Our test results, including what failed
Every number on this page comes from our own tests, with the date and the limits. Where a tool missed its pass mark, we say so.
At a glance
- Tip Check: flagged 92% and 98% of scam messages on two sealed test sets, no false alarms on honest ones.
- Truth Test: built on 76,913 strategy tests on past prices; indicator and crypto strategies are labelled "Unproven".
- Prop Coach: in 3,000 simulated challenges it passed 11.4% vs 4.9% for the best fixed size, at the same blow-up rate.
- Trade Planner: charges hand-checked on 7 markets.
- What failed: listed below with dates, and tools that missed their bar were removed.
How to read this page
Trading Checkup makes free tools that show trading risk and costs, and small fixed-price checking services. We never give tips. Pass mark = the score a tool must reach, written down before the test. Planted = errors we put in on purpose, so we know every right answer. Fresh = test material the tool has never seen.
Our testing record: what each number counts
These are the three numbers on our home page. They come from our own tests on past prices, not from customers. Counted 21-22 Oct 2026 from the test logs; shown on the site since then (page updated 5 Nov 2026).
76,913 strategy tests
What it counts: one trading rule (107 rule definitions) run on one market, kept only when it made at least 40 trades (30 for the indicator walk-forward split) and had a result on data the rule had not seen. Random "noise" strategies are counted separately (12,459).
Data: past daily and hourly prices up to October 2026, with costs taken off every trade.
Result, including failures: green verdicts held up 65-69% of the time on later data (about 42% for all strategies); crypto and always-in-the-market indicator strategies were not predictable, and the Truth Test labels them "Unproven".
5.87M forecasts tested
What it counts: 5,872,144 blow-up odds (the chance a simulated account breaks a loss limit), each checked against what then happened. A second check of 3.18 million more is not added to this number.
Data: simulated accounts on past hourly prices; tested 21-22 Oct 2026.
Result, including failures: shares within about 5 points, currencies about 6, crypto about 7-9. Simulated accounts, not customer results.
1,805 markets covered
What it counts: different instruments in our test data: 1,183 shares, 520 crypto coins (including coins that were later delisted), 63 currency pairs and 39 commodities, from 33 price sources.
Data: 2,925,763 market-days of daily prices up to October 2026.
Limits: past prices only. A market in the data does not mean every tool was tested on it.
Our test rules
The pass mark is written first
Before a test runs we write down what counts as a pass. We never lower it after seeing the result.
Fresh test material
Final tests use material the tool has never seen, used once. Practice material does not count.
A separate reviewer
Where quality is a judgement (write-ups, explanations), a separate reviewer scores it - not the person who built the tool.
Fails are published
If a tool misses its bar we say so, fix it or take it off the site. On 4 Nov 2026 we removed three tools that missed their bar, then retired tools that other sites already give away free.
Paid services
Four services passed their final test on fresh material before we offered them. Subtitles did not, so a native speaker checks every line.
- Planted errors found
- 93.1% · 93.8%
- Serious errors found
- 95.5% · 96.2%
- Clean models flagged by mistake
- fewer than 1 in 10
Two fresh test rounds on made-up financial models with errors put in on purpose (planted), so we know every right answer. Same version both times.
- Limits
- Synthetic test files, not client models. It finds formula and structure errors; it does not judge your assumptions. A person reviews every report.
- Before
- Versions 1 to 4 (28-29 Oct) did not pass every bar; the last two missed only on the quality of the plain-English explanations.
- Made-up facts in answers, our bot (round 1 · 2)
- 0% · 2.3%
- Same questions, plain AI chat
- 7.5% · 13.3%
- Questions the files don't answer
- refused
Two fresh test rounds on public pages and sample business documents; every bar passed in both.
- Limits
- Tested on our sample documents, not yours - you get your own test report before go-live. Speed was measured on a busy server.
- Before
- Versions 1 to 3 did not pass: re-worded questions were missed and some answers were too slow.
MT5 / MT4 EA code review
Passed- Planted bugs found
- 98 of 98
- Test rounds passed
- 2 of 2
- Extra check: bugs hidden by an outside writer
- 4 of 7
Expert Advisors (trading robots for MetaTrader) with bugs put in on purpose, two rounds in a row. The extra outside-writer check is not part of the pass mark - it is why a person reads every line.
- Limits
- Bugs written by someone else were found only 4 of 7 times, so a person reads every line of every report. A clean review is not a profit forecast.
- Before
- Version 1 passed 8 of 10 bars.
- Pass marks met, fresh set 1
- 18 of 18
- Pass marks met, fresh set 2
- 18 of 18
- Known problems found (round before)
- 97-100%
Fresh public test pages each round; write-ups rated by a separate reviewer.
- Limits
- Automated tools find only part of accessibility problems, so a person checks every finding. Not a legal certificate.
- Before
- Versions 1 to 5 (29 Oct - 3 Nov) did not pass: detection was good, but the written fixes scored just under our bar (and one round was too slow).
Checks (free)
The tools that stay on the site, with the result of their latest test.
Trade Planner
Exact maths · Oct 20267 of 7 markets
Charges matched our hand calculations for the fee schedules of 7 markets (India, US, UK, Australia, crypto) at the time of testing.
Limits: Fee tables change - check your broker's current schedule.
Prop Coach and Challenge Cockpit
Odds · 21-22 Oct 20265.87M forecasts
Blow-up odds were checked against 5.87 million forecasts on simulated accounts from past prices: shares within about 5 points, currencies about 6, crypto about 7-9. Coach sizing kept the blow-up rate close to the level chosen (aimed 10%, got 10.7%).
Limits: Simulated accounts, not customer results. Strict 1-step rules can be off by up to about 15 points, and intraday trailing-drawdown rules are a rough guide only.
Tip Check
Warning signs · 19 Oct 202692% · 98%
On two test sets of example scam messages in 20 languages it flagged 92% and 98%, with no false alarms on the honest examples.
Limits: The test messages were written for testing; real messages may differ. It shows warning signs - it cannot prove a message is safe.
Truth Test
Verdict · 21-22 Oct 2026held up 2 in 3
The verdict was checked on later data it had not seen: green verdicts held up 65-69% of the time (about 42% for all strategies), red 37%. So a green verdict is still wrong about 1 time in 3.
Limits: Crypto and always-in-the-market indicator strategies were not predictable, and the tool says so.
What we removed and why
On 4 November 2026 we took these off the site. Their old links now open the closest tool we stand behind.
Trade Label
Weakest measured result on the site (price lower 3 hours after a jump only 59-64% of the time); it reads like a sell signal, and the page promised a browser extension that does not exist.
TradingView scripts
The page told people to search TradingView for two scripts that were never published there, and the indicator marks buy-style signals. The Truth Test does the same check on your own trades.
Security Checkup message checker
Failed our own bar on fresh test messages (74% caught vs our 85% bar) and does the same job as Tip Check. Message checks now go to Tip Check.
What these numbers do not mean
- They are results of our own tests on test material and past data - not customer results and not a promise.
- Past results do not predict future results. None of this is financial advice.
- A passed test means a tool met its bar on that test. It can still be wrong on your case, so the limits are listed with each result.
Found a mistake or have a question? Use the contact details on our About page - we will check it.