Accumulation: price-tape context
What this helps you see
This house measure describes the directional character of price and trading activity against the stock’s own history. A smoothed phase helps distinguish persistent behaviour from a single unusual session.
What makes it different: DVPT describes the size of delivery activity; this lens adds a signed interpretation of the price tape. The interpretation does not establish the identity or intent of buyers and sellers. Its failure as a predictive feature remains part of the published record below.
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
Accumulation is DESCRIPTOR-ONLY. It characterises and confirms; it never ranks, picks, or sizes. This is a binding fence — the project keeps a falsified-approaches ledger and misrepresenting Accumulation's status is a blocking error.
Any predictive/ranking/position-sizing role was gated on a purged walk-forward + Deflated-Sharpe test, which FAILED (2026-06-22, the mep strategy design notes):
| Config | OOS return/vol | DSR | CAGR |
|---|---|---|---|
| Baseline panel | 0.76 | 0.45 | 23.0% |
| + Accumulation price-tape features | 0.68 | 0.36 | 18.7% |
| (Nifty 500 benchmark) | 0.86 | — | — |
The three price features lowered risk-adjusted performance and never cleared the Nifty-500 line (0.86), let alone the DSR ≥ 0.95 bar. The poison is close_vs_vwap — the in-sample #1-importance feature and the out-of-sample destroyer. The strategy ledger records it in the BLOCKING FAILURE MODELS table: *"Accumulation tape as alpha — DSR 0.45→0.36 when added — Descriptor-only; adds nothing. Do not re-test as alpha."*
The killer line (keep it): *the feature that best describes today's accumulation is exactly the one that fails to predict tomorrow* — which is precisely why Accumulation is a descriptor, and why the smoothed PHASE , however clean, is still a descriptor. Corroborating evidence: the 2026-07-05 footprint study found deliv_per barely elevates during genuine accumulation (δ≈+0.07), coherent with this failure — the delivery/price *level* is mostly noise.
6. Data & provenance
- Primary source, no delivery, no Screener. Accumulation reads
open/high/low/close/prev_close/avg_price[VWAP]/volume/num_trades/valuefrom the NSE bhav copy (bhavcopy_rows) — authentic/primary (guardrail #8), and deliberately **nodeliv_*** (it is explicitly *not* DVPT). Unlike DVPT'saccum_screenoverlay, Accumulation's core has no Screener/fundamentals dependency — it is clean of the copyright-remediation debt. - Within-stock history required. The z-scores need each stock's own trailing distribution, so Accumulation is full-history-recomputed (same doctrine as DVPT's ATH/first-ever). The smoothed phase reads only *trailing stored* daily scores → no look-ahead.
- F&O OI identity feed: the UDiFF F&O bhavcopy (same NSE host as the cash bhav) → stock-futures OI aggregated per underlying; primary-source. Pre-UDiFF (>2024) legacy backfill is a future 2nd parser (§9).