News Spread: is the published record edge or noise?
Does a member who sold every call the live signal labelled trade at the published credit and held it to expiration make money after costs, and more than rules that ignore the signal?
News Spread published 597 rows from 1 August to 10 November 2026: 539 labelled trade, 38 watch and 20 skip. This study sold every trade-labelled row at the published credit and held it to expiration. Of the 539 trade rows, 482 had expired by 10 November: they averaged -1.88% of the capital at risk per trade after costs, with a 95% interval of -9.07% to +4.63% and -1.31 dollars per contract. The interval spans zero, so the record cannot tell a loss from no effect; the mean is negative, so it does not meet the bar for a promising result. The study is exploratory: there was no pre-registration, the rules and verdict are recorded with the results, and no holdout was read. Its family N of 1 records one exploratory look, not a count of the 10 control, 8 sensitivity, 39 breakdown and 16 audit rows; no number is deflated against it. The record clears the study's sample floors, so the limit is the width of the interval, not a thin sample. The 57 trade rows still open on 10 November are left out. The 34 expired watch-labelled rows averaged +21.75% (interval +2.65% to +34.79%), but that is one descriptive cell among 39, not a tested result. The result is descriptive, not a pass or fail of a pre-registered test.
The report is labelled exploratory: this was one retrospective look at a live record, with no pre-registration, no counted trial and no holdout. The study's own rule labels the result UNDERPOWERED: the mean is -1.88% of the capital at risk and the 95% interval is -9.07% to +4.63%, which spans zero. The record clears the study's sample floors (482 calls, 50 tickers, 20 expirations); the width of the interval, not a thin sample, is the limit. The descriptive verdict records an exploratory reading, not a tested pass or fail.
Backtested by QS Research
Pre-registration
No pre-registration existed. This is a retrospective record written on 21 November 2026, after every result was seen. It records the question, data, rules and judging as they were run and changes none of them; nothing in it was fixed in advance, so no result counts as a pre-registered test. The question: does a member who sold every call the live signal labelled trade at the published credit and held it to expiration make money after costs, and beat rules that ignore the signal? The record is 597 published rows from 1 August to 10 November 2026: 539 trade, 38 watch and 20 skip. The scored set is the 482 trade calls that had expired by 10 November; 57 trade calls were still open. The trade is the published natural credit (short bid minus long ask), held to expiration and settled on the daily close of the expiration session. Costs are $2.60 per spread in commissions and fees, plus $0.05 a share to close a spread that finished between its strikes; a stress row moves the credit by $0.05 a share and the exit by $0.10. The return is net profit over the capital at risk (width minus credit plus the fee). The interval is a percentile cluster bootstrap with 4,000 resamples, seed 12, over tickers, run days and expirations; the interval used is the envelope of the three. The controls are the same calls with one choice removed: placebo dates, always bull put, always bear call, the other side, and a coin-flip side. Each control is anchored to the prior close and to the entry session's open, and the engine must beat it about both anchors. The overall reading was judged by that rule. The study's own rule labels the result UNDERPOWERED when the interval spans zero; it did: mean -1.88% of risk, interval -9.07% to +4.63%. The rule, costs and sample floors are in the same code version as the results; that version cannot show when they were written, and there is no record of whether they changed while the record was being read. The controls, sensitivities and 39 breakdown cells were not separate trials. Family N is 1 as one exploratory look, not a count of comparisons, and no number is deflated against it. No holdout was read.
Registered 2026-11-21Amendments
- 2026-11-21 Retrospective record written after the results: it sets out the question, data, rules and judging as run. No rule changed. approved: Recorded after the results; no rule changed
- 2026-11-21 Review fixes to the retrospective record: the interval floor now matches the code as run, and the over-strong power sentence is corrected. No rule or result changed. approved: Independent review, 2026-11-21; no rule changed
- 2026-11-21 Wording only: the retrospective record's verdict-rule sentence now says the code version cannot show when the rule was written, the scored bullet adds the expired and open row counts, and the report's section notes say what the overall reading was judged against. No rule or result changed. approved: Independent review, 2026-11-21; no rule changed
Trials · 0 trials · 0 passed · judged at N = 1 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| the published calls, base costs | RESULT | not counted |
| no costs (the published credit, held to expiration) | RESULT | not counted |
| fee only (the source's own basis) | RESULT | not counted |
| stress costs (entry slip and a dearer exit) | RESULT | not counted |
| first call of each position only | RESULT | not counted |
| without runs stamped on a non-session day | RESULT | not counted |
| without calls already through the short strike at the open | RESULT | not counted |
| without rows whose strikes were display-rounded | RESULT | not counted |
| every published row that had expired (trade, watch and skip) | RESULT | not counted |
| placebo dates | RESULT | not counted |
| always bull put | RESULT | not counted |
| always bear call | RESULT | not counted |
| other side | RESULT | not counted |
| coin-flip side | RESULT | not counted |
| placebo dates | RESULT | not counted |
| always bull put | RESULT | not counted |
| always bear call | RESULT | not counted |
| other side | RESULT | not counted |
| coin-flip side | RESULT | not counted |
| bear call | RESULT | not counted |
| bull put | RESULT | not counted |
| news read neutral | RESULT | not counted |
| no news read on the row | RESULT | not counted |
| side against the news read | RESULT | not counted |
| side agrees with the news read | RESULT | not counted |
| 0.20 to 0.30 | RESULT | not counted |
| 0.30 and over | RESULT | not counted |
| under 0.20 | RESULT | not counted |
| 14 days or fewer | RESULT | not counted |
| 15 to 30 days | RESULT | not counted |
| over 30 days | RESULT | not counted |
| 2026-08 | RESULT | not counted |
| 2026-09 | RESULT | not counted |
| 2026-10 | RESULT | not counted |
| first half | RESULT | not counted |
| second half | RESULT | not counted |
| rows from before the fee deduction | RESULT | not counted |
| rows published net of fees | RESULT | not counted |
| (fewer than 10 calls) | RESULT | not counted |
| ADBE | RESULT | not counted |
| APP | RESULT | not counted |
| COHR | RESULT | not counted |
| COIN | RESULT | not counted |
| CRM | RESULT | not counted |
| CRWD | RESULT | not counted |
| GLW | RESULT | not counted |
| HOOD | RESULT | not counted |
| INTC | RESULT | not counted |
| LRCX | RESULT | not counted |
| MRVL | RESULT | not counted |
| NOW | RESULT | not counted |
| ORCL | RESULT | not counted |
| PLTR | RESULT | not counted |
| TSLA | RESULT | not counted |
| VRT | RESULT | not counted |
| skip | RESULT | not counted |
| trade | RESULT | not counted |
| watch | RESULT | not counted |
| row written when it was generated | RESULT | not counted |
| news read before the call | RESULT | not counted |
| the side follows the news read | RESULT | not counted |
| the published credit was fillable | RESULT | not counted |
| the price the snapshot saw | RESULT | not counted |
| expiration after the run, DTE consistent | RESULT | not counted |
| one row per run and ticker | RESULT | not counted |
| strikes are the traded strikes | RESULT | not counted |
| the source's settlement reproduces | RESULT | not counted |
| every expired call is scored | RESULT | not counted |
| open calls are censored by the calendar only | RESULT | not counted |
| a run on every session | RESULT | not counted |
| record window | RESULT | not counted |
| one fixed rule over the record | RESULT | not counted |
| repeated exposure | RESULT | not counted |
| no control clears zero on its own | RESULT | not counted |
Limitations
- No pre-registration: the rules, costs, sample floors and verdict are in the same code version as the results; that version cannot show when they were written, and there is no record of whether they changed while the record was being read. No result here is a pre-registered test.
- One live window only: 1 August to 10 November 2026, 101 calendar days of runs, 63 run days with scored calls and 20 expirations. It is one market regime and says nothing about a different tape. The record also changed inside the window: a fee-adjusted credit and a one-word news read first appear on 20 August 2026, and a run publishes between 5 and 9 rows, so the study treats several rule versions as one.
- The published credit is a natural bid/ask quote from an option-chain snapshot taken before the options market opened, so no member could have hit it. The base result assumes a fill there. The stress row moves the credit by only $0.05 a share and the exit by $0.10, a token slip; the underlying moved a median 1.45% from the prior close to the open, and the stress does not re-price the spread for that move. A full gap stress was not run. The stress result is still negative, but its interval spans zero.
- The record does not hold the underlying price at the snapshot. Every control is run about the prior close and the open; neither is the snapshot price. About the prior close the engine's strikes look further away than they were, which flatters it, and that row must not be quoted alone. The open is the fairer anchor on the study's delta check, where the median gap is 0.020 at the open and 0.044 at the prior close; the prior close is nearer on 28% of calls.
- The side controls (always bull put, always bear call, other side, coin-flip side) use a mirrored spread at the same credit: a model price, not a quote. Equity skew would usually pay the put side more. The placebo control is a return-space model, not a traded spread.
- The 482 scored calls are 461 distinct positions on 50 tickers; the three most-called tickers hold 26% of the calls. The intervals resample whole tickers, run days and expirations, but the sample is concentrated.
- Of 597 published rows, 73 had not expired by 10 November 2026 and are left out: 57 trade, 4 watch and 12 skip. The 57 open trade calls are 20 October runs and 37 November runs. The October month cell covers 128 of 148 October trade calls, there is no November month cell, and the by-label table covers 524 expired rows, not all 597.
- Each individual ticker cell has no interval: it is a single ticker, and the study's interval routine needs at least 10 calls and 10 tickers before it forms one. The pooled ticker cell and the overall reading do have intervals.
- The 34 expired watch-labelled calls averaged +21.75% (interval +2.65% to +34.79%) while the trade calls lost. That is one of 39 descriptive cells and was not judged; with this many cells some clear zero by chance. It is a candidate for a new test, not a finding.
- The side agrees with the row's news read only 35% of the time and opposes it 37% (of the 455 trade calls with a news read); the news read is not confirmed as the reason for the side. The audit could not check which headlines were read: the record has no headline id or news timestamp. The agreeing cell was the worse one (-11.10%), the opposing cell the better one (+5.53%), and both intervals span zero; both are descriptive.
- Nine of 482 scored calls carry a strike gap consistent with a display-rounded strike. The study rebuilds the long strike from the row's width, but the short strike can still be half a point off. One sensitivity drops those rows; the base result keeps them.
- The record was appended at publication (median 2 minutes after generation, at most 5), but the file could have been edited afterwards; nothing in it proves otherwise. The history begins on 1 August 2026; anything published earlier is not in it. Four sessions in the window have no run, and the record cannot say whether a run failed or was withheld.
- The published page does not show a pinned data snapshot or a release commit. The result records name the daily bars and the record by content hash, but the report has no data-pin entry and no release commit, and the run date shown is the date of the commit that added the results, not a run log. The numbers are transcribed from the study's results write-up; the per-call rows are not published.
- This is a descriptive result. The mean is negative and the interval spans zero, so it shows no evidence of an edge; it also cannot rule out a small positive edge. The record clears the study's sample floors; the smallest true mean it could detect with 80% power is 9.79% of risk.
- A later change, approved on 11 November 2026, takes the option snapshot after the open and writes the snapshot price and headline fields into each row; it is a separate forward test, not part of this record. This study is of the earlier pre-open snapshot pipeline, and no result from the forward test is here.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
2 published versions · current 450909e258306e74, published 2026-11-21