We Tested The Rumers' SPY Opening-Bell Claim

1. Clock-only SPY claim and cannot-tell answer; spoken fixed comparators distinguished from printed rates and reported W/L/F counts; exact dates/ledger unknown and window mismatch inferred.

WE CANNOT TELL - within our predeclared tolerance on this window/proxy

The Rumers, a source who teaches trading, quote SPY opening-bell win rates in their video If You Only Have $50, Do This Every Morning (The 3R Rule) (published 2026-09-19): “You'd win 51% of the time. Two hours later, 53% a full session, at the end of the day, 54%.” Those words are at 1422.88-1429.52 seconds of their video's saved captions (T:613-615). Times quoted from their video are positions in their video, not in our film.

We tested only that clock claim: buy SPY at the 9:30 a.m. ET regular-session open, then sell after one hour, after two hours or near the close, as three alternative exits. The spoken 51%, 53% and 54% were registered as fixed comparators before any outcome was seen (P:17; P:176).

Their table, at 1424 seconds of the same video, prints 51.9% / 53.2% / 54.1% beside 10:30 a.m., 11:30 a.m. and 4:00 p.m. (1 hour, 2 hours, full session), with reported wins / losses / flat of 1,945 / 1,784 / 18, 1,991 / 1,742 / 13 and 2,025 / 1,704 / 13. Those rows sum to 3,747, 3,746 and 3,742; that arithmetic is consistent with the printed rates, but we have not verified a source formula or any of the trades. The table gives counts, not exact dates or the underlying trades. Our registered bands keep the spoken 51 / 53 / 54%: the registered bands were not re-centred on the printed rates. A separately labelled EXPLORATORY check after results, re-centring the same two-point bands on 51.9 / 53.2 / 54.1%, also gives cannot tell (EXPLORATORY). It was not preregistered, is not a second verdict and does not replace the lock (TBL:15-42; N:350-360; N:382-390).

Our registered window is 2011-07-01 through 2026-05-29. Their “last 15 years”, with the 2026-09-19 publication date, suggests roughly September 2011 to September 2026: that span is our inference from their wording and the publication date, not exact source dates. Our window starts and ends earlier and leaves out the later summer and September portion. Similar sample sizes do not identify overlapping dates, paired outcomes or independence (P:32-33; T:605).

This tests the clock rule, not The Rumers' full setup. They describe “$50 worth of risk”, not account size, and say a stop “has to be structurally located” (T:60; T:117-119). Their quoted clock rule has no stop, and neither does this test. We did not test their trend, liquidity or range setup, structurally placed stops, indicators, scaling, discretionary decisions, futures remarks, skill, account or course. The Rumers are credited as the source of the claim we tested.

Our near-close price is the last eligible print of the 15:59 regular-session minute: a proxy, not a certified 16:00 closing-auction fill. Recorded minute prices stand in for fills, the prices an order actually receives (P:55; R:1012).

2. All three point/range/band rows, counts and method limits; separate above-half question.

Wins require a strictly higher stored exit than entry. Losses and ties stay in the trade denominator. Only the three primary SPY rates receive the registered sampling bounds. The point label is WITHIN PREDECLARED TOLERANCE; the range label is OVERLAPS BAND (P:59; P:131-138; R:9-15). The point, range and aggregate labels here apply within our predeclared tolerance on this window/proxy; the Above-half column answers a separate question on the same window/proxy (below).

The hourly range reaches above its band's upper edge; the two-hour range extends below and above its band; the near-close range reaches above its band's upper edge. Each range allows values inside and outside the accepted tolerance; none is wholly inside and none is wholly outside. The registered aggregate is therefore cannot tell. All 3 of 3 measured points are within our predeclared tolerance on this window/proxy; that cannot be restated as the three rates being confirmed (R:15-17; P:135-150).

The film shows these rates at one decimal (51.8%, 53.0%, 54.3%) and the ranges as 49.7-53.8%, 50.8-55.3% and 52.2-56.4%. Each is the pinned fraction rounded half-up to one decimal, the precision of their printed table. The six-decimal values in the table above control.

Above half is a different question. The hourly range includes half, so that comparison is UNRESOLVED; the two-hour and near-close lower bounds exceed half, so those are ABOVE HALF. These labels reuse the same three bounds, not another inferential family, and they do not answer a money question (R:11-13; P:137).

The ranges use a calendar-block ratio cluster-t envelope with blocks of 63, 126 and 252 scheduled full sessions, keeping excluded-date coordinates and the final partial block. Each length requires at least 12 occupied blocks, and each of the three members keeps alpha = .05/6. This does not assume independent trading days; dependence beyond the registered approximation, including long-lived market changes, can invalidate the bounds (R:71; P:105-113).

Coverage: SPY 3,712 of 3,716 and QQQ 3,695 of 3,716 full scheduled sessions, both passing the registered overall and yearly floors. The 2011 and 2026 endpoint years are partial, and missing observations need not be random. The worst-case full-calendar frequencies below are missing-session arithmetic, not sampling intervals; no finite missing-return bound is established for mean money (R:19; R:40; P:45; Q1:2-4).

3. Measured money with exact saved branches, stored means, supplementary cents, size and estimated costs per exit.

Size and costs are our assumptions, not measured performance or source rules. Size fixes whole shares at 09:29 ET, before entry: q = floor(50/(0.01 x M)), where M is the immediately preceding scheduled session's last regular-minute close in the entry day's share units. The $50 is a hypothetical stress budget against a 1% adverse move; there is no stop order in this clock-only test, no $50 account and no maximum-loss guarantee. Same-day quantities are reused across the alternative exits. The model subtracts an estimated $0.04 per share round trip plus $0 commission, once: an estimated expense, not measured quotes or verified fills (R:91; P:71-81).

SPY's median quantity was 17 shares and its median entry notional $4,870.16, each over 3,712 primary trades. “Typical” means the stated sample median, not an example trade built from separate medians. Reference notional is capped at $5,000 by construction (q x M at most R/a); the film's “roughly” and “about five thousand dollars” refer to that cap. Eventual entry exposure can exceed the reference cap after a gap. The risk budget does not certify funding, margin eligibility, loss limits or an executable fill. MEASURED; DESCRIPTIVE; NOT INFERRED (DIST:9; DIST:12; P:75).

The registered branch sentences, quoted verbatim from the saved rendering: the win-rate clause is inferred under its conditional method; the money amount is measured; descriptive; not inferred; an enduring positive or negative money conclusion is unknowable from this study.

+60 minutes - cannot tell x measured below zero; full stored net mean = -0.3806646551724053 dollars/trade, N=3712 (FR).

We cannot tell whether all three SPY win-rate comparisons hold within our predeclared tolerance on this window/proxy. Measured after-cost average: $-0.380664655172 per trade over 3712 trades at the stated size and estimated costs; here is the measured shortfall; we cannot prove it is below zero.

+120 minutes - cannot tell x measured below zero; full stored net mean = -0.32539202586206134 dollars/trade, N=3712 (FR).

We cannot tell whether all three SPY win-rate comparisons hold within our predeclared tolerance on this window/proxy. Measured after-cost average: $-0.325392025862 per trade over 3712 trades at the stated size and estimated costs; here is the measured shortfall; we cannot prove it is below zero.

Near close - cannot tell x measured above zero; full stored net mean = 0.3891569773706881 dollars/trade, N=3712 (FR).

We cannot tell whether all three SPY win-rate comparisons hold within our predeclared tolerance on this window/proxy. Measured after-cost average: $0.389156977371 per trade over 3712 trades at the stated size and estimated costs; here is what is left; we cannot prove it is above zero.

These exits are alternative ledgers. Do not add their totals, and do not pick the near-close exit because its measured sign is positive. The full signed averages stay beside the verbatim sentences. The film's cents (-$0.38, -$0.33, +$0.39) are these means at cent precision, sign before the dollar sign (P:59; P:85-91). MEASURED; DESCRIPTIVE; NOT INFERRED.

Break-even cents per share and break-even constant basis points are two different cost models. Both use the gross-minus-commission numerator with their own share or notional denominator; they are not broker tariffs or evidence that a trade can be executed at that expense. Every registered cost cell remains in the result ledger; none is a menu for choosing a favourable result (R:103; P:81-83).

4. C01 corrected SPY conditional classes beside locked predecessors, counts, numerators, means, ratios, medians and residue cause.

C01 is a dated post-run descriptive correction (2026-10-18), not another market run or a registered-method change. An independent review of the findings found floating-point residue in the strict after-cost sign test: exact break-even trades could be labelled tiny wins or losses. The correction recomputes the after-cost classes in exact decimal arithmetic on the saved quantity, entry, exit, residual and commission strings, with no price rounding and the same net >0 / <0 / =0 rule, and with them the class counts, conditional win/loss means, ratios and signed medians; Q1 rates, bounds and labels, unconditional means and percentiles are unchanged. A saved example failed first, with float net -3.1041835768519377e-13 against an exact net of 0; after the correction it passed (C01M).

Extent: 44 of 11,136 SPY and 51 of 11,085 QQQ primary alternative-exit records change class, all to exact zero. These are alternative-exit records, not independent days or a portfolio. The corrected counts match the reviewer's independent derivation on 52 of 52 count fields across 6 of 6 fund/exit rows and both fund totals. The 30 of 30 gross win-rate count and rate fields are unchanged, and 6 of 6 unconditional net means are unchanged to cent precision; full-precision floating and decimal means can differ below a cent (C01:2-7; Q1:2-7; C01M).

For after-cost class counts, conditional win and loss magnitudes, ratios and signed medians, the C01 corrected figures below are the monetary facts. Each sits beside its LOCKED FLOAT predecessor, which is superseded and kept as a receipt. Gross statistics, unconditional averages and the locked percentiles keep their original scope; C01 does not recompute the percentiles, bounds or Q1 labels. Signed medians keep the registered inverse-ECDF convention. Conditional mean win divides by positive trades only, and conditional mean loss magnitude by negative trades only. MEASURED; DESCRIPTIVE; NOT INFERRED.

Every row of the six-row correction download (both funds, all three exits) appears below with its locked predecessor and its own denominators.

Download the six C01 rows (CSV)

5. Concentration: original counts and limits; removing large days never replaces primary.

SPY, primary size and estimated costs. MEASURED; DESCRIPTIVE; NOT INFERRED.

A negative top-ten / total ratio uses a negative total-net denominator; it is not a negative share of profits. A ratio above one means the selected positive sum exceeds the complete ledger's signed total. Removing the largest positive days is a concentration diagnostic, not a replacement sample or a requirement that a valid effect be free of large days, and no changed money sign earns a money verdict. The primary result above is unchanged by it (P:89; CONC:2-4).

6. Later buys (Q3), then B24 common-pair identity; descriptive, no extra verdict.

Later buys (Q3). The later-window grid starts every 30 minutes: 12 hourly windows including the open, or 10 two-hour windows including the open. Paired comparisons require every registered window of that duration to be executable on that date, and compare the opening outcome with the equal-weight mean of its 11 or 9 later windows, using the same daily pre-order quantity for dollars. They do not compare a full-sample open rate with an unmatched control sample (P:95-99). MEASURED; DESCRIPTIVE; NOT INFERRED.

The pooled window fractions keep each slot's valid observations, so their denominators differ from the matched-day denominators. Overlapping windows are correlated, not independent trading histories or a multi-position portfolio. A negative frequency difference beside positive mean differences is an observed distinction between frequency and magnitude; it does not show that opening entries are better, worse or equal (P:97-99).

Overnight and daytime (B24). Each pair uses the immediately preceding scheduled session, with both dates full, in the window and complete; it never jumps across an excluded date. On the common split-adjusted, dividend-exclusive basis, (1 + overnight) x (1 + intraday) = 1 + close-to-close for each admitted pair. The close-to-close leg includes overnight; it is not the same hold as open-to-close. These are parts of one price path, not competing theories, independent corroboration or buy-and-hold returns over the calendar (P:101). MEASURED; DESCRIPTIVE; NOT INFERRED.

7. QQQ is our descriptive extension, with C01/predecessor pairs and its own denominators.

The Rumers name SPY for the quoted clock-only rates. QQQ is our added fund, chosen before returns. There is no QQQ tolerance verdict, above-half label or money interval; every QQQ row is descriptive; not inferred, with the same registered size and cost conventions and QQQ's own pre-order marks, quantities and admitted dates. Similarity between the funds is not independent replication of a source claim about QQQ (P:17; P:142; T:605-607).

C01 also corrects QQQ's after-cost classes for the same floating-residue reason. The gross wins, losses and ties and the unconditional averages above are unchanged. Each row pairs the LOCKED FLOAT predecessor with the C01 figure; only the C01 classes, conditional means, ratios and medians are monetary facts. Positive and negative conditional means use their own class denominators.

QQQ concentration, later-window, pooled-window and overnight/daytime rows above use QQQ's own denominators (3,695 eligible trades; 3,694 of 3,716 paired days; 3,643 of 3,682 adjacent pairs). They carry the same descriptive limits as the SPY rows and add no verdict. MEASURED; DESCRIPTIVE; NOT INFERRED.

8. Locked-design summary, one run, correction custody, limitations, synthetic vs market distinction.

A summary of the plan we locked before the run: its main terms, not the full plan text. The lock commit acc81d8 is held in a private repository; no public ref pins it. Hashes identify records; they are not public timestamp proof.

Before the lock, two pre-run amendments are part of the registered plan. Amendment 1 tried one alternative interval method; it failed its calibration gate and was not adopted. Amendment 2 narrowed inference to the three SPY win rates before any market return was computed, and withdrew money members as verdict clauses while keeping every money output as a measured description (P:158-168; P:172-186).

Synthetic calibration, not market evidence: before the run, the three-member Q1 method passed a direct joint-coverage gate on 20 of 20 supplied synthetic scenarios (minimum 1979 of 2000 histories covered), while an ordinary independent-binomial comparison failed 12 of 20. These are supplied-history simulations, not trading observations (P:220; N:317).

One registered outcome run; no rerun or post-price method change (R:3). Registered sensitivities (boundary, delayed, endpoint-only, exclude-halt-days) were part of that run and showed no change in the primary point-band, point-versus-half or net-sign reading; the arms carry no bounds, so they do not re-test the above-half labels; that is not an agreement vote (N:311). A separate rebuild matched 48 of 48 requested quantities (six fund/exit rows by eight fields) within its stated scope; it does not certify every figure here or C01 (N:325).

After the run, two dated changes are labelled wherever they land. C01 (2026-10-18) recomputes the after-cost classes in exact decimal arithmetic, and with them the class counts, conditional win/loss means, ratios and signed medians; Q1 rates, bounds and labels, unconditional means and percentiles are unchanged, as section 4 shows. A findings amendment added the source's on-screen table: its printed rates and win/loss/flat counts replaced an earlier premise that the source counts were unknown. The measured results, money and C01 sections did not change, and the registered bands were not re-centred on the printed rates. A separately labelled EXPLORATORY check after results, re-centring the same two-point bands on 51.9 / 53.2 / 54.1%, also gives cannot tell (EXPLORATORY). It was not preregistered, is not a second verdict and does not replace the lock (N:350-360; N:382-390; TBL).

Limits: a named-fund window, not a survivorship-free stock-selection exercise or a claim across instruments. The store is split-adjusted and dividend-exclusive. Minute prices are proxies for fills; costs are estimates; the near-close price is not a certified auction print. Dependence beyond the registered block approximation can invalidate the ranges (N:309; P:67).

question

Q1, the only verdict question: do the three quoted SPY opening-bell win rates (51%, 53%, 54%) hold within our predeclared tolerance on this window/proxy? Q2-Q4 and every QQQ row are descriptive; not inferred.

Sources: P:17; P:176

source claim

The Rumers say, in their video published 2026-09-19, that buying SPY at the open and selling after an hour, two hours or at the end of the day wins 51%, 53% and 54% of the time. The spoken percentages are the registered comparators.

Sources: T:613-615; P:176

source unknowns

Exact source dates, the underlying trades, vendor, tie treatment, fill convention and whether costs were included are not supplied. Their on-screen table prints rates and win/loss/flat counts; their formula and trades are not verified.

Sources: TBL:15-24; TBL:47-54

window

2011-07-01 through 2026-05-29 inclusive, fixed before returns. "Last 15 years" with the 2026-09-19 publication date suggests roughly September 2011 to September 2026: our inference, not exact source dates.

Sources: P:32; P:33

eligibility and coverage

Full scheduled regular sessions only (09:30-16:00 ET); early closes excluded by calendar, never by price. Coverage floors: at least 99% of full sessions overall and at least 90% in each represented calendar year, for both funds.

Sources: P:36; P:37; P:45

entry and three alternative exits

Long SPY at the first eligible print of the 09:30 minute; exit at the first print of the 10:30 or 11:30 minute, or the last eligible print of the 15:59 minute (our near-close proxy, not a certified closing-auction fill). One hypothetical trade per session per exit; the exits are alternative histories, never added together.

Sources: P:53; P:54; P:55; P:59

scheduled coordinate blocks 63/126/252

Uncertainty ranges use a calendar-block ratio cluster-t envelope with blocks of 63, 126 and 252 scheduled full sessions, keeping excluded-date coordinates and the final partial block. One trade per session does not make signs independent.

Sources: P:105; P:107; P:109

12 occupied-block minimum

Each block length requires at least 12 occupied blocks, a positive denominator and finite nonzero variance; otherwise the range is withheld.

Sources: P:111

alpha .05/6 each

The family is the three SPY win-rate members only, each at alpha = .05/6; no alpha moved from withdrawn money members. Above-half labels reuse the same three bounds.

Sources: P:113; P:176

two percentage-point frozen tolerance bands

Tolerance: two percentage points absolute, inclusive bands [49%, 53%], [51%, 55%] and [52%, 56%]. A point inside its band is WITHIN PREDECLARED TOLERANCE; a whole range inside is WITHIN BAND; a range wholly outside is DISJOINT; otherwise OVERLAPS BAND.

Sources: P:131; P:135

conjunction and half reuse

The claim holds only if 3 of 3 ranges are wholly inside their bands; it does not hold if at least one range is wholly outside; otherwise cannot tell. Above-half classifications reuse the same three bounds and answer a different question.

Sources: P:136; P:137; P:138; P:182

quantity fixed before buy

Whole shares q = floor(50/(0.01 x M)), with M the immediately preceding scheduled session's last regular-minute close, fixed at 09:29 ET before entry. $50 is a hypothetical stress budget against a 1% adverse move, not account capital, a stop or a maximum loss.

Sources: P:71; P:75

estimated cost grid

Primary expense $0.04 per share round trip plus $0 commission, once per trade: an estimate, not a quote. The full registered sweep has 72 cells per fund and exit: risk $20 or $50, adverse 0.5%, 1% or 2%, $0 to $0.20 per share and $0 or $1 commission.

Sources: P:77; P:79; P:81

no money inference

Money amounts are measured; descriptive; not inferred. A positive or negative sign is an observed sign, never a money verdict; the nine branch sentences were frozen before returns.

Sources: P:141; P:144; P:146

primary vs descriptive extensions

SPY Q1 is the only inferential question. Later-window comparisons (Q3), the overnight/intraday decomposition (B24), concentration, every cost cell and the added fund QQQ are descriptive; not inferred.

Sources: P:139; P:140; P:142; P:178

missing observations

No forward fill, zero return or nearest-bar substitution. Four official market-wide halt records are exemptions, not imputed prices; unknown gaps stay unknown.

Sources: P:61; P:63; P:65

one run and C01 custody

One registered outcome run at the lock; no rerun or post-price method change. A dated post-run descriptive correction, C01 (2026-10-18), recomputes the after-cost classes in exact decimal arithmetic, and with them the class counts, conditional win/loss means, ratios and signed medians; Q1 rates, bounds and labels, unconditional means and percentiles are unchanged. It is shown beside its locked predecessors.

Sources: R:3; C01:2; C01M:1

reopen conditions

Holds needs 3 of 3 ranges wholly inside; any valid disjoint range would give does not hold. Exact source dates and the original ledger, actual order/fill and fee records, a verified error, or a fresh registration for a wider window or different method could change the reading.

Sources: P:138; P:148; N:331; N:333

summary scope

Summary, not the full plan text. Main terms only, paraphrased; not the complete private registry or a public immutable preregistration. The lock commit acc81d8 is held in a private repository; no public ref pins it. Hashes identify records; they are not public timestamp proof.

Sources: summary scope notice

9. Every film figure, the cited records, descriptive downloads, provenance.

Every figure the film says or shows, with its sources. The denominator is the film itself: 102 spoken sentences in twelve voice segments, 515 distinct on-screen text blocks (by slide, role, words and evidence context) covering 1634 state appearances across 46 slides and 153 states, focus and highlight states included, and 8 rows read by hand from 4 source-image placements: 625 rows. Repeated blocks are coalesced only when slide, role, words and evidence context agree; their state numbers stay listed. The mascot cutaways (12 states) carry no text or numerals.

Where the film's wording is looser than the findings, the literal words stay as spoken or drawn and a labelled clarification sits beside them in that row.

Downloads

• b30-c01-corrected-and-predecessors.csv - The six C01 rows, each corrected value beside its superseded LOCKED FLOAT predecessor
• b30-every-film-figure.csv - Every figure the film says or shows, with its sources and labelled clarifications
• b30-sources.csv - The source index: semantic ids, meanings, hashes and references
• b30-locked-design.json - The locked-design summary: main terms only, not the full plan text

Not published here: the per-length block counts, block denominators, standard errors and per-length bounds behind the three Q1 ranges (R:71); the remaining registered cost cells and distribution rows; the sensitivity, later-slot, annual-sum, extreme-day and ex-post rows; and the trade-level, daily-control, B24-pair and exclusion-waterfall ledgers (R:102-117). These stay in the private result files. The raw market inputs stay in our private read-only data store; their identities were checked 6 of 6 (R:1016).

C01 CSV columns

Film and source CSV columns

Film CSV columns: figure_id identifies one row; medium distinguishes spoken sentence, drawn text block and source image; location records the beat and sentence, or slide, role and states, or frame; words keeps the literal film text; context gives the full sentence, board or image; source_ids names the supporting records; evidence_contexts names the evidence meanings; provenance explains conditions and any labelled clarification. These are three row types, not independent measured observations.

Source index columns: source_id, meaning, sha256_lf and reference - a semantic identifier, what the record is, the SHA-256 of its LF-normalized text (or of an image's original pixels), and the pinned private commit or public external URL. A pixel hash is not a text hash. Line coordinates such as Q1:2 or R:11 mean physical line 2 or 11 of that record, header at line 1. The index does not publish the private records.

10. What would change the answer, standing notice, source credits, B29 next.

What would change our answer. The qualified holds branch needs all three SPY ranges wholly inside their frozen bands; any valid wholly-outside range would give does not hold. A verified error in the saved calculation, input basis, eligibility or source transcription would reopen the affected reading while preserving the old receipts, as C01 did for the after-cost classes. The exact source dates and the original ledger could allow a more faithful comparison; actual order/fill and fee records could replace assumed execution and expenses in a separately authorized, dated study. A wider window or a different uncertainty method would need a fresh registration before its outcomes. A method supporting a money verdict would need its own approved testing. A money sign, a favourable cost cell or removing large days cannot change the present reading (N:331-333; P:138; P:148).

The film says: “For every episode, we publish the full test report on our site, with the numbers, a summary of the plan we locked before the run and what would change our answer”, followed at once by “We are still filling in the back catalogue.” This page is the full test report on our site for this episode, with the numbers (section 9 lists the registered result rows, Q1 interval details and raw inputs not published here, and indexes every private record a cited figure comes from), a summary of the plan we locked before the run (section 8) and what would change our answer (above); the report compares the sizes of wins and losses in sections 4 and 7. For the back catalogue: offline records identify 44 released episodes, of which 27 have a report file in the library and 17 do not yet; the library says more reports are being prepared. This is a publishing practice with an unfinished archive, not a claim that the archive is complete.

Ground Truth is educational and entertainment content - not financial, investment, or tax advice. We are not licensed or registered advisers; we share original research and opinions. Nothing here is a recommendation to buy, sell, or hold any security, fund, or strategy. Backtests and simulations are hypothetical, model real-world costs imperfectly, and past performance does not guarantee future results. Do your own research and consult a qualified professional before making any financial decision.

Source: The Rumers, If You Only Have $50, Do This Every Morning (The 3R Rule) - https://www.youtube.com/watch?v=XBOGd67BJks . Caption times and the 23:44, 2:11 and 4:19 marks on our quote cards are positions in their video, not in this film. We credit The Rumers as the source of the clock claim we tested; this report makes no statement about their skill, honesty, account or course, and adds nothing personal beyond their public video.

Next: Can Luck Match a Trading Champion's Record? - https://youtu.be/H7iX0th2BBU . A different trader's contest record, tested separately.