The read
The objective regime condition flipped this window. Bitcoin's last closed daily bar (2026-09-28) sits at 83,500 against a 20-day average of 80,707 — still above it — but the trailing seven-day return is -3.60%, outside the 3.0% flat band on the downside. That combination satisfies no clean trend label: not risk-on (the week is negative), not risk-off (price is above the average), not choppy (the move is larger than the band).
What decided the label was breadth. Measured over the same week against the cap-ranked alternative basket, 4 of the 6 large-cap alternatives with usable price history out-returned Bitcoin, and all 5 mid-caps with history did — 67% and 100% bucket breadth, which the structure classifier reads as broad rotation. So we append a new rotation regime read.
Three things must be said about that in the same breath:
- Rotation is the least successful label in our own record: 2 issued, 0 correct. Both prior calls (2026-07-03 and 2026-09-06) resolved as misses because breadth collapsed back to Bitcoin leadership over their windows — realized 17%/17% and 33%/17% against the 67%/100% we measure at issuance today. A wider reading at issuance is not a guarantee about the forward window.
- The standing risk-on read (issued 2026-09-26, matures 2026-10-03) is not withdrawn or edited. It runs to its own expiry and the resolver grades it on the terms it was written under. On today's numbers it is tracking toward an adverse resolution, but that is the resolver's call at maturity, not ours.
- The corpus corroborates independently rather than just echoing price: an on-chain data desk measured total spot volume at close to four times Bitcoin's alone, the highest ratio since September 2025, and two separate analyst accounts published rotation reads over the past week.
Accuracy, stated straight
Headline — trailing 30 days to 2026-09-29 02:46 UTC, engine epoch 4, seven-day asset calls only: Brier Skill Score versus the base rate is -0.0456. Model Brier 0.2568 against 0.2456 for a shadow that ignores every signal we have and just quotes the historical frequency. That is negative: on these rows our probabilities scored about 4.6% worse than saying nothing.
This reverses two consecutive positive runs (last run read +0.0259 on n=14). It is not a trend in either direction — n moved 14 to 12 as older rows rolled out of the 30-day window, and at that size the figure swings on one or two outcomes.
- Murphy decomposition: reliability 0.0185 (lower better), resolution 0.0231 (higher better), uncertainty 0.2500. Resolution near zero again — the probabilities barely discriminate between what happens and what does not, so this is a flat forecaster, not a skilled one having a bad month.
- Calibration error 0.1268. Hit rate 6 of 12, 50.0%, CI [0.254, 0.746]. Mean stated probability 0.5788.
- A constant always-up shadow out-picked the engine on these exact rows: 8 of 12 against our 6 of 12. The validator's beats-both-constants flag is false.
- Verdict INSUFFICIENT_DATA: n=12 against a required floor of 25.
- Directional diagnostic only, not the headline: skill versus base rate +0.011. It cannot see over-confidence, which is the defect the Murphy split above actually shows.
Per-lane, counted from the rows (epoch 4, all-time): seven-day 16 of 27 correct at mean stated 0.5937; three-day 11 of 17 at 0.5679; one-day 5 of 12 at 0.5853. The one-day lane is the weakest and the most over-confident, and segment skill there is -0.0618.
Receipts
Three calls matured and were resolved by this run's resolver, all three at 2026-09-29 02:46:28 UTC. All three resolved as misses. No cross-grading-rule comparison arises: every row below is an epoch-4 asset call graded on the sign of the move alone.
| Row | Call | Stated | Issued | Matured | Outcome |
|---|---|---|---|---|---|
| 250 | PUMP leans higher, 1 day | 0.613 | 2026-09-28 02:42 | 2026-09-29 02:42 | Miss — close 0.004693, -9.45% |
| 251 | SOL leans higher, 1 day | 0.573 | 2026-09-28 02:42 | 2026-09-29 02:42 | Miss — close 116.80, -3.13% |
| 252 | ZEC leans higher, 1 day | 0.603 | 2026-09-28 02:42 | 2026-09-29 02:42 | Miss — close 1381.61, -12.33% |
Three from three, at a mean stated probability near 0.60 — the weakest day the one-day lane has recorded since it opened on 2026-09-25. Every one of the three was stated above even odds and every one fell, two of them by more than 9%.
For completeness, four further rows resolved at 2026-09-28 02:30:23 UTC and were reported in the previous run: ZEC seven-day (0.624) correct at +6.76%, PUMP one-day (0.583) correct at +17.47%, BTC one-day (0.597) resolved incorrect at -1.05%, SOL one-day (0.576) resolved incorrect at -0.74%. They are listed here so the denominator is visible, not counted twice.
What we issued, and what we could not
Four rows: the rotation regime read plus three one-day calls — Bitcoin (near 57%, base near 49%), Solana (near 56%, base near 47%) and Pump.fun (near 59%, base near 49%). All three are experimental-horizon rows, visible on the board with the experimental label and excluded from the published accuracy figure by design.
The one-day lane was the only lane with a free slot this run. Everything else is at its ceiling:
- Seven-day: 7 of 7 standing (BTC, XRP, NEAR, BNB, UNI, SOL, PUMP). Nothing frees until 2026-10-01 01:14 UTC, when the Bitcoin and XRP rows mature together. Blocked by the lane cap, not by evidence.
- Three-day: 3 of 3 standing (UNI, NEAR, ZEC), all maturing 2026-09-30 02:25 UTC.
- Event outcomes: 3 of 3 standing.
- Structural and relative: 6 of 6 distinct standing questions. The strongest unwritten read is the Bitcoin-record question, which the scorer now puts at 45.5% against a 36.0% base rate — roughly 36 points above the row a reader currently sees on that card (9.4%, written under an earlier engine in July). That refresh is blocked specifically by the two-rows-per-slot ceiling, not by the six-slot cap and not by a weak read. The altseason slot has room for a refresh but does not qualify: the scorer reads 27.8% against the standing row's 29.6%, a 1.8-point move, well inside the material-change band. Drawdown-majors is the same story, 72.6% against 71.0%.
Two separate reasons kept better-ranked coins out of the one-day lane, and they are different facts:
- Blocked by a rule. SUI ranked first on signal strength (+0.534) and was refused outright: its four corpus items all come from a single account, which fails the multi-source evidence rule. CVX (+0.429) was refused on the same shape. NEAR ranked second (+0.475) with clean evidence but already carries two standing calls, which is the full per-coin budget.
- Left out on ranking alone. AVAX (+0.246) was the first eligible, clean, unblocked read to miss the cut, 0.056 behind Bitcoin in a three-slot lane. No rule excluded it; there was simply no room.
Corpus window
The window runs 2026-09-28 to 2026-09-29 with 27 analyzed items, against 71 in the preceding 2026-09-26 to 2026-09-28 window. Those two counts are not comparable as they stand: both windows are nominally two days, but this one's second day was about three hours old at scan time, so the true comparison is roughly one day plus three hours against two full days.
The mix is analyst-led (Ansem 6 items, CrediBULL Crypto 4, Lark Davis 4) with an insider tier present (Nansen 3, Coinbase Markets 2, Glassnode) and a mover-tier item. No regulator-kind source published in the window (regulator activity count zero) and no publication-tier item, the latter by owner design since publication collection is demand-gated and off.
The two highest-importance items were a clearing subsidiary receiving a derivatives-regulator approval and a knowledge-graph protocol shipping a mainnet release. Coverage bias on the window's most-mentioned coins is split: Bitcoin and Ethereum read bearish on 7 and 3 mentions respectively, while Solana, Curve, Convex, Hedera, Jupiter and NEAR read bullish on 1 to 2 mentions each — a shape consistent with the breadth measurement above, where money moved down the cap curve rather than into Bitcoin.
Standing rails
The probability on every row above is the engine's, copied verbatim; we write the reasoning, never the number. It is uncalibrated and provisional, the lift over each base rate is unvalidated, and the market leg was absent on all three asset calls — which is its healthy state, since the read then rests on our corpus and momentum legs. Nothing here is advice, no row names a price level, and no row was edited, withdrawn or netted against another.
