The week-ahead regime: still choppy
The standing choppy regime read (issued 26 July, matures 2 August) stands unchanged, and this window's corpus backs it. Bitcoin carries by far our deepest coverage at 85 scoped mentions and nets mildly constructive: treasury accumulation at several listed holders against a three-day ETF outflow streak, a fund cutting its exposure to raise cash, and a fifth straight week without a purchase at the largest corporate holder. The market-level read is bearish, with coverage pointing at a slide ahead of the U.S. rate decision and one large trading firm publicly forecasting a surprise hike.
Alt breadth is thin: Ethereum, Solana, Binance Coin and Zcash all score sideways on the engine's 7-day scan, with no directional conviction at all — and Bitcoin scores sideways alongside them despite its constructive coverage. Nothing here is decisive in either direction, and nothing is broad enough to be a rotation.
What we issued this run
Storj, downside lean (7 days, and again on the experimental 3-day horizon). This is the widest signal gap on the slate, and it is idiosyncratic rather than a proxy for Bitcoin beta. Storj Labs filed for voluntary Chapter 11 restructuring; the token fell roughly 20 percent on the filing, coverage across three separate items is uniformly bearish, and on-chain reads show smart-money wallets flat rather than bidding the drop. Both of our available signals lean the same way. The modeled chance of a notable weekly decline is near 59 percent against a roughly 41 percent base rate - a real gap, but an unvalidated one.
XRP, upside lean (7 days). A thin one. Coverage is constructive on a run of spot ETF net inflows; momentum over the same stretch is negative. The two largely cancel and the engine nets roughly 29 percent against a roughly 26 percent base rate. We report it because it cleared the gate, not because we think it is strong.
Convex Finance, upside lean (experimental 3-day horizon). Two independent corpus items - an analyst bounce setup off multi-year lows, and a rumoured institutional bid for gauge voting power - line up with positive short-term momentum. Roughly 45 percent against a roughly 31 percent base rate.
The 7-day book carries forward existing leans on Convex Finance, Ethereum, Bitcoin and Curve, all issued earlier this week and all still running to their own expiries. We do not edit or retire a call once written.
Receipts: three matured, three wrong
All three calls that matured since the last run resolved against us.
| Call | Issued | Resolved | Outcome | What happened |
|---|---|---|---|---|
| XRP, 7-day upside lean (stated 30.8%) | 21 Jul 12:14 UTC | 29 Jul 03:03 UTC | incorrect | moved -7.57% |
| Solana, 7-day upside lean (stated 36.6%) | 21 Jul 12:14 UTC | 29 Jul 03:03 UTC | incorrect | moved -6.61% |
| Convex Finance, 3-day upside lean (stated 44.1%) | 26 Jul 02:38 UTC | 29 Jul 03:03 UTC | incorrect | moved +2.73% |
The third row is worth dwelling on, because it shows what these calls actually claim. Convex Finance rose over the window - the direction was right - but the move did not clear the size threshold the call was written against, so it graded as a non-event and the call is marked wrong. Our asset calls have three possible outcomes, not two: a notable move up, a notable move down, or sideways. Sideways is not a rounding case.
That is also why the stated probability on a call sits well below 50 percent and should: the reference point is the roughly one-in-three base rate for a notable move in a named direction, not a coin flip. And the complement is not the opposite direction - the residual on a 29 percent upside lean is mostly nothing much happening, not a fall.
Where the record stands
The headline figure is our Brier Skill Score against the base-rate reference, the only metric on this page that reads the probability we actually publish. On the current-engine claim pool it is -0.4598 across five resolved calls: our probabilities are scoring materially worse than simply quoting the historical base rate and ignoring every signal we have. The Murphy decomposition says where that comes from - reliability 0.1738, which is where over-confidence shows up, and resolution 0.0000, meaning the probabilities are not yet discriminating between outcomes at all.
We state this plainly rather than reaching for a friendlier number. A directional-accuracy diagnostic on the same calls reads -34 percent, which is no better. The claim pool is five resolved calls against a required minimum of twenty-five, so the formal verdict is insufficient data and no skill claim is established - but too early to judge is not the same as looking good, and on the evidence so far it is not looking good.
We do not hand-adjust the engine's output in response to that. The probabilities published above are exactly what the engine produced, and the miscalibration stays visible in the record until the engine itself is fixed.
What we did not call
The structural questions produced no new answer this run. The Bitcoin-record question emitted only a historical frequency with no directional signal, and it already carries two standing reads. The altseason and major-drawdown questions returned no scoreable read at all. A fresh relative read on Convex Finance versus Bitcoin scored well, but the relative slate is at capacity until the existing quarterly reads mature in October.
Dogecoin and Avalanche both cleared the score gate on momentum alone - with no directional corpus signal behind them, that is a single-signal read, not our multi-source edge, so neither was issued. Event calls remain offline. No read is a valid outcome, and five of the eleven catalog questions correctly show nothing today.
