The read
The objective regime read on the closed daily bar of 2026-09-30 lands inside the flat band: BTC closed 2.82% above its 20-day moving average, with a 7-day return of -0.92% against a 3.00% flat band. Breadth, measured on the frozen alt basket, reads 88.9% of the priced large-cap names and 80.0% of the priced mid-cap names ahead of BTC over the same seven days, which the structure classifier labels altseason. That is the same broadening the standing rotation row (id 253, issued 2026-09-29, runs to 2026-10-06) was issued on, so under the append-only rule no new regime row was written this run. The widget stays lit by 253; a second regime row two days into its window would be the same belief restated, not a new forecast.
Two regime rows are live at once because 237 (risk-on, issued 2026-09-26, runs to 2026-10-03) has not matured yet. On today's numbers 237 is tracking toward an adverse resolution - risk-on needs the 7-day return above the band and it is -0.92% - while 253 is tracking toward a favourable one. Both will be graded by the deterministic resolver at their own expiry; neither was touched.
What we issued
Four new calls, all from the ensemble engine, all direction up, none with a market leg present:
- Bitcoin (BTC), 7 days, about 52% against a 46% base rate for an up week. This is the broadest-based read on the slate: 66 corpus items over the scan window, 58 of them directional, across 12 distinct sources and four source kinds, with no single account above 29% of the directional share. The window itself is genuinely split - a 97% week-over-week collapse in BTC ETF inflows (to 31 million dollars) and an on-chain note flagging heavy resting supply overhead both read bearish, against constructive support reads and corporate-treasury commentary - and the centred corpus lean of +0.12 is what survives that split, not a consensus. Momentum carries one third of the blend weight but, with the stronger lean, contributes more of the displacement than the corpus leg.
- Avalanche (AVAX), 7 days, about 49% against a 42% base rate. Read this one carefully: the corpus leg is near-inert. Three items mention AVAX over the scan window from two distinct sources, two of the three from one account, and none of them are AVAX-specific - they are a broad alt-rotation piece, an undervalued-names roundup, and an exchange-exploit post-mortem that merely lists it. The engine falls back to the global baseline for the centring, leaving a centred lean of +0.03, and flags one of the two signals inert. Momentum carries this read almost entirely, on the back of a positive seven-day return for AVAX while BTC fell. The stated probability sits below 50% because the base rate does; the lean is the gap, not the level.
- Ethereum (ETH), 1 day (experimental horizon), about 55% against a 51% base rate. Coverage is deep and genuinely multi-source - 33 items, 28 directional, 10 distinct sources, largest share 39% - but it is also split almost exactly on ETH's own historical baseline, so the centred corpus lean is +0.0015, effectively nothing. The trend leg carries this read. Named items in the window run both ways: a leverage-shift note on options open interest dropping to 23% of futures OI reads neutral, two analyst reads lean up, one chart read leans down.
- Monero (XMR), 1 day (experimental horizon), about 57% against a 54% base rate - the thinnest read of the four and the only one where the two legs disagree. Three corpus items from three genuinely distinct sources, but no XMR-specific story: the closest is a reported KYC breach at a consumer fintech exposing 680 users with crypto balances, which the analysis tags as privacy-relevant rather than Monero news. Momentum points the other way on this one - the engine records a negative trend lean and one conflicting signal - so the corpus leg alone is what lifts the number above base. On a one-day horizon that is a small and uncertain lean.
What we did not issue, and the rule that blocked each
- Regime: no new row. The objective read matches the standing rotation call; append-only says leave it running.
- SUI (score 0.544, the highest on the whole slate), CVX (0.478), CRV (0.343), LINK (0.177): all four convicted and all four were refused on the single-directional-source rule. One account is not a cross-referenced read.
- NEAR (0.486), PUMP (0.450), SOL (0.282): all three convicted strongly and all three are at the two-standing-calls-per-coin ceiling already. No rule found fault with the reads; there is simply no slot.
- XRP (0.151): a 7-day XRP call convicted, but the 7-day book filled at seven with BTC and AVAX ranked above it, and a second same-day XRP call on the 1-day lane would be a twin of row 262, which matured minutes before this run and has not been graded yet.
- UNI (0.104): already carries a standing 7-day row (241); a second concurrent call on the same coin and horizon is barred.
- DOGE (0.426): the highest-scoring name that failed the conviction gate outright. The first name left out on rank alone rather than by a rule was XMR on the 7-day lane - it was issued on the 1-day lane instead, where it still had budget.
- Structural and relative questions: nothing emitted. The btc_ath scorer now reads 43.9% with a genuine signal edge, against a standing displayed row of 9.4% (row 121, issued 2026-07-18 under a prior engine) - a material gap, but that question already carries two standing rows and the two-per-slot ceiling blocks the refresh. The altseason scorer reads 29.3% against a standing 29.6%, inside the material-change band. Drawdown reads 70.7% against a standing 71.0%, also inside the band and at the slot ceiling anyway. All three rel_coin coins sit at two standing rows each.
- Events: the three-slot lane is full.
Receipts
Four rows were graded by the deterministic resolver this run. Two hits, two misses.
| Call | Issued | Resolved | Outcome |
|---|---|---|---|
| Bitcoin (BTC) leans higher over 7 days, 60.0% | 2026-09-24 01:14Z | 2026-10-01 03:09Z | Miss - closed -0.82% |
| XRP leans higher over 7 days, 46.0% | 2026-09-24 01:14Z | 2026-10-01 03:09Z | Miss - closed -0.33% |
| Solana's lead over Bitcoin this quarter looks firmer, 73.8% | 2026-07-02 17:25Z | 2026-10-01 03:09Z | Hit - SOL +46.3% vs BTC +35.8% |
| Ethereum's stance versus Bitcoin this quarter flips toward the upside, 50.5% | 2026-07-02 17:25Z | 2026-10-01 03:09Z | Hit - ETH +57.9% vs BTC +35.8% |
Both misses were near-misses in size and still count in full: a 7-day up call that closes 0.82% lower is graded on the sign of the move alone and is a miss, with no partial credit. The two 90-day relative calls were issued in July under engine epoch 1 and are graded by their own unchanged rule, so no cross-rule comparison arises between them and the two 7-day rows above, and no blended rate across them is computed.
Accuracy, stated plainly
Trailing 30 days to 2026-10-01 03:09Z UTC, engine epoch 4, 7-day asset calls only (the claim pool):
- Brier Skill Score against the base-rate reference: -0.100. Our probabilities scored about 10% worse than simply quoting each coin's historical base rate and ignoring every signal we have. Our Brier is 0.261 against the base-rate shadow's 0.237. This is the headline, and it got worse this run, not better - two 7-day misses landed and it moved from -0.046 to -0.100.
- Verdict: INSUFFICIENT_DATA. The claim pool is 14 scored rows against a minimum of 25, so no skill claim is published either way. Forty-four epoch-4 rows are graded overall across all horizons and kinds; only the 7-day asset calls count toward the claim.
- Murphy decomposition: reliability 0.046, resolution 0.039, uncertainty 0.245. Reliability is where over-confidence shows, and it is the larger of the two terms we control - our stated probabilities run hotter than the outcomes justify. Resolution near 0.04 against an uncertainty of 0.245 means the probabilities barely discriminate between what happens and what does not. Expected calibration error is 0.184.
- A constant always-up shadow out-picked us on these exact rows: 8 of 14 for the constant against 6 of 14 for the engine. The validator's beats-both-constants flag is false.
- Directional accuracy against the base rate reads -0.052 on the same 14 rows. That figure is a direction-picking diagnostic only - it never reads the probability we publish, so it cannot see over-confidence - and it is reported here as a note, never as the headline.
- Across all resolved rows in the trailing 30 days to 2026-10-01 03:09Z UTC, regardless of horizon: asset calls 19 of 39 correct at an average stated probability of 0.578, relative calls 7 of 14 at 0.617, regime 2 of 4, structural 0 of 1 at 0.822. Every one of those with a probability attached is flagged over-confident by the resolver - the stated numbers are consistently above the realised rate.
The engine's probability is uncalibrated and provisional. The reasoning above is the product; the number is a model output we publish so it can be scored against us.
