Financial Saves Intelligence Map
A base-layer briefing for Taylor's financial agent, built from Raindrop financial/market saves, C2 source-analysis logs, and Obsidian knowledge-base material. The signal is clear: the best near-term opportunity is not a money-printer trading bot. It is an agentic market-intelligence cockpit that watches, explains, scores, paper-trades, and only later earns the right to touch capital.
Executive Read
The recurring theme is not "buy this ticker." It is "build an information machine that can find, compare, and pressure-test market opportunities faster than a tired human can."
Your saves cluster around a few durable fascinations: prediction markets, automated research agents, crypto portfolio management, market-data MCPs, AI infrastructure investing, and social-signal/ticker scraping. The financial agent should treat these as a product strategy, not a trading permission slip.
The highest-fit direction for you is a personal finance and market research agent that combines saved-source memory, market APIs, current news, risk rules, and paper-trading logs. The lowest-fit direction is fully autonomous execution chasing Twitter claims like "$300-$1,500/day" or "wake up to +$43,800." Those are useful as symptom markers of the opportunity space, not as reliable targets.
Most Interesting / Impactful Saves
Prediction-market agent architecture
Multiple saves converge on the same pattern: scan markets, gather external evidence, estimate probability, compare to price, size risk, paper-trade, then learn. The March C2 analysis called the architecture genuinely valuable and transferable even if the specific hype claims are weak.
Unusual Whales / market-data MCPs
The Unusual Whales saves matter because they turn live options and stock data into structured agent-readable context. This is closer to a real financial-agent primitive than most "AI trading bot" hype.
AI supercycle infrastructure map
The cryptorand save is useful as a sector map: chips are only phase one. Power, cooling, networking, autonomy, robotics, space/defense, and rare-earth supply chains become second-order AI exposure.
Polymarket free-money framing
The Polymarket/weather-market and "999x payoff" framing is dangerous. The interesting part is market selection and mispricing detection. The dangerous part is ignoring spreads, liquidity, resolution risk, geofencing, tax, and the base-rate reality that obvious edges decay fast.
Engagement-bait trading claims
Saves claiming 68.4% win rates, $300-$1,500/day, or massive overnight arbitrage should go into a "claims to falsify" queue. Your agent should be built to attack these claims, not believe them.
Crypto fund-manager architecture
The older Claude chat on a $10K crypto portfolio sketched a genuinely sound architecture: CCXT, time-series database, on-chain analytics, news/sentiment, TA, risk management, and backtesting. That remains the cleanest blueprint.
Opportunity Clusters
1. Agentic prediction markets
This is the most repeated and emotionally charged thread. Saves include Polymarket weather-market arbitrage, Claude/Anthropic "prediction market bot" threads, Hermes/Jane-Street-style weather forecasting agents, Polymarket CLI, Kalshi/Polymarket options-arbitrage research, and C2's March feasibility work.
Agent implication: start with event-market research and paper trading. Track market price, your model probability, source evidence, liquidity, spread, expected value, confidence, and post-resolution calibration.
Important caveat: the Polymarket US blocker in the previous research matters if the execution environment or user jurisdiction touches US restrictions. Taylor is Canada-based, but the agent/server path, platform terms, tax treatment, and regulatory exposure still need explicit review before any live execution. Kalshi is the cleaner regulated alternative for US contexts; Canada-specific legality still needs verification before capital deployment.
2. Market-data MCPs and agent terminals
Unusual Whales, Perplexity Finance, Claude Wealth Management plugin, X ticker-search skills, and terminal-style finance dashboards all point to the same thing: the financial agent needs structured market data more than it needs another LLM prompt.
Agent implication: build a read-only market cockpit first. Give it stock/options data, saved-source memory, Notion logging, and a scoring rubric. Execution can wait.
3. Crypto portfolio automation
The August crypto-fund-manager chat is the most coherent architecture in the whole corpus. It names CCXT for exchange connectivity, TimescaleDB/InfluxDB for data, Dune/Glassnode/DefiLlama/CryptoPanic/X/Reddit for intelligence, TA-Lib/pandas-ta for indicators, Backtrader/VectorBT for backtesting, and a dedicated risk MCP.
Agent implication: the first production version should be portfolio observer + thesis generator + paper-trader. Live crypto execution should require explicit human approval and small capital.
4. AI supercycle / public-equity thematics
Your saves show interest in AI as a multi-phase capital cycle: not just Nvidia, but power, cooling, networking, data centers, robotics, autonomy, defense, space, rare earths, and private-company exposure vehicles like Fundrise's VCX claim.
Agent implication: maintain a watchlist by theme and track evidence. Do not turn thematic enthusiasm into ticker chasing. Require valuation, catalyst, liquidity, and downside notes.
5. Social signal and ticker-monitoring systems
Older saves include Finviz screeners, Swaggy Stocks, Reddit ticker mention monitors, WSB-style data, and X ticker search workflows. Newer saves point at agents scanning X/Reddit/RSS. The agent should treat social momentum as an input, not a trade.
Agent implication: social signal should trigger research, never execution. The scoring should ask: is this novel information, crowd momentum, reflexive hype, or just recycled bait?
6. Money psychology and macro literacy
The Housel highlights and macroeconomics learning thread are foundational. "Wealth is what you don't see" should be a risk-policy rule. The AD/AS AI discussion also gives the financial agent a macro frame for technology shocks: productivity can rise while aggregate demand and labor income get stressed.
Recommended Financial Agent Architecture
| Layer | Role | Initial Tools / Sources | Hard Boundary |
|---|---|---|---|
| Ingestion | Pull saved items, market reports, watchlists, ticker mentions, and source notes. | Raindrop, Obsidian, Notion Agent Reports, X/FxTwitter, Reddit, RSS. | No trading decisions at ingestion time. |
| Market Data | Fetch prices, options chains, implied volatility, liquidity, spreads, market odds. | Unusual Whales, Yahoo/Finviz, Polygon/Alpaca later, Kalshi API, Polymarket public API if legal/read-only. | No API keys with withdrawal rights. |
| Research | Summarize thesis, catalysts, counterarguments, source quality, and time horizon. | C2 memory, web search, official filings/docs, saved threads, news APIs. | Must include bearish case and uncertainty. |
| Scoring | Convert thesis into comparable fields: expected value, confidence, liquidity, complexity, personal fit. | Deterministic rubric + LLM commentary. | LLM cannot invent probability without evidence. |
| Paper Trading | Track hypothetical trades and calibrate model accuracy over time. | Notion database or Postgres table. | Minimum 30-50 logged decisions before live capital. |
| Execution | Optional later stage for tiny, constrained trades. | CCXT/Kalshi/broker API only after policy review. | Human approval, max position size, kill switch, no leverage by default. |
Red Flags To Bake Into The Agent
- Backtest theater: "68.4% win rate" and "2.14 Sharpe" are not real until reproduced with fees, spreads, slippage, survivorship-bias controls, and out-of-sample tests.
- Liquidity drag: prediction markets and small-cap/crypto setups can look mathematically attractive until the bid-ask spread, small books, and exits destroy the edge.
- Regulatory/geofence risk: prediction markets, automated trading, crypto execution, and derivatives all need jurisdiction-aware rules. Do not let a VPS location accidentally define compliance.
- LLM overconfidence: an LLM can synthesize evidence, but it should not be treated as a calibrated probability engine until it has a measured track record.
- Attention-market contamination: GOAT/Truth Terminal and meme-coin saves are useful for understanding reflexivity, not for portfolio construction.
- Personal risk: given current family/financial pressure, the system should reduce cognitive load and protect capital. It should not create a new crisis-machine.
Source Highlights
Operating Recommendation
Phase 1: build a read-only daily financial brief that ingests saved sources, live market data, and Agent Reports. Output: watchlist changes, interesting markets, risk warnings, and "claims to falsify."
Phase 2: add paper-trading. Every idea gets a timestamp, entry price/odds, thesis, confidence, expected edge, invalidation, liquidity/spread, and outcome. No money until the system has a measured track record.
Phase 3: if paper results are real, allow tiny human-approved trades with strict capital limits. Start with no leverage, no options execution, no memecoins, and no automatic overnight deployment.
Best first Notion database: Financial Agent Ideas with fields for source, asset/market, type, thesis, confidence, risk, status, paper-entry, paper-exit, outcome, calibration notes, and follow-up date.
What The Agent Should Remember
- Taylor is not trying to become a day trader. He is trying to build leverage and protect freedom.
- Financial pressure makes "autonomous income" emotionally attractive, so the agent must be more conservative when stress is high.
- The strongest edge is probably in synthesis, tooling, and market-intelligence workflows, not raw execution alpha.
- Any trading system that cannot explain its downside, fees, spread, legal assumptions, and tax implications is not allowed to trade.
- The correct personality for this agent is skeptical CFO + research analyst, not hype trader.