Quantify
Takes a trade log. Decides whether its numbers survive correction for search, sampling error, serial dependence and regime. Writes a report that names every test it could not run.
The tool is Quantify. The package is
qvalid. The namequantifywas already taken on PyPI by an unrelated project when this one was published, so the distribution and the import carry the shorter name and the tool carries the one it was always called. See D045 and D058.
uv run python examples/validate_full.py
One command, no network, no configuration. It reads two files from tests/fixtures and writes
examples/output/report.{html,json,tex}.
What it decides, and on what basis
A Sharpe ratio computed from a trade log is almost always reported without the three things that
determine whether it means anything: how many configurations were tried before this one, how wide
the sampling interval around it is, and whether the returns are serially dependent. qvalid
computes the ratio and all three corrections, and refuses to state a verdict when any of them is
missing.
| Question | Method |
|---|---|
| Is the Sharpe distinguishable from zero? | Delta method interval, Mertens (2002), with Newey and West (1987) HAC |
| Does it survive the search that produced it? | Deflated Sharpe and PBO, Bailey and Lopez de Prado (2014, 2017) |
| Is one strategy better than another? | White (2000) Reality Check, Hansen (2005) SPA |
| How bad can the drawdown get? | Stationary bootstrap, Politis and Romano (1994), against Magdon Ismail et al. (2004) |
| What is the chance of ruin? | First passage with the Broadie, Glasserman and Kou (1997) continuity correction |
| Does the edge live in one regime? | Causal labelling, Welch (1951) unequal variance test |
| Would it pass a proprietary desk? | Barrier model over daily paths, rules declared in YAML |
The example, verbatim
The sample log has 760 trades over 2.91 years. The report says:
Sharpe (sqrt q) -0.91 interval [-2.58, 0.76]
observed max drawdown 0.154 quantile 0.68 of the simulated distribution
probability of ruin 0.211 at a barrier of 85,000
regime equality of means p = 1.6e-11 over 9 states
deflated_sharpe NOT_REQUESTED the number of configurations tested was not declared
verdict SUPPRESSED required sections did not run
The interval contains zero, so the ratio is not distinguishable from zero. The regime test finds a real difference. And the verdict is suppressed, not negative: the run cannot say how many configurations were tried, so no correction for search was applied, and a strategy whose search correction did not run is not comparable with one whose did. Estimating that number would fabricate the input that determines the answer.
This is the design. An absent test is a typed state in the report, never a missing field and never a pass. No aggregate grade is produced, because collapsing heterogeneous evidence into one letter is the defect the tool exists to correct.
And the other direction
A tool that has only ever been shown to say no is indistinguishable from one that cannot say
anything else. tests/fixtures/trades_winner.csv is a log with real positive expectancy and the
twenty configuration sweep that produced it:
Sharpe (sqrt q) 1.89 interval [0.75, 3.03]
probability against zero 0.99931
deflated Sharpe 0.87707 against the best of 20 configurations
PBO 0.208
minimum track record 201 periods required, 760 observed
verdict certainty equivalent +0.396
The two probabilities are the whole argument in one place. Measured against zero this Sharpe is all but certain. Measured against the best of the twenty configurations that were tried before it, it is merely likely. That gap is what a backtest report normally omits, and the gap is not a rounding difference: it is twelve points of confidence that belong to the search rather than to the strategy.
Install
pip install qvalid
qvalid ui # drag the log in; the three config files get drafted
Or the same three steps on the command line:
qvalid inspect trades.csv # a column mapping to start from
qvalid probe trades.csv -m mapping.yaml # a symbology, with the multiplier your file implies
qvalid validate trades.csv --config run.yaml --out report.html
inspect reads the header of your export and prints a mapping, marking what it could not
work out rather than choosing:
columns:
entry_ts: Open Time
entry_px: # UNRESOLVED, another field also matched Price; decide which gets it
exit_px: # UNRESOLVED, another field also matched Price; decide which gets it
It prints and never saves. The mapping records which column was read as what, which is what lets somebody else reproduce a number, so the file has to be one you chose. See D060.
probe then reads your numbers and inverts the P&L identity to recover what multiplier each
symbol implies, printing it beside an empty slot rather than into it:
symbols:
ESZ4:
multiplier: # implied 50, from 745 trades
# ESZ4: consistent with pnl_convention: NET (spread 0.0e+00 against 2.3e-02 for the other)
That last line is worth a note. Whether a P&L column is net or gross of costs was documented
here as undetectable, because the error leaves a residual of exactly one fee per trade and no
single trade breaks any tolerance. Across trades it is detectable after all: under the wrong
convention the implied multiplier scatters, under the right one it is constant. The limit is
measured rather than assumed. Once costs fall below the rounding of the P&L column the
difference is gone, and probe says so instead of answering. See D061.
Developing it rather than using it? Install from the checkout so the qvalid command is the
code you are editing, and not a copy from PyPI under the same name:
pip install -e .
Python 3.12 or newer. Importers for generic CSV, TradingView and NinjaTrader; a symbology map supplies contract multiplier and tick size, which the P&L coherence check needs.
Reproducibility
Two runs on the same input with the same seed, in one environment, produce byte identical reports with the timestamp excluded. That is the claim about the seed governing everything, and it is checked exactly.
Across environments the claim is weaker and measured rather than hoped for. Changing the version
of numpy or scipy moves two of the report's 144 values by one or two units in the last place,
because a third moment and an F survival function are reductions whose summation order those
libraries choose. Every run of the test suite compares against the committed reference at 1e-9
relative, which is a thousand times tighter than the six significant figures the report renders,
so the report a person reads is identical. Text is compared with no tolerance at all: a regime
identifier or a suppression reason that changed is not a rounding difference.
Nothing inside core calls the BLAS, because the BLAS splits long reductions across threads and
the summation order would then depend on how many cores the machine has.
Licence
MIT. Design notes, the mathematical specification and the decision log are in docs/,
in Portuguese.
Metadata
Release files for qvalid 1.11.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| qvalid-1.11.0.tar.gz | 547.2 kB | With provenance | Details for qvalid-1.11.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| qvalid-1.11.0-py3-none-any.whl | Python 3 | none | any | Details for qvalid-1.11.0-py3-none-any.whl |
Total release size: 725.1 kB
Release files / qvalid-1.11.0.tar.gz
| Download URL | qvalid-1.11.0.tar.gz |
|---|---|
| Size | 547.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
eec384e2cd630873b8cf222114f2cebff4506ad07835c3495d4b09863d56220c
|
|
BLAKE2b-256 checksum How to use checksums |
440b198b83f9e44395b344e1fdb32e9950fdea531971d36ba57b8afcf6b565c5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 6, 2026.
Transparency logRelease files / qvalid-1.11.0-py3-none-any.whl
| Download URL | qvalid-1.11.0-py3-none-any.whl |
|---|---|
| Size | 177.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
aca614370ff2932a396dce5cab1260fa856b0f848d8de6d8e4aa3bdc6015d018
|
|
BLAKE2b-256 checksum How to use checksums |
05136a155192cf94e8c7d3491e0c5fc72d908cb19328b49b51b640e83a7312b4
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 6, 2026.
Transparency logSecurity for qvalid 1.11.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for qvalid-1.11.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 6, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.