Skip to content
Canli Capital

Research

Sharadar HDB zero-dividend quarantine

Short title: HDB dividend quarantine
Author: Arhan Canli
Decision: QUARANTINE_REQUIRED_NO_AUTOMATIC_REPAIR

Finding

The frozen Sharadar ACTIONS archive contains 11 dividends whose values are not positive. One survived the common-stock and liquidity filters and entered the executable lake: HDFC Bank ADR (HDB) on 2025-06-26 with value 0.0. The loader propagated that row unchanged. The event engine correctly rejects it, so the selected gross-profitability exact replay failed closed before producing a curve.

This is a vendor-originated source defect, not a backtest result. The raw archive, lake partition, loader, engine, and failed replay are all bound by SHA-256 in the machine-readable audit.

Primary-source cross-check

HDFC Bank's June 2025 Form 6-K exhibit states that the board recommended a ₹22 dividend per underlying equity share, with a 2025-06-27 record date and payment on or after 2025-08-11. Its 2025 Form 20-F states that one ADS represents three underlying shares. Sharadar's ACTIONS documentation defines the action value as numeric, but does not resolve the required ADS net-cash or date semantics for this anomalous row.

The issuer filing therefore contradicts 0.0 as the declared underlying dividend amount. It does not prove the net USD cash an ADS holder should receive after FX, withholding, and depositary handling. Moving a later vendor amount backward to the ex-date would be an imputation, not a verified correction.

Quarantine contract

Future Sharadar loads must reject any split or dividend with a non-finite or non-positive value before it reaches the executable lake. The existing raw archive and contaminated partition remain preserved by hash. No process may silently drop, move, or replace this row inside an identity-bound replay.

The correction gate unlocks only when an issuer/depositary record or corrected vendor record establishes all three executable fields: ADS ex-date, knowable-at timestamp, and net USD cash amount. Any repair then creates a new versioned lake and source-environment hash; the failed replay remains part of the public record.

Claim boundary

No return series was opened and no hypothesis was spent. This audit validates neither the historical curve nor any Sharpe, drawdown, capacity, diversification, or sleeve claim.