The Truth Twin

One decision, four versions of truth. The frozen research rule, what operations actually saw, what the executor suggested, and what the broker confirmed. Modeled P&L never becomes actual P&L by changing a label.

Current verdict: this was a blind suggestion machine, not an underperforming live account. Broker-confirmed P&L is ₹0.00. The healthy replay is −₹1,896.17, but it remains counterfactual.

Audit window 2026-07-13 to 2026-07-15. Generated 2026-07-15T18:45:38+05:30. Capital base ₹200,000. Probability, not certainty.

Which percent per day is which?

Advertised base

+1.331% per day

+₹2,662.00 on ₹2 lakh. Survivor-inflated research ceiling.

Advertised aggressive

+2.570% per day

+₹5,140.00 on ₹2 lakh. Configured Tier 2 model, never observed.

Robust-core research

+0.752% per day

+₹1,504.00. Weaker and unapproved headline sleeves removed.

Conservative planning base

+0.526% per day

+₹1,052.80. Seventy percent realization applied to robust-core research.

Derived aggressive planning

+0.854% per day

+₹1,707.04. conservative planning base multiplied by 1.8, less 0.094% per day funding estimate.

Current executable rate

NOT VALIDATED

Authorized capital is 0.0%. There is no defensible positive expected daily rate yet.

Why the rate fell: removing weaker and paper sleeves cuts 1.331% to 0.752%; the 70% realization prior cuts that to 0.5264%. Execution can reduce it further. The 3-session first-full-minute sample returned -3.469% in total, or -1.156% arithmetically per session, but it is not a shared-pool backtest or broker rate.

Decision funnel

Frozen specification97funded decisions
Operations saw2valid funded decisions
Executor suggested2valid intents
Broker confirmed0fills
94recorded healthy-replay entries
3specification entries missing from code
96closed replay trades
-0.948%counterfactual account return

Prospective truth recorder

WAITall four operational slots
WAITbid, ask and quote coverage
WAITDhan order and trade books
INCOMPLETEbroker-grade session truth

Session not started. A day becomes complete only when morning, cover, lunch and PM runs were captured, every intended ticket has a quote/depth snapshot, and Dhan orders plus trades reconcile. Missing authentication remains missing evidence.

SlotCapturedMachine sawExecutor keptSuppressedRun state
Waiting for the first prospective session.

Where decisions were lost or invented

CauseDecisionsSettledPendingModeled netRisk notionalEvidence

One shared ₹2 lakh account

−₹1,896.17unconstrained replay
−₹2,117.03strict shared-pool replay
−₹220.86capital-allocation difference
+₹191,641.15pending filled notional

Analytical fractional sizing, exits processed before entries at the same timestamp. This is a no-leverage safety case, not a claim about broker margin.

Extra round-trip slippageUnconstrained netUnconstrained returnShared-pool netShared-pool return
0 bps−₹1,896.17-0.948%−₹2,117.03-1.059%
5 bps−₹2,529.78-1.265%−₹2,666.12-1.333%
10 bps−₹3,163.39-1.582%−₹3,215.21-1.608%
15 bps−₹3,797.00-1.898%−₹3,764.30-1.882%

Did restored coverage add executable quality?

No, not on this sample after executable timing. The 75 resolvable restored decisions move from modeled +₹1,949.34 to a first-full-minute proxy of −₹3,433.73. The strict live-spec subset moves from +₹1,123.14 to −₹4,482.30. Restoring sight increased opportunity count, but did not preserve executable alpha.

Research-only restored result

+₹2,047.94

76 settled decisions at research timestamps. This is diagnostic evidence, not proof of quality.

Executable-timing restored result

−₹3,433.73

75 comparable decisions, a −₹5,383.07 change versus their modeled fills.

Strict live-spec timing result

−₹4,482.30

55 comparable live-spec decisions. Verdict: quality_not_added_after_executable_proxy.

Liquidity at ₹2 lakh

+0.249%

Declared daily-liquidity floors are not the binding problem. Opening timing, spread, queueing and broker shortability remain the execution gates.

Can the machine really get the displayed price?

Official-open fills are overestimated. On 95 of 96 closed trades, replacing research timestamps with the first full positive-volume minute after intent changes the comparable result from −₹1,994.77 to −₹6,938.78. This is still a proxy, not broker P&L.

NUVOCO example

+₹514.96 → +₹89.74

The replay books ₹308.65. The first full tradable-minute proxy is ₹318.80, a 3.29% adverse move before spread or queueing.

Downcap fade

−₹504.14

Modeled +₹4,608.23 on the same resolvable rows. The opening bounce happens before a post-open order can reach the market.

All comparable trades

-3.469%

Yahoo one-minute first positive-volume bar open in the first full minute after each intent; modeled stop exits retained

Remaining uncertainty

1

Unresolved trade. Queue position, spread and within-minute adverse selection are still absent, so the proxy may remain optimistic.

How far can we honestly backtest?

Common ten-sleeve research portfolio

2021-02-01 to 2026-04-08

Longest common window with mature daily features and exact retained F6 minute stop paths.

research_proxy One INR 2 lakh account, overlapping holding periods, family caps, deterministic deduplication and mark-to-market.

All 19 earning-sleeve research proxy

2024-01-01 to 2026-04-08

C11, S3A and W24A validated history starts in 2024; exact F6 minute paths end 8 April 2026.

research_proxy Treat calendar and dormant sleeves as valid zero-event periods, not missing trades.

Complete current operational machine

No historical window

Historical pre-open IEP, 15:22 tapes, surveillance state, results/ban state, bid-ask, exact selection order and broker fills were not all retained.

not_available_historically Begin prospectively only after every Truth Twin input recorder passes completeness. Earliest possible start after this audit is 2026-07-16.

Outer data boundary, not a claim: 2015-02-01 to 2026-04-08. NIFTY100 one-minute archive. Pre-2021 observations are regime context and exploratory stress data, not the production validation window.

Research exact means exact to retained rules and bars, not broker fills. Research proxy means at least one live input, timestamp, universe, spread or fill is approximated. A complete operational backtest cannot be created retroactively from data that was never recorded.

EdgeObserved tradesAvailable dataQualityReproducibilityBinding limitation
downcap_fade2021-01-28 to 2026-06-292021-01-01 to 2026-07-07research_proxycore_yes_complete_noOfficial open proxies pre-open IEP; historical surveillance and dynamic F&O membership are incomplete.
div_runup2021-01-15 to 2026-07-062021-01-01 to 2026-07-07research_proxypit_research_yes15:20 entry and exit use close proxies rather than broker fills.
midcap150_print_pop2021-03-26 to 2026-03-272021-03-01 to 2026-03-31research_exactcovered_cycles_onlyTen usable cycles; September 2021 source PDF is image-only and missing.
banentry2021-01-06 to 2026-06-022021-01-01 to 2026-07-08research_proxyresearch_yes_operational_partialPrice coverage about 84%; historical live-feed freshness cannot be reconstructed.
c11_straddle2024-01-04 to 2026-07-072024-01-01 to 2026-07-08research_proxyvalidated_2024_onwardFixed-strike paths are exact but bid-ask and fills are modeled. Raw 2021-23 data is outside the frozen validation scope.
dc_night2021-02-01 to 2026-07-062021-01-01 to 2026-07-07research_proxycore_yes_operational_proxyMachine entry is 15:22 rather than official close; historical results/surveillance vetoes are incomplete.
dc_pop2021-01-15 to 2026-07-022021-01-01 to 2026-07-07research_proxyresearch_yes_current_implementation_noCurrent at-most-six selection is not deterministically ordered; entry is 15:22.
banspell2021-01-06 to 2026-06-252021-01-01 to 2026-07-08research_proxystate_machine_yes_live_seen_prospectiveHistorical transitions exist, but the VM's frozen-on-feed-failure state can only be observed prospectively.
nightbook2021-02-01 to 2026-07-062021-01-01 to 2026-07-07research_proxycore_yes_complete_operational_noEntry is 15:22 versus research close; 2024 delivery and historical veto data have gaps.
exdiv2021-01-21 to 2026-07-072021-01-01 to 2026-07-07research_proxyresearch_yesCorporate-action and yield gates are PIT-clean, but opening and cover fills remain modeled.
special_div_exday2021-01-22 to 2026-02-062021-01-01 to 2026-07-07research_exactresearch_exact_unimplementedNo qualifier after February 2026; executor is not implemented.
event_basket2016-11-09 to 2026-06-192016-07-04 to 2026-07-07research_proxycore_a_yes_current_operational_proxyStatic universe survivorship and 15:22 entry versus research close; newer Friday veto differs.
f6_fade2021-01-01 to 2026-04-082021-01-01 to 2026-04-08research_proxyminute_path_yes_live_decision_noExact one-minute stop path exists, but historical auction IEP/quantity and current index/cost conventions do not.
fyend_taxloss2022-03-31 to 2026-04-302021-01-01 to 2026-04-30research_proxyresearch_exact_calendar_armedFive fiscal-year events on a survivorship-biased panel; matched controls share the bias.
w24a2024-01-12 to 2026-06-012024-01-01 to 2026-06-01research_proxypartial_only167 of 316 targets survived coverage/validity; no bid-ask and only 53% target coverage.
ipo_unlock2021-03-08 to 2026-06-222021-02-01 to 2026-07-07research_proxyresearch_yes_current_incompleteListing+30-day heuristic can differ from prospectus locks; live grader still uses next-open instead of T+10 close.
s3a_straddle2024-01-04 to 2026-07-072024-01-01 to 2026-07-08research_proxyvalidated_2024_onwardFixed-strike minute paths are exact but costs/fills are modeled; 2021-23 raw data was outside the frozen study.
ipo_listing_runner2021-02-03 to 2025-12-052021-01-01 to 2026-07-07research_proxyresearch_yes_dormantIPO census inferred from first appearance can include demergers/relistings; no 2026 qualifiers.
mpb_low2021-03-19 to 2026-07-012021-01-01 to 2026-07-07research_proxyresearch_yes_current_machine_noResearch uses full target/stop/15-session outcomes; live paper grader is only next-open interim.

How many independent bets?

Not 19 proven independent alpha streams. Only 5 earning sleeves have comparable dated returns in this matrix. They form 3 positive-dependence components. The remaining 14 sleeves are unproven, not uncorrelated.
3measured positive-dependence components
4.56statistical return dimensions
4broad economic hypotheses across 19 sleeves
0%capital authorized by correlation

The 4.56 figure is an eigenvalue participation ratio, not a count of executable edges. Idiosyncratic noise can increase statistical dimensions. F6 is the retained headline proxy cousin, not the exact live index-excess decision rule. Correlation cannot rescue the failed fill audit.

ComponentSleeves sharing riskEconomic originCo-active correlationBottom-decile overlap
1dc_night, nightbookovernight_price_action+0.3842.2%
2downcap_fade, f6_fademarket_microstructure+0.4740.9%
3exdivcalendar_forced_flowsingletonsingleton

Calendar correlation, inactive days set to zero

Sleevedowncap_fadef6_fadenightbookdc_nightexdiv
downcap_fade+1.00+0.24+0.08+0.09-0.14
f6_fade+0.24+1.00-0.01+0.05-0.00
nightbook+0.08-0.01+1.00+0.38-0.02
dc_night+0.09+0.05+0.38+1.00-0.05
exdiv-0.14-0.00-0.02-0.05+1.00

Common window 2021-02-01 to 2026-04-08, 1280 sessions. Unmeasured: div_runup, midcap150_print_pop, banentry, c11_straddle, dc_pop, banspell, special_div_exday, event_basket, fyend_taxloss, w24a, ipo_unlock, s3a_straddle, ipo_listing_runner, mpb_low.

What the 39 ships really are

19direct or candidate earning sleeves
6live-spec earning sleeves
11risk, veto or sizing rules
9meta, infrastructure or knowledge items
Rate honesty: paper sleeves contribute +0.572% per day, 42.9% of the displayed +1.332% sum. The published +2.570% Tier 2 number is modeled, not observed.
Economic clusterDisplayed contribution/dayConcentration
overnight_momentum+0.555%+41.7% of displayed headline
open_gap_fade+0.484%+36.3% of displayed headline
dividend_cycle+0.179%+13.4% of displayed headline
ban_cycle+0.096%+7.2% of displayed headline
panic_recovery+0.018%+1.4% of displayed headline

Five strongest research candidates, zero cleared for capital

PRESERVE_CASH_NO_SLEEVE_CLEARS_EXECUTABLE_GATE. Recommended unallocated capital: +100.0%. A high research rank is a queue for validation, not permission to scale. Related sleeves must ultimately share one component-level risk budget.
Research rankSleeveEvidence depthRobust contribution/dayExecutable sampleExecutable proxy netCapital stateShares future cap withNext gateAllocation
1downcap_fade18,121 qualifying observations across 234 days+0.144%24−₹504.14HOLD_EXECUTION_FAILEDdowncap_fade, f6_faderepair timing and fill assumptions0.0%
2f6_fade672 deployed-spec days+0.133%20−₹2,295.85HOLD_EXECUTION_FAILEDdowncap_fade, f6_faderepair timing and fill assumptions0.0%
3nightbook1,331 nights+0.213%17−₹2,411.29HOLD_EXECUTION_FAILEDdc_night, nightbookrepair timing and fill assumptions0.0%
4dc_night1,336 nights+0.206%18+₹55.98PAPER_ONLYdc_night, nightbookcomplete forward promotion gate0.0%
5exdiv1,278 scheduled observations+0.056%2−₹2,292.84HOLD_EXECUTION_FAILEDexdivrepair timing and fill assumptions0.0%

Paper promotion queue

RankCandidateStateForward countOutcome gateIndependence gateBlocker
1div_runupBLOCKED_INFRA0/20closed forward-paper fills across at least five event dates, net mean above zero, no missing commanded exitsREQUIRES_DATED_RETURN_SERIES_AND_DEPENDENCE_MAPno formal promoter and executable feed
2w24aBLOCKED_INFRA0/20mean premium P&L above zero and worst-loss to mean-win ratio at most 6REQUIRES_DATED_RETURN_SERIES_AND_DEPENDENCE_MAP15:20 schedule and live executor
3ipo_unlockBLOCKED_INFRA0/20full T+2-close to T+10-close forward mean above zeroREQUIRES_DATED_RETURN_SERIES_AND_DEPENDENCE_MAPfull-hold grader and router
4c11_straddleBLOCKED_INFRA0/20valid Thursday observations with exact 1.6x stop and 15:15 exit, net mean above zeroREQUIRES_DATED_RETURN_SERIES_AND_DEPENDENCE_MAPlive straddle executor
5mpb_lowBLOCKED_INFRA0/20full target, stop or 15-session outcomes with net mean above zeroREQUIRES_DATED_RETURN_SERIES_AND_DEPENDENCE_MAPfull swing-hold grader

Only closed forward-paper recorder outcomes count. Healthy replay, missed candidates, pending rows and interim proxy grades count as zero evidence. Every current candidate also needs a dated return series before the allocator can decide whether it adds a new bet or duplicates an existing family.

Decision-level audit

DateSymbolSleeveFamilySideModeled P&LResearchOpsExecutorBrokerCause

Validation

InvariantResultEvidence
event_id_uniquePASS142 events
decision_id_uniquePASS130 decisions
catalog_39PASS39 catalog items
headline_19PASSdirect/candidate edge count
shared_cap_never_exceededPASSmax 200000.00000000006
slippage_monotonePASS0/5/10/15 bps
no_modeled_pnl_in_broker_layerPASSbroker events 0
suggest_never_fillPASSsuggestions cannot have fill quantity
prospective_intent_has_operational_parentPASS0 prospective intents
broker_fill_has_exchange_identityPASS0 confirmed fill events
promotion_forward_onlyPASShealthy replay and missed paper rows excluded
family_map_scope_research_onlyPASScorrelations cannot be presented as broker evidence
family_map_missing_is_not_independencePASS14 unmeasured sleeves explicitly remain unproven
family_map_cannot_authorize_capitalPASSresearch diversification does not override execution gates