What are you working on today?
Multi-model extraction verification.
Dual-model with divergence flags.
Peer metrics cross-checked via FMP.
M-Score, forensic checks, BS+CF verified.
3 models construct the bear case independently.
Live EV/EBITDA, P/E, FCF yield, peer comps.
CEO statements cross-referenced over 4 quarters.
If peers flag a risk and you don't — silence is signal.
Automated multi-lens monitoring. Cron alerts on divergence.
| Metric | GPT-5.5 | Claude Opus | Status | FMP Verified |
|---|---|---|---|---|
| Group uPBT | c£92m | c£92m | Match | £92.8m |
| Retail uPBT | c£30m | c£30m | Match | ~£30m |
| Vet Group PBT | c£83m | c£83m | Match | ~£83m |
| Net Debt | c£20m | c£20m | Partial | £357m inc leases |
| DPS Change | -28% | -35% | Diverge | -43.1% |
| Check | Value | Finding |
|---|---|---|
| Goodwill/Equity | 98.7% | Tangible equity negative. Impairment = £96m write-down |
| Gross Margin (3yr) | 46.8→46.9→45.7% | 120bp structural compression |
| OpCF/PBT | 2.05x | Strong cash conversion |
| FCF/DPS | 4.3x | Dividend well covered |
| Net Debt Defs | £20m vs £357m | £337m lease liabilities excluded |
| Claim | Counter-Evidence | Severity |
|---|---|---|
| "Net debt c£20m" | Total debt £397m. 95% excluded | Misleading |
| "Comfortable with £99m" | FY26 guide was £115-125m, actual £92.8m | Credibility |
| "No CMA impact" | CMA text not analysed independently | Unverified |
| "£20m savings done" | Central costs rose £15.8→£21m | Unreconciled |
| Broker | Rating | Target |
|---|---|---|
| Jefferies | Buy | 265p |
| Canaccord | Buy | 245p |
| Berenberg | Hold | — |
| Extraction | 5/6 |
| Cross-Stmt | 2 Crit |
| Contradiction | 4 Flags |
| Market | Live |
Automated multi-lens monitoring. Cron jobs run every 6 hours across your coverage universe. Only surfaces findings that pass the materiality threshold.
Each node is a stateless function. Failures are isolated. The Synthesis Engine combines all lens outputs into a single narrative with flagged divergences. Self-improving: every divergence becomes training data for model fine-tuning.
Unlike Primer's Triggers (user-prompted, single model), Radar proactively scans with multi-lens verification. Only surfaces findings that exceed materiality thresholds — no noise. The M-Score breach on IWG would have been caught automatically.
| Layer | Primer | MultiLens | Advantage |
|---|---|---|---|
| Frontend | Next.js / Render | Next.js / Vercel | Faster edge deployment |
| Backend | Custom / Hetzner | Vercel Functions + Supabase | Serverless, no infra mgmt |
| Database | Unknown (Supabase ref) | Supabase (Postgres) | Integrated auth+storage+DB |
| Financial Data | Visible Alpha ($50K/yr) | FMP ($30/mo) | 1,666x cheaper |
| Consensus | Visible Alpha | FMP + MarketScreener | Comparable quality, fraction of cost |
| Models | OpenAI + Anthropic (1 at a time) | OpenAI + Anthropic + DeepSeek (parallel) | Verification via divergence |
| Orchestration | Monolith agent | Specialised stateless agents | Testable, isolated failures |
| LLM Ops | PostHog + Sentry | Helicone + PostHog + Sentry | Prompt versioning + cost tracking |
| Reports | S3 (no auth) | Supabase Storage (signed URLs) | Authenticated by default |
| Monitoring | Triggers (manual) | Radar (automated cron + materiality) | Proactive, not reactive |
4-lens verified. 2 critical flags.
M-Score breach. DSRI flagged.
Cross-statement clean.
PETS, TSCO, MKS, SBRY comps.
Independent regulatory analysis.
Vet sector peer. DCF + multiples.
| Company | Ticker | Sector | Price | EV/EBITDA | Radar | Flags |
|---|---|---|---|---|---|---|
| Pets at Home | PETS.L | Specialty Retail | 192p | 6.05x | 📡 Active | 2 Crit |
| IWG plc | IWG.L | Real Estate | 184p | 8.2x | 📡 Active | 1 Crit |
| Tesco | TSCO.L | Food Retail | 372p | 7.1x | 📡 Active | 1 Warn |
| CVS Group | CVS.L | Veterinary | 1,045p | 12.4x | 📡 Active | Clean |
| B&M European | BME.L | Discount Retail | 385p | 7.1x | 📡 Active | 1 Warn |
⚠ Multi-lens triggers cross-check across 2 models before alerting. Reduces false positives by ~60%.