This article explains a data-construction discipline central to our own research standards. Nothing here is investment advice. Decisions are yours.
A point-in-time universe answers, for every date in a backtest, one question with archival honesty: which assets qualified on that day, by the rules, using only what was knowable that day? It is the difference between simulating the market that existed and simulating the one that happened to survive into your download.
The fiction the default produces
The convenient backtest starts from today's top-100 list and fetches its history: every asset in the test is certified, by construction, to have survived until now. The strategy never meets the top-ten asset that went to zero in a week, the case the LUNA collapse made famous, never holds the exchange tokens that died with their venues, never touches the crypto graveyard at all. The inflation this buys is not subtle, and our own equity audit put a number range on its equity-market cousin; crypto's death rate is higher, so its flattery is larger. A momentum strategy is the worst case: it ranks on strength, and yesterday's doomed assets were often strong on the way up, so the current-list version systematically deletes exactly the trades that would have hurt most.
What PIT construction actually requires
Three archival ingredients. Historical rankings: market capitalization and volume as of each date, for every asset alive then, including the dead ones under their original identities. Rule evaluation in the past tense: every eligibility test the published universe rules impose, liquidity floors, listing age, exclusions, computed from that date's data only, with no peeking at what later happened. And event resolution: delistings, renames, splits of chains and tickers, each mapped so the asset that died is distinguishable from the asset that took its name. This is precisely the battery our vendor gate runs before any historical data is trusted, a commitment the build-in-public roadmap made in writing: a vendor that cannot enumerate the top-150 as of an arbitrary 2021 date, corpses included, cannot support an honest crypto backtest, whatever its marketing says.
Our specification, stated
The derivation program's research universe is point-in-time by construction: the top-150 by the published rules, re-evaluated at each rebalance date from that date's data. The number is deliberately wider than the monitor's current top-100, because a research universe must contain tomorrow's entrants and yesterday's exits to model turnover honestly: assets enter the test when they qualified, not when they became famous, and leave it when they failed the rules, not when they became embarrassing. The monitor's current-list top-100 is a different instrument doing a different job, live measurement under the monitor methodology, where survivorship is not a concept: the present has no survivors, only members.
What survives even PIT discipline
Honesty requires listing the residual biases, because PIT is necessary, not sufficient. Venue survivorship: exchange consolidation kills venues, and bars sourced from the survivors still miss the pairs that died with them. Fill fiction: an asset can be present in the universe while being untradeable in practice, halted chains, delisted pairs, and books too thin for the modeled size, so presence must be paired with honest cost and fill assumptions, and stops in backtests live or die on exactly those conventions. And data vintage: prices as recorded today are not always prices as displayed then; revisions, backfills, and symbol surgery leave scars only vendor archaeology reveals.
The honest posture is layered: PIT membership first, because it is the largest single correction, then each residual named and either modeled or disclosed, which is why the transplant experiment carries its own caveat list, and why every number this site publishes travels with its data-quality footnote until the vendor gate closes. Backtests do not fail from missing sophistication; they fail from simulating a kinder market than anyone was ever offered. Decisions are yours.