Last validated release
Passed
This is the status of the forecast release already retained for research.
Model · data · evidence
Research validation
See whether Quantiv forecasts beat matched benchmarks, how well the uncertainty estimates are calibrated, and whether the latest research state is publishable.
AssessmentOct 16, 2026, 08:14 ET
Performance
Does the model add information?
Historical out-of-sample results compare each admitted cohort with its matched benchmark on the same observations.
Weighted error reduction
16.8%
Lower mean absolute error than matched cohort benchmarks across the admitted horizon/cohort pairs.
Admitted cohort improvement · 8.6%–46.9%
Quantiv model5.10%
Matched cohort benchmarks6.12%
91,009holdout row-observations across 23 horizon models
83,762 matched benchmark row-observations across 30 admitted horizon/cohort pairs. Calibration uses all holdout rows; MAE uses matched rows. Events can appear at multiple horizons.
Performance by horizon
| Horizon | Cohort | Calibration rows | Matched rows | Benchmark | Model MAE | Benchmark MAE | Improvement | 50% interval | 80% interval |
|---|---|---|---|---|---|---|---|---|---|
| T-1 | Optionless | 5,466 | 4,917 | Historical move | 5.17% | 5.77% | 10.5% | 48.5% | 78.3% |
| T-2 | Optionless | 4,125 | 3,699 | Historical move | 5.14% | 5.76% | 10.8% | 50.2% | 78.3% |
| T-2 | Strict options | 1,355 | 1,355 | Market straddle | 4.31% | 6.27% | 31.2% | 48.3% | 72.8% |
| T-3 | Strict options | 861 | 861 | Market straddle | 4.25% | 6.26% | 32.0% | 47.2% | 74.3% |
| T-4 | Optionless | 2,537 | 2,250 | Historical move | 5.45% | 6.00% | 9.3% | 49.7% | 78.4% |
| T-5 | Optionless | 3,842 | 3,501 | Historical move | 5.30% | 5.81% | 8.7% | 50.0% | 77.7% |
| T-6 | Optionless | 5,903 | 5,366 | Historical move | 5.35% | 5.89% | 9.3% | 49.4% | 78.5% |
| T-7 | Optionless | 6,248 | 5,622 | Historical move | 5.28% | 5.87% | 10.0% | 50.5% | 79.6% |
| T-7 | Strict options | 1,926 | 1,926 | Market straddle | 4.34% | 6.65% | 34.7% | 47.9% | 75.6% |
| T-10 | Strict options | 887 | 887 | Market straddle | 4.33% | 6.71% | 35.5% | 46.7% | 76.4% |
| T-11 | Optionless | 2,457 | 2,165 | Historical move | 5.48% | 6.00% | 8.6% | 49.2% | 78.7% |
| T-12 | Optionless | 3,727 | 3,390 | Historical move | 5.31% | 5.84% | 9.2% | 49.5% | 78.8% |
| T-13 | Optionless | 5,788 | 5,245 | Historical move | 5.34% | 5.91% | 9.7% | 49.4% | 79.1% |
| T-14 | Optionless | 6,251 | 5,613 | Historical move | 5.29% | 5.87% | 9.8% | 49.8% | 78.9% |
| T-14 | Strict options | 1,922 | 1,922 | Market straddle | 4.29% | 6.80% | 36.8% | 48.0% | 77.0% |
| T-15 | Optionless | 5,449 | 4,893 | Historical move | 5.21% | 5.78% | 9.8% | 49.7% | 78.3% |
| T-15 | Strict options | 1,796 | 1,796 | Market straddle | 4.31% | 6.89% | 37.5% | 47.3% | 75.8% |
| T-16 | Strict options | 1,349 | 1,349 | Market straddle | 4.45% | 7.09% | 37.3% | 46.0% | 74.6% |
| T-17 | Strict options | 870 | 870 | Market straddle | 4.43% | 7.31% | 39.5% | 46.3% | 74.0% |
| T-18 | Optionless | 2,406 | 2,122 | Historical move | 5.57% | 6.14% | 9.2% | 48.5% | 78.3% |
| T-18 | Strict options | 652 | 652 | Market straddle | 4.09% | 7.35% | 44.4% | 48.0% | 79.3% |
| T-19 | Optionless | 3,623 | 3,294 | Historical move | 5.43% | 5.95% | 8.7% | 48.2% | 78.6% |
| T-20 | Optionless | 5,660 | 5,145 | Historical move | 5.39% | 5.96% | 9.6% | 49.1% | 78.9% |
| T-20 | Strict options | 1,680 | 1,680 | Market straddle | 4.49% | 7.55% | 40.6% | 46.4% | 75.6% |
| T-21 | Strict options | 1,854 | 1,854 | Market straddle | 4.37% | 7.60% | 42.5% | 45.9% | 75.9% |
| T-22 | Optionless | 5,514 | 4,959 | Historical move | 5.18% | 5.75% | 10.0% | 48.9% | 79.1% |
| T-23 | Optionless | 4,245 | 3,813 | Historical move | 5.16% | 5.75% | 10.3% | 50.3% | 79.5% |
| T-23 | Strict options | 1,280 | 1,280 | Market straddle | 4.38% | 7.67% | 42.9% | 46.5% | 74.3% |
| T-24 | Strict options | 784 | 784 | Market straddle | 4.47% | 7.88% | 43.3% | 47.2% | 76.1% |
| T-25 | Strict options | 552 | 552 | Market straddle | 4.19% | 7.89% | 46.9% | 49.3% | 79.0% |
MAE is absolute stock-move error. This is predictive validation, not a trading-P&L claim. Historical moves are a benchmark only; future forecasts have no historical-median fallback.
Uncertainty
Calibration
Observed quantiles should land close to their nominal targets. The marker shows observed frequency; the thin line is the target.
P1013.3%
Target 10.0%
P5057.2%
Target 50.0%
P9091.4%
Target 90.0%
80% interval78.1%
Target 80.0%
Additional checks
P25 observed31.1%
P75 observed80.1%
50% interval49.0%
Publication
Can new research publish?
The last validated release and the ability to publish the next one are different states. This is the only place both are summarized.
Next research release
Held
Awaiting eligible options data. The last validated forecast release is retained; new research publication remains blocked.
Controls checked13
Exceptions0
Forecast rows638
Events638
Model safeguards
passedServing model
Championactive
Driftwarning
Fallbackavailable
Shadow roleschampion_evidence, previous
Data safeguards
failedDecision universe
Event coverage63.1%
Eligible contracts57,637 / 96,562
Quarantine80,232
Duplicates0
Needs attention
7Why publication is constrained
criticalSnapshot 2026-09-30 is 11 market sessions behind expected 2026-10-15. This is the assessed snapshot state; new research stays blocked until fresh option evidence passes controls.16
warningCalendar events fall outside the current options decision universe2,826
warningIn-universe earnings events have no decision-eligible option chain468
warningAggregate option coverage is below the model activation limit; individually eligible events may publish468
warningOne or more model horizons lack sufficient eligible option coverage105
criticalOptions evidence did not pass the latest assessed controls. latest source date 2026-09-30 is 11 market session(s) behind expected 2026-10-15.
warningLatest source quotes for retired symbols are excluded from the decision universe4
These gates affect the next research release. They do not rewrite a previously validated forecast receipt.
Method
How the result is earned
The validation process can be reduced to three checks. Technical controls remain available in the audit trail below.
Separate past from future
Development labels fall strictly before 2026-03-26. Reserved earnings events run from 2026-04-18 through 2026-10-14.
Beat matched benchmarks
Strict-options results use the market straddle; optionless results use a historical-move benchmark. Each admitted pair passes point-error, calibration and statistical controls on reserved observations.
Verify before publication
Bundle identity, feature schema, shadow scoring, drift controls and rollback evidence all stay tied to the model that produced the stored forecasts.
Technical audit trailEvidence behind this page
Forecast receipt3db03b17a4…cfd9e871forecasts_2026-10-15.3db03b17a45f.receipt.json
Model artifact3370619ce2…8e9aa8ceapps/ml/model_trainer.py
Forecast artifacta8a218c261…43f7aff1scripts/daily_score.py
Active bundled14dc7f2aa…c869dd1bsigned champion
Control snapshotOct 16, 2026, 08:14end_of_day_research · publication blocked
Evaluation scopeHistorical out-of-samplesealed_temporal_exact_artifact_test
Admission receipt36751a5106…6c15e7fesigned model and cohort admission
Source bundle629386dcbf…bc12c2fffrozen before reserved evaluation
Serving bundled14dc7f2aa…c869dd1b23 horizons · 30 admitted horizon/cohort pairs
Reserved panelc48c61e803…c8bb2b33Reservation fd05c255ce…9d09669a
Reserved event window2026-04-18 – 2026-10-14Earliest snapshot 2026-03-26
Development checks
4 expanding walk-forward windows with 60d validation windows and a 5d purge are development evidence. They select the model before the reserved historical evaluation and are separate from the results above.
Admission proves exclusion of reserved labels from this candidate’s fitting and selection. The exact candidate excludes reserved labels from fitting, selection, early stopping and quantile calibration. Complete point-in-time calendar revision history and historical provider delivery timestamps are not established. Prior model-family exposure to this history is not established; this is not a claim of globally untouched history. Metrics describe admitted historical head/cohort cells; aggregate summaries do not establish aggregate statistical significance or live trading performance.
Decision scope
Read methodology →Quantiv outputs are end-of-day research evidence. Live stock prices may update spot-derived inputs, but options, IV, Greeks and other snapshot features remain frozen. The validation page does not present executable option quotes or live-trading signals.