Delivery footprint (DVPT)

What this helps you see

DVPT describes delivered rupee value per total exchange trade. It compares activity with the same stock’s history and places unusual delivery beside price behaviour and observed price zones. For a multi-session view, sum delivered value and trade counts before dividing; do not average daily ratios.

What makes it different: a within-stock delivery-footprint view, with normal activity, unusually large activity and price context kept distinct. It cannot identify the buyer, the seller or who initiated a trade. A large reading is not proof of institutional accumulation.

Explore the stock screen.

About this reference

This reference explains the reading, its evidence and its limits. Implementation and calibration notes are retained in the internal research archive. Published test definitions and unfavourable results remain available below.

4. Status, validation & honesty fence

The engine is LIVE; the original thesis is REFUTED. DVPT never ranks stocks for alpha — it is a within-stock confirmation/character/divergence lens on price. This is a binding honesty fence, not a hedge.

The v0.1 design (the dvpt picking strategy design notes, kept verbatim for history) premised DVPT as a leading smart-money detector that could drive a ranked 30–40-stock picking portfolio. Two independent lines of evidence refuted that premise:

• The mechanics (2026-06-22). DVPT = delivery value ÷ total num_trades. A trade prints on every order match, so num_trades collapses (→ high DVPT) only when both sides are concentrated — i.e. a block/bulk transfer that is already disclosed with client names. The case the strategy existed to catch — one informed buyer absorbing fragmented retail supply — generates many matches → a high trade count → a low, retail-looking DVPT. So DVPT reads the counterparty's fragmentation, not the accumulator's conviction; modern execution (VWAP/iceberg) fragments on purpose, so DVPT actively selects against sophistication; and it is side-blind (a high-DVPT day is equally consistent with distribution).
• The data ("counter-DVPT"). Reading the raw archive alone, a rising delivery footprint is NOT what precedes explosive moves — the winning rules prefer deliv_qty_trend ≤ ~1.5 (no surge) and lower delivery-%. Logged verbatim: "no stealth institutional-accumulation footprint before +10% moves in the EOD aggregate"; the whale-among-minnows ticket-dispersion hypothesis was real-data refuted, OOS both directions, every year 2012–26.

The accumulation-footprint calibration (the strategy ledger, 2026-07-05) nailed it quantitatively: against disclosed insider/SAST accumulation windows, f_deliv_per — DVPT's core — moved δ≈+0.072 / +0.083 (fail), i.e. "DVPT's core barely moves during real accumulation." The delivery level is mostly noise; delivered value + clip size (f_trade_size δ+0.329/+0.250, the one gate-passer) carry what little signature exists. This is coherent with the price-tape sibling Accumulation's alpha failure (the mep notes; DSR 0.45→0.36).

Doctrine (state it plainly): price strength is the only gross forward-return engine; delivery/accumulation reads are context/confirmation layers, never rankers or fundable alpha (the strategy ledger "BLOCKING FAILURE MODELS" corollary). Accordingly the surviving DVPT role is (1) confirmation — rising delivery alongside a rising price = a move being paid for with held stock; (2) divergence flag — price up + delivery collapsing = hollow; (3) within-stock relative only — never a cross-stock ranking input. The "who is transacting" decode moves to the named-flow channel (bulk/block + FII/DII + F&O OI), not the EOD aggregate.

Number-integrity : the last raw-close math in the signal path is gone — zones, hot-day averaged closes, and key-price weights are split/bonus-adjusted to the computing date's basis; there is ONE shared _hot_days_core; the >30% close-jump fallback rescales history only when the authoritative tape agrees. Golden regression: an automated test.

6. Data & provenance

• Primary source (guardrail #8-clean). NSE bhav copy: sec_bhavdata_full_DDMMYYYY.csv (delivery, 2020→present) and, pre-2020, the **MTO ⋈ legacy cm*bhav.csv.zip merge that reconstructs delivery back to ~2005** (and carries ISIN for the security_master). NSE is authentic/primary — no vendor, no Screener in the DVPT core.
• EQ-only, T2T excluded. Every delivery measure reads the EQ series only; names under trade-to-trade surveillance (BE/BZ — delivery is 100% by rule, so it carries no information) show no delivery signals — excluded, not polluted.
• Survivorship by construction. The backtest/ignition universe on any date is the raw bhav membership of that date (delisted names included), gated through security_master (renames stitched on ISIN; demergers/mergers flagged). Point-in-time, never today's listed set.
• ⚠ Screener touchpoint (disclose): the accum_screen sweet-spot overlay pulls the point-in-time patearn fundamental tier via fundamentals_asof → research.db.fundamentals_history, which is Screener-derived. That is the one non-primary dependency in DVPT's orbit; the standing remediation is the Screener→BSE/NSE XBRL migration (guardrail #8). DVPT's own delivery signal has no Screener dependency.