What has changed
Every change to the course since it was first built. Buyers get all of it free, forever — this page is the evidence that the promise is being kept rather than a sentence on a sales page.
Newest first.
2026-11-15 80 more tested
- The ledger now covers 946 approaches: 932 rejected, 13 waiting on forward data. It was 964, 948 and 15.
- Round 489 added; every one is in the ledger with its verdict.
- Twenty earlier rounds are no longer on the public ledger, and these counts leave out their strategies: their rules or data came from sources whose terms do not allow republishing them. One more round keeps its verdict but loses its chart, which was drawn from trial data.
- The ledger now covers 964 approaches: 948 rejected, 15 waiting on forward data. It was 961, 945 and 15.
- Rounds 485-487 added; every one is in the ledger with its verdict.
- The ledger now covers 961 approaches: 945 rejected, 15 waiting on forward data. It was 866, 854 and 11.
- Rounds 384-484 added; every one is in the ledger with its verdict.
- Three new bugs in Module 7, all of which made a result look better: profit divided by a bot's first small buy instead of its stake, one coin symbol that joined two different coins, and index members picked before they had joined the index. Each has its cost and the check that catches it.
- Every module now goes by one name, the title it carries in the course, on the sales page, in Module 1 and in the Bug Catalogue.
- The ledger opens with four outcomes (passed, waiting on forward data, rejected, not a full test); the individual verdict words and what each means are one click away.
- On a phone, a calculator's verdict scrolls into view after you load an example or run a test; the course index says so when the data toolkit is not running yet, instead of opening a dead page.
2026-11-05 19 more tested
- The ledger now covers 866 approaches: 854 rejected, 11 waiting on forward data. It was 857, 845 and 11.
- Rounds 375-383 added; every one is in the ledger with its verdict.
- The ledger now covers 857 approaches: 845 rejected, 11 waiting on forward data. It was 847, 835 and 11.
- Rounds 364-372 added; every one is in the ledger with its verdict.
- Round charts in the ledger are easier to read. The line is green above where it started and red below, with the final result labelled at its end, and the yearly return and worst fall come first.
2026-11-03 43 more tested
- The ledger now covers 847 approaches: 835 rejected, 11 waiting on forward data. It was 804, 795 and 8.
- Rounds 313-363 added; every one is in the ledger with its verdict.
- Modules 2-8 reviewed end to end, and what the review found is fixed. Module 8's results table had a stale rejected count and a row no longer backed by the ledger; both now read from it. Module 7's drawdown bug is shown as it actually happened, Module 6's percentiles use the exact formula Module 7 teaches, and Module 5's calculator now agrees with the module.
- Smaller fixes: wrong "next module" pointers, broken sentences, a 15% that should have been 17%, an unlisted 19th data source, and new terms explained where they first appear.
- The calculators ask for a number instead of reading an empty box as zero.
- New: the letters. Every newsletter that has gone out, in full, at /letters/.
2026-11-01 148 more tested
- The ledger now covers 804 approaches: 795 rejected, 8 waiting on forward data. It was 800, 791 and 8.
- Rounds 309-312 added; every one is in the ledger with its verdict.
- The ledger now covers 800 approaches: 791 rejected, 8 waiting on forward data. It was 656, 649 and 6.
- Rounds 151-308 added; every one is in the ledger with its verdict.
2026-10-25 3 more tested
- The ledger now covers 656 approaches: 649 rejected, 6 waiting on forward data. It was 653, 647 and 5.
- Round 150 added; every one is in the ledger with its verdict.
2026-10-24 27 more tested
- The ledger now covers 653 approaches: 647 rejected, 5 waiting on forward data. It was 637, 631 and 5.
- Rounds 135-149 added; every one is in the ledger with its verdict.
- The ledger now covers 637 approaches: 631 rejected, 5 waiting on forward data. It was 626, 620 and 5.
- Rounds 120-134 added; every one is in the ledger with its verdict.
2026-10-23 91 more tested
- The ledger now covers 626 approaches: 620 rejected, 5 waiting on forward data. It was 535, 529 and 5.
- Rounds 113-119 added; every one is in the ledger with its verdict.
- Rounds in the ledger now open into their backtest. A Chart button on a round shows its curve against the benchmark it was judged by, its Sharpe, noise floor, return, worst fall and trades, and the reason for its verdict in a sentence. The newest rounds have one already; the rest are being added round by round.
- The site has a lab page, rebuilt from the research record at every update: how many strategies have been tested, round by round, where they ended up, and the newest rounds.
2026-10-22 2 more tested
- The ledger now covers 535 approaches: 529 rejected, 5 waiting on forward data. It was 533, 527 and 5.
- Rounds 111-112 added; every one is in the ledger with its verdict.
- Every tool and course figure answers nonsense with a plain message. A blank box, a few extra zeros, or a price file that jumps a hundredfold in a day gets told what is wrong, in the free tools, the Judge and the figures in Modules 1, 5, 6 and 8. The strategy test says plainly when a rule never trades, a stake eaten by costs shows a loss of 100% and stays there, and chart keys and the Judge's notes sit inside a phone screen.
2026-10-20 6 more tested
- The ledger now covers 533 approaches: 527 rejected, 5 waiting on forward data. It was 531, 525 and 5.
- Rounds 109-110 added; every one is in the ledger with its verdict.
- The course is now called Honest Strategy Testing. The name says what the course teaches rather than how many strategies passed, which is a number that can change. Every place that states the pass count now reads it from the ledger, like every other number on the site.
- The ledger now covers 531 approaches: 525 rejected, 5 waiting on forward data. It was 529, 524 and 4.
- Rounds 107-108 added; every one is in the ledger with its verdict.
- The ledger now covers 529 approaches: 524 rejected, 4 waiting on forward data. It was 528, 523 and 4.
- Round 106 added; every one is in the ledger with its verdict.
- The ledger now covers 528 approaches: 523 rejected, 4 waiting on forward data. It was 526, 522 and 3.
- Rounds 104-105 added; every one is in the ledger with its verdict.
- The ledger now covers 526 approaches: 522 rejected, 3 waiting on forward data. It was 522, 518 and 3.
- Rounds 100-103 added; every one is in the ledger with its verdict.
- The ledger now covers 522 approaches: 518 rejected, 3 waiting on forward data. It was 521, 517 and 3.
- Round 99 added; every one is in the ledger with its verdict.
- The ledger now covers 521 approaches: 517 rejected, 3 waiting on forward data. It was 520, 516 and 3.
- Round 98 added; every one is in the ledger with its verdict.
- The ledger now covers 520 approaches: 516 rejected, 3 waiting on forward data. It was 514, 511 and 2.
- Rounds 92-97 added; every one is in the ledger with its verdict.
- The strategy test needs no download now. Pick one of six markets: the S&P 500, the Nasdaq 100, US small caps, gold, long Treasuries or Bitcoin. The prices load with the tool, adjusted for dividends and splits and refreshed with the site. Your own file still works. Every menu choice is explained in plain words as you make it, every condition in the result says why it matters, and each chart carries a colour key drawn the way the chart draws it.
- Easier to use with a keyboard, a screen reader or a thumb. Every page has a skip link and a main region, the sales page shows its sections even if its script never runs, and every standalone link, slider and file picker is at least 44 pixels tall on a phone. In the ledger, a selected filter is now readable in the dark theme.
- One type scale. The pages used over fifty different text sizes and twelve corner radii; they now use seven fluid sizes, fourteen fixed steps and six radii, the same on the site and in both packages.
- Nothing scrolls sideways any more. Long formulas break at their own "where" and "and" instead of hiding behind a scrollbar, and on a phone wide tables reflow rather than cut off columns.
2026-10-19 4 more tested
- The ledger now covers 514 approaches: 511 rejected, 2 waiting on forward data. It was 510, 507 and 2.
- Rounds 88-91 added; every one is in the ledger with its verdict.
- The ledger now covers 510 approaches: 507 rejected, 2 waiting on forward data. It was 506, 503 and 2.
- Rounds 85-87 added; every one is in the ledger with its verdict.
- A fifth calculator: test a strategy written as one sentence. Load a file of daily prices, describe a rule in one line, and the page runs it with the fair benchmark, the placebo sweep, the settings next to yours, doubled costs and the noise floor, then says what happened and when: the deepest fall with its dates, the longest dry spell, and whether the second half of the file still agreed with the first. The verdict is the ledger's: fail, or candidate. It ships in both packages as well, and its arithmetic is checked against a Python rerun before every deploy.
2026-10-16 1 more tested
- The ledger now covers 506 approaches: 503 rejected, 2 waiting on forward data. It was 503, 500 and 2.
- Rounds 83-84 added; every one is in the ledger with its verdict.
- The ledger now covers 503 approaches: 500 rejected, 2 waiting on forward data. It was 501, 499 and 1.
- Rounds 81-82 added; every one is in the ledger with its verdict.
- The ledger now covers 501 approaches: 499 rejected, 1 waiting on forward data. It was 500, 498 and 1.
- Round 80 added; every one is in the ledger with its verdict.
2026-10-15 6 more tested
- The ledger now covers 500 approaches: 498 rejected, 1 waiting on forward data. It was 497, 495 and 1.
- Rounds 77-79 added; every one is in the ledger with its verdict.
- The ledger now covers 493 approaches: 491 rejected, 1 waiting on forward data. It was 490, 489 and 0.
- Round 76 added; it is in the ledger with its verdict.
- Candidates are their own state. A strategy that cleared the backtest gate and is waiting on forward data is neither rejected nor a survivor, and the page, the dot grid and the social card now show it as such, read from the ledger's own CANDIDATE rows.
- The FINRA short-interest source states its real floor. The API serves nothing before December 2017, whatever start date you ask for; the preset, the README and the function's own trap note now say so, instead of promising 2010 and delivering 2018 after a long wait.
- The bug count is read from the catalogue at build time, like every other count, so adding a bug to Module 7 moves the sales page, Module 1, Module 8, the tools page and both READMEs with it.
- The Bug Catalogue's toolkit README describes what that tier ships: one file beside the note, no launcher, and the recomputation harness inside it.
- Module 5's trend-filter case states both halves of its result: drawdown cut against buy-and-hold in 7 of 7 windows, and the fair benchmark beaten in 0 of 7. The sales page says the same.
- Reading time is one figure everywhere: about three hours, which is what the eight modules add up to.
- The social card lays its dots out from the count, so the band fits whatever the ledger reaches.
- The Judge accepts a count typed as text in a JSON file, and treats 150, 150.0 and "150" as one setting in both places it compares the sweep.
- The toolkit's cache folder is fixed, and the panel says why: the server only ever reads files from inside it, which is what stops a page in another tab using it to read your disk.
- The toolkit README describes all five tabs, and says that contact.json lives in your user config folder rather than beside the course, so sharing the course cannot share your address.
- The ledger now covers 490 approaches, 489 of them rejected. It was 484 and 483.
- Round 75 — IBS on four fresh indices — FAIL on all four, and it revises Round 74 (2026-10-14)
- Round 74 — RSI(2) and IBS on the S&P 500 — both FAIL, one by decay and one by size (2026-10-14)
2026-10-14 The rejection, drawn
- The ledger now covers 484 approaches, 483 of them rejected. It was 482 and 481.
- Round 73 — routine vs opportunistic insiders — FAIL, and it is not the costs (2026-10-14)
- The ledger now covers 482 approaches, 481 of them rejected. It was 479 and 478.
- Round 72 — Treasury October-March seasonality — FAIL, and the family dies with it (2026-10-14)
- Round 71 — VIX/VIX3M spot slope as a risk-off filter — FAIL (2026-10-14)
- A twelfth figure, in Module 6 and on the sales page. The opening-range breakout against the twenty arbitrary anchors it was tested against: the real 09:30 rule at +0.30, the median of its own placebos at +0.37, the best of them at +1.15. The numbers were already in the table beside it; the picture is there because the shortest bar being the real one is the entire finding and a table does not say that at a glance.
- The 479 dots on the sales page now finish before the one that survived arrives, instead of it appearing mid-wave like any other.
- Fixed, in Tool 2: a list of returns that empties the account was scored rather than refused. Pasting percentages into the decimals box -- 2.1 meaning +2.1% read as +210% -- produced a Sharpe of -0.87 beside a worst fall of -158.0%, and an account cannot lose more than everything. Past -100% the balance is negative and every later return changes sign. The equity-column and trade-list modes already refused this; the two returns modes now do too, naming the period it happens on and which option was probably meant.
2026-10-07 The arithmetic, in full
- Every number derived, not asserted -- and placed in the lesson that uses it rather than in an appendix. Why a Sharpe annualises by the square root of time, and the full derivation of the noise floor from the distribution of the maximum of N draws (Module 1); why "it was profitable" carries almost no information (Module 5); the placebo sweep as an exact permutation test (Module 6); the standard error that decides how long you must watch a strategy, and the seven conditions stated formally (Module 8). Including a worked example of a formula that erred in the flattering direction, and how you would notice.
- Two new figures, bringing the course to eleven: the distribution of the best of N attempts, and the confidence interval around a measured Sharpe as the forward record lengthens.
- Two data sources repaired. FINRA had switched its default response from JSON to CSV, so the parse broke while the endpoint kept returning 200. HistData's download form grew a token that has to be scraped. Both now fetch; a new
verify_sources.pychecks every source through every preset in one command. - The toolkit judge fills itself in. Paste your returns and your benchmark's and it works out six of the nine figures it needs, instead of asking you to compute them elsewhere.
- The data panel is grouped and preset-driven. 37 named presets — "3-month T-bill", "Fama-French 3 factors, daily" — rather than raw argument names.
- Fixed: SEC quarterly data reported its date range as 1970 because the dates are stored as YYYYMMDD integers.
2026-10-04 Interactive figures in every module
- Nine live figures added across the eight modules. Module 1 now runs a thousand coin-flippers on demand and shows you the luckiest. Module 4 writes your pre-registration and computes the noise floor into it. Module 8 tells you how many years of forward data a given Sharpe needs.
- The toolkit opens with a double-click.
START-TOOLKIT.batandSTART-TOOLKIT.commandlaunch a local control panel: score a candidate against the seven conditions as you type, fetch any of the twelve data sources by clicking, and see what you have already cached and how stale it is. - The thirteen pre-registrations are searchable in one page rather than one file at a time.
2026-10-03 The scoring tools, rewritten
- Tool 2 takes four kinds of input, including profit-per-trade and a pasted broker CSV with a column picker. The old version assumed you had a tidy list of daily returns, which almost nobody does.
- Tool 3 builds the comparison list for you. Pick what your strategy depends on, type the value, and it generates twenty arbitrary alternatives to test against — plus instructions for running them in TradingView, MetaTrader or Python.
- Tool 4 has market presets, so you no longer have to source benchmark figures yourself.
- A full explainer of what a Sharpe ratio is, where every platform already prints yours, and what each range actually means.
2026-10-02 The ledger became searchable
- The whole search in one filterable page — every family, round, price series and screened source, with its verdict and the reason where one was reached.
2026-09-28 First complete build
- Eight modules, the data map, the scoring toolkit, the ledger and the pre-registrations, packaged as a single download that works offline.