Ten protocols you can review before collecting data. No dataset is admitted to this public catalog and no findings are reported.
PROTOCOL · DATA NOT ADMITTED
PCR and next-session NIFTY movement
Hypothesis
A predeclared end-of-day open-interest PCR bucket has a different next-session mean return from the unconditional benchmark.
Definitions
PCR = selected put OI / selected call OI. Fix instrument, strike universe, expiry policy and timestamp. Volume PCR is a separate experiment.
Dataset required
Timestamped contract OI, exchange calendar and index returns
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Store both components and actual publication time. Set thresholds in training data only. Join to a later, nonoverlapping next-session return. Purge training labels spanning the test boundary.
Calculation and benchmark
Compute the selected put/call OI ratio. Compare next-session open-to-close returns by frozen bucket against all eligible dates.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Zero denominators, missing contracts or reports, unresolved report revisions; report exclusion counts rather than silently filling.
A predeclared normalized FII index-futures net position is associated with later returns in held-out observations.
Definitions
Choose report units. Define net = long minus short, and optional net/gross normalization within a consistent instrument bucket.
Dataset required
Participant OI reports, publication timestamps and index returns
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Retain report revision and availability time. Lag until available, not merely by printed report date. Freeze normalization and buckets before the held-out period.
Calculation and benchmark
Compute long minus short and, separately, net divided by gross. Compare later returns only after effective report availability.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Unresolved report categories, missing denominators, unavailable revisions and nontrading sessions under a documented calendar.
Sensitivity checks
Raw versus gross-normalized position, changes versus level, expiry rolls, holding horizon and report availability delays.
Limitations
Hedges and positions outside the chosen bucket are unobserved. Do not describe net contracts as an intention or a full directional book.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.
The failure frequency of a frozen breakout definition differs across predeclared market conditions.
Definitions
Define a prior range, close-based crossing and failure as a later close back inside a fixed window.
Dataset required
Point-in-time OHLC bars, calendar and event timestamps
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Detect breaks from prior completed bars. Require a complete forward window and evaluate on a later holdout.
Calculation and benchmark
Count later closes back inside the range among triggers with a complete declared horizon. Report missing windows separately.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Incomplete windows, missing bars and overlaps under the declared rule.
Sensitivity checks
Range length, horizon and close versus intrabar definition.
Limitations
Event frequency does not measure executable returns.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.
A frozen opening-range protocol has different held-out outcomes across regimes defined before entry.
Definitions
Specify exchange calendar/timezone, opening interval, crossing timestamp, entry/exit and a regime based only on prior available information.
Dataset required
Time-stamped intraday bars and executable quotes
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Keep opening-range data separate from the later outcome. Fit regime cutoffs on training observations. Use actual contract availability and purge boundary-overlapping labels.
Calculation and benchmark
Freeze the opening range and entry timestamp. Compare matched held-out groups using declared fill and cost assumptions.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Observed option premium changes differ across predeclared windows after controlling for contract, underlying movement and IV assumptions.
Definitions
Specify instrument, exercise style, strike selection, expiry distance, bid/ask convention, calendar time and synchronized quote availability.
Dataset required
Synchronized option and underlying bid/ask observations
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Separate observed premium changes from European-model attribution. Freeze window selection, exclude crossed/stale quotes using prior rules, and preserve repricing residuals.
Calculation and benchmark
Compare synchronized premium changes and full model repricing. Keep residuals and quote-side sensitivity visible.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Missing synchronized quotes, unresolved expiry or contract multipliers, stale/crossed quotes and unlicensed redistribution.
Sensitivity checks
Time windows, expiry buckets, moneyness, rate/dividend assumptions, IV estimator, liquidity and quote-side convention.
Limitations
Time-of-day associations do not isolate theta causally. No preferred premium-buying or selling window is asserted.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.
A regime classifier frozen using prior available data has measurable held-out coverage and transition stability; no return advantage is presumed.
Definitions
Declare features, prior windows, label vocabulary, training-only cutoffs, feature availability and abstention conditions. This evaluates labels, not an ORB strategy.
Dataset required
Chronological price features and session calendar
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Freeze classifier and calendar before holdout. Recompute each label from data available at that decision only. Compare coverage, transition frequencies and unknown-label rates against an unconditional declared benchmark; keep economic returns separate.
Calculation and benchmark
Compute prior-window path efficiency under declared features. Evaluate coverage and transitions before any return comparison.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Insufficient past window, missing sessions, future-adjusted features, unresolved revisions and unavailable historical universe. Record counts without forward filling.
Sensitivity checks
Training-window choice, threshold stability, prior-only feature alternatives and holdout shifts; report multiplicity and small groups.
Limitations
Stability does not imply predictive value or profitable execution. Feature extraction is investigator-specific and is not supplied by the generic descriptive adapter.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.
Under a frozen breakout/failure definition, events with a high prior-normalized volume have a different failure frequency.
Definitions
Prior level excludes the event candle. Define event availability, volume normalization, failure horizon, stop and event overlap policy.
Dataset required
Time-stamped OHLCV and point-in-time instrument universe
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Build a point-in-time universe including delisted names where applicable. Use only event-time volume information; full-session volume cannot filter an earlier intraday event. Compare held-out groups and counts.
Calculation and benchmark
Normalize event-time volume against comparable prior windows, then compare complete-event failure counts.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Equal-window realized variance differs by time of day in held-out sessions.
Definitions
Use fixed IST windows and sum squared log returns at a declared sampling interval.
Dataset required
Timestamped intraday midquotes or consistently sampled bars
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Use equal-length windows and one price convention. Hold out later whole sessions.
Calculation and benchmark
Sum squared log returns at one fixed interval across equal windows. Hold out whole sessions and retain missing-window counts.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Missing bars, partial sessions and unresolved price discontinuities.
Sensitivity checks
Sampling interval, stale quotes, auctions and session-level clustering.
Limitations
Realized variance depends on sampling and does not predict direction.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.
A predeclared event class has a different held-out return distribution in fixed post-publication windows from matched non-event observations.
Definitions
Freeze event eligibility, original release timestamp/timezone, revised versus initial values, instrument universe, window endpoints and overlap policy. No label may depend on the subsequent price reaction.
Dataset required
Historical expiry calendars, intraday quotes and matched sessions
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Use permitted original-time event records and reviewed exchange calendar. Make the decision only after release and realistic availability. Match benchmarks with prior-known information, purge overlaps across splits and separately review multiple-event clustering and costs if executable returns are claimed.
Calculation and benchmark
Match historical expiry sessions to eligible non-expiry controls on prior-known features. Compare equal IST windows and report exclusions.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Unknown original publication times, after-the-fact event classification, overlapping announcements under the frozen rule, missing quotes and unreviewed revised data.
Sensitivity checks
Release latency, fixed window lengths, scheduled versus unscheduled events, matched-control rules and clustered uncertainty; all alternatives must be disclosed.
Limitations
No event dataset is admitted and no causal or trading claim is available. The generic runner validates precomputed observations; it does not construct or authenticate event timestamps.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.
Date-effective explicit execution assumptions change a fixed, predeclared trade sample's net outcomes relative to its gross outcomes.
Definitions
Freeze trade selection, quantities, units, execution side, matched fills, fee schedule evidence and slippage convention. Distinguish costs already embedded in fills from additional modelled costs.
Dataset required
Permitted de-identified fill pairs and date-effective fee evidence
Coverage and sample period
Pending a versioned dataset, recorded availability timestamps and a declared split.
Method
Admit permitted versioned fill records without customer identifiers. Apply verified date-effective charges per leg/order/turnover as applicable, reconcile against known arithmetic and compare identical gross/net sample memberships. Reserve a later validation period; never optimise sample selection on net results.
Calculation and benchmark
Compute net P&L from matched fills and dated charges. Compare gross and net outcomes on identical trade membership.
Point-in-time and bias checks
Record effective availability before the decision, freeze definitions before holdout, and disclose missing and overlapping observations. Never tune on held-out outcomes.
Empirical status
NOT_AVAILABLE. No dataset has passed admission for this catalog.
Exclusions
Missing fills/quantities, unverifiable fee dates, ambiguous units, duplicate orders, refunds not reconciled and records lacking usage permission. No private account data is automatically imported.
Sensitivity checks
Tick rounding, partial fills, spread/slippage double counting, flat/per-order costs, turnover rates and nonlinear charges. Report all assumptions and unmatched records.
Limitations
A cost reconciliation is not evidence of strategy predictability. The free tool has an explicit simplified charge model, not an authoritative broker tax engine; a dedicated reviewed adapter is still needed for real fills.
To reproduce a future result, retain dataset hashes, cutoff dates, parameters, exclusion counts and runner version.