8 The Platform as a Research Substrate
Author: Todd B. Adams Reinforces: Proposal §1 — Executive Summary & §3 · Reading order: doc 01 of the research-platform package
The claim of this doc. Most factor-network research is validated on static, survivor-biased, cost-free academic panels — precisely the conditions the proposal (§1.1) criticizes traditional risk models for assuming. SBFoundation removes those conveniences. It is a live empirical instrument built to be hard on itself, which is exactly what makes it a credible testbed for a claim as strong as “we can detect market phase transitions.”
8.1 Why a production trading platform, not a notebook
The proposal’s model must survive four failure modes that kill most quant-network research. The platform addresses each as a standing architectural property, not a one-off cleanup.
| Failure mode | How amateur research fails | How SBFoundation is built | Status |
|---|---|---|---|
| Look-ahead leakage | Features accidentally use future data; the “signal” is time travel | Gold is point-in-time by construction — a historical date sees only what was knowable then | BUILT |
| Survivorship illusion | Dead companies dropped; predictability manufactured, not returns | 4,858 delisted names spliced back with realistic terminal losses; the evidence gauntlet subtracts the measured haircut | BUILT |
| Cost fantasy | Gross edges that die after the spread | Per-name spread + capacity model; every result reported net-of-cost | BUILT |
| Overfitting | Best-of-many-tries kept and reported as if tested once | Deflated-Sharpe, PBO, purged-CV, honest family-wise trial counts | BUILT |
A network model that clears these conditions is making a far stronger claim than one fit on a clean CRSP panel. That is the substrate’s entire value to the research.
8.2 What the substrate offers the network model
- Point-in-time tensors for free. The proposal’s Supra-Adjacency tensors (factor loadings, dynamic edge weights) would be built from Gold — inheriting its no-look-ahead guarantee without new engineering. See doc 02.
- A real cross-section, not N=100 synthetic nodes. The proposal’s simulator (§6) uses 100 synthetic stocks to proof the physics; the platform offers the empirical graduation target — a full, liquid, survivorship-corrected equity universe with ~2 decades of history.
- An executability verdict. A phase-transition signal is only useful if it can inform a tradeable book. The platform terminates in cost-aware, risk-capped, broker-submitted orders (doc 02), so a network signal is judged on executed P&L, not in-sample fit.
- A validation gauntlet already calibrated to the literature. The anti-overfitting machinery (doc 08) applies to any new producer — including a network one — so GNN claims face the same bar as every factor.
- Reproducible experiments. Every run is immutable and audited (doc 09).
8.3 The honest boundary
SBFoundation does not implement multiplex networks, GNNs, supply-chain ingestion, or 13F ownership today. Those are the research (PROPOSED). What it provides is the substrate and the discipline — the reason the research can be done credibly here rather than started from zero. Where each proposed piece attaches is doc 06; what is missing and in what order it lands is doc 07.
Cross-links: platform whitepaper · business capability map · proposal §1