This platform sits in a gap between two categories of tools most investors use: brokerage-provided research (Schwab, Fidelity, Robinhood) and general-purpose AI assistants (ChatGPT, Claude). Neither was designed to do what this does.
Schwab Analyst Ratings, Fidelity Research, Robinhood Snacks — these give you pre-packaged research written for a mass audience. They’re useful for basic due diligence but share the same structural limitations:
| Capability | Retail Platforms | This Platform |
|---|---|---|
| Earnings Analysis | Beat/miss headline, maybe a summary paragraph | Structured extraction of every metric, margins, guidance, management tone, themes — compared quarter-over-quarter automatically |
| Investment Thesis | Third-party analyst reports (if available) | Auto-generated theses with conviction scoring, delta tracking, consistency checks — plus ingestion of your own thesis notes and external documents, with bias flagging |
| Scoring | Star ratings or buy/hold/sell from a single analyst | Multi-dimensional composite (Fundamental, Thematic, Valuation, Catalyst) with transparent formulas and configurable weights |
| AI Exposure | Not tracked | Five-dimension AI Resilience framework scored per company, with infrastructure reinterpretation and calibration anchors |
| Signals | Basic price alerts | Estimate revisions, PE momentum (velocity + acceleration), theme lifecycle tracking, peer rank drift, thesis consistency flags |
| Automation | None — you check manually | Full autopilot: pre-market recaps, post-market recaps, earnings processing, signal alerts, weekly outlook — all scheduled |
| Peer Context | Side-by-side comparison tables | Percentile rankings across 5 dimensions within your actual peer groups, with composite scoring |
You can ask Claude or ChatGPT to analyze a stock. You’ll get a thoughtful response. But that response has no memory, no data pipeline, and no way to track whether its analysis was right. Here’s the structural difference:
| Capability | General-Purpose AI | This Platform |
|---|---|---|
| Data Freshness | Training cutoff + whatever you paste in | Live pipeline ingesting FMP, SEC XBRL, SEC 8-K, FRED, analyst ratings, Reddit sentiment — refreshed on schedule |
| Persistence | Ephemeral — each conversation starts blank | Persistent signal store with TTL-based expiration, historical snapshots (PE, estimates, conviction), version-tracked theses |
| Scoring Discipline | Subjective, varies by prompt | Deterministic formulas with calibration anchors, soft caps, cross-signal penalties, and sector-aware thresholds |
| LLM Calibration | Uncalibrated — scores drift across sessions | Reference anchors (PLTR, ORCL, CEG, etc.) provided in every scoring prompt to ground output. 92-point soft cap requires cited vulnerabilities. |
| Consistency Checks | None — it will confidently contradict itself | Thesis consistency engine that flags when conviction contradicts estimates, ratings, price action, or peer rank |
| Longitudinal Tracking | Cannot track changes over time | Theme momentum lifecycle (new → accelerating → stable → decaying), estimate revision velocity, conviction history sparklines |
| Portfolio Awareness | Doesn’t know your holdings | Portfolio and watchlist distinction flows through every view: scoring, peer rankings, previews, rebalancing, alerts |
| Automation | You drive every interaction | Background scheduler runs 8+ daily tasks: signal refreshes, earnings processing, recap generation, alert checks |
This platform uses LLMs as a component inside a structured pipeline — not as the interface itself. Claude scores AI resilience, extracts transcript data, generates theses, and writes recaps. But every LLM output flows through deterministic scoring formulas, consistency checks, and calibration constraints before it reaches you. The result is AI-assisted analysis with guardrails, not AI-generated opinions.
The system doesn’t just generate its own theses — it ingests yours. You can attach thesis notes, external research documents, and personal conviction rationale to any company. The pipeline then treats your input as a first-class signal: it incorporates your reasoning into thesis generation, but simultaneously flags potential biases by cross-referencing your narrative against quantitative signals (estimate revisions, peer rankings, price action, financial health trends). If you’re bullish on a name where the data is deteriorating, the system will surface that tension explicitly rather than silently agreeing with you.
When user-supplied thesis notes are present, the system runs a structured check across several dimensions:
| Check | What It Flags |
|---|---|
| Conviction vs. Estimates | High user conviction but consensus estimates are falling — are you seeing something analysts aren’t, or anchoring to a stale view? |
| Conviction vs. Price Action | Bullish thesis but price is below your own bear case target — the market is pricing in something you may be dismissing |
| Conviction vs. Peer Rank | High conviction on a name ranking in the bottom quartile of its peer group across multiple dimensions |
| Narrative vs. Financial Trends | Growth thesis but margins contracting, FCF declining, or leverage increasing over the trailing 4 quarters |
| Confirmation Bias | User notes emphasize the same bullish themes already captured by the system — flags absence of bear-case consideration |
The goal isn’t to override your judgment. It’s to ensure that when you hold a strong view, you’re doing so with full awareness of what the data says — not in spite of it accidentally.
Once deployed, the platform operates on a daily schedule without manual intervention:
| Time (ET) | Task | What It Does |
|---|---|---|
| 7:00 AM | Morning Alerts | Scans all signals for threshold breaches, generates alert digest |
| 8:40 AM | Pre-Market Recap | LLM-generated briefing: overnight movers, key levels, day’s earnings calendar |
| 9:00 AM | Early Scan (AM) | Quick-process any earnings reported pre-market |
| 4:16 PM | Early Scan (PM) | Quick-process post-close earnings (16-min lag for megacap movers) |
| 4:30 PM | Post-Market Recap | End-of-day briefing with session performance, after-hours earnings |
| 7:00 PM | Evening Alerts | Second alert pass after full day’s data settles |
| 10:00 PM | Earnings Autopilot | Full pipeline: prep upcoming (7-day lookahead), process reported, refresh scores |
| Sun 6 PM | Weekly Outlook | Week-ahead briefing with earnings calendar, macro events, thesis updates due |
| Source | What It Provides |
|---|---|
| FMP | Transcripts, analyst ratings, estimates, price history, treasury rates, VIX, equity risk premium |
| SEC EDGAR (XBRL) | Quarterly financial statements (revenue, margins, balance sheet) |
| SEC EDGAR (8-K) | Same-day earnings press releases, parsed by Claude |
| FRED | Macro economic indicators (interest rates, CPI, unemployment, GDP) |
| Retail sentiment on covered tickers, Claude-scored for signal | |
| Anthropic Claude | Transcript analysis, thesis generation, scoring, recap writing, 8-K parsing |