proof #006gold / public live record31 Oct 2026
Umar Punjabi LIVE said:
#36 · 13 Apr 2026
He invited viewers to check his record against his streams. We checked all 61 published rows, then tested the sweep rules he states. Two questions, two separate sets of evidence.
Our reconciliation of his list with his public streams. These counts are not a strategy win rate.
12 were on stream but listed differently, including two rows against himself. The 19 include off-air calls, unavailable streams and trades he says he did not take. 13 remain unresolved and are not counted as disproven.
Check all 61 rows ↓Record captured 29 October 2026. Labels use his own words on that day's public stream, or his first review if a trade was still open. They do not claim a trade never happened elsewhere.
These are the actual files that produced the test. Open a step to see the implementation.
id,date_as_listed,pair,direction,result,rr_as_listed1,Jan 21,XAU/USD,Buy,TP,1:22,Jan 27,XAU/USD,Sell,TP,1:23,Jan 28,XAU/USD,Buy,TP,1:24,Feb 17,XAU/USD,Sell,TP,1:45,Feb 19,XAU/USD,Buy,SL,—6,Feb 27,XAU/USD,Sell,TP,1:27,Mar 03,XAU/USD,Sell,TP,1:28,Mar 05,XAU/USD,Sell,TP,1:29,Mar 06,XAU/USD,Sell,TP,1:510,Mar 09,XAU/USD,Sell,TP,1:211,Mar 10,XAU/USD,Buy,TP,1:712,Mar 11,XAU/USD,Sell,TP,1:213,Mar 17,XAU/USD,Sell,TP,1:214,Mar 18,XAU/USD,Sell,TP,1:315,Mar 19,XAU/USD,Sell,TP,1:3def v2_label(e: dict, result: str, direction: str, rr) -> tuple[str, str]: """ONE rule for every row (RECORD_AUDIT.md §1). His words on that day's public stream decide; price never does. MATCH = he took that side and his final stated outcome equals the list (RR within his partial convention: the list may book the RR his runner reached while he still held part of the position).""" V = rr if result == "TP" else (-1.0 if result == "SL" else 0.0) if e["taken"] == "N": return NOTON, SUB_NOTON[e["not_taken_why"]] if e["needs_frame"] == "Y": return PEND, SUB_PEND[0] if e["side_on_stream"] and e["side_on_stream"] != direction.upper(): return DIFF, "other side" o = e["his_outcome"] B, H = _num(e["his_rr_booked"]), _num(e["his_rr_reached"]) if o in ("OPEN", ""): return PEND, SUB_PEND[1] if o in ("TP", "MANUAL_PROFIT", "MANUAL_LOSS") and B is None:"""Umar Punjabi's liquidity-sweep strategies, coded as he states themon his live streams. One engine, two presets: ASIA_EXACT S1, the "Asia sweep" he defends in stream #136: Asia high (low) swept, a 5-minute break of structure, with the 4-hour bias, before London. LONDON_SWEEP_NY S2, his Aug-Oct live checklist on gold: London high (low) swept, a 5-minute break, a retest entry, in the first two New York hours.His words sit next to the rule they define: S<n> = live streamnumber and time (projects/.research/umar-live/04_strategy_rules.md;[W] = Whisper reading), VA = "How To Backtest... (My EXACTStrategy)", m_xfJNbi6_Y. Our readings are marked "ours". All ofthem, and every variant, were written down before the first run(projects/umar-live-audit/prereg/2026-10-29_S1_S2.md)."""The same sweep rules, written a second time on purpose: plain Python lists and minute-by-minute loops, nonumpy tricks. It imports nothing from the `backtest` package: not the rules class, not the data loader, not thenews calendar, not the session clock, not the fill code. It was written from the pre-registration(projects/umar-live-audit/prereg/2026-10-29_S1_S2.md), not from sweep.py.What it reads for itself:- the raw 1-minute file (`np.load`: the integer prices and their scale, nothing else);- the calendar CSVs (`usd_high_impact.csv`, `holidays.csv`), each release time turned into UTC with its own zoneinfo;- the session times (Asia box, London open, NY open, 17:00 flat), again with its own zoneinfo.The rules and the costs come in as plain objects read by field name only (`rules.window_end`, `costs.slippage`,...), so a `types.SimpleNamespace` works as well as the clean module's own classes.tests/test_sweep.py runs both over the whole gold history and requires the same trades, field by field, withexact float equality. Two separately written programs agreeing on every trade is the check that the clean moduledoes what the pre-registration says. (It cannot show that the pre-registration reads his rules correctly.)Gold: Dukascopy one-minute BID and ASK, January 2015–October 2026. Session clocks use their local daylight-saving rules.
Full costs include the spread at each fill, $0.05 slippage and $0.07 per ounce round-trip commission. These model a raw-spread account, not his exact broker fills.
Curves use weighted_R from the original CSV, including half-risk equal-sweep trades. Chart CSVs add date, entry_t, exit_t and result_R aliases; original trade files remain under data/backtest/trades/all_versions.
The complete variant result JSON and every available version's trade CSV are downloadable. No losing variant is hidden. Raw multi-year price archives are not bundled; reproduction requires sourcing the documented Dukascopy data.
Full method and limitations ↗One version at a time. Select another reading to see what changes. Every losing version stays in the table.
His stated sweep rules, with named alternative readings. Full cost, zero cost and optimistic fills are separate versions. These are mechanical tests, not a replay of his discretionary personal trading.
244 trades · 35.2% net-positive · -0.179R per trade · PF 0.716
Curve = cumulative R, not dollars or a compounded account. Prepared from the downloadable CSV, with bucket extrema retained. Win-rate reference uses a separate percentage scale.
| Receipt | |||||
|---|---|---|---|---|---|
| 268 | 39.2% | -0.06 | 0.896 | CSV ↓ | |
| 837 | 36.2% | -0.102 | 0.831 | CSV ↓ | |
| 407 | 37.6% | -0.091 | 0.845 | CSV ↓ | |
| 560 | 29.3% | -0.184 | 0.748 | CSV ↓ | |
| 366 | 36.1% | -0.122 | 0.802 | CSV ↓ | |
| 348 | 38.5% | -0.096 | 0.825 | CSV ↓ |
Yearly totals in R. Green above zero; red below. The last year is partial (data ends 2026-10-26). The raw JSON keeps every year’s win rate and trade count.
Exact counts from this CSV. Longest consecutive stop-loss run: 15. A day-end loss is not counted as a stop.
His stated sweep rules, with named alternative readings. Full cost, zero cost and optimistic fills are separate versions. These are mechanical tests, not a replay of his discretionary personal trading.
436 trades · 24.1% net-positive · -0.353R per trade · PF 0.569
Curve = cumulative R, not dollars or a compounded account. Prepared from the downloadable CSV, with bucket extrema retained. Win-rate reference uses a separate percentage scale.
| Receipt | |||||
|---|---|---|---|---|---|
| 453 | 23% | -0.39 | 0.531 | CSV ↓ | |
| 583 | 22.5% | -0.398 | 0.522 | CSV ↓ | |
| 375 | 28.8% | -0.213 | 0.723 | CSV ↓ | |
| 1,372 | 25.4% | -0.313 | 0.611 | CSV ↓ | |
| 660 | 26.1% | -0.294 | 0.632 | CSV ↓ | |
| 649 | 21.7% | -0.462 | 0.472 | CSV ↓ |
Yearly totals in R. Green above zero; red below. The last year is partial (data ends 2026-09-21). The raw JSON keeps every year’s win rate and trade count.
Exact counts from this CSV. Longest consecutive stop-loss run: 14. A day-end loss is not counted as a stop.
His stated sweep rules, with named alternative readings. Full cost, zero cost and optimistic fills are separate versions. These are mechanical tests, not a replay of his discretionary personal trading.
34 trades · 35.3% net-positive · -0.169R per trade · PF 0.783
Curve = cumulative R, not dollars or a compounded account. Prepared from the downloadable CSV, with bucket extrema retained. Win-rate reference uses a separate percentage scale.
| Receipt | |||||
|---|---|---|---|---|---|
| 36 | 33.3% | -0.194 | 0.756 | CSV ↓ | |
| 39 | 30.8% | -0.343 | 0.614 | CSV ↓ | |
| 34 | 23.5% | -0.539 | 0.424 | CSV ↓ | |
| 115 | 28.7% | -0.369 | 0.578 | CSV ↓ | |
| 52 | 28.8% | -0.379 | 0.565 | CSV ↓ | |
| 49 | 28.6% | -0.517 | 0.481 | CSV ↓ |
Yearly totals in R. Green above zero; red below. The last year is partial (data ends 2026-10-19). The raw JSON keeps every year’s win rate and trade count.
Exact counts from this CSV. Longest consecutive stop-loss run: 6. A day-end loss is not counted as a stop.
Net-positive trades in our mechanical readings, 2015–October 2026, after costs. His private key level and discretionary overrides are not reproduced. The 61-row stream reconciliation is separate.
Short source excerpts, not our paraphrase. Hinglish quotations are translated; timestamps let you hear the original. Re-uploads are labelled.
agar aap per trade 2% risk kar rahe ho, to aapka 36% returns banta hai
#100 · 7 Aug 2026
Watch on YouTube ↗ 22:02YouTube loads only after you press play. A source may disable embedding or become unavailable; the original timestamp link remains available.
The direction below comes from his page's own data; the public page does not show a Buy/Sell column. Original Hindi words, readings and caveats are retained. English renderings in brackets are our translations.
The same standard is applied to wins and losses. At least 11 groups of wins and at least 6 losses taken on air have no row either.
Unresolved chart frames and speaker checks remain documented in the downloaded audit. New evidence can change a label.
| Row | His list | Our reading | Stream and timestamp | His words and limits |
|---|---|---|---|---|
| #1 · Jan 21 | Buy · TP · 1:2 | MATCH | #2 eYEBAl01m_Q · 00:59:39.7; 01:06:07.9 (spot); 01:07:37.2 (spot) ↗ | 00:59:37 'देखो भाई 36 पिप्स का था हमारा, 68 पिप्स का हो गया, 1-2 भी हो गया' [36-pip risk, now 68 pips, 1:2 done too] … 01:06:05 'मैंने आपको 1:2 दिला दिया है, इनफैक्ट 1:3 दिला दिया है … 93 पिप्स' [I got you 1:2, in fact 1:3, 93 pips] … 01:07:37 'चलो ब्रेक इवन हिट हो गया, I am out at breakeven, I have booked some profits here' [BE hit, out at BE, booked some profit] reached = 93 ÷ 36 pips (his numbers) = 2.58; list books the RR reached while he held the runner (his partial convention). v2.1: W main pass agrees at 00:59:36.9–00:59:43; it looped at 01:05:48–01:08 → spot re-read spot_r1a (hi+en) agrees: '1:2 … in fact 1:3 … 93 pips', 'I am out at breakeven, I have booked some profits'. |
| #2 · Jan 27 | Sell · TP · 1:2 | NOT ON THAT DAY'S PUBLIC STREAM · setup left for viewers, not taken | #3 WLdNf6hK28k; #4 LdYeQKl9yk0 · #3 00:50:02.1; #3 00:50:25.8; #4 00:05:29.5 ↗ | #3 00:50:02 'अभी एक दो घंटे में मिलेगा या ऐसा भी हो सकता है आज पूरा दिन trade ना मिले' [maybe no trade all day] … 00:50:26 'short मार सकते हैं … but again you are in a long term bullish market, short मारना भी risky होगा, unless … London low break' [conditional short only] … #4 00:05:29 'कितने लोगों ने कल का शॉर्ट ट्रेड लिया एंड प्रॉफिट बुक किया … yesterday was a solid trade' [who took yesterday's short and booked profit?] v2.1: no entry word by him on #3 (W+C: only a conditional short 'unless London low breaks'; #3 ends 00:53:31 with a Telegram update promised). The refresh had compared the #4 quote with #3 (wrong stream, recall 0.12); on #4 Whisper agrees at 00:05:29.5. |
| #3 · Jan 28 | Buy · TP · 1:2 | PENDING FRAME · needs a frame | #4 LdYeQKl9yk0 · 00:36:39.8 ↗ | 00:32:30 'चलो भाई 1 1 डन है … आई एम स्टिल इन' [1:1 done, still in] … 00:36:29 'मैंने थोड़े पार्शियल प्रॉफिट्स बुक कर लिए' [booked some partials] 1:2 never stated. ; refresh with tools/record_v2_refresh.py) Whisper hi agrees (auto check, recall 0.71). |
| #4 · Feb 17 | Sell · TP · 1:4 | PENDING FRAME · open at stream end, no later word | #9 tzJ7Rzgu07I · 01:08:27.0 ↗ | 01:02:27 'अब भाई 1:1 हो गया है रिस्क फ्री कर दो' [1:1 done, make it risk-free] 01:05:35 'फुल डीपी आएगा भाई वेट करो' [full TP will come, wait] 01:08:39 'बस भाई डन फॉर द डे' [done for the day] final outcome never stated; 1:4 never spoken (red team §0.2). ; refresh with tools/record_v2_refresh.py) Whisper hi agrees (auto check, recall 0.75). |
| #5 · Feb 19 | Buy · SL · — | LISTED DIFFERENTLY · other side | #10 d6CJrdrt584 · 00:40:14.0; 00:41:08.8; 00:44:32.2; 00:44:55.1 (spot) ↗ | 00:40:12 'यह कैंडल बेरिश क्लोज होगी, एंट्री ले लेना' [this candle closes bearish, take entry] 00:41:08 'चलो मैंने एंट्री बना ली है' [I have taken the entry] … 00:44:32 'आज एसएल गया भाई … today's SL day' [SL hit today] … 00:44:55 'this is our first SL in the last 10 streams' his entry was a SELL (after a bearish candle close at the upper zone, W+C); list Buy SL (outcome agrees, side differs). v2.1: W main pass looped 00:44:01–00:44:16 and 00:44:58; spot re-read spot_r5 (hi+en) agrees 'first SL in the last 10 streams'. |
| #6 · Feb 27 | Sell · TP · 1:2 | PENDING FRAME · needs a frame | #12 QDYscNS5DkA · 00:42:28.3 ↗ | 00:40:29 'I don't think I will take a trade' … 00:42:28 'यह यू टीपी है भाई' [this is TP] RR not spoken; whether he was in is unclear. ; refresh with tools/record_v2_refresh.py) Whisper hi agrees (auto check, recall 0.67). |
What is promoted, what is promised, and what the primary sources actually say. These are published-source observations, not findings about anyone’s private accounts.
Our source audit counted 116 live descriptions containing both links. In stream #138, he says to join VIP and deposit $200 first. That documents a condition he states, not his commission amount or how he is paid.
On 11 September he says a $1,000 Fortress deposit gives access to his key-level indicator. We cannot test its undisclosed construction or assume the public sweep checklist reproduces his private trades.
Primary-source pages checked October 2026. Their contents can change; the receipts record what this audit used.
Umar Punjabi LIVE or their team are welcome to respond. We will publish the response here. Found a mistake in a quote, rule or trade? Send the source and the row; we will check it and log the correction publicly.
Send evidence or a response ↗Published 31 October 2026 from the 30 October v2.1 pack. Corrections are welcome with a row number, stream and timestamp.
The rules. The code. Every trade. Browse small previews here, or take the full research pack with you.
Public receipts · published 2026-10-31
# realshyt / Umar Punjabi LIVEPublished 31 October 2026. Record reconciliation v2.1: 17 matches, 12 listed differently, 19 not on that day's public stream, 13 pending frames.This preserves working research notes and their outstanding checks. See METHOD.md, NUMBERS_TRACE.md and data/record_audit/RECORD_AUDIT.md for the limits. No claim that an email was sent. Umar Punjabi is welcome to reply at bilal@realshyt.com.Files- data/his_record: his published lists, read 29 October.- data/record_audit: all 61 row labels, words, sources and unresolved checks.- data/backtest: original results, calendars, pre-registration and original trade CSVs.- trades: chart-ready copies of every available version; result_R is weighted_R, date is day, entry_t/exit_t alias the original UTC times.- code: actual implementation, runner scripts and tests. Supply the documented raw Dukascopy price archives to re-run; those large archives are not bundled.76 versions have CSVs and charts. Each trade count and weighted result was checked against its source JSON. No backtest was rerun or tuned to publish this page.Not financial advice. Our test, our opinion. His private key-level indicator and discretionary decisions are outside the mechanical test.Copy copies the visible page of source. Raw and Download contain the entire file.
# realshyt / Umar Punjabi LIVE
Published 31 October 2026. Record reconciliation v2.1: 17 matches, 12 listed differently, 19 not on that day's public stream, 13 pending frames.
This preserves working research notes and their outstanding checks. See METHOD.md, NUMBERS_TRACE.md and data/record_audit/RECORD_AUDIT.md for the limits. No claim that an email was sent. Umar Punjabi is welcome to reply at bilal@realshyt.com.
Files - data/his_record: his published lists, read 29 October. - data/record_audit: all 61 row labels, words, sources and unresolved checks. - data/backtest: original results, calendars, pre-registration and original trade CSVs. - trades: chart-ready copies of every available version; result_R is weighted_R, date is day, entry_t/exit_t alias the original UTC times. - code: actual implementation, runner scripts and tests. Supply the documented raw Dukascopy price archives to re-run; those large archives are not bundled.
76 versions have CSVs and charts. Each trade count and weighted result was checked against its source JSON. No backtest was rerun or tuned to publish this page.
Not financial advice. Our test, our opinion. His private key-level indicator and discretionary decisions are outside the mechanical test.
Start with a row in the 61-row table. Open the timestamp and compare his words with the preserved list. For a backtest trade, use its UTC minute, direction, bid/ask convention and weighted risk in METHOD.md.
The server samples uniformly from every trade in this version. The complete CSV stays out of your browser.
Times: All normalized trade timestamps are UTC. Session boundaries use London and New York local clocks with daylight saving.. Prices are CFD prices; costs can explain differences from the chart.
Open every trade in this version ↓A name is a start. A public claim with its source is better. Comment RECORD on the audit reel and tell us what you want checked next.
Our test and our opinion of it. Short public excerpts used for commentary and review. Educational content, not financial advice. Past results do not predict future results.
All investigations ←