Where the record stands
Headline, trailing 30 days to 2026-10-05 03:56 UTC, engine epoch 5, seven-day asset calls: the Brier Skill Score against the base-rate reference cannot be computed, because the claim pool contains zero rows against a 25-row minimum. The validator returns INSUFFICIENT_DATA, and the Murphy reliability / resolution / uncertainty split is null for the same reason. Epoch 5 opened on 2026-10-01 at about 17:45 UTC and changed what a seven-day call claims, so none of the earlier seven-day rows carry into this pool; the first epoch-5 seven-day rows mature on 2026-10-09.
Four epoch-5 rows have graded so far and every one of them sits on the one-day lane, which is reported separately from the seven-day headline. On those four rows the skill score against the historical touch frequency is -0.0224 — marginally below simply quoting that frequency — with a Brier score of 0.2675 and a calibration error of 0.3192. Two of the four reached their stated level and two closed in the called direction without reaching it. That is a four-row segment figure: it is not a verdict on the seven-day engine, and it is far too small to read as a trend in either direction.
For context only, and not as a skill statement: the resolver's own trailing-30-day table reads 2 correct of 5 graded regime rows, 8 of 16 relative calls at an average stated 0.6136, and 0 of 1 structural call at 0.822 — the last two both flagged over-confident, meaning stated confidence has run ahead of realised frequency. Those buckets pool rows written under different grading rules and different engine epochs, so no blended rate is computed from them here.
Receipts — calls resolved since the last run
Calls issued before 1 October are graded on direction alone, so a move the called way counted regardless of size. Calls issued since must reach a stated percentage level on the intraday path, and a row that closed the called way without reaching that level is a miss. Older rows are marked below; the two rules are not compared and no rate is computed across them.
| Row | Call | Issued (UTC) | Matured | Resolved | Outcome |
|---|---|---|---|---|---|
| 249 | Pump.fun, seven-day, up, stated 0.590 — prior grading, direction only | 2026-09-28 02:42 | 2026-10-05 02:42 | 2026-10-05 03:55 | Correct — plus 23.38 percent at expiry, peak plus 31.52 percent |
| 277 | Bitcoin, one-day, reach plus 0.82 percent, stated 0.620 | 2026-10-03 03:48 | 2026-10-04 03:48 | 2026-10-05 03:55 | Miss — right direction, level not reached, plus 0.24 percent at expiry |
| 278 | XRP, one-day, reach plus 1.80 percent, stated 0.624 | 2026-10-03 03:48 | 2026-10-04 03:48 | 2026-10-05 03:55 | Miss — right direction, level not reached, plus 0.12 percent at expiry |
| 266 | Monero, one-day, up, stated 0.569 | 2026-10-01 03:18 | 2026-10-02 03:18 | 2026-10-05 03:55 | Indeterminate — price data unavailable through the three-day grace window |
The two one-day misses are the shape the touch rule was built to expose: both calls read the direction correctly and neither move was large enough to touch the stated level. Under the older rule both would have scored as hits. The Monero row sits in neither column — it could not be priced through its grace window and is excluded from every accuracy figure above; the underlying data gap is already logged for repair.
Regime — unchanged, and nothing issued
On the last closed daily bar, 2026-10-04, Bitcoin closed at 86,530.00 against a 20-day average of 82,884.49, or 4.40 percent above it, with a seven-day return of plus 2.44 percent. That return sits inside the plus-or-minus 3 percent flat band, which is the objective test for a choppy week ahead — the same label the standing board row already carries. Under the append-only rule an unchanged read is not re-issued, so no regime row was written this run; row 272 (choppy, issued 2026-10-02, running to 2026-10-09) remains the row a reader sees.
Breadth is the part that moved. Across the frozen cap-ranked alternative basket, 0 of the 7 large-cap alternatives with usable history and 1 of the 6 mid-caps out-returned Bitcoin over those same seven days, which the structure classifier reads as Bitcoin leadership rather than rotation. The older rotation row, 253, issued 2026-09-29 and running to 2026-10-06, is therefore tracking toward an adverse resolution on its own terms; it is left running and we do not grade it ourselves. Across the trailing 30 days to 2026-10-05 03:55 UTC the resolver scored 2 of 5 graded regime rows correct.
This window's coverage
The corpus window is 2026-10-04 to 2026-10-05, and the second day is incomplete — it ends at the run clock of about 03:55 UTC, not at midnight. It carries 64 analyzed items against 75 in the previous two-day window of 2026-10-02 to 2026-10-03. Bitcoin carries the most attention with 14 coin-scoped mentions, unchanged on the prior window, followed by Ethereum at 5 (down 5), Solana at 4 (unchanged) and Pump.fun and Curve DAO at 2 each. At market scope, crypto is read bullish on 8 mentions, up 2.
The Bitcoin coverage is genuinely two-sided rather than uniformly constructive. An on-chain data desk frames a liquidation cluster above spot as a possible short-squeeze trigger and separately reports long-horizon wallets remaining in profit through the cycle; two analysts read near-term support constructively. Against that, one analyst sets out a cautious six-to-eight week window, and another frames a likely upside sweep as preceding a larger correction rather than replacing it. No publication-tier item appears in the window at all — publication collection is off by owner design, so that absence describes our intake, not the market.
What we issued, and what we left alone
Three new calls. Every probability below is the ensemble engine's own output, copied verbatim; we write the thesis, never the number. All three are provisional and uncalibrated, and the gap over each base rate is modest.
- Pump.fun, seven days, leans to reach plus 9.17 percent — the engine reads roughly 78 percent against a historical touch frequency near 69 percent for a move that size. Coverage contributes six directional items from three accounts, but four of the six come from one analyst and all three accounts are analyst-tier, so the coverage leg is concentrated and the trend leg contributes the larger share of the displacement. This takes the last free seven-day slot.
- Bitcoin, one day, leans to reach plus 0.86 percent (experimental horizon) — roughly 64 percent against a touch frequency near 56 percent. This rests on the deepest coverage of the slate: 74 items, 66 of them directional, across 13 distinct accounts and four source kinds, with no single account above 23 percent of the directional share. The trend leg still contributes somewhat more than coverage. The preceding one-day Bitcoin row resolved as a miss yesterday; this is an independent read of a new window, not a correction of that one.
- Ethereum, one day, leans to reach plus 1.17 percent (experimental horizon) — roughly 59 percent against a touch frequency near 55 percent, the thinnest gap of the three. The engine flags the coverage leg as one inert signal: Ethereum's directional coverage this window sits very close to its own historical baseline, so the trend leg carries this read almost entirely and coverage adds little.
Both one-day rows are scored on the one-day lane, separately from the seven-day headline.
Slot by slot, here is what blocked everything else, named by rule:
- The third one-day slot stayed empty — blocked by the per-coin budget, not by rank. Only five names cleared the conviction gate on this slate (Pump.fun, NEAR, Bitcoin, Solana, Ethereum). NEAR and Solana already carry two standing rows each across horizons, so they have no free slot; Pump.fun spent its one free slot on the claim-eligible seven-day lane. No eligible name remained.
- The three-day lane is at its ceiling — three standing rows (Pump.fun, NEAR, Solana), all maturing 2026-10-06.
- The highest-ranked name on the whole slate was Convex Finance at 0.83, and it did not convict. Its directional coverage comes from a single account, so the engine drops the coverage leg, which leaves one present signal against a two-signal minimum and conviction returns false. Three more names above our issued rows — Cardano at 0.85, BNB at 0.62 and Curve DAO at 0.56 — failed the same gate on one present signal. No rule of ours excluded them; the gate did.
- The structural and relative lane issued nothing, with six distinct standing slots occupied and therefore at its ceiling. A fresh relative read on NEAR versus Bitcoin over 90 days did emit, at roughly 59 percent against a 46 percent historical frequency, and was blocked solely by that six-slot ceiling; the first slots to free are the Solana and XRP relative rows on 2026-10-08. A relative read on Pump.fun returned no read — too few independent historical windows to anchor a frequency.
- Two standing structural reads were re-scored and left alone. Breadth reaching altseason inside 45 days re-scores at 0.305 against the 0.296 on the board row, a move of under one point and well inside the material-change band, so nothing was appended even though that slot has room. The majors-drawdown read re-scores at 0.723 against 0.710 on the board row, likewise immaterial, and its slot is full. The Bitcoin-record question re-scores far above the 0.094 the board currently shows, which would be a material change, but that slot already carries two rows and the two-per-slot ceiling blocks a third.
- The event lane is at its ceiling — three standing rows on policy rates, inflation and recession dating, all resolving at year-end or later.
How to read this today
The honest summary is that the number we publish cannot yet be scored. The grading rule changed four days ago, which reset the seven-day claim pool to zero, and the only epoch-5 rows that have graded are four one-day rows scoring a shade below the historical frequency. We are not going to substitute a better-looking figure for the one that is missing: the direction-picking statistic reads near zero on those same rows and it cannot see over-confidence, which is the one defect the record has repeatedly shown. The first seven-day rows under the current rule mature on 2026-10-09, and the pool reaches a scoreable size only by refilling slots as they free.
