What we got wrong

That a casefile’s Why-the-score section carries a profile factor, and that the reader finds it there and reads which of the three states it names.

That the casefile’s Levels section gives the value high, the point of control and the value low, and how far the price sits from the point of control, from which the reader reads the price’s state.

Why the score lists the seven factors of the conviction score, and the volume profile is not one of them; the casefile shows the profile under Levels. Found while correcting the factor lessons for S-2.

That each factor under a score is printed as a score from 0 to 100, the number above its bar (such as “63/100”) being that factor’s own score and the fill being the score, with the quiz keying “Its score out of 100”; and that the Why-the-score section shows what each factor’s band has been worth in the record.

That each factor is printed as a reading, high, mixed or low, with the exact score kept on the server, and its bar fills a third, two thirds or the whole way; that the quiz keys “Its reading: high, mixed or low”; and that below the list the section shows how calls that scored in the same band have resolved.

From 6 October 2026 (S-2) /api/conviction-score sends each factor as a reading instead of its score, because the seven exact scores beside the exact composite let the weights be solved, and the casefile and the conviction page print the word. The section never showed a record for each factor’s band; the record beside the list is the score’s.

That the casefile’s Why-the-score section shows the zone factor as a score beside the other six factors, a number standing for the zone position, and that the reader reads the supply demand factor’s score there.

That the section shows the zone factor as a reading, high, mixed or low, which comes from the zone position the lesson lists, and that the reader reads that reading there.

From 6 October 2026 (S-2) the casefile prints each factor as a reading rather than its score.

That at $1,000 a month, closing a $12,000 gap with new contributions alone takes about a year.

That with nothing sold the account grows as the money goes in, so reaching the target takes about twenty months.

The accuracy read redid the arithmetic: after twelve months the account in the example is still at 62% stocks.

A quiz offered "a record of every name considered" as a wrong answer, though the lesson describes a never-emptied list in nearly those words.

That option is replaced with one the lesson does not support, so only one answer is right.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

That monthly execution-quality reports are published by trading venues and not by brokers.

That since August 2026 the largest brokers publish them as well as the venues.

The accuracy read checked the amended disclosure rule, which took effect in August 2026.

That subtracting inflation from the nominal return always overstates the real return, by more as rates rise.

That the shortcut overstates a positive real return and overstates the size of a negative one.

The accuracy read found the lesson's own third example running the other way.

That trading in US stocks runs from four in the morning to eight at night Eastern, stated as the whole of it.

That those are the exchanges' hours; some brokers also offer an overnight session on a separate venue, and the large exchanges have announced overnight sessions of their own.

The accuracy read checked the current state of overnight trading.

That this platform's chart draws the fifth retracement level at 76.4%.

That the chart's drawing tool draws 78.6%, while the Fibonacci levels the platform works out by itself use 76.4%.

The accuracy read compared the lesson with what the chart actually draws.

A quiz explanation said that the age of a zone is its freshness.

That how long ago a zone formed is not one of the four readings, and freshness is about revisits, not age.

The accuracy read found the explanation contradicting the lesson's own definition of a fresh zone.

That an evaluation asking for ten trading days requires the trades to fit inside ten days.

That ten days is a floor, not a ceiling: the trades may take longer, and passing in about ten days means 8 to 16 trades a day.

The accuracy read found the sentence contradicting the lesson's own definition of the ten days as a minimum.

That exchange margin requirements are changed without notice.

That the exchange changes them at short notice, often about one business day, and that it is a broker's own added margin that can rise with no notice at all.

The accuracy read checked how the exchange publishes margin changes.

A quiz explanation said the difference between a parked answer and a data failure was "the next question", which it was not.

It now says the difference is a separate question further down the quiz.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

That a central bank's inflation goal is usually written against the price index inside the national accounts; and that month 1 was the largest month in the example.

That the United States writes its goal that way while most other central banks use a consumer price index; and that month 1 was the second-largest month in the example.

The accuracy read checked the published mandates of the large central banks, and the lesson's own table.

A scenario answer referred to "the two years" around a six-month record, in a case where the service had existed for 20 months.

It now refers to the 20 months as a whole.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

That the reclaim after a sweep is a close back inside the range on the same bar or the next one.

That the reclaim is a close back inside the range on the same bar; a bar that closes beyond the level is a break.

The accuracy read found the next-bar case contradicting the lesson's own rule that a sweep is a wick and never a close.

The lesson pointed to "the previous lesson" for the volatility index, after another lesson had been placed between the two.

It now names the lesson, Reading volatility, where the volatility index is explained.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

That a large special cash dividend adjusts an option contract by a ratio, and that crossing a $0.60-wide quote twice costs $1.20 a share.

That the strike is reduced by the amount of the dividend with the deliverable left at 100 shares, and that in at the ask and out at the bid costs the width once, $0.60 a share.

The accuracy read checked the options clearing organisation's published adjustment policy and redid the arithmetic; the worked example and a quiz answer changed with it.

That the 5.0R in the table is one extreme of the range to the other, divided by the distance to the stop.

That it is the distance from the entry to the far extreme, divided by the distance to the stop.

The accuracy read redid the division: extreme to extreme gives a different figure from the one in the table.

That all four ratios in the table are computed per period and annualised by the square root of the number of periods.

That only the first two are per-period ratios of that kind; the other two are computed from totals.

The accuracy read compared the sentence with how the lesson's own table computes each ratio.

That this platform marks price as stretched by its distance from the point of control.

That the distance is measured from the edge of the value area, as a share of the point-of-control price.

The accuracy read compared the quiz with how the platform makes the measurement.

A quiz offered "a discount of $2.00" as a wrong answer for a fund holding $20.00 a share and trading at $18.00, which is also true.

That option is replaced with a wrong percentage, so only one answer is right.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

In two places, that a profitable replay is weak evidence about the method.

That a replay is not evidence about the method at all, which is what the rest of the lesson already said.

The accuracy read found the lesson landing on both sides of its own distinction.

That a shrinking MACD histogram means the distance between the two moving averages is closing.

That the histogram is the gap between the MACD line and its signal line, so when it shrinks the distance between the two averages has stopped growing as fast, or has begun to narrow.

The accuracy read recomputed a case where the histogram shrinks while the two averages are still moving apart.

The lesson cited the journal article under the title of its earlier working-paper version.

The title is now the one the article was published under in the journal named.

A citation check against the journal's own listing.

A quiz explanation said the results of passed signals exist only because the passes were written down on the day.

That the results can be scored afterwards from the price; what only a record made on the day holds is the reason each signal was passed.

The accuracy read found the explanation contradicting the lesson body.

A quiz offered selling "whenever the price reaches a target" as a wrong answer, though a pre-set price is one of the plan terms the lesson itself lists.

That option is replaced with one the lesson does not support, so only one answer is right.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

The lesson gave a working-paper number with the title of a different, earlier paper by the same authors.

The title now matches the working paper cited: it carries the subtitle "Evidence from Factors and Transactions Data".

A citation check against the published record of both papers.

A quiz explanation said volume, range and the close are all measured against the name's own recent bars.

That volume and range are measured against the recent average, and the close is read inside the bar itself.

The accuracy read found the explanation contradicting the lesson body.

That when a company is moved to a different sector, the published histories of both sector indices are rebuilt on the new membership.

That the change normally applies from the review forward, and the history already published stays as it was, built on the membership of its time.

The accuracy read checked the index providers' own announcements of the 2018 sector changes; the quiz answer was wrong with the text and is corrected too.

A quiz offered "over a single day of holding" as a wrong answer, though the study did report daily returns.

That option is replaced with one that is false, so only one answer is right.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

A sentence about the cost of "keeping the last of those back" pointed at the wrong item in the list above it.

It now says what it meant: keeping the method back has a cost, and the cost falls on the reader.

A line-by-line accuracy read of every lesson on October 5, 2026, with each suspected error checked a second time independently before it was changed.

That a dividend paid during an account transfer goes to whoever held the shares on the payment date.

That it goes to the firm that held the shares on the record date, and is passed on afterwards.

The accuracy read checked how a dividend entitlement is fixed.

That the overnight-return study covered a decade.

That the study covered the years 1993 to 2006.

A citation check against the working paper's own sample period.

That OPEN "means conditions historically favoured long picks" and CLOSED that they "did not", and that "the gate itself is under live testing; its record is on the experiments pages".

That OPEN means the gate's conditions for long picks were met and CLOSED that they were not, and that the gate's reads so far do not agree: one confirmed that gated long picks beat the index; a later one failed, with the caveat that the gate never closed in its window. Both rows are on the receipts page's ledger.

A Fable review on 2026-10-02: "historically favoured" said more than the record shows. The ledger carries one read that confirmed the gate (E2-GATE) and a later one that failed (E3-H1). The glossary and the title page carried the same claim and were corrected the same day.

That a call's direction means "the data favours a rise" (or a fall), that its plan's target is "where it has historically resolved", and that its factors are "trend, momentum, relative strength, patterns, regime".

That the direction is which way the factors lean, taken together, a description and not a forecast; that the target is a reference level in the call's direction, set by a fixed rule and framed in ATR, not a forecast that price reaches it; and that the factors are the seven readings behind the score — trend, momentum, range position, setup quality, volume, confluence and the supply/demand zone.

A Fable review on 2026-10-02 checked the plan code: a target comes from a fixed rule per setup (a multiple of ATR, a pattern's measured move, or a multiple of the risk), not from a record of where such setups ended, and the platform has not measured that. The factor list named five things that are not the score's seven. The order-types lesson's take-profit line said the same about targets and was corrected with it.

That a name with no earnings date on file is read one way on a screen registered from October 6, 2026 and another way on every other screen, where it is always left out of an "earnings not within N days" screen.

That from October 6, 2026 every screen reads it one way: after a night the earnings calendar answered, no date on file means no report in the calendar's window and the name passes; after a night it did not answer, the name is left out and the rankings page says so in a line. A screen registered before that date keeps the earlier reading for its record.

A Fable review on 2026-10-02 found that leaving every undated name out drops exactly the names whose reports are furthest away, and that a live screen has no record to protect, so live screens follow the same rule from the same date. The scenario now takes place on a night the calendar did not answer, where the name is still left out.

That the casefile's earnings-drift table "records what a name's price did after each of its own past reports" — and, in the earnings-report lesson, that the table "measures this per name".

The same words, true from Sep 29, 2026: the table now measures from the report date. Until then it measured from the day each fiscal quarter ENDED, about a month before the report, so the lesson described a table the platform did not yet have.

A read of the EPS history feed's dates on Sep 29, 2026 (they are quarter ends, not report dates) while moving the casefile's price history to a second source. The route was corrected the same day (e96856a) and the error is on the public ledger's integrity list (SEP-29). The lesson's wording stands; what was wrong was the platform behind it.

That “a technical grade has been worth different amounts in different names, and the fundamentals are the most likely reason”, and that the casefile shows “the grade, the plan, the record of that grade”.

That two calls with the same score can sit on very different businesses and the platform makes no claim about what that difference has been worth, and that the casefile shows the score, the plan, the record of past calls and the fundamental context.

Since the E4-H3 read on 2026-09-25 the grades are internal-only pending a recalibration charter and the grade-based record is no longer shown, so the lesson could neither say what a grade has been worth nor that the casefile shows it.

That the fundamental context the casefile shows beside the score is “the reason the same grade has been worth different amounts in different names”.

That it is context shown beside the score and one of the things a reader weighs that the score leaves out, with no claim about what a grade has been worth.

The sentence rested on grades having a measured worth. The E4-H3 read on 2026-09-25 found the grades did not rank outcomes, and they are internal-only pending a recalibration charter.

That the record of what followed each trend state is on the track-record pages, “and it is that record, not the name of the cross, that the grade uses”.

That the trend factor reads the stack of averages and its slope, that what followed each state is on the track-record pages, and that the score describes the reading, not what followed it.

A check of the lessons after the E4-H3 read on 2026-09-25, which found the grades did not rank outcomes and left them internal-only pending a recalibration charter. The score is worked out from how the factors read on the night a call is logged, and the grade was a band of that score, so neither drew on the record of what followed.

That a long call in a stacked uptrend and a long call against a falling 200-day are different animals, “and the grade reflects it”, which read as the grade telling the stronger call from the weaker.

That the trend factor reads the two differently, and that the score describes that difference and does not say which call will do better.

Since the E4-H3 read on 2026-09-25 the grades are internal-only pending a recalibration charter: the sentence pointed to a letter the pages no longer print, and the read found the higher bands of long calls had not done better than the lower ones.

That a call’s grade is “a measured category with a historical hit rate”, that a reader should know “what the grade has actually been worth in the record”, and that a SHORT call’s grade of A describes “How the factors line up, and the band’s record”, which the quiz marked correct.

That a call carries a conviction score out of 100 describing how its factors read in the call’s direction, and that the score describes “How the factors line up”, now the correct answer, with no claim about what any band has done.

The E4-H3 read on 2026-09-25 found the grades did not rank outcomes, so there is no band record for the lesson to point to; the grades are internal-only pending a recalibration charter. The lesson had also kept the old answer after the quiz in What BUY means was corrected, so the two disagreed.

That a replay opens a resolved call with its “symbol, side, grade and the night it was logged all shown”.

That it shows the symbol, the side and the night the call was logged. The record still keeps the grade it stamped at the time; the replay does not print it.

Since the E4-H3 read on 2026-09-25 the grades are internal-only pending a recalibration charter, and the replay no longer prints the grade.

That an event alert reports a name’s grade changing, worded like “NVDA grade B → C”, and that a screen can filter on the grade.

That the alert reports the conviction score moving far enough to cross one of the internal bands, worded like “NVDA conviction score 71.4 → 63.9”, and that a screen filters on the other readings a casefile prints.

Since the E4-H3 read on 2026-09-25 the grades are internal-only pending a recalibration charter: the alert now carries the score’s move instead of the letters, and a new screen no longer offers a grade filter.

That the conviction score is banded into six grades from A+ to D, set out in a table that called A+ “historically the highest-hit-rate band”; that Best Available considers only grade A and above because lower grades “showed no reliable edge at the July read”; and that a band’s word is replaced by its actual hit rate once enough calls have closed.

That the conviction score is a weighted seven-factor total out of 100 that describes and does not forecast; that its letter grades are internal-only pending a recalibration charter, because the E4-H3 read, a test of long calls, found they did not rank outcomes, the higher bands of long calls having done no better than the lower ones; and that no band is claimed to have done better than another.

The pre-registered E4-H3 read on 2026-09-25 tested long calls and found that the grades did not rank their outcomes at seven days (A −2.17, B+ −0.78, B +0.33 points against the index). The ledger row demotes the grades to internal-only pending a recalibration charter and removes the public grade-based win-rate context, so the lesson could no longer teach the bands or what they had done.

That each row of the public ledger shows the night, the symbol, the side, the grade, the logged price, the seven- and thirty-day results and the verdict.

That each row shows the night, the symbol, the side, the logged price, the seven- and thirty-day results against the index and the verdict, with no grade.

Since the E4-H3 read on 2026-09-25 the grades are internal-only pending a recalibration charter, and the public ledger no longer carries a grade column.

That a long call on this platform is a description made of “a grade, the factors behind it, and a plan with a stop and a target”.

That it is made of “a direction, the factors behind it, and a plan with a stop and a target”, and the last quiz question asks about a high conviction score rather than a grade of A.

The grades have been internal-only since the E4-H3 read on 2026-09-25, pending a recalibration charter, so a call no longer carries a letter on the page.

That the grade on a long call “says how many factors agree and what that band has historically done”, in the answer to the lesson’s last quiz question.

That a high conviction score describes how the factors line up for the name: a description, not a forecast of what the price will do. The letter grades themselves have been internal-only since the E4-H3 read, pending a recalibration charter.

The pre-registered E4-H3 read on 2026-09-25 found that the grades did not rank the outcomes of long calls at seven days (A −2.17, B+ −0.78, B +0.33 points), so the answer could no longer point to what a grade band has done.

That once the plan is written the platform’s readings become inputs to it, and that “a conviction grade is part three”, the part of the plan that states the setup the reader is looking for.

That a conviction score is a reading to check against part three, the setup the plan states, rather than a verdict on it.

The E4-H3 read on 2026-09-25 found the grades did not rank outcomes, and they are internal-only pending a recalibration charter, so a plan could not take a grade as its setup. The score describes how the factors read; it does not stand in for a setup the reader has written down.

That the SEC's bulletins describe an annuity's surrender period as typically six to eight years, sometimes ten or longer.

That the SEC's bulletin describes it as typically six to eight years, sometimes as long as ten.

A read of the SEC's Updated Investor Bulletin: Variable Annuities (30 October 2018) found the wording "after six to eight years or sometimes as long as ten years"; the lesson's "or longer" went past the source.

That Cowles (1933) classified Hamilton's editorials from 1902 to 1929, and that both studies read one writer over about twenty-seven years.

That Cowles's sample ran from December 1903 to November 1929, about twenty-six years, and found about 12% a year from following the editorials against 15.5% from holding; 1902 is the year Hamilton took the editorship, which is how Brown, Goetzmann and Kumar date his record.

A read of the text of Brown, Goetzmann and Kumar (1998): they quote Cowles on "the 26 years under review" and date their Table I from December 1903 through November 1929. The lesson had given Cowles the 1902 start that belongs to Hamilton's editorship.

That Brown, Goetzmann and Kumar (1998) found Hamilton's calls earned less return than holding the market, with less risk taken.

That Brown, Goetzmann and Kumar (1998) found Hamilton's timing had value once risk is counted: his calls kept the portfolio out of the market for long stretches, so that in their own simulation it earned about the same as holding the market with less volatility, and measured against the risk it carried the record showed high Sharpe ratios and positive alphas. The lower-return finding is Cowles's (1933).

A check of the paper's abstract (Journal of Finance 53(4), 1311–1333) found no statement that the calls earned less than holding; that result belongs to Cowles, and the lesson had attached it to the later paper. Confirmed the same evening against the paper's text: its Table II puts the Hamilton portfolio at 10.73% a year against 10.75% for the all-stock portfolio, with a Sharpe ratio of 0.559 against 0.456 and a Jensen alpha of 4.04%.

That no widely accepted study known to the academy shows prices turning at the retracement levels more often than chance, that this was the honest state of the evidence, and that it was why nothing was cited — which read as though no test of the ratios had been published.

That Batchelor and Ramyar, in Magic numbers in the Dow (2006), tested turning points of the Dow Jones Industrial Average from 1914 to 2002 and found ratios near Fibonacci values no more often than chance; that Prechter replied in Elliott Waves, Fibonacci and Statistics (2006) that filtered trends are not Elliott waves; and that no widely accepted study shows the levels marking turns better than chance.

A survey of what the major curricula teach turned up a published test of Fibonacci ratios in Dow swings, and a published reply to it, that the lesson had said nothing about.

That the daily loss limit adds up the realised losses of the whole trading day.

That it adds the day's realised profits and losses together and pauses the session when that running total falls to the limit, which is what the simulator does.

A read of the simulator's code (lib/academy/sim.js): the limit nets the day's closed trades rather than summing losses alone.

That the scheduled releases which move everything at once are government releases, with the ISM manufacturing survey listed among them without distinction.

That most are from government statistical agencies and the ISM survey is published by the Institute for Supply Management, a private association, so a lapse in federal funding delays the federal releases and not the private surveys.

The Institute for Supply Management is not a government agency; the lesson had grouped its survey with the federal releases.

That the reporting rules do not require the compensation element to be included in the basis a broker reports for employer shares.

That for shares acquired through an equity plan since 2014 the rules do not let the broker include it: Treas. Reg. §1.6045-1(d)(6)(iii).

"Not required" understated the rule; the regulation forbids the inclusion for post-2013 equity-plan shares, which is why the reported basis is commonly nothing or the discounted price paid.

That Credit Suisse announced the XIV notes would be valued on 15 February 2018 and repaid on 21 February "at the closing indicative value of that date", which read as the value on 21 February.

That the notes were valued at their closing indicative value on 15 February 2018, the accelerated valuation date, and repaid at that value on 21 February, the acceleration date.

A read of Credit Suisse's 6 February 2018 release (SEC EDGAR, exhibit 99.1) found the payment set at the closing indicative value on the accelerated valuation date, 15 February; the sentence had pointed "that date" at the payment date instead.

That under Rule 605 the venues that execute orders publish monthly execution-quality statistics, with nothing said about brokers.

That the SEC's 2024 amendments to Rule 605, in force from 1 August 2026, also require broker-dealers introducing or carrying 100,000 or more customer accounts to publish the same monthly statistics for their customers' orders.

The lesson described the rule as it stood before the 2024 amendments; the compliance date was moved to 1 August 2026 (Federal Register, 2 October 2025) and has now passed.

That a Roth withdrawal is qualified once the account has been open five years.

That it is qualified at least five tax years after the first contribution to any Roth IRA, and that a Roth 401(k) counts its own five years.

A check against the tax authority’s own guidance found the lesson attaching the clock to the wrong thing — the account rather than the taxpayer’s first Roth contribution.

That a charge may not be called non-recurring when something similar is likely within two years, stated as though it applied to everything a company says.

The same rule, framed as applying in filings.

A check found the lesson applying a filing rule to a company’s speech generally.

That paying the whole of last year’s tax is enough to avoid a penalty, with no condition attached.

The same, provided a return was filed for last year and it covered all twelve months.

A check found the lesson stating a safe harbour without the condition that makes it available.

That a fund in the top quarter over one five-year period has rarely stayed there over the next.

That few funds in the top quarter of their category in one year have stayed in the top quarter in each of the years that followed.

A citation check found the lesson describing a different measurement from the one the published persistence tables actually report.

A description of the publisher’s exclusion from adviser registration that was broader than the case supports.

That a bona fide publication of general and regular circulation offering impersonal advice falls outside the definition of an adviser, citing Lowe v. SEC (1985).

A check against the decision found the lesson dropping the two conditions the ruling turns on.

That after a listing the underwriting banks’ analysts do not publish research for a period, stated as a general rule without its source or its exceptions.

That FINRA’s rules set a minimum quiet period of ten days before the underwriting banks’ analysts publish research, that the minimum does not apply to emerging growth companies, and that coverage often starts later still.

A read-only check against FINRA’s own rule found the lesson had turned a minimum with a named exception into a flat rule.

A citation of Brav and Gompers that reported the underperformance without reporting what the authors concluded from it.

That they found the gap concentrated in the smallest listings without venture-capital backing, that established firms of similar size and book-to-market did about as badly, and that they concluded it was not an effect of listing as such.

A citation check found the lesson using a paper to support a claim the paper argues against.

That a margin account’s maintenance requirement is measured at the close.

That the requirement must be met throughout the trading day — twenty-five percent for long positions in marginable stock and more for short positions — and that a deficit left unmet for several business days can restrict the account’s credit.

A check against FINRA’s margin rule found the lesson describing the old end-of-day test rather than the intraday one.