CRYPTO MECHANICS
Signal laboratory: market mechanics → testable hypotheses → signals with public track records. dashboard · bulletins
We don't sell confidence. We earn it — and show the process.
Every signal block below exists in code and data, not in a pitch deck. Each has its own sample size, its own validation threshold, and a public counter. Until a block passes its threshold it is marked validating — that's not embarrassing, that's method.
STAGE 1 · HYPOTHESISMechanics + dataformula, thresholds, observation log
STAGE 2 · VALIDATIONSample accumulationshadows logged every cycle, n grows
STAGE 3 · SIGNALn ≥ threshold + half-samplespublished with full outcome history
House rule: a block becomes a signal when its sample passes the threshold (usually n≥100) AND the edge holds on both independent half-samples AND survives costs. We never sell a "maybe" as a "will".
Signal blocks
1 · Compression detectorvalidated · 570 days
What it does: detects daily Value Area squeeze and flags "double compression" — two narrow days in a row. Predicts a zone breakout within ≤3 days, with direction from the 30-day channel slope + weekly CVD.
570 daysvalidation span (3 BTC regimes)
96–100%breakout after compression
60%directional accuracy (axes agree)
36/36signals in the latest regime
Where it lives: "VA migration" chart on the dashboard — yellow/orange/green marks with verdict tooltips.
Honesty: raw POC-migration direction without a regime filter is only 42–44%; we trust the arrow when channel and CVD agree, otherwise "direction unreadable".
2 · Ready-made plans + outcome ledgerlive · since #006
What it does: bulletins with pre-defined trigger/entry/stop/targets/expiry. After publication every plan is checked against the market: ACTIVATED / TP1 / TP2 / STOP / EXPIRED / UNVERIFIABLE.
6 issuesbulletins published
1st verifiable plan#006: entry/stop/expiry set before the market
100%outcomes in the open ledger
Where it lives: public bulletins — the "honesty ledger" with full history, including UNVERIFIABLE statuses we don't hide.
3 · Fuel verdictvalidating · n=28/30
Idea: before entry, compute whose stops already burned: CLEAN (fuel untouched) / OWN_BURNT / FOE_BURNT / MIXED. Hypothesis: entering right after a major harvest works against you.
60% win · +858CLEAN segment (<$5k burned)
−23415 trades after major burns
28 / 30scored trades to threshold
validation progress: 28/30 tradesupdates automatically with every EXIT
Shadow is appended to every scanner entry post-factum. Gate into trading — after n≥30 per segment and half-sample confirmation.
4 · State map 3×3validating · 345 slices
Idea: every hour the market is in one of 9 states: {whose burn: longs/shorts/mixed} × {OI tank: filling/flat/draining}. Hypothesis: the future price distribution depends on the state, not on news.
5/5 majorsagree on L_BURNT+OIfill (buyback)
64–100% upbest cell across coins
14–17% upworst (MIXED+flat) — "don't trade"
345 / ~2000slices by next week
recorder writes every 5 minutesthreshold: n≥100 per cell
Two cells already deliver a management "don't trade" — and that's a product too: capital saved instead of a false signal.
5 · H6 setup classesvalidating · canon n=206
Idea: every scanner entry is labeled with a TANK:FLOW:COH class (OI tank × flow direction × cascade coherence). The canon showed a full R of difference between coherent and anti-coherent entries.
+1.72RFULL:INFLOW:COH (n=39, PF 9.4)
+0.72Rsame tank+flow, but ANTI (n=39)
0 / 30live entries in classes (shadows since Aug 23)
live labeling restored Aug 24threshold: n≥30 per class
6 · Cascade detector (ignitions)material · canon collected
Idea: a minute of forced flow ≥$200k, ≥80% one-sided = "ignition". Canon of 3348 events (2.08M ticks) collected, forward returns computed, trade specification not closed.
Will be promoted to a signal only after pre-registration and full-procedure runs. Rushing this costs more than silence.
The data behind it
82.6 GBHyperliquid block-level fills (125 days, per-address closedPnl)
2.08MBybit liquidation ticks (Tardis, ignition canon)
130symbols with hourly OI/funding history
522symbols in the live scanner (hourly)
Sources: Binance/Bybit/Hyperliquid/Coinalyze. All research is reproducible: formulas and thresholds are published in the open log.
Why so slow and so open: every block was once an n=5 hypothesis. We could show you an "AI consensus of 5 out of 7 models" today — but we couldn't show you its outcome ledger. So: ledger first, then signal, then product. Boring. Verifiable.