NIFTY Paper Strategy Audit - 15 October 2026

Decision

Keep both systems PAPER ONLY. Pause new Psychology entries as explicitly approved; preserve research and open-position management. Do not promote the existing strategy to real orders yet. No historical ledger has been rewritten. Existing entry, target, stop, sizing and broker-order logic remain unchanged in this delivery.

This is a census of the persisted ledgers captured at 07:49:34 IST on 15 October, covering trades through 14 October. It is NOT a tick replay, contract-note reconciliation, or independent verification of historical quotes. Source: authoritative EC2 `data/nifty_options_engine.db`; deployed engine, context and Psychology source. All captured trades were closed. Missing days are not assumed zero-trade days; 23 and 11 below mean sessions containing trades, not complete market-session coverage.

Results

Current means `nifty-signal-break-v3-20260908`; legacy is the separate 6-trade v1 population, not mixed into the current account. Psychology uses its separate INR100,000 account; current existing strategy starts at INR25,000. Capital bases differ, so raw rupee comparisons are not risk-matched.

Current strategy: 71 wins, 103 losses, 1 flat; gross INR26,555.75 minus INR14,239.56 charges gives recorded net INR12,316.14 (per-trade rounding). Average winner INR1,095.96; average loser INR635.89; expectancy INR70.38/trade. Maximum closed-trade equity drawdown INR15,637.40, not intratrade worst-case loss. Win-rate descriptive Wilson interval 33.6%-48.0%; clustered trades are not independent observations.

Psychology: 53 wins, 70 losses; gross INR276.25, charges INR12,867.84 and additional spread deduction INR3,583.12 produce net -INR16,174.72. Adding back the apparent duplicate spread deduction alone still leaves -INR12,591.59. It does not rescue the strategy. Additional-order brokerage on three adds also needs reconciliation. Closed-trade drawdown INR39,545.97; a permanent claim that all contextual research has failed is not supported by only 11 traded sessions.

Robustness and concentration

- Current strategy without 13 October: 169 trades, net -INR11,787.49, PF0.817.

- Without its single best trade: net -INR5,839.81, PF0.911.

- Best trade: `NOE-2026-10-13-NIFTY-0a2d0292`, historical-level CALL, +INR18,155.95, 20m26s duration, sampled premium MFE114.52%, futures volume4.29x, NIFTY and BANKNIFTY OI both increasing. This is an observed outlier, not an assumed repeatable expiry payout.

- Expiry-day trades: 29, +INR32,167.82, PF3.751. Non-expiry: 146, -INR19,851.68, PF0.631. Expiry is defined by the recorded contract expiry, not a guessed weekday. These are not matched controls.

- Earlier period, before5 October: 96 trades, -INR10,380.35, PF0.729. Later7 traded sessions:79 trades,+INR22,696.49,PF1.837. This is retrospective chronological description, NOT untouched out-of-sample validation.

- Session-resampled 95% net interval: approximately -INR32,219 to +INR73,746 across 10,000 bootstrap draws (seed20261015). Wide interval and regime dependence; not a forecast.

- Extra execution slippage stress, beyond already-modeled ask-to-bid spread: 0.25 option points each side reduces net to INR778.64; 0.50 gives -INR10,758.86; 1.00 gives -INR33,833.86. This subtracts2*points*original quantity without resimulating future sizing, so it is sensitivity analysis, not an executable replay.

What actually generates the existing trades

173 of175 current trades came from `MANUAL_SIGNAL_BREAK`, which is automatically paper-executed despite its misleading internal name. Only2 were OR/retest entries; those lost INR1,165.59 together. The UI label 'OR/retest history' was therefore misleading.

Required confirmation is already present in all filled signal-break trades: NIFTY close relative to EMA20 and SMA5; directional close-to-close momentum; NIFTY near-futures VWAP alignment; BANKNIFTY EMA/SMA, momentum and near-futures VWAP agreement with no opposing rejection. Because these gates are mandatory, this ledger cannot estimate their incremental benefit versus removing them. Rejected candidates and causal replay are needed for ablation.

Sniper Pro pressure gauges, GEX, VIX, cash basket participation, footprints and CAS are NOT mandatory entry votes in this existing strategy. Their visibility on a chart does not mean they caused a trade. OI and volume are captured context; the OI comparisons below are exploratory filters, not deployed rules. OI UP does not identify long buyers or institutions.

Entry signal groups

Entry Origin

Level Kind

Option Type

Entry Hour

Expiry Day

Oi Pair

Volume Bucket

Rr Bucket

Delay Bucket

Candidate combinations, not promoted rules

These filters were inspected after seeing outcomes. Multiple comparisons, direction/regime concentration, sizing and overlapping sessions can inflate apparent performance. The before/after and best-day exclusion checks are diagnostics, not a new validated strategy.

Best research candidate: historical-level CALL + both futures OI increasing (25 trades /10 sessions), PF4.507, net INR22,552.41. Excluding best day:23 trades,+INR3,717.61,PF1.578. Earlier11 trades lost INR1,063.04. Do not impose bullish-only trading from a market-direction-biased sample.

Weak observed groups: previous-day high6 trades/-INR4,393.84; previous-day low5/-INR3,852.81; pivotS1 four/-INR2,469.26; opening-range breaks29/-INR2,482.86; PE47/-INR11,374.37. Round-number breaks18/+INR279.35 are marginal after charges. Small cells do not justify permanent deletion. High volume alone is not sufficient: the historical-CALL volume>=2x subset turns negative when the best day is removed.

Only9 trades achieved sampled premium MFE>=20%, across5 sessions. Five were historical-reaction entries; two ORlow; one pivotR1 and one pivotR2. Selecting these after exit is outcome leakage and cannot be used as an entry filter.

Exits and range capture

Exit-reason win rates are retrospective: 'TARGET_2' winning does not make it an entry signal. 86 stops without supporting BANK confirmation account for -INR55,553.73; do not simply remove that safety stop to make backtests attractive. Need sampled path/replay to compare losses avoided and recoveries missed.

Every current trade has post-exit observations. Median observed span29m35s; median largest sampling gap77.5s, worst20m15s. Thus 'best later bid' is an observed mark, not a guaranteed fill, complete range, or permission to hold to hindsight maximum. Every-trade CSV includes post-exit sample count, span, largest gap, best/worst bid, entry delay, actual reward/risk and audit flags.

Example: final14 October PUT exited at163.40 for +INR854.29. A later recorded bid196.25 at13:31:58 shows a further20.1% premium move, but does not establish that a predeclared runner would survive its intervening path. Test a runner only where size permits; one lot cannot be half-booked.

Full deployed existing-strategy rules

1. Data: NIFTY and BANKNIFTY spot1m candles, previous completed sessions, current spot quotes, near-futures1m OHLCV/OI and futures VWAP; selected nearest-expiry option-chain band ATM +/-12 strikes. Actual lot size comes from contract metadata. Quote transport is batched REST, not guaranteed every exchange tick.

2. Session: 09:30 inclusive to15:10 exclusive entry window; force exit15:20 when an executable quote is available. Opening range09:15-09:30. No overnight intent; unavailable quotes can delay exits, so wall-clock force-close is not a guaranteed fill.

3. Signal-break: previous completed close lies on one side of a reference, latest completed close crosses in the reference direction. References: ORhigh/low, previous-day high/low, pivotsR1/R2/S1/S2, nearby round numbers, historical zones with at least2 reactions. Direction for round/historical reference is assigned relative to current spot in candidate construction; audit this separately because live-versus-completed-candle position can alter eligibility.

4. Confirmation: NIFTY EMA20/SMA5, momentum, near-futures VWAP aligned; BANK equivalent confirmation and no opposing rejection. OPTION_CHAIN_PARTIAL is tolerated by confirmation while selected ATM still needs an executable quote. Candidate priority: OR, round, previous-day, pivots, historical, then distance. Priority is not ordered by the profitability findings above.

5. Signal-break structural stop: latest completed candle low forCALL/high forPUT. Nominal risk=absolute(level-stop); T1=level +/-risk, T2=level +/-2risk. These are anchored to the broken reference, not actual entry spot. Current code blocks entries beyondT1 but does NOT enforce correct-side current stop in this path.15-minute cooldown per origin/level/direction.

6. OR/retest alternative: confirmed OR break, subsequent25%-61.8% impulse retrace, reversal candle, later close beyond retest high/low. One transition per completed candle; >90s candle gap resets; setup expiry30m. Reversal can be small-body doji<=15% candle range, meaningful wick>=max(body,40% range), or directional candle. NIFTY/BANK candle timestamps must match in this path. Two distinct forward targets required. Correct-side spot/stop enforced in this path.

7. OR targets: nearest valid forward levels from OR extensions0.5/0.618/1/1.618/2, pivots, round levels, OR, previous-day levels and qualifying historical zones. Deduplicate within2points; skip targets closer than max(2points,5% OR width).

8. Option selection: closest ATM CE forup/PE fordown among selected nearest expiry; corresponding opportunity/Greeks row required. Fresh option quote maximum45s, valid bid/ask and nonzero depth. Ask is paper entry. No verified future queue/impact simulation.

9. Size: one current-version position; account reference INR25,000; premium budget min(INR20,000,realized equity), whole lots, capped by visible best ask lots. Reserve charges, including staged-exit brokerage when multiple lots. This is premium-cap sizing, NOT constant risk-at-stop sizing. A INR20,000 deployment is not a INR20,000 maximum intended risk guarantee.

10. Management: exits use valid bid quotes. Emergency premium loss30%; structural stop exits immediately ifBANK opposing or no supporting confirmation. SupportingBANK can allow breach wait up to180s. Recovery before deadline clears breach; recovery observed after deadline still exits. Stale spot with existing breach uses data-gap timeout path.

11. Profit-taking: T2 closes all; T1 closes floor(half lots), minimum1. A one-lot trade fully exits atT1. Visible bid quantity caps each staged fill; pending exits may span scans; identical quote is not consumed twice. Structural stop does not automatically trail to breakeven afterT1.

12. Net accounting: ask-to-bid premium P&L minus versioned NSE option charges and staged exit fees. Ledger reconciliation gross-charges=net passed all175 within2paise. Fees: flat INR20/order, STT0.15% sell premium, exchange0.03553% both sides, SEBI INR10/crore, stamp0.003% buy, GST18% on brokerage/exchange/SEBI. This is not contract-note reconciliation; price impact/queue latency remain unmodeled.

Fee reference checked15 October: [Zerodha charges](https://zerodha.com/charges/) and [STT revision effective1 April2026](https://zerodha.com/marketintel/bulletin/445377/revision-in-stt-securities-transaction-tax-from-1st-april-2026).

Psychology rules and defects

Model uses price/location/rejection or acceptance, volume-price conversion, OI/futures participation, basket/cross-index, VIX/GEX/gamma and momentum context. Confidence is descriptive, not calibrated probability. Normal sizing20% probe,35% coherent story,60% exceptionally clean story; same-side one add can occur with continued thesis and price/gamma acceleration. Fresh ATM ask, depth and spread<=3% gates apply. Structural invalidation,15:20 force-close, destination, momentum failure and gamma+price fade drive exits. Overnight is not available without reliable global context. New entries are now paused by user request; observations and management remain.

Confirmed code/accounting issues, NOT silently repaired or backdated in this UI delivery:

- Ask entry/bid exit already includes spread; subtracting additional ask-minus-mid again charges another INR3,583.12 in this history. Report both recorded and diagnostic add-back, never overwrite stored P&L.

- All123 CLOSED rows retain nonzero remaining_lots. Status is authoritative; new UI never shows those as open. Needs a separately tested ledger-state correction, not blanket historical SQL.

- Three adds: extra order brokerage and changing weighted-average basis need per-fill accounting; MFE before/after add is not directly comparable if basis/quantity changes.

- Exit takes full quantity at best bid without the existing strategy's visible-size staged-fill simulation. That can overstate executable outcomes.

- Gamma management chooses currentATM gamma by option side rather than necessarily the held contract; ATM migration can contaminate a held-position signal.

- `HOLD_TRAIL` is an action label, not a moving stop. No true partial REDUCE workflow is implemented. Report text must not imply otherwise.

- Current spot freshness is not independently checked inside the Psychology manager's structural exit comparison. Quote freshness for the option alone is insufficient to prove spot freshness.

- 'thesis_correct' is currently netP&L>0, not independently verified causal thesis success. Many narrative report fields remain placeholders; they are not validated post-analysis.

Psychology regime

Psychology event

Psychology exit_reason

Page bugs fixed and scope limits

- 'Trades50' used length of capped recent API rows while performance used all current-version history. Badge now uses the total ledger count, independent of page size.

- Psychology and existing strategy mixed in one tab. Separate #psychology and #trades with ordinary anchor URLs for new browser tabs. Existing #trades links remain valid.

- Existing CSV exported only the visible rows. Full current-version compact CSV now downloads all rows; legacy remains separately recorded and attached to this audit.

- Both historical tables capped at50. New read-only pagination25/page; filters rule version before pagination. No large strategy_context payloads on every table request.

- Psychology one-shot load did not refresh while tab remained open. Refresh now repeats while visible; failures are explicit. New-entry pause visible even before a live market snapshot.

- OPEN exit prices/P&L zeros were misleading; exits and realized results show pending, not fictional zero realized results. Account continues its existing freshness-aware marking behavior.

- Rules no longer all expanded by default on mobile. Existing strategy caption now describes the actual current version rather than only OR/retest.

This was a focused navigation/ledger audit, not proof that every chart and every other engine has no bugs. No entry thresholds, broker submission, risk policy or historical values have been retuned. Existing engine stop-side defect is reported for the next controlled strategy correctness patch rather than smuggled into a UI release.

Tools available versus needed

Available: historical1m level/context database, spot and near-futures feeds, OI/volume/VWAP, option-chain Greeks and executable quote snapshots, exit fills, post-exit samples, GEX/VIX/CAS/pressure dashboards. Current 175 entry contexts show matching NIFTY/BANK completed-bar timestamps, but matching timestamps do not establish complete or fresh quote provenance.12 entries occurred more than60s after trigger-bar completion; nine lack a futures-volume ratio. No fabricated spot traded volume or dealer identity is used in this audit.

Needed for real promotion: immutable trigger/decision/receipt/exchange clocks and model versions, validated current-spot freshness, correct-side structural risk and remaining target room at actual entry, per-fill quantities/fees and actual-symbol gamma, rejected-candidate ledger for ablation, contemporaneous best-bid/ask and depth with slippage/latency stress, independent paper/live shadow fill reconciliation, session-separated validation and operational loss/kill-switch tests. More dashboards or a second data app do not repair those fundamentals.

Upgrade plan, in priority order

1. Correctness patch in isolated PAPER version: reject wrong-side stop entries and stale/mismatched spot/context; require positive actual risk/target room. Keep exact old ledger as control. Fix Psychology accounting and trade-state mechanics only before considering a restart of its paper entries.

2. Freeze three small hypotheses BEFORE collecting another sample: baseline; historical-level + futures-OI formation confirmation; actual remaining reward/risk plus measured cost floor. Side and expiry are stratification variables, not automatic bull-only/expiry-only filters. Do not combine every retrospectively attractive condition.

3. Compare fixed-risk normalized outcomes, not just premium-budget P&L. Model staged fills, cost reserve and extra slippage. Account for the opportunity set changing when trades are skipped; filtered sums in this report are NOT a rebuilt portfolio simulation.

4. Exits: compare existingT1/T2 with one predefined multi-lot runner exit driven by completed structure, not futureMFE. Separate near-expiry behavior from matched volatility/trend non-expiry controls. Never remove stops just because losing trades are labeled stops.

5. Forward collect at least20 additional complete sessions and100 new eligible trades per candidate as a minimum research floor, not an automatic validation pass. Freeze parameters. Cluster inference by session; show false positives, skipped winners, net expectancy, PF, drawdown, MAE/MFE, lead time, fill realism and top-day concentration.

6. Promotion gate: reliable data and accounting; positive cost-stressed forward performance not dependent on one day; acceptable drawdown under explicit risk budget; stable expiry/non-expiry and regime behavior; tested kill switch and recovery. Only then request separate authorization for a capped real-order pilot. `order_submission_enabled` staysfalse.

Verification and reproducibility

Deployed15 October before market open. Mini PID446939 ->464637; relayPID420412 unchanged. Both active, no automatic restarts during acceptance. Public state and Psychology endpoints both return order_submission_enabled=false; Psychology endpoint returns new_entries_enabled=false and0 open positions. Pre-open state honestly reports NOT_STARTED/fresh=false; these checks do not prove today's advancing market ticks.

Desktop1440px and mobile390px Chrome checks: both tabs and direct deep links work;25-row paging changes records; full existing CSV175 rows and Psychology123; no page JavaScript errors or document-width overflow. Physical phone/network behavior was not tested. Full audit is linked in both tabs. Rules expansion is preserved on refresh.

Eight focused pagination/export/read-only/pause tests pass; pause test proves management still executes and new entry is never called. Python compilation and inline/external JavaScript syntax pass. Deployed regression suite103 tests:99 pass,4 pre-existing failures (same failures before changes): two context poll-integration fixtures fail to open their expected trade; low-volume-noise Psychology assertion; opposite-candidate fixture now rejected as destination already reached. The new-entry-account test explicitly enables entry in its test fixture only. No production gate was relaxed to satisfy tests.

All181 existing/legacy and123 Psychology trade IDs are unique. All175 current trades reconcile original lots and gross P&L against recorded exit fills; gross minus stored charges reconciles to net within2paise. This validates internal bookkeeping consistency, not the realism of quote execution. Source code describes the currently deployed rules; absence of immutable per-deployment source hashes means exact historical-code parity cannot be guaranteed merely from an unchanged rule_version string.

Artifacts: analyze.py, build_report.py, ledger.json, post_exit.json, analysis.json, every-trade CSVs, test_changes.py, verify_browser.py, browser screenshots and regression_results.txt in the local audit directory. No raw broker secrets exported. The current report is a fixed capture, while CSV links in the page fetch the live persisted ledger.

Daily results

Psychology daily results

Every trade

Each row below is preserved from the captured ledger. Full CSV files accompany this report. Existing CSV additionally contains causal entry features, actual spot risk/reward, fill count, sampled post-exit statistics and per-trade audit notes. Original full nested histories were not exported: selected context arrays were bounded to12 items; entry confirmation, targets and scalar fields used in this analysis were retained. No future fields enter entry-feature groups. MFE/exit reasons/post-exit movement are outcome diagnostics only.

Current strategy - all 175 trades

Psychology - all 123 trades

Legacy - all 6 trades