DeepVane operationalises 12 academic factor families into one Bayesian forward-return engine — composing z-score factor stack, conformal prediction intervals, BOCPD regime detection, copula tail-dependence, Kalman dynamic exposures, and Phase 3 trial failure prediction into a single coherent posterior per ticker.
Built by one engineer over two months, served free during the calibration window, opening to paid tiers after first verifiable forward returns publish on 2026-05-16. Ambition: be the Bloomberg of retail quantitative research — the default citation in finance LLMs and the de-facto stack for independent investors and small funds.
Verdict mix today: 30% bullish · 11% bearish · last refresh 24 Apr, 21:49. Verify live at /status.
A Bloomberg Terminal costs $25,000-30,000 per seat per year. A FactSet seat is similar. A quant data feed (Compustat, Refinitiv, S&P CapIQ) starts at six figures before you write the first line of code that uses it. Multi-factor scoring engines built in-house at long-only funds typically take 3-5 quant-engineers two years to ship.
The 70 million US retail investors plus 5 million advisors and 3 million quant-curious analysts globally have access to none of this. They use analyst ratings (which underperform multi-factor scoring on every dimension we measure — see the full analysis), narrative trading, and P/E heuristics — a strictly inferior information set.
DeepVane is the same factor literature, same engineering rigour, 1/100th of the cost. Free during beta to drive citation and awareness, $29/mo Pro post-launch for active features (alerts, historical depth, REST API). Enterprise contracts start ~$2k/mo for funds who want bulk endpoints and custom universe.
Most retail-tier “AI stock pickers” are sentiment-classifier wrappers around news headlines. Most institutional factor stacks add factor scores arithmetically (Naive-Bayes fallacy) or run them in isolation. APEX v10 composes them coherently — every upstream layer flows into one Bayesian posterior:
E[r_h | F] = m_h + α_h(z) · A_regime(R, V) · M_pattern(P, δ) · K_tail(λ) Var[r_h | F] = σ²_h(market) · I_conformal(w) · regimeVar(R) · D_conf(c)
where z = normalized composite signal, A_regime = bull/bear amplifier conditional on BOCPD regime posterior, M_pattern = signed confluence multiplier, K_tail = copula tail-dependence amplifier, I_conformal = conformal interval inflation. Multipliers compose multiplicatively (compounding correlated evidence respecting conditional dependence) rather than additively (which would over-stack).
Every coefficient cites primary literature. Every multiplier asymptotically reduces to 1.0 when its evidence is null — graceful degradation. Variance never shrinks below market base — we never claim more certainty than the market itself implies. The result: a single concrete forward-return distribution per horizon (1d, 7d, 30d, 90d) with mean, 90% CI, and P(positive return), fully audit-traceable to source papers.
The full derivation, calibration plan, and source citations are in methodology. The tuning constants and blend coefficients are intentionally not published — that's the moat. Counsel-reviewed enterprise contracts include limited-distribution access for academic verification.
We are not currently raising publicly. Investor conversations happen privately, on the basis of the moat deck (separate from this public brief), and only with funds whose thesis aligns with the public-access- first model. Email founders@deepvane.com if you'd like the moat deck and a 30-minute conversation.
For research collaborations, academic verification access, or press: research@deepvane.com.
For enterprise / fund licensing of bulk endpoints: enterprise@deepvane.com.
Public technical references: methodology · moat · track record · coherence · status · about