Historical data
Everything the layer has ever observed, kept. Observations are written once and never edited, which is what makes a result computed today reproducible next year.
Append-only by design
Observations are immutable. A row is written for an asset and an hour, and it is never updated, backfilled or corrected in place. If a provider revises a figure after the fact, the revision arrives as the next hour's observation — the original stays as it was.
This is the property that makes the store usable for research. A backtest run against last November's data gets the values that were actually observable last November, not a retrospectively cleaned version of them. Silent restatement is how backtests end up describing a past that never existed.
What is retained
| Series | Grain | From | Breadth | Contents |
|---|---|---|---|---|
| Market observations | hourly | Nov 2025 | 1,000 assets/snapshot | price, cap, volume, volatility, dominance, rank |
| Social observations | hourly | Nov 2025 | 1,000 assets/snapshot | volume, interactions, dominance |
| Sentiment | hourly | Nov 2025 | 1,000 assets/snapshot | one value per asset per hour |
| Score | hourly | Nov 2025 | 1,000 assets/snapshot | stored alongside its inputs |
| Channel reach | daily | 2024 | 69 assets | followers, posts |
| Repository activity | daily | 2024 | 69 assets | commits, PRs, issues, stars |
| Contributors | daily | 2024 | 69 assets | contributors, contributions |
| Topic mentions | daily | 2024 | 69 assets | mention counts |
| Narrative snapshots | hourly | 2025 | 10 topics/snapshot | topic, count, category |
News flow is the one category with no long history: it is served as a rolling window of the most recent 200 items and is not retained as an archive.
What it is good for
Research
The hourly grid means market, social and sentiment values for the same asset share a timestamp exactly. Lead-lag questions — does attention move before price, and by how long — can be asked without interpolating between series sampled on different clocks.
Backtesting
Scores are stored alongside the inputs that produced them, so a strategy can be replayed against the score as it stood at the time rather than against a score recomputed with today's formula. Read the caveats below before relying on this.
Academic and quantitative analysis
The combination of a broad hourly cross-section and slower daily fundamentals for a curated subset supports panel work: cross-sectional at the hour, longitudinal at the day.
Caveats that matter for backtests
- Survivorship in the covered set. Each hourly snapshot covers the top 1,000 assets by the provider's combined rank. Assets drop out and return. A naive study of "assets with continuous history" selects for assets that stayed ranked, which is a performance filter in disguise.
- Gaps are real and must be handled. Missing hours exist, through boundary rotation and through upstream outages. Forward-filling across them fabricates observations. Leave them missing.
- The series starts in November 2025. That is roughly one market regime. Any conclusion drawn from it is a conclusion about that regime.
- Score definitions can change. Stored scores reflect the formula in force at write time. Methodology changes are documented on moonboard.io; a series spanning a change is not internally consistent.
- Hourly is not intraday. Nothing here supports execution modelling, slippage estimation or anything that depends on sub-hour price paths.
How to get it
Historical series are available to the applications built on the layer. There is no public download, no bulk export endpoint and no self-serve API key today. Packaged historical extracts are the subject of the datasets programme, and the interface is covered on the API page. Researchers with a specific question are welcome to get in touch — that is currently the honest answer rather than a form that promises more.
Related
- Building historical crypto timeseries — the design decisions behind this store.
- Datasets — packaged extracts and their licensing constraints.
- Research use case — who uses this and how.