Systems operational

847 sources polling now Last verdict 3s ago Uptime 99.97%

Data sources & methodology

The verdict layer
for the public market

12,847 named sources, 30+ AI models, 847,000 entities monitored across the registry. Every live counter, big-win row, RTP scan, flagged scam, and ranking position on this site sources back to a public-market signal we can name. Read this page like a regulator.

12,847Named upstream sourcesevery one cite-able

2.4BData points per dayingested at the edge

30+AI models active12 frontier, 18 specialist

847KEntities monitoredacross the registry

99.97%Uptime, trailing 90dsee /status

9Regions, 4 continentsedge + reconciliation

Trust commitments

Four commitments, every verdict

These apply to every surface ShouldEye ships, regardless of vertical. Per-vertical sections below extend them with named upstream sources and polling cadences.

We read the public market

Operator info panels, public big-win streams, on-chain settlement, regulator filings, retailer feeds. Every signal is cite-able. Every source is named on this page.

We never sit between you and the operator

ShouldEye does not proxy spins, deposits, trades, baskets, or any operator-side action. The verdict layer reads. It does not transact.

We do not collect your activity at the operator

Spin history, basket, wallet, positions, and operator-side identity stay with the operator. Site telemetry is page-level, scoped to your ShouldEye account, never sold, never shared with operators.

We mask what the source masks

Operator-side player handles stay as the operator publishes them. Wallet addresses truncate. Email and phone never display. PII never leaves the source.

The Advisory Board

30+ models. One verdict.

Every ShouldEye verdict is the synthesis of multiple frontier models, in-house specialist fine-tunes, and a routing classifier that decides which engines see which query. No single engine ever gets the final word. The registry below names twelve of the engines in current production. The full Board, including standby engines and shadow models in evaluation, is published in the operating-agreement appendix.

GPT-5 TurboFrontier

OpenAI

Synthesis and complex reasoning. Lead author on contested verdicts.

1.8Mqueries / day

1.4sp50 latency

Claude 4 OpusFrontier

Anthropic

Nuance, multi-step analysis, regulator-grade copy review.

1.2Mqueries / day

1.7sp50 latency

Gemini 2 UltraFrontier

Google

Multi-modal. Reads PDFs of regulator filings and pricing screenshots.

640kqueries / day

1.9sp50 latency

DeepSeek V3Frontier

DeepSeek

Cost-efficient synthesis for high-volume verticals.

4.2Mqueries / day

0.9sp50 latency

Llama 4 405BFrontier

The Router

Every query is classified by intent before any frontier engine sees it. Slot questions fan out to SE-Slots-Math plus three frontier models. Rug-check queries trigger the on-chain pull, SE-Rug-Forensics, and four frontier models in parallel. The router itself is a small in-house classifier optimized for sub-50ms decisioning so the fan-out is invisible to the user.

The Reconciliation

Engines vote. When they agree, the verdict ships immediately. When they disagree, SE-Trust-Graph-1.5 drills into the source data, names the dissenting engines, and surfaces the disagreement on the verdict artifact. The verdict ships with the dissent visible. No single engine ever gets the final word.

The Audit Log

Every verdict is logged with the engines that voted, the sources they pulled, the polling timestamps, and the dissent summary at issue time. The log is append-only and powers the public /track-record ledger. If a verdict is later contradicted, the original log entry stays in place, with the contradiction appended.

The verdict pipeline

From public signal to public verdict, in five stages

Every verdict travels the same five stages. The pipeline is the same whether the query is a slot RTP scan, a rug-check on a fresh contract, a deal-verification pull, or a scam check on a typed domain. The work each stage does is named, the latency budget is named, the artifacts at every stage are auditable.

Ingestion

Public-market signals are pulled into the system at the edge. Polling cadences are public, listed per vertical below, falsifiable against the source.

continuous

WHOIS / RDAP
Polymarket on-chain settlement
Operator info-panel scrape
Retailer price feeds
Regulator filings

Enrichment

Raw signals are normalized, deduplicated, and stamped with the entity they refer to. Stamping is what makes a verdict cite-able later. Every observation carries the source URL, the timestamp, and the entity ID.

< 200ms per record

Entity resolution
Deduplication
Geolocation
Tokenization
Source-URL retention

Cross-check

Each fact is reconciled against at least one independent source before it can drive a verdict. Single-source facts can inform a verdict but cannot be the only basis for it. Cross-check is what kills false positives.

< 800ms typical

Cross-vendor price
Cross-regulator filings
Cross-RTP info-panel scrape
Cross-on-chain explorer

Verdict

The Advisory Board synthesizes. Frontier models read the reconciled facts and produce a draft. Specialist engines weigh in on domain math. SE-Trust-Graph reconciles dissent and ships the final call with a confidence band.

1.2s p50

Frontier model fan-out
Specialist model weigh-in
Dissent reconciliation
Confidence calibration

Eight surfaces, eight source registers

Each register names the upstream sources, the polling cadence, the Board engines that actually issue the verdicts, the masking policy, and what we explicitly do not collect for that surface.

Companies & entity registry

The master registry every verdict reads from

Before any verdict is issued, the system identifies the entity. The companies registry is the canonical record: 847,000 companies, 12,000 brands, 4 million domains, reconciled against eight independent registers. Every other vertical on this page reads off this registry first.

847KCompanies indexed

4.1MDomains in registry

8Independent registers

Board engines on call

SE-Trust-Graph-1.5 GPT-5 Turbo Claude 4 Opus Gemini 2 Ultra (filings OCR)

Surfaces this lineage informs

/directory/companies /company-check /global-trust /rooms

Upstream sources (7)

WHOIS and RDAP feeds via authoritative registrars. Domain registration, ownership transfer, registrar history.
Companies House (UK), SEC EDGAR (US), OpenCorporates global registry, KvK (NL), Bundesanzeiger (DE), AFIP (AR), ASIC (AU).
LinkedIn Company pages, public profile data only. Crunchbase open API for funding and founder records.
Better Business Bureau active and inactive listings. Trustpilot business pages and verified-reviewer profiles.
Wikidata company entries and Wikipedia article corpora, refreshed daily off the dump.
OpenSanctions and OFAC consolidated lists for sanctioned and politically-exposed entities.
Government open data: FCC license register, FAA operator register, USPS business locator, FTC enforcement archive.

Polling cadence

WHOIS once per hour per tracked domain. Public registers refreshed daily. Social profile pages weekly. OFAC and OpenSanctions every six hours. New-entity discovery rolling, two-minute floor.

What we mask

Director and beneficial-owner names display only as the source publishes them. PII never leaves the source. Truncated handles ("J*** S***") stay truncated.

What we do not collect

Private corporate filings, internal email addresses, employee-list scrapes, or any data behind a paywall or login. The registry only ingests public-record material.

Methodology last reviewed May 2026

Browse the registry

Scam intel & rug detection

How we know a site or contract is dangerous

Scam, rug, and account-locked verdicts are reconciled from regulator paperwork, registration histories, on-chain forensics, and public consumer outcomes. No private complaint pipelines. No paid removals. Median time-to-flag against MetaMask blocklist: 11 days ahead.

4.1MScam patterns trained on

142KDomains flagged all-time

11dMedian lead time vs MetaMask

Board engines on call

SE-Scam-3.2 SE-Rug-Forensics-2.1 GPT-5 Turbo Claude 4 Opus SE-Trust-Graph-1.5

Surfaces this lineage informs

/is-it-a-scam /rug-check /company-check /global-trust

Upstream sources (7)

WHOIS and RDAP feeds for domain registration history and ownership churn. Lookalike-domain detection via Punycode-normalized fuzzing.
Public regulator registers. FTC consumer-protection orders, FCA Warning List, ASIC Investor Alert, SEC EDGAR enforcement, ESMA warnings, MAS Investor Alert List, JFSA alerts.
Better Business Bureau and Trustpilot public review corpora. Reddit and X public posts on tracked entities via firehose APIs.
On-chain rug indices. TokenSniffer, GoPlus, Honeypot.is, RugDoc cross-reference for ERC-20, BEP-20, Solana SPL, and Tron TRC-20 contracts.
Wallet blocklists. MetaMask, Phantom, Trust Wallet, Rabby phishing and scam lists, polled hourly.
Archived policy snapshots. Wayback Machine plus our own daily archive of every tracked merchant's terms, refund policy, and contact page.
Phishing kit fingerprint database, hosted in-house, regenerated weekly off observed attack infrastructure.

Polling cadence

WHOIS once per hour per tracked domain. Regulator filings once per day. Review corpora rolling 60 seconds. On-chain rug indices rolling 30 seconds. Wallet blocklists hourly. Policy archive once per day.

What we mask

Reporter usernames stay as the source masks them. Original review handles where Trustpilot or BBB has already redacted them. Wallet addresses display truncated. Email and phone never display.

What we do not collect

Your search queries outside your own ShouldEye account. Anonymous visitor pastes into the chat are session-local and never written to disk.

Methodology last reviewed May 2026

See a scam verdict

Top rankings & trust scores

How we rank, and why our rankings can be audited

Rankings on /top-rankings are not opinion. Every position is a function of measured signals. The Top Rankings layer reads off the companies registry and the verdict ledger, then applies a weighted scoring model that is itself public. No paid placement. No affiliate-influenced positions.

127Ranking categories live

24hScore recompute cadence

0Paid placements, ever

Board engines on call

SE-Trust-Graph-1.5 GPT-5 Turbo Claude 4 Opus DeepSeek V3

Surfaces this lineage informs

/top-rankings /top-rankings/all /trending-companies /global-trust

Upstream sources (6)

Verdict ledger. Every verdict ShouldEye has ever shipped on the entity, weighted by recency and outcome confirmation.
Trust-score components: regulator status, refund-signal volume, public complaint velocity, ownership transparency, policy clarity, response latency.
Cross-vendor reviews aggregated. Trustpilot, BBB, Google Reviews, Reddit subreddit sentiment scored.
Real consumer outcomes where public: chargeback rates published by card networks, regulator settlements with restitution amounts.
Press citations across cited outlets (Reuters, Bloomberg, AP, BBC, FT, WSJ, plus accredited national press).
Self-published transparency reports where the entity publishes them.

Polling cadence

Rankings recomputed daily at 04:00 UTC. Trust-score components refreshed hourly. Press-citation scan rolling 5 minutes. Ledger weight recompute live on each new verdict.

What we mask

Nothing on the scoring side. The scoring model is public. Every ranking row carries a "why this position" link that opens the score breakdown.

What we do not collect

Paid placement. ShouldEye has never accepted payment to influence a ranking position, will not, and the policy is in the operating agreement, not just the marketing.

Methodology last reviewed April 2026

See live rankings

Deal & price intelligence

How we know a price is actually a deal

Every deal verdict is anchored against the price-history archive. We do not take a vendor's word for "60% off". We compare today's price against three years of every observed price for that SKU. Median fake-markup detection: 8 minutes from claim posting to flag.

16MSKUs in price archive

3yPrice history depth

60sTracked-SKU refresh

Board engines on call

SE-Deal-Pricing-2.4 DeepSeek V3 GPT-5 Turbo SE-Trust-Graph-1.5

Surfaces this lineage informs

/directory/deals /instant-checks

Upstream sources (6)

Retailer public price feeds. Amazon Product Advertising API, Newegg, Best Buy, B&H, Walmart Open API, Target Open API.
Cross-vendor aggregators. PCPartPicker for component categories. Skinflint and Pricerunner for EU pricing.
Manufacturer MSRP sheets. Apple, Samsung, NVIDIA, AMD, Sony, LG, Dell, HP, plus the rolling tracked-brand list.
Internal price-history archive. Three years of every observed price for every tracked SKU, stamped at observation time, retained for audit.
Coupon-code repositories with verification metadata: code, observed expiration, observed stack policy, observed cashback compatibility.
Stock-availability feeds for the in-stock and restock alerts (retailer-published).

Polling cadence

Tracked SKUs every 60 to 120 seconds. Full catalogue refresh once per day. MSRP sheets once per week and on every detected manufacturer announcement. Coupon codes verified on first use and every six hours thereafter.

What we mask

Nothing on the deal side. Retailer name, observed price, and observation timestamp are the disclosure.

What we do not collect

Your basket, wishlist, or purchase history outside your ShouldEye account. Affiliate clicks route from your browser direct to the retailer. We never intercept your checkout.

Methodology last reviewed May 2026

See live deals

Prediction markets

How we know where the smart money is leaning

Predict verdicts read off public, on-chain, regulated, or self-published market data. We do not custody a position or settle a trade. ShouldEye is a read layer over the markets that already exist. Settlement data is on-chain where possible so any reader can independently verify.

2,400+Live markets indexed

15sPrice tape refresh

4Settlement venues

Board engines on call

GPT-5 Turbo Claude 4 Opus Grok 3 SE-Trust-Graph-1.5

Surfaces this lineage informs

/predict /predict/all

Upstream sources (6)

Polymarket public on-chain markets, USDC settlement on Polygon. Block-by-block reconciliation.
Kalshi public market data API, CFTC-regulated event contracts.
Manifold Markets public API, prediction-market data.
PredictIt public market data where active.
News sentiment from cited outlets only. Reuters, Bloomberg, Associated Press, BBC, FT, WSJ. Editorial corpora excluded.
Regulator and event filings. SEC EDGAR, US federal court dockets, the Federal Register, UK Companies House, ECJ judgment archive.

Polling cadence

Market prices every 15 seconds. News sentiment every 5 minutes. On-chain settlement live, block-by-block. Regulator filings polled once per day.

What we mask

Nothing. Markets and price tape are public.

What we do not collect

Your positions on any third-party prediction market. We do not custody, route, or settle prediction-market trades.

Methodology last reviewed May 2026

See live markets

Slots & gaming

How we know which slot is paying right now

Big-win rows, live RTPs, and bonus-EV verdicts read off operator-published streams and slot info-panels. Player handles stay as the operator publishes them. We never sit between you and the casino. The verdict layer reads. It does not transact.

4,800+Slot titles tracked

47Operators in coverage

24hFull RTP rescan cycle

Board engines on call

SE-Slots-Math-1.8 SE-Trust-Graph-1.5 GPT-5 Turbo DeepSeek V3

Surfaces this lineage informs

/gaming /directory/games

Upstream sources (6)

Operator big-win streams. Stake.com, BC.Game, Roobet, Rollbit, BetFury, Wildz, HighFlyer published live wins feeds.
Provider public top-wins. Pragmatic Play, Hacksaw Gaming, Nolimit City, Relax Gaming, Push Gaming, Big Time Gaming.
On-chain provably-fair settlement on operators that sign each spin to chain.
Slot info-panel scraping for RTP variants at every tracked operator. The info-panel is the regulator-disclosed surface and is the canonical source for the deployed variant.
Provider math sheets for slot certification, used to reconcile the operator-published variant against the certified payout curve.
Operator bonus terms feeds. Welcome offers, reload offers, wagering multipliers, max bet during bonus, contribution table by game.

Polling cadence

Operator big-win streams every 30 seconds. RTP info-panel rolling scan, full coverage every 24 hours per operator. Provider top-wins every 5 minutes. Bonus terms polled hourly.

What we mask

Operator-side player handles stay as the operator publishes them. The "D**4f" you see is the operator's mask, not ours. We do not unmask, deanonymise, or correlate handles across operators.

What we do not collect

Your gambling session. Your deposits, withdrawals, spin history, balance, or operator account identity. We never sit between you and the casino.

Methodology last reviewed May 2026

See live slots

Side-by-side product analysis

How we know one product really beats another

Every comparison verdict reconciles three independent registers: the manufacturer's spec sheet, certified test reports, and the owner-review corpus. When the three disagree, the verdict surfaces the disagreement. We do not pretend consensus exists when it does not.

38KSKUs comparable

12Test-lab partners

3Independent registers

Board engines on call

Claude 4 Opus Gemini 2 Ultra (spec PDFs) GPT-5 Turbo SE-Trust-Graph-1.5

Surfaces this lineage informs

/compare /products /instant-checks

Upstream sources (6)

Manufacturer-published spec sheets, scraped on release and on every detected spec update. PDF OCR for filings.
Certified test reports. DXOmark, RTINGS, AnandTech lab data, Tom's Hardware lab data, Notebookcheck, Wirecutter test logs.
Retailer SKU registries for live availability and bundle composition.
Owner-review corpora. Amazon verified-purchase reviews, manufacturer-hosted reviews, Trustpilot, Reddit subreddit threads.
Repair indices. iFixit repairability scores, manufacturer parts catalogs.
Warranty and recall registers. NHTSA, CPSC, FDA, plus international counterparts.

Polling cadence

Spec sheets once per week per SKU and on every detected manufacturer announcement. Reviews rolling daily. Availability rolling hourly. Recall and warranty registers daily.

What we mask

Nothing on the spec side. Manufacturer name, SKU, observed availability, and observed review count are the disclosure.

What we do not collect

Your past purchases or browsing history outside your ShouldEye account.

Methodology last reviewed April 2026

Compare anything

The trust graph

How the verdict you see is actually built

The trust graph sits above every vertical above. Every claim from every Expert Engine, on every verdict, is reconciled against the public-market source it rests on before the verdict ships. When sources disagree, dissent is named. When sources are stale, the verdict is flagged.

2.4MVerdicts archived

96.8%Confirmed accuracy

∞Retention, append-only

Board engines on call

SE-Trust-Graph-1.5 SE-Router-2.0 Every Board engine, on every call

Surfaces this lineage informs

/how-shouldeye-works /track-record

Upstream sources (3)

The union of every named source on this page.
Internal verdict archive. Every verdict ShouldEye has ever shipped, with the source list, agreement rate, and dissent summary as of issue time, retained for audit.
External outcome reconciler. Watches for confirmation signals on every pending verdict and flips status when one lands.

Polling cadence

Live. The trust graph reconciles on every verdict request and stamps the verdict with the source list, polling timestamps, agreement rate, and dissent summary at the moment of issue. The outcome reconciler runs continuously.

What we mask

Nothing in the audit log itself. The audit log is the receipt.

What we do not collect

Any data not already collected by the per-vertical layers above. The trust graph is a reconciliation surface, not a separate ingestion path.

Methodology last reviewed May 2026

Read the ledger

Infrastructure & compliance

The boring stuff a procurement team will ask

Where the system runs. What it complies with. How much it processes. The audit cadence. Read this row before you ask us to fill in your vendor questionnaire.

Global footprint

9 regions, 4 continents
Edge ingestion in us-east, us-west, eu-west, eu-central, ap-south, ap-northeast, sa-east, af-south
Reconciliation primary in us-east-1, hot standby in eu-central-1
Live status feed at status.shouldeye.com

Compliance & audit

SOC 2 Type II, audit in progress (target completion Q3 2026)
GDPR-aligned. Article 30 records of processing maintained
CCPA + CPRA aligned. Consumer requests inbox at privacy@
ISO/IEC 27001 controls implemented. Certification target 2027
Public security disclosure policy at /security

Throughput & retention

2.4B data points ingested per day, p99
12.4M model invocations per day across Board engines
Verdict archive: append-only, retained indefinitely
Source-pull retention: 18 months, gzipped, queryable
Replication factor 3, RPO < 60s, RTO < 5m

SOC 2 in audit

GDPR aligned

CCPA aligned

ISO 27001 controls

99.97% uptime, 90d

Have a question about a verdict you saw?

Open it in the Advisory Board chat and we will show you the named source list, the polling timestamp, the Board engines that voted, and the agreement rate behind that exact verdict.

See the track record

Four commitments, every verdict

We read the public market

We never sit between you and the operator

We do not collect your activity at the operator

We mask what the source masks

30+ models. One verdict.

The Router

The Reconciliation

The Audit Log

From public signal to public verdict, in five stages

Ingestion

Enrichment

Cross-check

Verdict

Archive

Eight surfaces, eight source registers

The boring stuff a procurement team will ask

Global footprint

Compliance & audit

Throughput & retention

Have a question about a verdict you saw?