# wavelab ledger A public, append-only, hash-chained record of what 88 pre-registered trading strategies said about BTCUSDT on the 4-hour chart, bar by bar, each paired with a random control, and how every one of them turned out. ## Closed This record was closed openly on 2026-10-17. Its intake stopped that day: the bar of 2026-10-17 16:00 UTC was refused (seq 21703) and the closure was declared in the record itself (incident at seq 21704). The plans already open kept resolving and were published until the last outcome (seq 25144). The CLOSED incident (seq 26760) was then written with its counts, anchored in Bitcoin and published, and the final segment ends at seq 26944. Nothing will be appended. Its successor is **wavelab-ledger-v2**, whose genesis row is `e6f0146f31307793428e2e0f8e20c9d06fdd08917972f68cef1a4aa4897e643d`. The data here stays public and verifiable, as it was: clone it and run the verifier (version 1.0.0) as shown in [Verify it](#verify-it). It must print `PASSED`, with warnings that the operator closed the record early and that no bar has been promised since the closure. ## What it is for It measures, and nothing more. The operator's stated prior, sealed in the pre-registration before the first signal, is that **none of these strategies has an edge**. The ledger exists so that whether that prior is wrong can be settled by anyone, from public data, without trusting the operator. It is not a signal service, and nothing in it is a recommendation to trade. ## How it is built - **Pre-registered.** Before the first bar, the genesis ceremony sealed the 88 strategies (the whole source of the modules they live in, their parameters, and the numpy and TA-Lib versions they run under), the exit geometry (a stop at 1.5 ATR, a target at 2R, and a close at 48 bars), the cost model, the random-control algorithm, the two endpoints, the sample size and the stopping rule. Nothing about how a trade is scored can change without a new pre-registration in the chain. - **Promised before it is known.** Each 4h bar is announced while it is still open, with the roster of strategies that must answer for it. At the close, every strategy on that roster gets a row: a plan, a watch, a no-trade or an abstention with its reason. A promise that goes undelivered stays visible as a shortfall. - **Controlled.** Every plan is paired with a control that enters at the same bar and price with the same risk and reward. The control's side is a coin flip drawn from the bar's sealed promise, the strategy, its direction and the close, none of which anyone could vary once the bar had closed. A pair measures whether choosing the side beat a coin. It does not remove the market's drift for a strategy that is almost always on one side, so each strategy's share of longs is reported, and one more than three-quarters on one side gets no verdict. - **Judged by sealed rules, in a sealed order.** The PRIMARY question is the family's: pooled over every strategy's independent episodes (positions that never overlap), did the 88 choose sides better than a coin? It is decided once, over the episodes entered before the three-year deadline and once all of their outcomes are written, by whether the lower bound of a one-sided 95% cluster bootstrap over 8-day blocks is above zero. Later plans are never pooled in. It is a single test, so it carries no multiplicity correction. The SECONDARY question is each strategy's, on its first 30 independent episodes, by a Student t bound at one-sided 0.05/88. That test is underpowered by construction: it needs about 1.13 R per episode, more than any strategy in this roster has produced, so a strategy it does not flag is not shown to have no edge. The rules, their critical values and their measured power are in the pre-registration; the verifier implements both. - **Scored from public data.** Outcomes are computed from Binance's public daily kline archive, and each resolution cites the archive files it used and their published checksums. - **Sealed every half hour.** A head row seals everything written since the previous one, and the rows it covers are published here as one new file. Heads are never more than an hour apart, counting the genesis as the first; a late one is preceded by an incident that says so. - **Anchored in Bitcoin.** Each head's hash is timestamped through OpenTimestamps, which needs no account: the proof is checked against Bitcoin's own block headers. Anchoring takes a few hours after a head is sealed. The proofs are in the chain themselves, so even a rewritten history here could not move a head that was already anchored. ## Layout ``` segments/YYYY/MM/DD/-.ndjson ``` Each line holds one row: `{"seq": …, "row_hash": …, "body": …}`. The `row_hash` is the SHA-256 of the `body`, and each body names the hash of the row before it. The body of each row is its full content in a canonical line format. Files are only ever added. You can check that no published segment was ever changed or removed: ```bash git log --diff-filter=MDR --name-only -- segments # must print nothing ``` That check reads the history you cloned, so by itself it cannot see a rewrite that happened before you cloned it. Two things close that gap. The server that publishes this repository refuses force-pushes and deletions on `main` (`receive.denyNonFastForwards`, `receive.denyDeletes`); that is a promise of the operator's, who runs the server. What does not depend on trusting the operator is that every head is timestamped in Bitcoin, so a rewritten chain would have to produce a hash that was already in a block; that covers each head once it is anchored, a few hours after it is sealed, and the verifier tells you which is the latest anchored head. Clone with full history (no `--depth`, which this server does not offer anyway): the verifier warns when it cannot read it. ## Where it lives The operator's own host publishes it, read-only: - `https://wavelab.dr-techsolutions.com/git/wavelab-ledger.git`, to clone; - `https://wavelab.dr-techsolutions.com/files/wavelab-ledger/`, the same files to browse; - `https://wavelab.dr-techsolutions.com/ledger`, a page that reads it. Keeping your own clone is the strongest thing a reader can do: a later rewrite of anything you hold would no longer match it, whoever hosts the copy then. ## Verify it ```bash git clone https://wavelab.dr-techsolutions.com/git/wavelab-ledger.git git clone https://wavelab.dr-techsolutions.com/git/wavelab-verify.git python3 wavelab-verify/verify.py wavelab-ledger ``` The verifier uses only the standard library and shares no code with the operator. It recomputes the chain, every plan's stop, target and cost, every control's coin flip and every outcome from the exchange's archive, and it reports each disagreement. See [wavelab-verify](https://wavelab.dr-techsolutions.com/files/wavelab-verify/README.md) for what it checks and what it does not. ## Licence The data is published under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Operated by DR Markets.