# Weekly Calibration #12

**WEEKLY · CALIBRATION** · October 28, 2026 · MarketMania Research · Weekly series

Window: **Oct 19-25, 2026 UTC** · Cutoff: **Mon Oct 26, 2026, 16:00 UTC** · Methodology **v1.1 (2026-08-10)**, hash `e66c7e8c864a2233`

PDF: https://marketmania.ai/research/reports/weekly-calibration-2026-10-19.pdf · Open data (JSON): https://marketmania.ai/research/reports/weekly-calibration-2026-10-19.json

> Research question: *"when a model says 70, does it hit 70% of the time?"*

## Research Snapshot

| RESEARCH SNAPSHOT |  |  |  |
|---|---|---|---|
| **5,856** directional calls scored | **7** models (stable lineup) | of **9,171** mature forecasts | window **Oct 19-25, 2026 UTC** |
| hit = direction rule: tp1/tp2->hit, sl->miss, expiry->sign of gross pnl | FH gate **1h / 4h / 1d** | base field hit **57.0%** vs **61.5** stated | cutoff **Mon Oct 26, 2026, 16:00 UTC** |
| market: BTC net **-2.39%** (prior -1.98%) | ann vol **29.3%** (was 27.4%) | TOP5 volume **$14.02B**, -12% w/w | pairwise corr **0.93** (was 0.94) |

## Key finding

> **KEY FINDING · OBSERVATION (one weekly window)**
>
> ### The field's overconfidence gap moved +15.4pp -> +4.5pp week over week, 7 of 7 model lines narrowed, and field Brier went 0.2759 -> 0.2484.

## TL;DR

- **OBSERVATION -- The gap read +4.5pp this week.** Directional calls hit 57.0% against 61.5 stated mean confidence (issue #11: +15.4pp on 46.6% vs 61.9). Field Brier is 0.2484 (was 0.2759) and 6 of 7 model lines scored below 0.25, the score of an uninformative always-50% predictor. One window, descriptive.
- **7 of 7 models narrowed their gap.** Gap moves ran from gpt-6.1-sol's -15.9pp (+21.5pp -> +5.6pp) to deepseek-v4-pro's -7.1pp (+11.6pp -> +4.5pp); 7 of 7 model lines narrowed week over week. claude-opus-5.5 is the lowest-Brier line (Brier 0.2443); gemini-3.1-pro leads on hit-rate (58.3%). claude-opus-5.5 is the only line whose gap is negative (-1.5pp: stated confidence below its hit-rate).
- **The high-confidence buckets, N>=10 only.** gpt-6.1-sol 70-80 61.4% (n=44), deepseek-v4-pro 70-80 62.5% (n=16), gemini-3.1-pro 70-80 59.7% (n=342), gemini-3.1-pro 80-100 50.0% (n=26). Cells below N=10 are marked insufficient and rank nothing.
- **Trading, kept separate from every calibration table.** 3 of 7 lines finished net-positive; net PnL ran +$9.86 (qwen-3.8-max, best) to -$23.29 (gemini-3.1-pro, worst), field -$43.00 against issue #11's -$352.09.

### Week-over-week chart (see PDF for the grouped bar chart)

Overconfidence gap in percentage points (mean stated confidence minus hit-rate), week over week, sorted by this week's gap. Short names: opus-5.5 = claude-opus-5.5, fable-5 = claude-fable-5, gemini-3.1 = gemini-3.1-pro, gpt-6.1 = gpt-6.1-sol, qwen-3.8 = qwen-3.8-max, deepseek = deepseek-v4-pro. The opus-5.5, gpt-6.1 and grok-4.7 columns compare this window's single-build line against issue #11's spliced line. "Was" values are the numbers published in issue #11 for the back-to-back window Oct 12-18.

## Why it matters

Every MarketMania forecast carries a model-stated confidence from 0 to 100 alongside its direction call. This report checks whether that number tracks reality: on a well-calibrated forecaster, calls made at 70% confidence should hit their direction about 70% of the time. With issue #12 the series has twelve points on every model's gap -- enough to see movement, not enough to claim a trend: calibration and market regime move together, and durability claims belong in the Monthly series, not here.

## How to read this

Gap pp = mean stated confidence minus hit-rate, in percentage points; positive means overconfident. Brier is the mean squared error of the stated probability against the realised outcome, so lower is better and 0.25 is what an uninformative always-50% predictor scores. Coverage = calls scored divided by the mature calls available to that model. Confidence buckets are the models' own natural breakpoints, not an arbitrary binning; cells below N=10 are marked insufficient and never used to rank anything. Prediction and trading metrics live in separate sections and are never combined. Week A (Oct 12-18, the issue-#11 window) is the spliced week: the three lines flipped at Oct 17, 2026 09:01 UTC inside it (gpt-5.6-sol -> gpt-6.1-sol, claude-opus-5 -> claude-opus-5.5, grok-4.6 -> grok-4.7), so each week-over-week row for those lines joins this week's single-build line to last week's spliced line. This window (Oct 19-25) is the first full week on the new builds. In issue #11 those lines carried the labels "claude-opus-5.5 (incl. opus-5-era, Oct 12 00:00-Oct 17 09:01 UTC)", "gpt-6.1-sol (incl. 5.6-era, Oct 12 00:00-Oct 17 09:01 UTC)" and "grok-4.7 (incl. 4.6-era, Oct 12 00:00-Oct 17 09:01 UTC)".

## Calibration by model (sorted by Brier, best first)

| Model | N | Cov. | Hit rate | Mean conf | Gap pp | Brier | 95% CI |
|---|---|---|---|---|---|---|---|
| claude-opus-5.5 | 700 | 52.6% | 58.0% | 56.5 | -1.5 | 0.2443 | 54.3%-61.6% |
| gpt-6.1-sol | 962 | 72.3% | 57.6% | 63.2 | +5.6 | 0.2468 | 54.4%-60.7% |
| grok-4.7 | 836 | 69.4% | 55.9% | 60.1 | +4.3 | 0.2472 | 52.5%-59.2% |
| qwen-3.8-max | 829 | 62.4% | 56.3% | 59.9 | +3.5 | 0.2480 | 52.9%-59.7% |
| deepseek-v4-pro | 891 | 67.0% | 56.5% | 61.0 | +4.5 | 0.2481 | 53.2%-59.7% |
| claude-fable-5 | 768 | 57.7% | 56.4% | 60.9 | +4.5 | 0.2486 | 52.8%-59.9% |
| gemini-3.1-pro | 870 | 66.1% | 58.3% | 67.5 | +9.3 | 0.2551 | 55.0%-61.5% |
| **Field (all models)** | **5,856** | - | **57.0%** | **61.5** | **+4.5** | **0.2484** | - |

N = directional calls scored this window. Cov. = coverage, scored / mature calls available to that model. Gap pp = mean stated confidence minus hit-rate (positive = overconfident). 95% CI is the descriptive Wilson interval on hit-rate. The claude-opus-5.5, gpt-6.1-sol and grok-4.7 week-over-week rows join this week's single-build line to issue #11's spliced line (the three lines flipped at Oct 17, 2026 09:01 UTC, inside week A) -- see How to read this.

## Calibration curve: hit-rate by stated-confidence bucket

| Model | 50-60 | 60-70 | 70-80 | 80-100 |
|---|---|---|---|---|
| claude-fable-5 | 52.5% (217) | 58.0% (550) | 0.0% (1) * | n/a (0) |
| claude-opus-5.5 | 58.9% (660) | 42.5% (40) | n/a (0) | n/a (0) |
| deepseek-v4-pro | 55.2% (257) | 56.9% (617) | 62.5% (16) | n/a (0) |
| gemini-3.1-pro | 100.0% (1) * | 57.7% (501) | 59.7% (342) | 50.0% (26) |
| gpt-6.1-sol | 50.0% (154) | 58.9% (764) | 61.4% (44) | n/a (0) |
| grok-4.7 | 52.0% (325) | 58.4% (510) | n/a (0) | n/a (0) |
| qwen-3.8-max | 55.2% (353) | 57.0% (468) | 100.0% (1) * | n/a (0) |

n in parentheses. * = N<10 (insufficient) -- no conclusions drawn from these cells. "n/a (0)" = no calls landed in that bucket this week. Sub-50 bucket (n greater than 0 only): grok-4.7 0.0% (n=1) *; qwen-3.8-max 57.1% (n=7) *; deepseek-v4-pro 0.0% (n=1) *. Mid-scale ranking: 5 of the 6 models with N>=10 in both 50-60 and 60-70 out-hit 50-60 from 60-70 this week (issue #11: 0 of 6).

## Hit-rate by forecast horizon (FH)

| Model | 1h | 4h | 1d |
|---|---|---|---|
| claude-fable-5 | 55.4% (486) | 57.3% (239) | 62.8% (43) |
| claude-opus-5.5 | 54.3% (440) | 62.7% (228) | 75.0% (32) |
| deepseek-v4-pro | 55.2% (563) | 58.0% (286) | 61.9% (42) |
| gemini-3.1-pro | 58.4% (570) | 57.4% (258) | 61.9% (42) |
| gpt-6.1-sol | 56.7% (598) | 57.2% (313) | 70.6% (51) |
| grok-4.7 | 54.1% (519) | 55.9% (272) | 75.6% (45) |
| qwen-3.8-max | 52.9% (539) | 61.0% (254) | 75.0% (36) |

n in parentheses; * = N<10 (insufficient). 7 of 7 model lines hit better on 1d than on 1h this window (issue #11: 5 of 7).

## Trading result (a different question)

A different question: is any of this profitable to trade. "Accurate" and "profitable" are not the same thing, and per methodology this section is never mixed into the prediction/calibration metrics above.

| Model | Trades | WR | Net PnL | Gross PnL | Max DD |
|---|---|---|---|---|---|
| qwen-3.8-max | 829 | 45.6% | +$9.86 | +$92.76 | -$29.21 |
| claude-opus-5.5 | 700 | 48.0% | +$7.89 | +$77.89 | -$33.38 |
| deepseek-v4-pro | 891 | 48.6% | +$0.54 | +$89.64 | -$36.55 |
| grok-4.7 | 836 | 45.7% | -$2.57 | +$81.03 | -$43.89 |
| gpt-6.1-sol | 962 | 47.0% | -$13.57 | +$82.63 | -$44.75 |
| claude-fable-5 | 768 | 45.2% | -$21.86 | +$54.94 | -$44.08 |
| gemini-3.1-pro | 870 | 46.3% | -$23.29 | +$63.71 | -$37.55 |

3 of 7 model lines finished net-positive and 7 gross-positive this week (issue #11: 0 of 7 net-positive, 7 gross-positive). Ranked best-to-worst net PnL. Per methodology this table is never combined with the calibration tables above; a week's PnL is regime as much as skill.

## Platform note: TP/SL calibration

> **Context, not a finding.** Since Aug 13 the public sandbox can trade with model-specific TP/SL multipliers learned from this same weekly history (v1); since Aug 19, v2 adds per-ticker and confidence-bucket (50-70 / 70-100) resolution. That feature consumes calibration history; it does not feed back into any table in this report, which measures stated confidence vs. direction-hit only. No effect size is claimed for v2. The toggle lives at marketmania.ai/indices.

## Market check: smaller daily moves on lower volume

> **MARKET CHECK · OBSERVATION (two adjacent calendar weeks)**
>
> OBSERVATION — mean |1d move| 2.05% -> 1.35% (-34% rel), BTC realized vol 34.9% -> 27.5% (ann., hourly); field directional accuracy 46.6% -> 57.0% (+10.4pp), 7 of 7 models improved, sim win-rate up for 7 of 7, field sim PnL -$352.09 -> -$43.00 (adjacent calendar weeks Oct 12-18 vs Oct 19-25; descriptive, one pair of weeks, not a claim).
>
> Robustness: raw price-sign accuracy 44.1% -> 56.0% — the same direction as the trade-based hit rule.

| Measure | Oct 12-18 (week A) | Oct 19-25 (week B) | Change |
|---|---|---|---|
| Mean \|1d move\| (5 assets) | 2.05% | 1.35% | -34% rel |
| BTC realized vol (ann., hourly) | 34.9% | 27.5% | -7.4pp |
| Field directional accuracy | 46.6% | 57.0% | +10.4pp |
| Models improving hit-rate | -- | 7 of 7 | -- |
| Raw price-sign accuracy | 44.1% | 56.0% | +11.9pp |
| Field sim win-rate | 39.2% | 46.6% | +7.5pp |
| Field sim net PnL | -$352.09 | -$43.00 | -- |

Week A is exactly the issue-#11 window, so this slice and every week-over-week column in this report compare the same two back-to-back weeks. Descriptive, one pair of weeks; the hit rule is methodology v1.1 (trade-based) and the raw price-sign row is the robustness check (hourly closes at :00 against slots at :01). BTC realized vol here is the hourly-return series from the alive slice; the snapshot row's figure is the audited daily-candle estimate -- different estimators, both reported as measured. Lineage note: the model cutover (2026-10-17T09:01:00+00:00, inside week A, day 6 of 7) splices three lines -- gpt-5.6-sol -> gpt-6.1-sol, claude-opus-5 -> claude-opus-5.5, grok-4.6 -> grok-4.7. Week A (2026-10-12..2026-10-18, the published issue-#11 window) holds the cutover: its old-id rows are aggregated under the heads (it reproduces the published issue #11); week B (2026-10-19..2026-10-25) is the first full week on the heads (0 old-id row(s) remapped in it). The grok 4.5 -> 4.6 flip (2026-08-24T09:22:00+00:00) sits eight windows back (inside issue #4's window).

## Practical implications

- Stated confidence is still a ranking hint at best, not a probability: the field's gap is +4.5pp this week against +15.4pp in issue #11.
- Gap and regime move together. The field hit 57.0% against 46.6% in issue #11 and stated mean confidence read 61.5 against 61.9, so the gap came out at +4.5pp: a field that hits 57.0% instead of 46.6% closes or opens a confidence gap without a single stated number changing.
- Trading direction this week (3 of 7 lines net-positive) is a regime read, not a strategy result; the three net-positive lines were all net-negative one week earlier (issue #11: 0 of 7 net-positive).

## Limitations

- Prediction metrics (this report) and trading metrics are kept in separate sections per methodology -- they are never combined into a single score.
- 95% Wilson CIs shown are descriptive, not inferential: observations inside one window are dependent, so read them as a range, not a formal coverage guarantee.
- All 5,856 scored calls sit inside one market regime. This window had trend on 1 of 7 days (issue #11: 3 of 7); every comparison with issue #11 is a comparison across regimes as well as across weeks. Twelve week-over-week points cannot separate drift from regime; no durability claim is made.
- Models report confidence at discrete levels, not a continuous scale; the buckets reflect those natural breakpoints. Cells below N=10 are marked insufficient.
- Underlying price series are reconstructed from trade entry prices (median per symbol-slot), not an independent tick feed. Series density and lineage notes (wave 12): (1) rolling-1w forecasts emit daily since Aug 22, 2026 and rolling-1M daily since Aug 23, 2026, so inside this window (Oct 19-25) daily coverage is FULL, as it was in issues #4, #5, #6, #7, #8, #9, #10 and #11: the 1w series has slots on 7 of 7 days and the 1M series on 7 of 7 days; both sit outside this report's FH gate (1h/4h/1d) and appear only in the exclusions counter (1w 480, 1M 478). (2) This window is the first full week on the builds that took over at Oct 17, 2026 09:01 UTC: claude-opus-5.5, gpt-6.1-sol and grok-4.7 carry no rows from the old builds, and all 7 lines carry 1,470 rows each. Week A (the issue-#11 window) is the spliced week -- "claude-opus-5.5 (incl. opus-5-era, Oct 12 00:00-Oct 17 09:01 UTC)" (1,155 rows from the claude-opus-5 era, 315 from claude-opus-5.5); "gpt-6.1-sol (incl. 5.6-era, Oct 12 00:00-Oct 17 09:01 UTC)" (1,155 rows from the gpt-5.6-sol era, 315 from gpt-6.1-sol); "grok-4.7 (incl. 4.6-era, Oct 12 00:00-Oct 17 09:01 UTC)" (1,155 rows from the grok-4.6 era, 315 from grok-4.7) -- so each of the three week-over-week rows joins this week's single-build line to last week's spliced line, and a remapped comparison is not the same model build. (3) No roll-out slot of the flip falls inside this window: the two hourly slots before it (Oct 17, 07:01 and 08:01 UTC) lie in week A, so this window's invalid counter holds none of their rows. (4) Past lineage flips: the Oct 17, 2026 09:01 UTC flip sits one window back (inside the issue-#11 window), the grok 4.5 -> 4.6 flip (Aug 24, 2026 09:22 UTC) sits eight windows back (inside the issue-#4 window), and qwen-3.8-max succeeded qwen-3.7-max on Aug 5, 2026 (fully before this window).
- Market-state row is computed on a single-exchange (binance) daily candle series, as in issue #11 after the audit that found the raw table mixes two exchanges; issue #11 values are as published.
- Research-to-date counter: directional forecasts resolved inside the FH gate (1h/4h/1d) since Jul 11, counted at the issue cutoff (Oct 26 16:00 UTC) -- the definition pinned in issue #3 and carried by issues #4, #5, #6, #7, #8, #9, #10 and #11, which printed 71,945 at the Oct 19 16:00 cutoff. The pack reproduces that pin at the previous cutoff (control OK), so this issue's 77,801 is an additive step under one definition; counter deltas against issue #10 and earlier remain definitional.

## Counters & lineage

| Counter | Value | Counter | Value |
|---|---|---|---|
| Forecasts total | 10,290 | Mature (scored pool) | 9,171 |
| OK in gate | 9,171 | -- of them directional | 5,856 |
| Out of gate (1w / 1M) | 480 / 478 | -- of them sideways | 3,315 |
| Invalid | 161 | Pending (next issue) | 0 |
| Late closes | 0 | Uptime, all grids | 266 / 266 slots |
| Source file | weekly_metrics_2026-10-19.json | Report cutoff | Mon Oct 26, 2026, 16:00 UTC |
| Generated at (pipeline) | 2026-10-27 11:37 UTC | Methodology | v1.1 (2026-08-10) · hash e66c7e8c864a2233 |

**THIS REPORT: 5,856 scored observations.** MARKETMANIA RESEARCH TO DATE (as of Oct 26, 2026 cutoff): 77,801 directional forecasts resolved since Jul 11 · 7 models tracked (current line-up; earlier versions folded into their successors' lineage) · 5 assets · 5 horizons · hourly · 60 published reports.

> **Issue #12.** Weekly Calibration is a living series. Issue #11's open questions -- "the gap moved -2.8pp while the field base moved +2.8pp -- the sixth issue running in which the two move nearly one-for-one in opposite directions; does that hold for a seventh?" and "The best-Brier line changed hands again, to deepseek-v4-pro -- does it hold for a second issue, or change hands again?" -- read this issue as: field gap +15.4pp -> +4.5pp (7 of 7 lines narrowed) on a field base that moved +10.4pp, a seventh issue running in opposite directions, and the best-Brier line changed hands again: claude-opus-5.5 at 0.2443 (issue #11: deepseek-v4-pro at 0.2657). Also on the record: gpt-6.1-sol's 70-80 bucket read 61.4% (n=44) against its own 60-70 at 58.9% (n=764). The claude-opus-5.5, gpt-6.1-sol and grok-4.7 rows are the first full week on the Oct 17 builds; week-over-week rows compare against the spliced issue-#11 week, where those lines carried the labels "claude-opus-5.5 (incl. opus-5-era, Oct 12 00:00-Oct 17 09:01 UTC)", "gpt-6.1-sol (incl. 5.6-era, Oct 12 00:00-Oct 17 09:01 UTC)" and "grok-4.7 (incl. 4.6-era, Oct 12 00:00-Oct 17 09:01 UTC)". The grok 4.5 -> 4.6 flip (Aug 24, 2026 09:22 UTC) sits eight windows back. Engine 1.1 has powered the sandbox since Aug 18, i.e. before this window; no cross-engine PnL comparisons are claimed.

## What we're testing next

- Next issue: the gap moved -10.9pp while the field base moved +10.4pp -- the seventh issue running in which the two move nearly one-for-one in opposite directions; does that hold for an eighth?
- The best-Brier line changed hands again, to claude-opus-5.5 -- does it hold for a second issue, or change hands again?
- Monthly series: is the overconfidence gap stable across market regimes (trend vs flat) at monthly n?

## Related research

| Report | Direct PDF link |
|---|---|
| Consensus Watch #12 | https://marketmania.ai/research/reports/consensus-watch-2026-10-19.pdf |
| Weekly Model Watch #12 | https://marketmania.ai/research/reports/model-watch-2026-10-19.pdf |
| Weekly Calibration #11 | https://marketmania.ai/research/reports/weekly-calibration-2026-10-12.pdf |

The three weekly reports publish together as one issue each week; each links straight to the others' PDF and to its own previous issue. Direct links are the posting rule from wave 2 on.

## Cite this report

```bibtex
@misc{mm_calibration_2026w43,
  title  = {Weekly Calibration #12: confidence vs. direction-hit, Oct 19-25 2026},
  author = {{MarketMania Research}},
  year   = {2026}, month = {October}, day = {28},
  url    = {https://marketmania.ai/research/reports/weekly-calibration-2026-10-19.pdf},
  note   = {Methodology v1.1, hash e66c7e8c864a2233; source weekly_metrics_2026-10-19.json}
}
```
