Family 01
Volatility
How much the market expects things to move. Realised and implied, across assets, across horizons.
28 signals·VIX · MOVE · SkewHot +16
Pick any index and watch the deterministic pipeline build the signal from raw inputs. No black boxes, no hidden weights — every step is published.
Schema Score blends three lenses so a single number can answer three questions at once: is the backdrop supportive, is the price attractive relative to that backdrop, and are current narratives reinforcing or fighting the setup?
Structural support
How supportive the current macro environment is for the asset before price or narrative enters the picture. Comes from MarketSchema's index system.
Price vs. thesis
Whether price is stretched or attractive relative to structural support. A strong thesis that's pulled back beats an equally strong thesis that's already extended.
Story confirmation
Narratives overlay, not backbone — deliberately sized smaller so headlines don't overpower structure and price. They can strengthen or weaken a setup.
Plain English: macro is the backbone, entry value can improve or penalize timing, and narrative tells you whether the tape is reinforcing or undermining the setup. If a sleeve has no evidence, it does not get invented.
Every reading is the top of a four-step pyramid. Each step is deterministic and published — you can rebuild the current composite from public data and these formulas.
Asset Schema Scores use a 0–100 scale. Indexes are normalized in z-score, percentile, rebase, or raw space according to their registry definition; the display bands below are deterministic labels, not hidden optimized thresholds. See the normalization contract.
Stressed readings are rare by design — roughly one trading day in nine. When the composite lives above 65 for consecutive weeks, we call it a regime shift and flag it on every dependent page.
Every signal belongs to a family. Families are the atoms of attribution: when we say "COIN is up because of Crypto vol disconnect," we can point to the family responsible and the signals inside it that moved.
How much the market expects things to move. Realised and implied, across assets, across horizons.
The price of risk in corporate debt — IG, HY, EM, distressed. Where refi stress shows up first.
How easily dollars move through plumbing. SOFR, RRP, basis swaps — the early-warning system.
Who's buying, who's selling, how crowded it is. ETF flows, CoT, dealer gamma.
Growth, inflation, policy. Curves, breakevens, nowcasts — the slow engine under everything.
How assets co-move under a story. The layer that links AI capex, Japan normalization, fiscal stress.
Does MarketSchema’s scoring actually predict forward returns?
Track record under review — 58 days collected.
Evaluation window started and public evaluation metrics are being QA’d before release.. Every score, setup packet, and regime call is being recorded and will be scored against realized outcomes.
Do higher-scored assets outperform lower-scored ones?
In reviewHow often do setup packets call the right direction?
In reviewDoes the regime engine correctly identify market states?
In reviewMarketSchema records every score computation and setup packet. After the review window, we publish only track-record numbers backed by enough forward return data to compute meaningful hit rates and decile performance. This is a deliberate choice: we would rather show real numbers than premature statistics.
Average 30-day forward returns grouped by Schema Score decile, published only after QA.
Checking evaluation maturity.
The chart stays empty until verified decile returns are published. We do not show illustrative bars or inferred performance.
A methodology is only as honest as the things it refuses. Here are the four rules we've adopted and intend to keep.
Every index on MarketSchema has a published formula, a public constituent list, and a documented rebalance cadence. If we can't show you the math, we don't ship it.
Show you the exact weights and the code that produces them.
Hide a model behind a brand or a “proprietary score.”
Scores don't contain "what we think should happen." Drivers are observations. When we publish commentary, it's labelled as commentary, never baked into the score.
Label every score with its input recipe.
Lift or tilt a score “because it feels right.”
If a source goes dark, we hold the last valid value and flag it. We never synthesize missing data to make a chart look clean. Carry-forward data is labeled, not hidden.
Show a carry-forward badge and the age of the last tick.
Spline or forward-fill through outages.
When weights or constituents change, we version the index and publish a diff. You'll see the old weight and the new weight. Historical charts keep the old regime on the timeline.
Log every weight change with a reason.
Re-score history without a version bump.
Because a percentile is intuitive where the registry declares percentile normalization. Other indexes publish z-score, rebased, or raw-normalized values. The normalization contract spells out the window, fallback cadence, and threshold family instead of hiding every series behind one generic scale.
Freshness follows the source and the surface. Equity and crypto-linked inputs can refresh faster than macro series, but methodology panels report the accepted observation or last-check timestamp. Macro publishers such as FRED and EIA are usually daily or weekly. If you need a faster series, it's a source-license and ingestion-policy question, not a methodology question.
We hold the last valid value and flag the contribution. The composite continues to publish with a visible carry-forward badge on the affected family. We never interpolate and we never mark-to-model a missing tick. If the outage is longer than 24 hours on a series that matters, we pause the index and say so on the status page.
No. Weights are derived from rolling diagnostic power — how much each signal explained index variance over the trailing window. The formula is public and reproducible. A human reviews every monthly rebalance for sanity, but cannot tilt a weight beyond the formula's output without bumping a version.
Yes. Every timestamp is reproducible — the API exposes the vintage of every input used at that timestamp, the weight in force, and the version of the formula. You can re-run the composite from a published vintage and get the same number to three decimal places.
No. We publish observations, not trade signals. We don't run a book, we don't sell a newsletter of "what to buy tomorrow." Drivers and narratives are map-making tools for experienced investors — the interpretation stays with you.