Crypto Strategy Research Method
A rule, a data identity, a causal fill policy and a complete ledger form one experiment.
Define Before Measuring
Freeze venue, instrument, decision clock, candidate-selection rule and exit policy. Preserve the parameter variants explored and the chronological evaluation split. A strategy result without those inputs cannot be independently reconstructed.
Keep Different Results Separate
A research reference price, a paper order and an exchange fill need different labels. Cost assumptions are not invoices. A scenario sensitivity chart is not a performance series. This publication does not manufacture combined scores from unavailable evidence.
Review Decision
The useful outcome is a documented reason to proceed with more research, revise an implementation assumption or reject the idea. No worksheet on this site provides an investment recommendation or authorises an order.
Primary Sources
A Crypto Backtest Needs the Tokens That Disappeared
Define historical eligibility, listings, delistings and price availability before testing a strategy on today's surviving markets.
Crypto Backtest Data Workbench
Freeze venue history, token membership, product terms and causal data joins. A current crypto directory does not define the universe available in the past.
Define an experiment that could fail
Record the input instrument, venue, eligible history, observation clocks, causal rule, exclusions and evaluation question before choosing attractive results. Entry candidates retain the raw causal boundary-recovery objective. Freeze entry identity before a single predeclared exit; exit-adjusted returns, profit factor and win share do not select the entry model. Keep every trial and inspected period, including failures and missing inputs. More explanatory pages are not additional independent validation samples. These teaching calculations do not complete the canonical factory or master for a genuine crypto instrument.
Separate source identity from arithmetic identity
The browser engine has a narrow, reproducible contract: manually declared labels and prices, initial quote notional, fee basis points, signed settlement funding and a valuation multiplier. It uses decimal.js; a separate Python Decimal implementation supplies software reference cases. Matching those calculations verifies equations, not that an actual contract uses those quantities or that an exchange filled them. Minimum lots, contract multipliers, partial closes, borrowing, margin, liquidation and tax are outside this tool. Their absence remains explicit rather than being hidden inside a reassuring precise net number.
Retain the unknown branch
A reconciliation journal records missing observations instead of inventing a clean sequence. Client request, acknowledgement, unique execution, cancellation, remaining order quantity and open position are different objects. Duplicate messages do not create more fills. When the request outcome is unknown, the next task is authoritative state reconciliation with the responsible service owner, not a favourable assumed execution. The paper drill is fully authored with null real fills and no incident-frequency measure. Its completion means the exercise is inspectable, not that any live account was recovered successfully.
Bound the product comparison
Resource profiles explain tasks rather than financial rankings. A chart rule, paper automation environment, news feed, signal message and copying subscription have different producing and receiving contracts. Current terms must support each selected product and jurisdiction. No price, crypto coverage, refund or trial is inferred from a familiar platform name. The desk has no paid-account cohort or representative results population. Material marketing context is disclosed, and the old model headline remains a supplied historical summary until dated source records support a stronger interpretation.