# Preset protocol: three retail crypto decisions

Version `retail-decisions-2.0`, effective 21 September 2026. Version 2.0 keeps every primary rule from 1.0 and adds horizon, regime, path-risk, uncertainty and cross-venue checks. No primary threshold was changed after reading the 1.0 results.

## Shared evidence

The studies use completed Binance BTCUSDT and USDT spot daily candles. The BTC-only studies use the frozen 17 August 2017–20 September 2026 BTC series already published with the sentiment/ATH edition. The altcoin study uses the frozen Binance directory and 591 discovered target-market series from the four-decisions edition. It excludes products marked as leveraged, stable/fiat/wrapped/equity-like, benchmark or ambiguous by that edition's product classifier.

All timestamps are UTC. A condition observed at a completed daily close can execute no earlier than the following daily open. The base proportional cost allowance is 0.15% of notional on each purchase or sale. Cash earns zero in the primary results. Taxes, spread outside the allowance, market impact, staking, lending and personal restrictions are not modelled.

## Study A: DCA plus buying the dip

- Start on eligible first-of-month UTC opens with a complete prior 365-day price history and a complete following 365-, 730- or 1,460-day study path.
- Contribute USD 1,000 on 12 dates, 30 UTC dates apart. Every policy receives the same USD 12,000 nominal contributions on the same dates.
- Compare plain DCA with policies reserving 10%, 25% or 50% of every contribution.
- Split each reserved amount equally into trailing-365-day closing-drawdown triggers of 10%, 20% and 30%.
- A completed close meeting a trigger buys at the next UTC open. A newly contributed bucket may buy immediately if the prior completed close already meets its threshold.
- A bucket can deploy once. Untriggered capital remains cash at the common terminal open and is included in terminal value.
- Publish terminal return on total contributions, paired difference from plain DCA, cash left, deployment, transactions, costs and cash-flow-adjusted unit-value drawdown.
- Main sample uses all eligible monthly starts. A greedy non-overlapping 365-day sensitivity is disclosed separately.
- Publish 365-, 730- and 1,460-day endpoint summaries. Classify each start by its prior-close distance from the trailing 365-day high: within 10%, 10–30% below, or more than 30% below.
- Sensitivity only: credit waiting cash with 3% or 5% annual compound yield. This does not change the zero-yield primary result.
- Estimate a 95% interval for the paired median reserve-minus-plain-DCA result with 2,000 deterministic circular moving-block resamples of 12 consecutive monthly starts. Seed and code are public.

## Study B: after an altcoin doubles

- For each eligible non-BTC market, use the minimum close of the preceding 90 completed UTC dates as the reference low.
- The main event is the first completed close crossing from below to at least 100% above that fixed reference low. Preset 50% and 200% thresholds are sensitivity checks.
- Require a trailing-30-day average quote volume of at least USD 1 million and apply a 90-day cooldown per symbol and threshold.
- At the next UTC open, compare five decisions applied to one unit of existing altcoin value: continue holding; sell, buy BTC and later sell BTC; sell to zero-yield cash; retain half and move half to BTC; or retain half and move half to cash.
- Use common 30-, 90- and 180-day terminal opens. Holding pays only the terminal sale allowance; switching pays altcoin sale, BTC purchase and BTC terminal sale allowances; cash pays the initial sale allowance.
- Pending, missing-exit and missing-path events remain in the candidate ledger and are not converted to zero returns.
- Publish maximum favourable and adverse held-altcoin excursion over each path.
- For the main +100%/90-day outcome, publish: first qualifying event per symbol; equal-symbol-weight medians; BTC prior-90-day direction; signal-volume ratio above/below 2×; listing age below/above 180 days; and distance within/more than 10% below the prior available high.
- Estimate a 95% interval for the median of symbol-level median BTC-minus-hold differences by 3,000 deterministic symbol-level bootstrap resamples. It addresses repeat observations within symbols, not market-wide dependence.
- The frozen directory is not a complete point-in-time history of every delisted Binance market. Results are descriptive for the eligible discovered universe and may contain survivorship and selection bias.

## Study C: how long to hold Bitcoin

- Use every eligible completed BTCUSDT UTC daily open as a start.
- Buy at that open and hypothetically sell at the open exactly 30, 90, 180, 365, 730 or 1,460 UTC dates later.
- Charge 0.15% at entry and exit. Calculate terminal net return, sampled daily maximum drawdown, maximum favourable/adverse excursion, first break-even touch, permanent break-even day and share of dates underwater.
- Publish loss counts and distributions for all daily rolling starts. These windows overlap heavily; they are not independent trials.
- Publish a greedy non-overlapping sensitivity for every horizon. The four-year claim is limited to the available Binance BTCUSDT history and must not be described as a law, guarantee or proof that every investor can wait through the observed drawdown.
- Publish a 30-day-step curve from 30 through 1,460 days. A zero-loss point at one grid horizon must not be described as a permanent threshold if losses return at a later point.
- Stratify 365-, 730- and 1,460-day starts by prior trailing-high position and prior-day Fear & Greed zone. These are descriptive associations, not causal estimates.
- Estimate moving-block 95% intervals for median return and loss rate with 1,000 deterministic resamples. Block length equals the tested horizon up to a 365-day cap.
- Independently calculate the same endpoints from Coinbase BTC-USD daily candles on matched dates. Venue agreement is a market-data robustness check, not an independent Bitcoin cycle.

## Publication rules

Every page must lead with a direct Quick Answer, show denominators and negative cases, expose observations/candidates/summary files, provide visible FAQ and matching Schema, include contextual internal links, and state that historical frequencies are not future probabilities or personalized advice.