# Quarterly Calibration #1

**QUARTERLY · CALIBRATION** · October 8, 2026 · MarketMania Research · Quarterly series

Window: **Jul 15-Sep 30, 2026 UTC** · Cutoff: **Thu Oct 1, 2026, 16:00 UTC** · Methodology **v1.1 (2026-08-10)**, hash `e66c7e8c864a2233`

PDF: https://marketmania.ai/research/reports/calibration-quarterly-2026-q3.pdf · Open data (JSON): https://marketmania.ai/research/reports/calibration-quarterly-2026-q3.json

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

## Research Snapshot

| RESEARCH SNAPSHOT |  |  |  |
|---|---|---|---|
| **59,606** directional calls scored | **7** models (lineage spliced: opus, grok, qwen) | of **122,770** mature forecasts | window **Q3 2026 · Jul 15-Sep 30 · partial** |
| hit = direction rule: tp1/tp2->hit, sl->miss, expiry->sign of gross pnl | FH gate **1h / 4h / 1d** | base field hit **45.6%** vs **62.6** stated | cutoff **Thu Oct 1, 2026, 16:00 UTC** |
| market: BTC net **+28.56%** (prior n/a) | ann vol **38.9%** (was n/a) | TOP5 volume **$173.94B**, n/a q/q | pairwise corr **0.77** (was n/a) |

## Key finding

> **KEY FINDING · OBSERVATION (one quarterly window)**
>
> ### Over the quarter the field's overconfidence gap read +17.0pp against 62.6 stated mean confidence, the monthly gaps ran +16.3pp to +17.8pp inside it, and the field slope across the three months is -0.40 pp/month (flat).

MONTHLY EVIDENCE -> QUARTERLY VERDICT (rule below)

| Metric | M1 Jul 15-31 | M2 Aug 1-31 | M3 Sep 1-30 | Quarter | Verdict |
|---|---|---|---|---|---|
| Field gap (conf - hit), pp | +17.8 | +16.3 | +17.0 | +17.0 | Confirmed (3/3) |
| Lowest-Brier model | grok-4.6 | opus-5 | qwen-3.8 | grok-4.6 | Rejected (1/3) |
| Widest-gap model | gpt-5.6 | gpt-5.6 | gpt-5.6 | gpt-5.6 | Confirmed (3/3) |

Verdict rule: each row states one reading of the quarter; a month agrees when it shows the same reading (same sign, or the same name). Confirmed = 2 or 3 of the 3 months agree with the quarter; Rejected = 0 or 1. M1 holds Jul 15-31 only: the platform's first forecast slot is Jul 15 18:01 UTC. Monthly cells come from the frozen monthly packs (7-line roster, lineage spliced); the quarter is the pooled Jul 15-Sep 30 pull.

## TL;DR

- **OBSERVATION -- The quarterly gap read +17.0pp.** Directional calls hit 45.6% against 62.6 stated mean confidence over 59,606 calls. Field Brier is 0.2803 and 0 of 7 model lines scored below 0.25, the score of an uninformative always-50% predictor. One quarter, descriptive.
- **The field gap slope across the three months is -0.40 pp/month (flat).** Across those months the field gap ran +17.8pp -> +16.3pp -> +17.0pp and the per-model slopes split 3 narrowing / 2 widening / 2 flat under the pack's own rule (|slope| < 0.5 pp/month = flat). grok-4.6 (spliced) is the lowest-Brier line (Brier 0.2716); gemini-3.1-pro leads on hit-rate (47.2%).
- **The widest monthly swing belongs to qwen-3.8-max (spliced).** Its monthly gap moved over a 4.0 pp range inside one quarter, against a field range of 1.5 pp. There is no prior quarter; by month the per-model gaps ran +14.2pp to +22.1pp (Jul 15-31), +13.2pp to +21.8pp (Aug 1-31), +14.7pp to +23.5pp (Sep 1-30).
- **Trading, kept separate from every calibration table.** 0 of 7 lines finished net-positive over the quarter; net PnL ran -$835.51 (gemini-3.1-pro, best) to -$1141.58 (gpt-5.6-sol, worst), field -$6703.95.

### Quarter-over-quarter chart (see PDF for the grouped bar chart)

Overconfidence gap in percentage points (mean stated confidence minus hit-rate) over the whole quarter, sorted by the quarter's gap. Short names: opus-5 = claude-opus-5, fable-5 = claude-fable-5, gemini-3.1 = gemini-3.1-pro, gpt-5.6 = gpt-5.6-sol, qwen-3.8 = qwen-3.8-max, deepseek = deepseek-v4-pro. Field gap +17.0pp on field mean stated confidence 62.6; by month the field gap read +17.8pp / +16.3pp / +17.0pp (Jul 15-31 / Aug 1-31 / Sep 1-30). First quarterly issue: there is no prior quarter (platform data begin Jul 15, 2026), so the chart carries no gray series and every prior cell in this report reads n/a; a month inside the quarter is not a prior window.

## 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. The monthly series carries one point per model per issue. A calendar quarter carries three monthly points plus a pooled quarterly figure (this first issue has no prior-quarter baseline), which is what the slope table and the month-by-month confidence buckets below are computed from. Descriptive for this window.

## 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. 3 model lines are lineage splices this quarter (claude-opus-5 (incl. 4.8-era, Jul 15-30); grok-4.6 (incl. 4.5-era, Jul 15-Aug 24); qwen-3.8-max (incl. 3.7-era, Jul 15-Aug 5)). The quarterly rows are one pull at the quarterly cutoff (Thu Oct 1, 2026, 16:00 UTC); the monthly rows are the frozen monthly packs at their own cutoffs, so the two do not have to agree to the last decimal.

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

| Model | N | Cov. | Hit rate | Mean conf | Gap pp | Brier | 95% CI |
|---|---|---|---|---|---|---|---|
| grok-4.6 (spliced) | 7,244 | 45.9% | 45.0% | 60.0 | +15.0 | 0.2716 | 43.9%-46.2% |
| claude-opus-5 (spliced) | 8,261 | 46.3% | 45.9% | 60.9 | +15.0 | 0.2720 | 44.8%-46.9% |
| claude-fable-5 | 8,447 | 47.3% | 45.5% | 60.7 | +15.1 | 0.2722 | 44.5%-46.6% |
| deepseek-v4-pro | 7,799 | 43.7% | 45.3% | 60.9 | +15.6 | 0.2754 | 44.2%-46.4% |
| qwen-3.8-max (spliced) | 9,537 | 53.7% | 45.4% | 61.5 | +16.1 | 0.2784 | 44.4%-46.4% |
| gemini-3.1-pro | 9,130 | 51.4% | 47.2% | 66.0 | +18.9 | 0.2874 | 46.1%-48.2% |
| gpt-5.6-sol | 9,188 | 51.4% | 44.9% | 67.4 | +22.5 | 0.3014 | 43.9%-45.9% |
| **Field (all models)** | **59,606** | - | **45.6%** | **62.6** | **+17.0** | **0.2803** | - |

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. There are no prior gaps to compare against and the page-2 chart has no gray series because this is the first quarterly issue: there is no prior quarter (platform data begin Jul 15, 2026). claude-opus-5 (spliced) = claude-opus-5 (incl. 4.8-era, Jul 15-30). grok-4.6 (spliced) = grok-4.6 (incl. 4.5-era, Jul 15-Aug 24). qwen-3.8-max (spliced) = qwen-3.8-max (incl. 3.7-era, Jul 15-Aug 5).

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

| Model | 50-60 | 60-70 | 70-80 | 80-100 |
|---|---|---|---|---|
| claude-fable-5 | 45.6% (2,742) | 45.5% (5,703) | 50.0% (2) * | n/a (0) |
| claude-opus-5 (spliced) | 47.1% (2,139) | 45.5% (6,108) | 16.7% (12) | n/a (0) |
| deepseek-v4-pro | 46.0% (2,552) | 45.0% (4,743) | 43.5% (432) | 53.3% (15) |
| gemini-3.1-pro | 35.4% (65) | 47.2% (6,279) | 47.0% (2,727) | 56.9% (58) |
| gpt-5.6-sol | 43.4% (106) | 45.9% (6,537) | 42.3% (2,530) | 40.0% (15) |
| grok-4.6 (spliced) | 45.8% (3,561) | 44.3% (3,650) | 42.3% (26) | n/a (0) |
| qwen-3.8-max (spliced) | 46.6% (3,077) | 44.9% (5,120) | 44.3% (1,006) | 50.0% (46) |

n in parentheses. * = N<10 (insufficient) -- no conclusions drawn from these cells. "n/a (0)" = no calls landed in that bucket this quarter. Sub-50 bucket (n greater than 0 only): grok-4.6 (spliced) 28.6% (n=7) *; claude-opus-5 (spliced) 50.0% (n=2) *; deepseek-v4-pro 54.4% (n=57); qwen-3.8-max (spliced) 43.4% (n=288); gemini-3.1-pro 100.0% (n=1) *. Mid-scale ranking: 2 of the 7 models with N>=10 in both 50-60 and 60-70 out-hit 50-60 from 60-70 this quarter.

## Hit-rate by forecast horizon (FH)

| Model | 1h | 4h | 1d |
|---|---|---|---|
| claude-fable-5 | 46.8% (5,347) | 44.0% (2,603) | 39.8% (497) |
| claude-opus-5 (spliced) | 47.0% (5,085) | 44.8% (2,666) | 39.6% (510) |
| deepseek-v4-pro | 46.4% (4,875) | 44.5% (2,482) | 37.8% (442) |
| gemini-3.1-pro | 48.1% (5,832) | 46.0% (2,774) | 42.9% (524) |
| gpt-5.6-sol | 45.8% (5,744) | 44.2% (2,931) | 38.4% (513) |
| grok-4.6 (spliced) | 46.2% (4,579) | 44.0% (2,292) | 35.9% (373) |
| qwen-3.8-max (spliced) | 46.2% (6,018) | 44.7% (2,978) | 39.2% (541) |

n in parentheses; * = N<10 (insufficient). 0 of 7 model lines hit better on 1d than on 1h this quarter.

## 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 |
|---|---|---|---|---|---|
| gemini-3.1-pro | 9,130 | 36.3% | -$835.51 | +$77.49 | -$835.70 |
| claude-fable-5 | 8,447 | 35.6% | -$888.90 | -$44.20 | -$888.90 |
| deepseek-v4-pro | 7,799 | 36.5% | -$910.72 | -$130.82 | -$913.30 |
| claude-opus-5 (spliced) | 8,261 | 35.4% | -$911.08 | -$84.98 | -$911.08 |
| grok-4.6 (spliced) | 7,244 | 34.1% | -$929.15 | -$204.75 | -$929.15 |
| qwen-3.8-max (spliced) | 9,537 | 35.0% | -$1087.02 | -$133.32 | -$1087.02 |
| gpt-5.6-sol | 9,188 | 34.8% | -$1141.58 | -$222.78 | -$1141.65 |

0 of 7 model lines finished net-positive and 1 gross-positive over the quarter; by month 0 / 0 / 0 lines finished net-positive and the field net PnL ran -$3095.02 / -$1054.79 / -$2554.15 (Jul 15-31 / Aug 1-31 / Sep 1-30). Ranked best-to-worst net PnL. Per methodology this table is never combined with the calibration tables above. The pack labels one of the three months calm and two volatile. The prior-quarter comparison reads n/a because this is the first quarterly issue: there is no prior quarter (platform data begin Jul 15, 2026).

## 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 history (v1); since Aug 19, v2 adds per-ticker and confidence-bucket (50-70 / 70-100) resolution. Both switches landed INSIDE this window, so the quarter spans three feature states. 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 v1 or v2. The toggle lives at marketmania.ai/indices.

## Overconfidence drift, month by month

The three months of the quarter are the windows of the published Monthly #1 (its prior window and its month) and Monthly #2, and their monthly gaps read +17.8pp -> +16.3pp -> +17.0pp. Slope is an ordinary least-squares fit over months 1..3; the pack calls a line "flat" when |slope| is under 0.5 pp/month. 3 of 7 lines read narrowing, 2 widening and 2 flat, and the field line reads flat at -0.40 pp/month.

| Model | M1 gap | M2 gap | M3 gap | Quarter gap | Slope pp/mo | Trend |
|---|---|---|---|---|---|---|
| claude-opus-5 (spliced) | +15.6 | +13.2 | +15.9 | +15.0 | +0.15 | flat |
| grok-4.6 (spliced) | +14.2 | +15.1 | +15.8 | +15.0 | +0.80 | widening |
| claude-fable-5 | +15.9 | +13.9 | +15.6 | +15.1 | -0.15 | flat |
| deepseek-v4-pro | +16.6 | +15.4 | +15.3 | +15.6 | -0.65 | narrowing |
| qwen-3.8-max (spliced) | +18.7 | +14.7 | +14.7 | +16.1 | -2.00 | narrowing |
| gemini-3.1-pro | +20.0 | +18.8 | +18.0 | +18.9 | -1.00 | narrowing |
| gpt-5.6-sol | +22.1 | +21.8 | +23.5 | +22.5 | +0.70 | widening |
| **Field (all models)** | **+17.8** | **+16.3** | **+17.0** | **+17.0** | **-0.40** | **flat** |

Months are M1 Jul 15-31, M2 Aug 1-31, M3 Sep 1-30. Slope = OLS over months 1..3, in pp per month; "flat" means |slope| < 0.5 pp/month. Source: the frozen monthly packs (7-line roster, lineage spliced), i.e. the windows of the published Monthly #1 (its prior window and its month) and Monthly #2 at their own cutoffs. The Quarter gap is the pooled quarter, not the mean of the three months; the prior-quarter field gap reads n/a (first quarterly issue). claude-opus-5 (spliced) = claude-opus-5 (incl. 4.8-era, Jul 15-30). grok-4.6 (spliced) = grok-4.6 (incl. 4.5-era, Jul 15-Aug 24). qwen-3.8-max (spliced) = qwen-3.8-max (incl. 3.7-era, Jul 15-Aug 5).

| Bucket | M1 | M2 | M3 |
|---|---|---|---|
| 50-60 | 46.5% vs 56.8 (4,374) | 46.1% vs 57.0 (4,952) | 45.7% vs 56.9 (4,916) |
| 60-70 | 44.5% vs 63.3 (12,588) | 46.9% vs 63.0 (12,120) | 45.5% vs 62.9 (13,432) |
| 70-80 | 44.4% vs 73.0 (2,262) | 43.6% vs 72.9 (2,213) | 45.5% vs 72.9 (2,260) |
| 80-100 | 58.6% vs 81.8 (58) | 50.0% vs 81.0 (42) | 44.1% vs 80.5 (34) |

Field level: every model's calls pooled into the stated-confidence bucket. Cells are hit rate vs mean stated confidence, with n in parentheses; * = N<10 (insufficient), "n/a (0)" = no call landed in that bucket that month. Source: the quarter pull split by slot month at the quarterly cutoff, so the three columns are one consistent pool.

## Market check: a calm July, then two volatile months

> **MARKET CHECK · OBSERVATION (three months of the quarter)**
>
> OBSERVATION -- mean |1d move| 1.54% / 1.98% / 1.91% by month (1.81% over the quarter), BTC realized vol 33.5% / 36.9% / 34.9% (35.1%; ann., hourly); field directional accuracy 45.0% / 46.3% / 45.5% (45.6%), sim win-rate 31.9% / 35.8% / 38.4% (35.4%), field sim PnL -$3095.02 / -$1054.79 / -$2554.15 (-$6703.95) (Jul 15-31 / Aug 1-31 / Sep 1-30, quarter in parentheses; descriptive, three windows, not a claim).
>
> Robustness: raw price-sign accuracy 46.8% / 46.0% / 48.4% by month (47.1% over the quarter) against 45.0% / 46.3% / 45.5% (45.6%) under the trade-based hit rule; month to month the two moved in the same direction in 0 of 2 steps.

| Measure | Jul 15-31 | Aug 1-31 | Sep 1-30 | Quarter |
|---|---|---|---|---|
| Mean \|1d move\| (5 assets) | 1.54% | 1.98% | 1.91% | 1.81% |
| BTC realized vol (ann., hourly) | 33.5% | 36.9% | 34.9% | 35.1% |
| Field directional accuracy | 45.0% | 46.3% | 45.5% | 45.6% |
| Models improving hit-rate | -- | 5 of 7 | 1 of 7 | -- |
| Raw price-sign accuracy | 46.8% | 46.0% | 48.4% | 47.1% |
| Field sim win-rate | 31.9% | 35.8% | 38.4% | 35.4% |
| Field sim net PnL | -$3095.02 | -$1054.79 | -$2554.15 | -$6703.95 |

There is no prior quarter, so this slice prints the three months and the quarter side by side instead of a prior-against-current pair; each month is its own pack's window at its own frozen cutoff and the Quarter column is the quarter pull. Descriptive, three windows; 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). "Models improving" compares a month with the month before it (Aug 1-31 against Jul 15-31; Sep 1-30 against Aug 1-31). 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 splices in this window: claude-opus-5 (incl. 4.8-era, Jul 15-30); grok-4.6 (incl. 4.5-era, Jul 15-Aug 24); qwen-3.8-max (incl. 3.7-era, Jul 15-Aug 5). 24,922 rows were remapped inside the quarter (19,532 / 5,390 / 0 in Jul 15-31 / Aug 1-31 / Sep 1-30).

## Practical implications

- Pooled over the quarter the field's gap is +17.0pp on 59,606 scored calls: stated confidence averaged 62.6 while the direction hit 45.6%.
- The field gap spans 1.5 pp across the three months inside the quarter, against a pooled quarterly gap of +17.0pp on field mean stated confidence of 62.6 (by month 62.8 / 62.6 / 62.6). The prior-quarter gap reads n/a because this is the first quarterly issue: there is no prior quarter (platform data begin Jul 15, 2026).
- Trading over the quarter finished 0 of 7 lines net-positive (by month 0 / 0 / 0); the pack labels one of the three months calm and two volatile. The prior-quarter comparison reads n/a because this is the first quarterly issue: there is no prior quarter (platform data begin Jul 15, 2026).

## Limitations

- No prior quarter: this is the first quarterly issue (platform data begin Jul 15, 2026), so every "prior" cell reads n/a and the page-2 chart carries no gray series; a month inside the quarter is not a prior window. Partial quarter: the window is the calendar quarter Jul 1-Sep 30; the platform's first forecast slot is Jul 15 18:01 UTC, so 78 of 92 days carry data. Asset universe: the quarter's rows also carry 1,592 directional calls on AVAX (414), DOGE (347), LINK (436), PEPE (394) and TON (1), all in Jul 15-31; they are counted in the field totals but not ranked in the asset tables.
- Two cutoffs live in this issue. The quarter is frozen at Thu Oct 1, 2026, 16:00 UTC; the three monthly blocks are frozen at their own published cutoffs (M1 Jul 15-31, M2 Aug 1-31, M3 Sep 1-30). A forecast still pending at its month's cutoff is immature in the monthly pack and mature in the quarter pull, so the quarter need not equal the sum of the months; here the two agree (59,606 vs 59,606 directional; no day of the quarter falls outside the three months): the quarter pull split by slot month gives 19,541 / 19,383 / 20,682, the same as the frozen monthly packs.
- 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 59,606 scored calls sit inside one calendar quarter, and three monthly points are three points: an OLS slope over them is a description of this quarter, not evidence of drift. This window ran 24 trend days of 77 with a BTC daily move (Jul 1-15 carry none), and the pack's own volatility rule splits the three months of the quarter as M1 calm, M2 volatile, M3 volatile; every month-to-month comparison inside the quarter is a comparison across regimes as well as across windows. Models report confidence at discrete levels, not a continuous scale; the buckets reflect those natural breakpoints and 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 (quarterly #1): (1) rolling-1w forecasts emit daily since Aug 22, 2026 and rolling-1M daily since Aug 23, 2026, so across this 92-day window daily coverage is PARTIAL as measured -- the 1w series has slots on 55 of 92 days and the 1M series on 41 of 92 days; both sit outside this report's FH gate (1h/4h/1d) and appear only in the exclusions counter (1w 4,001, 1M 2,787). (2) 3 model lines are lineage splices inside this window: claude-opus-5 (incl. 4.8-era, Jul 15-30); grok-4.6 (incl. 4.5-era, Jul 15-Aug 24); qwen-3.8-max (incl. 3.7-era, Jul 15-Aug 5). (3) Every per-model aggregate in this issue sees the merged head id; the raw id is kept in the pack, and 24,922 rows were remapped inside the quarter (19,532 / 5,390 / 0 in Jul 15-31 / Aug 1-31 / Sep 1-30).
- Market-state row is computed on a single-exchange (binance) daily candle series, as in the weekly series after the audit that found the raw table mixes two exchanges. The quarter block covers 78 days (Jul 15-Sep 30) of the 92-day calendar quarter, and there is no prior-quarter block, so the TOP5 volume figure is a 78-day sum with no quarter-over-quarter change.
- Research-to-date counter: directional forecasts resolved inside the FH gate (1h/4h/1d) since Jul 11, counted at the issue cutoff (Thu Oct 1, 2026, 16:00 UTC) -- the D2 definition pinned in the weekly series; this issue prints 59,755. The pack carries no prior-cutoff pin for a quarterly window (not applicable: there is no prior quarter, and the prior-cutoff counts are 0 by construction), so the control this issue is the monthly reproduction: the M3 block of this pack (Sep 1-30) reproduces the published Monthly #2 on 7 of 7 checks, the M2 block (Aug 1-31, cutoff Tue Sep 1, 2026, 16:00 UTC) reproduces the published Monthly #1 on 7 of 7 checks, and the M1 block (Jul 15-31, cutoff Sat Aug 1, 2026, 16:00 UTC) reproduces the prior window of Monthly #1 on 7 of 7 checks. Counter deltas against pre-wave-3 issues remain definitional.

## Counters & lineage

| Counter | Value | Counter | Value |
|---|---|---|---|
| Forecasts total | 133,039 | Mature (scored pool) | 122,770 |
| OK in gate | 122,919 | -- of them directional | 59,606 |
| Out of gate (1w / 1M) | 4,001 / 2,787 | -- of them sideways | 63,164 |
| Invalid | 3,332 | Pending (next issue) | 149 |
| Late closes | 0 | Uptime, all grids | 2,929 / 3,496 slots |
| Source file | quarterly_metrics_2026-q3.json | Report cutoff | Thu Oct 1, 2026, 16:00 UTC |
| Generated at (pipeline) | 2026-10-07 11:01 UTC | Methodology | v1.1 (2026-08-10) · hash e66c7e8c864a2233 |

**THIS REPORT: 59,606 scored observations.** MARKETMANIA RESEARCH TO DATE (as of cutoff Oct 1, 16:00 UTC): 59,755 resolved since Jul 11 · 7 models tracked (current line-up; earlier versions folded into their successors' lineage) · 5 assets · 5 horizons · hourly · 48 published reports.

> **Issue #1.** Quarterly Calibration is a living quarterly comparison; each issue appends one more calendar quarter of confidence-vs-hit data. This first issue sets the baseline on a partial quarter (Jul 15-Sep 30): field gap +17.0pp on 59,606 scored calls, field Brier 0.2803, monthly gaps +17.8pp -> +16.3pp -> +17.0pp, field slope -0.40 pp/month (flat); there is no prior quarter. 3 model lines are lineage splices this quarter (claude-opus-5 (incl. 4.8-era, Jul 15-30); grok-4.6 (incl. 4.5-era, Jul 15-Aug 24); qwen-3.8-max (incl. 3.7-era, Jul 15-Aug 5)). Engine 1.1 has powered the sandbox since Aug 18, i.e. inside this window; no cross-engine PnL comparisons are claimed.

## What we're testing next

- Quarterly #2 (Q4 2026): the same three series over Oct 1-Dec 31, cut on Jan 1 plus the maturity lag, so its quarter-over-quarter comparison runs against this partial Q3.
- Quarterly Config Watch #1: the same Q3 window through the config search engine (w309), published as the fourth quarterly report of this issue.
- Quarterly Stability Index: Run 3, the next repeat-prompt stability run, is the first cross-quarter point on model stability.

## Related research

| Report | Direct PDF link |
|---|---|
| Quarterly Consensus Watch #1 | https://marketmania.ai/research/reports/consensus-watch-quarterly-2026-q3.pdf |
| Quarterly Model Watch #1 | https://marketmania.ai/research/reports/model-watch-quarterly-2026-q3.pdf |
| Monthly Calibration #2 | https://marketmania.ai/research/reports/calibration-monthly-2026-09.pdf |

The three quarterly reports publish together as one issue each quarter; each links straight to the others' PDF and to the last monthly issue of the same series. Direct links are the posting rule from wave 2 on.

## Cite this report

```bibtex
@misc{mm_calibration_quarterly_2026q3,
  title  = {Quarterly Calibration #1: confidence vs. direction-hit, Jul 15-Sep 30 2026},
  author = {{MarketMania Research}},
  year   = {2026}, month = {October}, day = {8},
  url    = {https://marketmania.ai/research/reports/calibration-quarterly-2026-q3.pdf},
  note   = {Methodology v1.1, hash e66c7e8c864a2233; source quarterly_metrics_2026-q3.json}
}
```
