful: even at peak external pressure, median report-to-removal held ~5.1 hours.
Reports from removed-link pages
Since 3 August 2026 the 404 page of a removed link asks the visitor where the link reached them, and opens their mail client with a report to abuse@tg.pe. The page offers this only for links taken down by the operator, by a ban, or by the Safe Browsing re-check, and for links past their expiry; an author's own deletion and older removals recorded without a reason show a plain not-found. Every such report therefore arrives after the link is dead. None of them triggered a removal, and none is counted in the table above.
| How the page asked | Emails | With an answer |
|---|---|---|
| Questions pre-filled in the email (3 Aug – 2 Oct 2026) | 36 | 2 |
| Questions answered on the page first (from 2 Oct 2026) | 9 | 8 |
Counted by hand from the operator's mailbox on 2026-10-26. The page sets the subject line (Scam link: tg.pe/{code}), so a subject search finds this channel's mail, short of anyone who rewrote the subject. “With an answer” means at least one question answered with something about the link or how it arrived, or a screenshot attached. A greeting, a signature, or one unintelligible reply counts as no answer.
- 45 emails from 38 sender addresses about 12 links. One link drew 13 of them, and the 8 answered emails after the change came from 6 people. The before/after split rests on 9 emails, so read it as a direction, not a rate.
- Reaching the page is not evidence of abuse. Of the 12 links, 8 had been removed (7 by operator review, 1 by a ban) and 4 had expired on the short automatic lifetime set at submission (two by a VirusTotal verdict, one by the new-domain check, one by the ad-smartlink pattern). Those 4 drew 13 of the 45 emails. The subject line is the page's wording, not the visitor's verdict.
- The answers name the delivery, not the destination: SMS from a short code posing as Peru's FISE gas subsidy or a SUTRAN traffic fine, a betting game pushed from phone numbers, a job offer on a website, a loan repayment, and one screenshot of the scam text.
- Reports outlive the takedown. The two most-reported links were still being reported 38 and 40 days after removal.
- There is no denominator. Visits to a removed link's page are not recorded in the database, so the share of visitors who report cannot be computed.
Onward reporting: channels and case detail
When the operator confirms a removed destination is malicious and classifies it, the indicator is reported onward to the abuse-handling community, so the underlying page can be acted on wherever else it is reaching people.
Each report goes only to the channels that fit it; a mis-routed report costs reporter standing with that channel. Counting only successful sends: 35 via API and 81 assisted-manual.
MISP events carry machine-readable taxonomy tags: TLP:GREEN or TLP:CLEAR, source:tg.pe, an incident classification, and the destination's registration age, the same datum the new-domain check resolves (the measurement section). When a reported destination is later taken down, the event is updated in place with the outcome and timeline.
CSAM is never sent to these phishing/malware feeds; it is referred out of band to NCMEC, the IWF, or law enforcement.
| Channel | Method | Reports |
|---|---|---|
| Browser & takedown vendors · 38 | ||
| Google Safe Browsing | manual | 15 |
| Netcraft | API + manual | 13 |
| Microsoft SmartScreen | manual | 5 |
| Phishing Initiative | manual | 3 |
| phish.report | API + manual | 2 |
| Threat-intel feeds & public scans · 27 | ||
| CIRCL MISP | API | 12 |
| abuse.ch ThreatFox | API | 8 |
| urlscan.io (public scans) | API | 7 |
| Hosting & network providers · 21 | ||
| Framer (website builder) | manual | 7 |
| Cloudflare | manual | 6 |
| Origin hosts and their upstreams: OMEGATECH, VIRTUO, AWS, GMO Pepabo, IPVendetta, InterServer, UltaHost, First Colo (one each) | manual | 8 |
| DNS registration layer · 14 | ||
| NetBeacon (→ registrar) | manual | 5 |
| NameSilo (registrar, directly) | manual | 3 |
| Other registrars directly: GoDaddy, Namecheap, Spaceship (one each) | manual | 3 |
| Registries directly: .top, .blog, .qpon (one each) | manual | 3 |
| Impersonated brands · 10 | ||
| REA Group, 4-72, EasyPark, Charter (Spectrum), Comcast (Xfinity), Orange, La Poste, A1 Telekom, UCS, UFSC (one each) | manual | 10 |
| Everyone else · 6 | ||
| colCERT, Colombia's national CERT | manual | 2 |
| CERT-In (India) and CAIS (Brazil's academic-network CERT), one each | manual | 2 |
| Telegram (the scam's bot backend) | manual | 1 |
| A card-lookup API whose key was embedded in the scam kit | manual | 1 |
All were phishing or scam; no malware or CSAM has been routed to these feeds. Ten of the sends were made by email outside the reporting tool and logged afterwards from the sent-mail archive (October 2026); the count includes them.
13 August 2026, hour by hour
The case summarised on the transparency report, as logged. All times UTC.
| Time | Event |
|---|---|
| 02:12 | validationreq.top registered (registrar PDR Ltd.) |
| 03:29 | Link submitted to tg.pe; automated checks flag the hours-old destination and contain the link; evidence captured automatically |
| 05:57 | CIRCL MISP event published (event 508641) |
| 06:29 | Reports filed: registrar (via NetBeacon), .TOP registry, Cloudflare, Google Safe Browsing, Microsoft SmartScreen, Netcraft |
| 06:47 | Cloudflare URL-level block verified; sibling URLs on rotated paths still live |
| 09:46 | .TOP registry applies serverHold (ticket #1022402) |
| 09:59 | NS resolution verified dead network-wide |
The page itself showed nothing but a brand mock-up and a "Verification" button; the payment-card form appeared only after a click. An automated capture therefore scored the page benign, and it took operator review to spot the fraud behind the button.
The impersonated brand's security team and the origin hosting provider were notified the same day, and the MISP event was updated afterwards with the takedown outcome and this timeline.
The registrar comparison is one data point, not a pattern; routing through a third-party intake can simply be slower.
An earlier loop, closed more slowly
On 25 May 2026 a tg.pe link to a Flipkart-impersonation phishing page on gauseva.life was removed and reported to five channels. On 2 June 2026 phish.report confirmed the domain suspended by its registrar (case case_fn4ae2bso6ps), eight days from removal to suspension, with an external reference the operator did not generate.
Not every report closes. A cloaked affiliate smartlink reported to Netcraft came back "no threats found", because monetisation redirectors serve benign content to datacentre scanners.
Two more domains have closed since (below), and more have not.
What each layer did, and how fast
Every response observed so far on a reported destination, by the party that responded. The clock starts when that party was sent the report. Two different clocks appear and are labelled: to action is when the site visibly changed, to reply is when a person or system answered. A fast reply with no action is common, so the two must not be compared.
| Case | Who responded | Clock | What happened |
|---|---|---|---|
validationreq.top13 Aug 2026 | Cloudflare | 18 min to action | Blocked the one reported URL; the kit's rotated paths stayed live. |
| .TOP registry, directly | 3 h 17 m to action | serverHold: the whole domain left DNS. | |
| Registrar, via NetBeacon | none in window | No action before the registry's hold landed. | |
paquetes-colombia.blog and envio-colombia.blog16 Aug 2026 | Cloudflare (paquetes) | 2 h 05 m – 2 h 24 m to action | A phishing warning page in front of the site. The range runs from the last check that saw it untouched to Cloudflare's reply saying it had acted. |
| Cloudflare (envio) | 28 m 55 s to reply | Warning page on 5 of the 12 reported URLs; the other 7, including the card form, kept serving through nine re-checks over 20 minutes. | |
| Origin host (AWS) | 3 m 40 s to reply | Declined: the site sits behind a CDN, so no AWS resource could be identified; referred back to Cloudflare. | |
| .blog registry, directly | 7 h 23 m to reply | Forwarded to the registrar; no action of its own. | |
| Registrar (NameSilo), email | 2 h 35 m to reply | Declined the email and asked for its web form. No action came from the email in the next 42 hours. | |
| Registrar (NameSilo), web form | 12 h 37 m to action | clientHold on paquetes and a sibling domain in the same second; envio was not touched. | |
| NetBeacon | 66 s to reply | Automated receipt. The only follow-up that reached the reporter was a Cloudflare ticket; no registrar reply. | |
mittpanel.qpon21 Oct 2026 | Netcraft | same minute | Judged malicious. |
Seven *.framer.app sites23 Oct 2026 | Netcraft, via API | 1 m 50 s – 5 m 17 s to verdict | All seven judged malicious (one was already known). |
| Framer, the platform | 26 h 19 m to reply | Removed all seven. The first reply was automated; the removal notice came the next day, and every site returned 404 when re-checked on 26 Oct. |
Counting domains rather than responses: three reported domains are confirmed suspended (gauseva.life, validationreq.top, paquetes-colombia.blog), the platform removed the seven sites on its subdomains, and one cloaked smartlink never closed. Every row above is a single observation, often the only one of its kind, so none of it is a rate, and nothing here should be read as "typically" or "usually". Most onward reports never send an outcome back to the reporter, which is why the fraction that end in removal is still not measured (see the gaps list).
How a correct report still fails
The cases above failed in ways that a count of reports sent, or of replies received, would not show. Each was seen once or twice; they are listed as mechanisms, not frequencies.
- Fast, at the wrong size. A CDN acts on the exact URLs it was given. It does not extend the block to paths nobody reported, and it may not finish the list it was given: on
envio, 7 of 12 reported URLs kept working after a reply saying access had been restricted. A reply is not a verification. - The intake form is the channel. The same evidence sent by email was declined; sent through the registrar's web form, it led to a hold 12.6 hours later. That is the registrar narrowing intake to one channel, not refusing to act, and it means an unanswered email says little about whether the registrar would have acted.
- A referral loop between CDN and host. The CDN names the origin host but not the server's address; the host needs a resource it can identify, and points back to the CDN. Both sides were polite and opened case numbers, and nothing was removed through that path.
- Scanners see the wall, not the site. Behind a CDN's bot challenge, automated scanners record only the challenge page and report the site as clean: for
paquetes, urlscan.io's page verdict was 0 and a single VirusTotal engine called it malicious, while urlscan's machine-learning engine, which does not judge by the page body alone, scored it 96 and malicious. A near-zero score on a challenge-walled site means "not seen". It stops once the CDN swaps the challenge for its own warning page. - Some reports have no right address. A page on a platform subdomain such as
*.framer.appshares its domain with every other customer, so the DNS-layer channels (registrar, registry, NetBeacon) can only act on the whole platform. Only the platform can remove one site, and here it did, about a day later. - The reporter's cost grows per domain. Several web forms put a human challenge in front of each submission. Every extra domain in a cluster costs the reporter another round; it costs the attacker nothing extra to register.
Bans & domain lists
222 submitters are currently banned, all Telegram users; a ban soft-deletes every link that user ever created. No Web submitter has ever been individually banned; Web abuse is stopped upstream by the safety pipeline, and Web IPs are too mutable to ban usefully. The verified central figure is a median of 2 links before a ban.
Domain lists
| List | Count | Examples | Behaviour |
|---|---|---|---|
| Block list | 135 | Phishing landing pages, malware distribution sites | Refused on both Web and Telegram. |
| High-risk list | 383 | Other URL shorteners, free hosting platforms | Refused on Web only; identified Telegram users pass. |
Links before a ban
| Median (50th percentile) | 2 |
| 75th percentile | 3 |
| 95th percentile | 13 |
Methodology & definitions
- “Removed for safety / abuse” =
short_links.deleted_at IS NOT NULLminus authors deleting their own links: automated re-check removals, operator removals, ban cascades, and older removals recorded without a reason. Bulk cleanups count too, which is why the August 2026 flood is called out wherever the figure appears. - “Banned submitters” = entries in
banned_users. - Country code =
CF-IPCountryat submission time (Web only); VPS/VPN endpoints read as rented infrastructure, not the submitter's residence. - Clicks are counted only since the April 2026 SQLite→MySQL migration; earlier clicks are not counted.
- Cache TTLs a reimplementer needs: a registration date is immutable, so it is cached 30 days; reputation is mutable, so it is cached 1 day.
This report is generated from a snapshot of the production MySQL database plus the operator's mail archive.
What's not in this report (yet)
| Not measured | Why it is absent | What would close it |
|---|---|---|
| Per-stage catch funnel | Per-stage block counts are published live on Stats, but they are numerators only: nothing records how many submissions each stage saw, so the 216 background tightenings have no rate, and a link stopped by three stages is indistinguishable from three links stopped once each. | An integrator query carrying the total scan count, which is what turns that count into a rate and each stage's overlap into a number. At this n it must be clearly labelled as small. |
| Phished-brand attribution | The large majority of reports are phishing, but which brands are most often impersonated has not been tagged. | Per-message body analysis, in progress. |
| Closed-loop takedown rate § | Three reported domains are confirmed suspended (two by a registrar, one by a registry serverHold), a platform removed seven sites on its subdomains, and the responses with an observed outcome are tabled under what each layer did. Most reports never send an outcome back, so there is no denominator. |
Recording the outcome of every onward report, split between throwaway phishing registrations and cloaked monetisation redirectors. |
| What threat feeds would block | In observe mode a feed match is written only to the server's error log, not to a table, so how many submissions enforcement would have refused cannot be counted. The one measured slice is the August 2026 backtest: 28 matches among 30,409 live links, 10 removed on review. | A table row per observe-mode match, which is also what switching to enforcement should wait for. |
| Registrable-domain age at scale RDAP coverage | Legacy removals scrubbed the destination, so the usable sample is small. | Retaining the destination and its source-tagged registration date on every removal going forward, which would turn the two-population shape into a population statistic. |
| Per-lookup RDAP outcome RDAP coverage | resolveDomainAge() computes the source tag, but it is only persisted when a submission also trips the new-domain honeypot, and malicious_detections.source only ever carries virustotal. The outcome survives in a cache that overwrites itself and drops failures, not in a table, which is the wrong instrument for the job. |
Logging one row per age resolution, with the source, the TLD, and the miss cause. That would replace every disclosure under the coverage figure with a clean denominator. |
One correction to the record belongs with that last row: an earlier version of this report claimed the source tag recorded RDAP-hit versus pDNS-fallback per lookup. It does not.
If any of this is useful, for research, for policy work, or because you run a similar service and want to compare notes, the operator would rather share than keep it private: abuse@tg.pe.