The read
The standing week-ahead regime call remains choppy (issued 18 July, matures 25 July). Nothing in this cycle argues for a change: Bitcoin's blend still leans mildly higher on modestly constructive coverage and firm short-term momentum, but the modeled odds of a decisive weekly move stay below even, and alt breadth is thin - a handful of large caps lean up with the tape rather than leading it, and two of the coins we scan show no directional read at all.
The corpus window for this run (20-21 July) is unusually thin: only two items cleared analysis, covering Bitcoin, Ethereum, Solana and the memecoin sector, all neutral. The engine's own corpus signal reads a longer trailing window, which is why it still finds a lean where the daily snapshot shows almost nothing. Read the calls below with that thinness in mind.
Receipts - what resolved since the last run
Three calls matured and were scored by the deterministic resolver this cycle. Two hits, one miss.
| Call | Issued | Matured | Resolved | Outcome |
|---|---|---|---|---|
| XRP - 7-day downside lean (0.641) | 14 Jul 01:16 UTC | 21 Jul 01:16 UTC | 21 Jul 12:06 UTC | incorrect (+4.63%) |
| Ethereum - 3-day upside lean, experimental (0.432) | 18 Jul 08:10 UTC | 21 Jul 08:10 UTC | 21 Jul 12:06 UTC | correct (+4.76%) |
| Bitcoin - 3-day upside lean, experimental (0.412) | 18 Jul 08:10 UTC | 21 Jul 08:10 UTC | 21 Jul 12:06 UTC | correct (+3.38%) |
The XRP miss is the notable one: the model called a downside week and XRP rose more than four and a half percent instead. That reversal is also what flipped the momentum leg, and the model now leans the other way on XRP - a new, independent call, issued on its own merits, not a repair of the one that failed. The failed call stands in the record exactly as written.
Accuracy - where the current engine actually stands
The headline metric we measure ourselves by is the Brier Skill Score against the base-rate reference - the only figure that reads the probability we actually publish. For the current recalibrated engine on claim-eligible seven-day calls it is still unavailable: zero resolved calls in the claim pool, against a minimum of 25. The verdict remains INSUFFICIENT_DATA. There is no skill claim to make yet, and we will not manufacture one from a friendlier statistic.
For completeness, and clearly labelled as not the headline:
- The recalibrated engine has four resolved calls in total, all on the experimental three-day horizon, two of them correct. Four is far too few to mean anything, and three-day calls are excluded from the public accuracy claim by pre-registration.
- Every one of the 27 resolved seven-day calls in the book was issued by the previous engine, whose skill score was negative - its probabilities scored worse than simply quoting the historical base rate. That is precisely why the engine was recalibrated on 14 July, and why its numbers are materially lower now. We report that negative record rather than retire it quietly, but it is not the current engine's record and we do not carry it forward as one.
New calls this cycle
Four calls cleared the conviction gate on freed slots.
- XRP, seven-day upside lean, roughly 31 percent against a roughly 25 percent base rate. Coverage leans mildly positive and momentum has turned firmly higher; both signals point the same way.
- Solana, seven-day upside lean, about 37 percent against a roughly 32 percent base rate. Constructive coverage does most of the work here - tokenized-asset launches, a new actively managed multi-token product and fresh stablecoin issuance on the chain - while momentum is barely positive. The thinnest of the four.
- Ethereum, three-day upside lean, about 45 percent against a roughly 34 percent base rate. Experimental horizon, excluded from the accuracy claim.
- Bitcoin, three-day upside lean, about 45 percent against a roughly 36 percent base rate. Experimental horizon, excluded from the accuracy claim.
The four standing calls from earlier in the week - Ethereum and Bitcoin on seven days, Zcash on seven and three days - were re-scored and came back within a hair of where they were issued, so they run on unchanged. Records here are never edited.
What we did not call, and why
- No event calls. Our external event feed has not refreshed since 12 June, so no event has a defensible probability behind it. This is the thirty-second consecutive cycle blocked this way; it is escalated and awaiting a human decision.
- Structural questions. The Bitcoin-record question produced a read a shade above its historical frequency, which is where the current engine's standing call on it already sits, so nothing new was written. The older call on the same question - issued at 46 percent by the previous engine - stays on the board as written: a prior-engine record, not the current read, and the append-only rule means we do not revise it. The altseason and drawdown questions produced no directional read at all, and the relative-strength slots for Ethereum, Solana and XRP are already occupied. Those cards stand as they are.
- Skipped despite a passing score. One memecoin-infrastructure token scored up, but only because momentum overwhelmed coverage that leans the other way - a single-source read by another name, and we do not publish those. Two more coins had only one live signal each, one had no usable price history, and two showed no directional read.
- No social-trend input. The trends feed returned a billing error again this cycle. It does not feed the probability, only the candidate slate, so the effect is limited to slate breadth.
Standing state
One regime call, five seven-day asset calls, three experimental three-day calls, and ten long-horizon structural and relative calls - four structural and six relative, across six questions - are live. All of them will be scored by the resolver when they mature, hits and misses alike.
