Methodology
Most weeks it says nothing.
The Delta Engine answers one question about Bitcoin, once a week: do current conditions resemble the conditions that have marked past bottoms. The honest answer is usually no, and a tool built to find something every week would have been a worse tool.
Bottoms are noisy. Price is falling, sentiment is poor, volume is heavy, and every one of those things is also true in the middle of a decline that has a long way left to run. Reading them individually gives you a wall of indicators and no answer.
The Delta Engine reduces that to one weekly state. It reads the closed weekly candle: how the week finished, whether buyers or sellers dominated the actual traded flow, how heavy the volume was, where price sits against its recent range, and how bullish or bearish investor sentiment is. Those combine into a single reading.
The design choice underneath all of it is conservatism. The tool would rather stay silent than call a bottom on a weak week, because a bottom signal that fires often is a bottom signal that means nothing.
Earlier versions published weaker readings alongside the strong ones. Testing showed the weak signals carried no forward edge, and the weak top-side readings were actively misleading. They are still computed and stored, and they are no longer shown. A signal with no edge does not become useful by being displayed with a caveat.
The tool publishes Strong Bottom, and otherwise publishes No Signal, which is by far the more common of the two. A top-side watch flag exists and is labelled on screen as unvalidated, because it has fired forward once. The tool's own copy calls it a starting point rather than a track record, and that framing is deliberate.
Every week carries a confidence marker showing whether all inputs were available. The sentiment survey it uses publishes mid-week and cannot be collected automatically, so the Monday reading goes out flagged as preliminary and is recomputed once sentiment lands. The alternative was publishing a clean-looking state built on a missing input.
When the rules improve, past weeks are relabelled under the new rules. One week previously counted as a bottom no longer qualifies and is no longer counted. That means any historical tally moves. Freezing a flattering old count and quoting it forever would have been the easier option and it is not what the tool does.
The thresholds were tuned on the same history the record comes from
This is the most important caveat on the page. The rules that decide what counts as a bottom were developed using the same weekly history that any performance figure would be measured against. There is no held-out period that the tool has never seen. That is a real limitation on how much weight any historical count deserves, and stating it plainly is more useful than a percentage that quietly ignores it.
The record is short and the affirmative signal is rare
The weekly archive runs from the middle of 2023, and only a handful of those weeks carry the bottom signal at all. Rarity is the design working as intended. It also means the sample behind any statistic about it is small enough that the statistic should be read as description rather than as evidence.
Part of the history was reconstructed after the fact
A portion of the stored weeks were backfilled from historical data rather than captured live as they happened. Each row is tagged with which it is. A backfilled week and a live-captured week are not equivalent, and the tool keeps the distinction visible rather than blending them into one clean series.
One input depends on a person
The sentiment survey blocks automated collection, so it is entered by hand each week. If it is not entered, that week runs a category short and is flagged as such. It is a known fragility in an otherwise automated pipeline.
Silence is the honest answer most weeks.
Members see the weekly state, the conditions behind it, the confidence marker and every stored week since the record began.
Discussion