Belief in. Structure out.

Two modes, one spine

Wellow ships as two products behind a top-nav toggle. Different content, the same engineering grammar: a typed schema, structured generation, and a read-only share page with a 90-day TTL.

• Research turns a future-facing conviction ("travel will become more frequent", "data centers are the bottleneck for AI") into an investment thesis with ranked stock and ETF exposure, a hypothetical-return backtest, a four-step causal chain, the risks that would break the thesis, and live macro data points to track.
• Philosophy turns a personal belief about how to live ("career is more about what you enjoy than what you're good at", "lifestyle is a self-fulfilling prophecy") into a structured worldview: philosopher alignment across the canon, behavioral exposure across career / financial / lifestyle / relationships, the contradictions you may be living through, and the self-trackable signals that show whether you actually live the belief.

Why structured

Most thinking tools end at the paragraph. You ask a chat model, you get prose back, you have to do all the decomposition yourself. Wellow's defining property is structure: every artifact comes back as the same shape, so you can compare across beliefs and we can guarantee certain things about every output.

In Research that means counter-arguments per pick, polarity-aware exposure, validated tickers, and no "buy / sell / guaranteed" language. In Philosophy that means alignment direction is named, contradictions are surfaced in second-person, signals are behavioral and falsifiable, and the canon is curated rather than hallucinated.

How a thesis is built (Research)

1. Polarity classification. Your conviction is read as bullish, bearish, or structural. Bearish theses route through replacements, hedges, and inverse-tracking ETFs. Nothing gets labeled a "direct beneficiary" if it would lose under the thesis.
2. Structured generation. A frontier model produces the thesis against a strict schema: refined thesis, opinionated take, ranked exposure cards, causal chain, risks, data to watch. Each field has a length budget and a specific role.
3. Ticker validation. Every symbol the model proposes is matched against a market-data company database via fuzzy name overlap. Hallucinated tickers are dropped. If too few validate, the thesis isn't saved.
4. Data enrichment. Sparkline price history comes from Twelve Data. Macro data points come from FRED with their latest values, units, and source links. The hypothetical-return basket is computed against SPY at 1y / 3y / 5y lookbacks. When the belief maps to a live prediction market, Kalshi market-implied odds are attached and charted.
5. Rendering + sharing. The result lives at an unlisted, read-only URL with a 90-day TTL. Shareable, not searchable, not used for training.

How a worldview is built (Philosophy)

1. Axis classification. Your belief is placed on six values axes (individualism / collectivism, structure / freedom, stability / change, rational / experiential, meaning / nihilism, risk / safety). A fast model handles this in a few hundred milliseconds.
2. Canon match. The belief's vector is compared to a curated database of 35+ philosophers, each with hand-assigned axis vectors. Top-aligned and most-opposed names get fed to the generator so opposed alignments are real challenges, not strawmen.
3. Structured generation. A frontier model produces the worldview against a strict schema: refined claim, philosopher alignment, behavioral exposure, causal chain, risks, contradictions, signals to watch. Voice is opinionated and second-person; no academic hedging.
4. Rendering + sharing. Same read-only, unlisted-URL pattern as Research. 90-day TTL, not searchable, not used for training.

Data sources

Gemini 2.5 Pro

Frontier model used for structured generation in both modes. Output validates against a Zod schema before we persist it.

Financial Modeling Prep (FMP) — Research only

Company name validation and the logo CDN. The ticker-hallucination filter.

Twelve Data — Research only

Daily close prices for sparklines and the 1y / 3y / 5y basket backtest. Delayed end-of-day data, refreshed daily.

FRED (St. Louis Fed) — Research only

Macro indicator values for the "data to watch" cards. Updates on each indicator's publishing cadence.

Kalshi — Research only

Live market-implied probability for the event behind a thesis, when the belief maps to a Kalshi prediction market. The YES price is the market's odds; polled live on the page. Shown only when a confident match exists.

Curated philosopher canon — Philosophy only

35+ thinkers spanning ancient, modern, and contemporary traditions. Editorial axis vectors, not embedding-based retrieval. Source of every alignment in the output.

What this is not

• Not advice. Wellow does not give financial, investment, tax, legal, medical, or life-coaching advice. Theses and worldviews are research artifacts, starting points for your own thinking.
• Not a chat. Every output is the same shape. Consistency is the whole point.
• Not real-time (Research). Sparkline prices are end-of-day data. Macro values update on their cadence. Wellow is research-grade, not a trading terminal.
• Not a personality test (Philosophy).Philosophy mode doesn't classify you. It takes a single belief and shows what it commits you to behaviorally. The decision is still yours.

Try it

Pick a mode and start with one sentence.

• Research — turn a conviction about the future into a structured investment thesis.
• Philosophy — turn a belief about how to live into a structured worldview.