The findings we can name: 145 flags across 26 audited backtests

Lizard carries seven named findings, CryonX carries two, and Logan carries ten. All three went through the same engine on the same kind of tester report. Across our 26 default audits that engine has raised 145 flags under 22 distinct names, and the list of names it is able to raise at all is finite, short and published. This page is that list.

What a flag is

A flag is not a sentence we write. It is a record with five fields and it is produced by a rule that either fires or does not.

• A dimension. One of six, namely data quality, structure, costs, concentration, regime and prop fit.
• An id. A short stable name such as K-8020 or S-GRID. The prop fit ids carry the profile name inside them, so one rule evaluated against three rule sets can appear under three different ids.
• A severity. One of info, caution or red. There is a fourth level called ok and no flag ever carries it, because a rule that finds nothing writes nothing.
• A title and a detail. Both are generated from the measured numbers of that audit, which is why two audits with the same id rarely read the same.

The lamp on an audit page is then the highest severity among the flags of its dimension, and every lamp starts at ok. That single sentence is the whole aggregation rule, and it has a consequence worth holding on to. An ok lamp does not mean we looked and found the dimension healthy. It means none of the named rules for that dimension fired.

The twenty two names, ordered by how often they appear

The count is the number of the 26 default audits that carry the id. The 145 flags break down into 45 red, 59 caution and 41 info.

• P-iqcapital_classic-CONS, 19 audits. The consistency rule of the prop profile is breached in at least one year. This is the single most common finding in the catalog.
• K-8020, 18 audits. Eighty percent of the profit was made on a small number of days.
• R-NEGYEARS, 16 audits. Two or more calendar years closed negative.
• S-STACK, 15 audits. More than one position open at the same time, either as heavy stacking or as plain parallel entries.
• S-HEDGE, 12 audits. Long and short held at the same moment.
• DQ-SILENT-HEAD, 11 audits. The robot refused to trade the opening stretch of the window it was tested on.
• S-GRID, 7 audits. Entries added against an open losing basket, which is the averaging signature.
• C-SHARE, 6 audits. Costs eat a large share of the gross profit.
• C-SWAP, 6 audits. Financing costs exceed commission.
• R-ONEYEAR, 5 audits. One year carries more than 60 percent of the whole result.
• DQ-FAILS, 5 audits. More than half a percent of intended entries never filled in the simulation.
• DQ-TICKS, 4 audits. Less than half the window ran on real ticks.
• P-ftmo_challenge-SIZING, 3 audits. Death free only at a quarter of the tested size or smaller, measured against that profile.
• P-iqcapital_classic-MPL, 3 audits. The per position loss limit is breached even at one third size.
• DQ-RANDOM, 3 audits. The tester log contains randomization prints, so a single run understates the spread of outcomes.
• R-LONGBETA, 3 audits. The profit leans on the long side while the instrument itself rose.
• S-MARTINGALE, 2 audits. Volume rises after losses more often than after wins.
• P-iqcapital_classic-SIZING, 2 audits. As above, against the IQ Capital rule set.
• P-generic_6pct_trailing-SIZING, 2 audits. As above, against the generic trailing rule set.
• P-iqcapital_classic-DEATH, 1 audit. No sizing down to one eighth survives the drawdown floor.
• P-generic_6pct_trailing-DEATH, 1 audit. The same finding against the second profile, and it is the same audit, Scalping Robot Pro.
• EQ-FLOAT, 1 audit. The floating drawdown ran far deeper than the closed trade curve shows.

Two things in that list are worth a second look. The most common finding is not about the trading at all, it is about a funding rule, and the rarest finding is the one that measures risk while a position is still open.

The thinnest and the thickest audit

CryonX and Gold Snap carry two flags each and no default audit carries fewer. Logan carries ten, and so does Pulse Engine, which is measured but not listed. No default audit carries none at all.

Logan is worth naming because five of its ten flags are prop fit findings across the three profiles. A single structural property, in its case sizing that cannot be scaled down far enough, is counted once per rule set it collides with. Reading the raw flag count as a ranking would therefore punish it three times for one trait, which is exactly why the audit page shows six lamps and not one number.

Thirteen checks have never fired, and most of them cannot

Counting each prop fit rule once rather than once per profile, the engine defines 32 named checks. 19 of them have appeared in a published audit. 13 have not, and for most of them the reason is our own test setup rather than the robots.

• The missing log check. It fires when no tester log was supplied. All 26 default audits record a log, so it has never had an opportunity.
• The zero delay check. It fires when the test ran without execution delay. All 26 ran at 10 milliseconds, which is the fixed value of our protocol.
• The two pairing checks. They fire when exits could not be matched to entries from the log. All 33 stored audits record a pairing confidence of exact.
• The three integrity checks. They fire on a failed balance reconstruction, on deals out of chronological order and on duplicated deal numbers. No default audit records any of the three.
• The zero commission check. It fires when commission is zero and no retrofit was applied. 22 of 26 carry a retrofit, and the remaining four carry commission natively at -22.14, -5.59, -697.86 and -41.43 USD.
• The zero swap check. It fires when swap is zero although positions were held over a rollover. Exactly two runs show a swap of zero, Lizard and Prop Firm Gold EA, and both show 0 overnight trades. Not paying financing you never owed is not a modelling gap.
• The carrier check. It fires when one sub strategy produces the result while the others are ballast. 17 of 26 audits detect more than one sub strategy, and Logan detects 76, so the rule had plenty of chances and stayed quiet in all of them.
• The overnight ban and the weekend ban. Both fire only against a profile that forbids the holding. All three profiles we run allow overnight and weekend positions, so these two can never fire until a fourth profile is added.
• The data head check. It fires when the window opens before the symbol has history. No default audit records that classification.

That is the honest shape of a clean data quality lamp in this catalog. It is partly a statement about the report and partly a statement about the fixed conditions we test under.

Run these checks yourself

• Ask what the absence of a warning was allowed to mean. A review with no complaints has either checked something and found it clean or never had a rule for it. The two look identical from the outside and only a published list of checks tells them apart.
• Count the findings before you weigh them. Ten flags in one audit and two in another is not a ranking when five of the ten are the same trait measured against three rule sets.
• Read the detail line, not the title. Two audits under the same id can differ by orders of magnitude, because the numbers in the detail come from that run alone.

Every figure above is public in the audit files linked in place, and the category pages list the verdict lamps for free. Your own tester report? The browser check is free.