Personal Analyst
Monitor →Research tooling. Every output is a distribution of historical outcomes with an explicit sample size — not a forecast, not advice.
Early build. The Router (below) turns a question into a spec, and the Librarian/Synthesiser/Critic can rank analogs and draft a brief once a study has run — but the registry is still 23 curated events, and every number always comes from the deterministic engine, never from the model. See “How this works” for what's real today.
The Router only ever picks from the 23 events in the registry below — it cannot invent one. Review the spec it proposes before running anything.
Mega-IPO clustering signal plan §10.5 — curated cohort, not a full historical census
Ranks the 11 curated mega-IPOs by proceeds ÷ real market cap (FRED Wilshire 5000) and flags the top percentile as "hot." Checked against 5 known S&P 500 peak dates and self-reports the resulting hit rate — including false positives — rather than asserting the signal works. See “How this works” for why this is a curated-cohort rank, not the plan's original "since 1980" percentile.
Study spec
Cohort
Immediate impact spaghetti overlay, indexed to t0
Duration days to trough vs. days to recovery
Cross-asset heatmap median return / bp-delta by window, n on hover
Index mechanics sources & uses, forced passive demand — code only
Analog ranking by structural similarity to today
Why today differs per cohort event, code-computed
Policy reaction what officials can do, and what it achieved before
Vulnerability where the defence contradicts itself
Brief Synthesiser, then a separate adversarial Critic pass
Watch a condition from this study
Turns into a tripwire proposal for review — not a live tripwire yet.
No study run yet. Pick a preset above (or edit the spec JSON) and click Run study.
No historical episode matches these conditions.
Try a different cohort tag/class or an explicit event id from the registry — or build a synthetic analog below.
No strong historical match. A synthetic analog decomposes this into component shocks that DO have history, composes them with stated assumptions, and reports a deliberately wider error band than any single observed study.
Proposed decomposition review before anything runs — you can adjust weights, code re-validates either way
What the library can't capture
What the library can't capture
Composed range, every series hatched fill — never mistake this for an observed study
Component contribution which component drives each series' composed figure, selected window
Live reweight recomposed client-side, no server call — additive mode only
Sign in required. Personal Analyst is a members tool — sign in from the top bar to run a study.