# Weekly Model Watch #1

**WEEKLY · MODEL WATCH** · August 10, 2026 · MarketMania Research · Weekly series

Window: **Aug 3-9, 2026 UTC** · Cutoff: **Mon Aug 10, 2026, 16:00 UTC** · Methodology **v1.1 (2026-08-10)**, hash `e66c7e8c864a2233`

> Research question: *"who led the week, which tickers were readable, who saw the turn first, and does a model agreeing with itself across timeframes/horizons make its call stronger?"*

## Research Snapshot

| RESEARCH SNAPSHOT |  |  |  |
|---|---|---|---|
| window **Aug 3-9, 2026 UTC** | **8** models (7 current + qwen-3.7-max legacy to Aug 5) | **4,042** directional calls scored of **9,121** mature | FH gate **1h / 4h / 1d** |
| hit = direction rule: tp1/tp2->hit, sl->miss, expiry->sign of gross pnl | base field hit **42.4%** | cutoff **Mon Aug 10, 2026, 16:00 UTC** | methodology **v1.1 (2026-08-10)**, hash **e66c7e8c864a2233** |

## Key finding

> **KEY FINDING · OBSERVATION (one weekly window)**
>
> ### The week's only trend reversal went uncalled: zero of 8 models caught BTC's +2.1% Monday turn.

## TL;DR

- OBSERVATION -- **No weekly title awarded.** The top four models sit within ~1.9pp of hit-rate (qwen-3.8-max 44.1%, gemini-3.1-pro 43.7%, gpt-5.6-sol 43.0%, qwen-3.7-max 42.2%) with overlapping 95% CIs -- the title rule (>=5pp gap AND non-overlapping 95% CI) was not met. qwen-3.8-max leads descriptively.
- **Ticker of the week: SOL; hardest: ETH.** SOL was the field's most readable coin (48.2%, n=852) and ETH the hardest (34.8%, n=724) -- the widest gap in the table, and the one split where 95% CIs land clean of each other (44.9-51.6% vs. 31.4-38.4%); every adjacent pair in the ranking overlaps its neighbor. Pair of the week: qwen-3.8-max x SOL 52.9% (n=104) -- the same model also owns the worst pair (x ETH, 28.4%, n=88).
- **Self-agreement added nothing this week.** When a model's two timeframes (or neighboring horizons) agreed on direction, hit-rates ran 1.9-4.5pp BELOW the weekly base; the rare intra-model disagreements did far better (71.4% / 70.6%) but on just 7 and 17 calls -- descriptive only, not a strategy.
- **A one-regime week.** All 7 days were flat (max |BTC daily move| 0.98%, methodology flat/trend threshold 1.5%); the field skewed long over short (average of each model's own long:short ratio, ~2.2:1) in a sideways-heavy tape (55.7% of mature calls); its single most confident miss was a conf-85 BTC long (qwen-3.7-max, Aug 4).

### Weekly leaderboard chart (see PDF for the bar chart with CI whiskers)

| Model | Hit rate | 95% CI |
|---|---|---|
| qwen-3.8-max | 44.1% | 39.5%-48.7% |
| gemini-3.1-pro | 43.7% | 39.9%-47.6% |
| gpt-5.6-sol | 43.0% | 39.1%-46.9% |
| qwen-3.7-max *(legacy)* | 42.2% | 36.4%-48.2% |
| grok-4.5 | 42.2% | 38.3%-46.1% |
| deepseek-v4-pro | 41.6% | 36.8%-46.7% |
| claude-fable-5 | 41.4% | 37.7%-45.3% |
| claude-opus-5 | 40.9% | 36.5%-45.4% |

Dashed line in the PDF chart = field base (42.4%).

## Why it matters

MarketMania scores 8 models against the same market, hour after hour. This first issue of Weekly Model Watch asks four practical questions about that week of calls: Who led the week, which tickers were readable, who saw the turn first, and does a model agreeing with itself across timeframes/horizons make its call stronger? Each question gets its own block below, with its own table, its own N, and its own honesty caveats -- this is the fullest report in the weekly series (8 analysis blocks); later issues compress as the format settles.

## Leaderboard of the week

The full 8-model field, ranked by this week's directional hit-rate.

| Model | N | Coverage | Hit rate | 95% CI |
|---|---|---|---|---|
| **qwen-3.8-max** | 447 | 53.9% | 44.1% | 39.5%-48.7% |
| gemini-3.1-pro | 629 | 48.3% | 43.7% | 39.9%-47.6% |
| gpt-5.6-sol | 612 | 46.9% | 43.0% | 39.1%-46.9% |
| **qwen-3.7-max**<br>_legacy, to Aug 5_ | 263 | 56.0% | 42.2% | 36.4%-48.2% |
| grok-4.5 | 607 | 46.7% | 42.2% | 38.3%-46.1% |
| deepseek-v4-pro | 377 | 28.9% | 41.6% | 36.8%-46.7% |
| claude-fable-5 | 642 | 49.2% | 41.4% | 37.7%-45.3% |
| claude-opus-5 | 465 | 35.6% | 40.9% | 36.5%-45.4% |

No weekly title is awarded this issue. Title rule: **>=5pp gap AND non-overlapping 95% CI, N>=10** (N>=10 per side). The top four sit within ~1.9pp of hit-rate (44.1% down to 42.2%) and every adjacent pair of 95% CIs overlaps, so neither condition is met -- qwen-3.8-max leads descriptively at 44.1% (n=447), not by title. Movement vs. prior week: first issue -- starts next week (issue #2). Hit-rate rank and Brier (calibration) rank diverge sharply for several models this week (e.g. claude-fable-5 ranks 7th of 8 by hit-rate but 2nd by Brier) -- see the calibration bridge in Counters &amp; lineage, and the full picture in Weekly Calibration #1.

## Ticker of the week

This week's 5-ticker field, ranked by directional hit-rate.

| Symbol | N | Hit rate | 95% CI |
|---|---|---|---|
| **SOL** | 852 | 48.2% | 44.9%-51.6% |
| XRP | 800 | 46.5% | 43.1%-50.0% |
| BNB | 782 | 43.6% | 40.2%-47.1% |
| BTC | 884 | 38.4% | 35.2%-41.6% |
| **ETH** | 724 | 34.8% | 31.4%-38.4% |

SOL was the field's most readable ticker this week (48.2%, n=852) and ETH the hardest (34.8%, n=724). Their 95% CIs are well separated (44.9-51.6% vs. 31.4-38.4%) -- the widest gap in the table -- while every adjacent pair in the ranked list (SOL-XRP, XRP-BNB, BNB-BTC, BTC-ETH) overlaps its neighbor. Read this as a one-week readability signal, not a durable per-asset skill claim.

## Pair of the week

Top 6 of the 12 model x ticker pairs tracked this week (best hit-rate), and the 5 worst. Pairs shown require N>=8 calls.

**Top 6 pairs**

| Model | Symbol | N | Hit rate |
|---|---|---|---|
| **qwen-3.8-max** | SOL | 104 | 52.9% |
| **qwen-3.8-max** | XRP | 96 | 52.1% |
| deepseek-v4-pro | SOL | 80 | 50.0% |
| gpt-5.6-sol | XRP | 119 | 49.6% |
| claude-opus-5 | SOL | 110 | 49.1% |
| gemini-3.1-pro | XRP | 141 | 48.9% |

**5 worst pairs**

| Model | Symbol | N | Hit rate |
|---|---|---|---|
| gpt-5.6-sol | ETH | 113 | 34.5% |
| gemini-3.1-pro | ETH | 108 | 34.3% |
| deepseek-v4-pro | BTC | 81 | 33.3% |
| claude-opus-5 | ETH | 73 | 32.9% |
| **qwen-3.8-max** | ETH | 88 | 28.4% |

qwen-3.8-max owns both extremes this week: its best individual pairing, x SOL (52.9%, n=104), is also the week's best pair overall, and its worst pairing, x ETH (28.4%, n=88), is also the week's worst. Pair history accumulates across issues -- read this week's top/bottom-6 as a first data point, not a ranking.

## Who saw the reversal first

Reversal definition: 4h grid; trend = sign of prior 24h; counter-move >=2.0% (BTC/ETH) sustained 8h. A model 'calls' the reversal if it has a matured directional hit call in the new direction, FH 4h or 1d, in the slot window [T-12h, T+FH].

| Symbol | Time (UTC) | New side | Move (8h) | Callers | First caller |
|---|---|---|---|---|---|
| **BTC** | Mon Aug 3, 08:00 | Long | +2.10% | 0 | **NOBODY** |

This week's only qualifying reversal was BTC's turn to long at 08:00 UTC on Monday, Aug 3 -- an 8-hour move of +2.10%. Zero of the 8 active models had a matured hit call in the new (long) direction inside the [T-12h, T+FH] window; first caller: **NOBODY**. This is a single event -- the reversal detector's first live week -- so it is reported descriptively, not as a claim about any model's turn-detection skill. Scope in v1 is BTC/ETH only; expansion to more assets is planned after 2-3 issues.

## Cross-TF confirmation: does agreeing with yourself help?

Within one forecast horizon (FH), does a model's call from a longer input timeframe (TF) agree with its call from a shorter one -- and if so, does the call do any better?

| FH | vs. | Agree n | Agree hit | Dis n | Dis hit | Mixed | Base hit | Lift pp |
|---|---|---|---|---|---|---|---|---|
| 4h | TF 4h vs 1h | 441 | 39.7% [35.2%-44.3%] | 7 | 71.4% [35.9%-91.8%] \* | 208 | 41.7% (n=665) | -2.0 |
| 1d | TF 1d vs 4h | 53 | 34.0% [22.7%-47.4%] | 0 | n/a (0) | 25 | 38.5% (n=78) | -4.5 |

n in Base hit column is the base-rate call count. `\*` = N<10 (insufficient); `#` = N<20 (thin); "n/a (0)" = no disagreeing calls. Insufficient/thin cells are never used to rank models.

**Regime control (degenerate this issue).** Methodology calls for a trend-vs-flat split on this table (by_regime). All 7 days this week were classified flat: 08-03 +0.09% · 08-04 +0.93% · 08-05 +0.98% · 08-06 -0.57% · 08-07 +0.88% · 08-08 +0.16% · 08-09 +0.09% (max |move| 0.98%). With zero trend days, the split is degenerate this issue: the flat bucket shown above IS the whole sample, and the trend bucket is n=0 throughout. We report the numbers as-is rather than fabricate a split; this becomes informative in a mixed (trend + flat) week.

## Cross-FH confirmation: does a shorter horizon confirm a longer one?

Does a model's call at one forecast horizon (FH) agree with its call at a shorter FH on the same timeframe -- and if so, does the call do any better?

| FH | vs. | Agree n | Agree hit | Dis n | Dis hit | Mixed | Base hit | Lift pp |
|---|---|---|---|---|---|---|---|---|
| 4h | by FH 1h | 410 | 39.8% [35.1%-44.6%] | 17 | 70.6% [46.9%-86.7%] # | 238 | 41.7% (n=665) | -1.9 |
| 1d | by FH 4h | 44 | 36.4% [23.8%-51.1%] | 1 | 0.0% [0.0%-79.4%] \* | 33 | 38.5% (n=78) | -2.1 |

The 1d cells are thin across both this block's fh_1d_confirmed_by_4h (44 agree / 1 disagree) and Block 5's fh_1d_tf_1d_vs_4h (53 agree / 0 disagree) -- 44-53 agree, 0-1 disagree. Disagree cells with N<10 (both 1d disagree cells here and in Block 5, plus Block 5's 4h TF-pair disagree cell, n=7) are marked insufficient and are never used to rank models. The 4h FH-confirmation disagree cell below (n=17) clears the N>=10 floor but is still thin -- read descriptively only.

## Side-mix and the week's failure

How each model split its calls this week across long / short / sideways, and the week's single costliest miss.

| Model | N | Long | Short | Sideways | L:S |
|---|---|---|---|---|---|
| claude-fable-5 | 1,305 | 37.9% | 11.3% | 50.8% | 3.34 |
| claude-opus-5 | 1,305 | 28.1% | 7.5% | 64.4% | 3.74 |
| deepseek-v4-pro | 1,305 | 16.9% | 12.0% | 71.1% | 1.40 |
| gemini-3.1-pro | 1,302 | 31.7% | 16.6% | 51.7% | 1.91 |
| gpt-5.6-sol | 1,305 | 30.5% | 16.4% | 53.1% | 1.86 |
| grok-4.5 | 1,300 | 26.9% | 19.9% | 53.3% | 1.35 |
| **qwen-3.7-max**<br>_legacy, to Aug 5_ | 470 | 29.6% | 26.4% | 44.0% | 1.12 |
| qwen-3.8-max | 829 | 39.3% | 14.6% | 46.1% | 2.69 |

L:S = each model's own long-share divided by its short-share. Averaging that ratio across the 8 models (not pooling raw counts) gives the field's ~2.2:1 long-over-short skew quoted in the TL;DR; the field also called 'sideways' more than half the time overall (55.7% of mature forecasts, 5,079 of 9,121).

**Consensus skew.** 471 symbol/FH/TF cells this week had 5 or more of the (up to 7) active models on the same side. Those skewed cells' mature calls hit 41.7% (n=2,901) -- below the week's 42.4% field base rate, so leaning with a 5-of-7+ skew did not outperform the field this week either. (A stricter 6-of-6+ unanimity threshold is Consensus Watch's own metric -- see that report for the full herding analysis.)

| Model | Symbol | Side | Conf | FH | Slot (UTC) | Exit | Net PnL |
|---|---|---|---|---|---|---|---|
| **qwen-3.7-max** | BTC | Long | 85 | 1h | Aug 4, 19:01 | Expiry | **-$0.28** |

**Fail of the week.** The week's single highest-confidence individual miss: qwen-3.7-max, BTC, long, confidence 85, FH 1h, slot Aug 4 19:01 UTC, exit by expiry, net -$0.28. Cross-reference: that same slot was a 7-of-7 unanimous long across the active field that day (see Consensus Watch #1) -- the whole field, not just this one model, was wrong-footed.

## Counters and the calibration bridge

Gate totals, uptime and data-quality for this window, plus a bridge to this week's calibration read.

| Metric | Value |
|---|---|
| Forecasts total | 9,194 |
| OK in gate | 9,121 |
| Invalid | 3 |
| Out of gate (1w) | 70 |
| Mature (scored pool) | 9,121 |
|   - directional | 4,042 |
|   - sideways | 5,079 |
| Pending (next issue) | 0 |
| Late closes | 0 |
| Uptime, 1h slots (tf=1h) | 165 / 168 |
| Uptime, 4h slots (tf=4h) | 41 / 42 |
| Uptime, 4h slots (tf=1h) | 41 / 42 |
| Uptime, 1d slots (tf=1d) | 7 / 7 |
| Uptime, 1d slots (tf=4h) | 7 / 7 |
| Source file | weekly_metrics_2026-08-03.json |
| Generated at (metrics pipeline) | 2026-08-11 14:24 UTC |
| Methodology | v1.1 (2026-08-10), hash e66c7e8c864a2233 |
| Report cutoff | 2026-08-10 16:00 UTC (frozen) |

**Data quality.** Per-model ok-rate (share of a model's forecasts that passed gate validation) was at or near 100% for every model this week: gemini-3.1-pro 99.85%, qwen-3.8-max 99.88%, and 100.0% for the other six (claude-fable-5, claude-opus-5, deepseek-v4-pro, gpt-5.6-sol, grok-4.5, qwen-3.7-max).

**Calibration bridge.** qwen-3.8-max was also this week's best-calibrated model by Brier score (0.2713, lowest of the 8-model field) -- the same model leading this issue's hit-rate leaderboard (Block 1), though for most other models Brier rank and hit-rate rank point in different directions (e.g. claude-fable-5 ranks 7th of 8 by hit-rate but 2nd by Brier). Full confidence-bucket and by-horizon calibration analysis lives in Weekly Calibration #1.

## Practical implications

- Do not read this week's leaderboard as a skill ranking. The top four models overlap in both point estimate (within ~1.9pp) and 95% CI, and the title rule was not met -- qwen-3.8-max's lead is descriptive, for this window only.
- Treat SOL as this week's most-readable ticker and ETH as the hardest with real caution: it is one week, and the field also missed its only reversal outright (0 of 8 models), which points to a general weak spot at turns rather than an ETH-specific one.
- Do not use within-model cross-TF/cross-FH agreement as a confidence booster yet. In every tested cell, agreement underperformed the base rate this week (by 1.9-4.5pp); the small number of disagreements did better but on too few calls (7, 17) to act on.

## Limitations

- Single week (Aug 3-9, 2026 UTC). Every finding in this issue is descriptive for this window only; no claim is made about next week or about any model's underlying skill.
- This week sat inside a single market regime: all 7 days were classified flat (max |BTC daily move| 0.98%). The trend-vs-flat regime control called for by methodology (Block 5) could not be exercised this issue -- there were no trend days to compare against.
- Observations inside one window are not independent: the same models watch overlapping symbol/FH/TF cells hour after hour, so agree/disagree calls made close in time are correlated, not i.i.d. draws.
- 95% Wilson CIs shown throughout are descriptive, not inferential, for the reason above -- read them as a range, not a formal coverage guarantee.
- The reversal detector (Block 4) is v1 and BTC/ETH-only. Its underlying price series is reconstructed from trade entry prices (median per symbol-slot), not an independent tick feed, per this week's metrics notes. One event this week is not enough to characterize detector performance either way.
- Cross-TF and cross-FH disagreement cells are tiny this week (n=0, 1, 7, 17) -- too small to rank models on; cells below N=10 are marked insufficient and are never used to rank models.
- This is issue #1 of Weekly Model Watch -- there is no prior issue to compare against, so no week-over-week deltas are reported (Block 1's 'movement vs prior week' begins with issue #2).

> **THIS REPORT: 4,042 scored observations**
>
> MARKETMANIA RESEARCH TO DATE (as of Aug 10, 2026 cutoff): 14,151 resolved forecasts since Jul 11, 2026 · 9 models tracked (7 current frontier + 2 archived legacy generations) · 5 assets · 5 forecast horizons · hourly cadence · 4 published research reports

**Snapshot principle.** Every figure in this report is frozen at the Monday 16:00 UTC cutoff for the Aug 3-9 window. Any later data correction is handled as a note in a future issue, not by silently editing this one.

## What we're testing next

- **Next issue:** does the reversal blind spot repeat, and does cross-TF agreement stay flat-to-negative?
- **Monthly test:** cross-TF/FH lifts with regime control across mixed weeks (trend vs. flat), block bootstrap.

## Related research

- Stability Index Run 2 — marketmania.ai/research/reports/si-run-2.pdf
- Weekly Calibration #1 — marketmania.ai/research/reports/weekly-calibration-2026-08-03.pdf
- Consensus Watch #1 — marketmania.ai/research/reports/consensus-watch-2026-08-03.pdf

*These three weekly reports -- Consensus Watch, Weekly Calibration and Weekly Model Watch -- publish together as one wave each week; links are stable.*

## Cite this report

```bibtex
@misc{mm_model_watch_2026w32,
  title  = {Weekly Model Watch #1: weekly leaderboard, ticker/pair reads and cross-confirmation, Aug 3-9 2026},
  author = {{MarketMania Research}},
  year   = {2026}, month = {August}, day = {10},
  url    = {https://marketmania.ai/research},
  note   = {Methodology v1.1 (2026-08-10), hash e66c7e8c864a2233; window Aug 3-9, 2026 UTC; source weekly_metrics_2026-08-03.json}
}
```