phit

Data & Workflow

Statistics, Modeling, & Visualization

Test suites

Verification

Models

Numerics

Test suites

On this page

Scope

Every PHP test suite in the release check: saved references from R, lme4, mpmath, Python, and phibble, and phit's own contracts.

References

Reference
Saved reference fixtures; the reference manifest lists each one's generator, versions, and hash

Data

Fixture provenance
generators, versions, hashes, and tolerance policies Source

Method

Each suite runs against the release build of the extension. Its exit status and its own summary line are recorded; each row is one suite.

Check
Comparison with a reference implementation
Recorded
2026-11-01, PHP 8.5.10, macOS 27.0.1 (arm64), commit 066e2e1

Tolerances

exact equality

Applies to: Exact results

Results, row accounting, parsed values, types, errors

|x − r| ≤ 1e-9 + 1e-7 × |r|

Applies to: Model fixtures

OLS and GLMs

|x − r| ≤ 1e-12 + 1e-12 × |r|

Applies to: Model matrices

Columns, terms, and rows exact; lm and glm fits 1e-9 + 1e-7·|r|

|x − r| ≤ 1e-5 + 1e-5 × |r|

Applies to: Mixed-model estimates

Likelihood 5e-6 + 1e-9·|r|; glmer as in the binomial mixed model

|x − r| ≤ 1e-7 + 1e-6 × |r|

Applies to: Predictions

Fits, intervals, standard errors

|x − r| ≤ 1e-7 + 1e-6 × |r|

Applies to: Weighted and offset GLMs

|x − r| ≤ 1e-10

Applies to: Exposure intercept and standard error

Likelihood and counts 1e-9

|x − r| ≤ 1e-300 + 1e-12 × |r|

Applies to: Distribution functions

Logarithms relative to max(1, |r|), quantiles 1e-15 + 1e-12·|r|

|x − r| ≤ 1e-12 × |r|; at r = 0, x = 0 exactly

Applies to: Descriptive statistics and smoothing

Zeros, ECDF counts, and type-7 identities exact

|x − r| ≤ 1e-15 + 1e-12 × |r|

Applies to: Sampling

ChaCha uniforms and discrete samples exact

|x − r| ≤ 1e-11 + 1e-11 × |r|

Applies to: Posterior-draw kernels

Moments, covariance maps, affine draws, log determinants

as stated in the record

Applies to: Extreme inputs

Per-case tolerances stored with the fixture

x: package value; r: reference value.

Results

24 comparisons: 24 pass, 0 fail, 0 skipped.

Every comparison
Case Quantity Criterion Values Status
OLS against saved R lm() fixtures PHP OLS integration tests passed |x − r| ≤ 1e-9 + 1e-7 × |r| pass
Logistic regression against saved R glm() fixtures PHP logistic integration tests passed |x − r| ≤ 1e-9 + 1e-7 × |r| pass
Probit regression against saved R glm() fixtures PHP probit integration tests passed |x − r| ≤ 1e-9 + 1e-7 × |r| pass
Poisson regression against saved R glm() fixtures PHP poisson integration tests passed |x − r| ≤ 1e-9 + 1e-7 × |r| pass
Deprecated aliases return their replacement's result PHP 0.4 names: 10 deprecated aliases and fit_ols dispatch passed exact equality pass
Formula designs against R model.matrix(), and lm()/glm() fits PHP formula tests passed (33 model.matrix, 14 lm, 15 glm fits; 30007 R comparisons) |x − r| ≤ 1e-12 + 1e-12 × |r| 30,007 pass
Formula mixed models against lme4 lmer() and glmer() PHP formula mixed-model tests passed (7 lmer and 3 glmer lme4 fits; 782 comparisons) |x − r| ≤ 1e-5 + 1e-5 × |r| 782 pass
predict() against R predict() PHP predict tests passed (25 R predict() cases; 877 comparisons) |x − r| ≤ 1e-7 + 1e-6 × |r| 877 pass
Random intercept against saved lme4 fixtures PHP mixed-model integration tests passed |x − r| ≤ 1e-5 + 1e-5 × |r| pass
Weighted and offset GLMs against base R PHP base-R compatibility tests passed |x − r| ≤ 1e-7 + 1e-6 × |r| pass
Mixed-model structures against four lme4 optimizers PHP expanded mixed-model tests passed (15 models, four lme4 optimizers) |x − r| ≤ 1e-5 + 1e-5 × |r| pass
Binomial random intercept against lme4 glmer() PHP binomial random-intercept integration tests passed (28 lme4 fits) |x − r| ≤ 1e-5 + 1e-5 × |r| pass
Result contracts and exception categories PHP stability, result contracts and exception categories passed exact equality pass
Automatic intercepts and Poisson exposure offsets PHP automatic intercepts and Poisson exposure offsets passed |x − r| ≤ 1e-10 pass
Distribution functions against R and mpmath PHP distribution tests passed (8618 comparisons, 25 independently adjudicated references, 43 documented boundary differences) |x − r| ≤ 1e-300 + 1e-12 × |r| 8,618 pass
Descriptive statistics against R PHP descriptive tests passed (145 R comparisons, missing policies and validation) |x − r| ≤ 1e-12 × |r|; at r = 0, x = 0 exactly 145 pass
Kernel densities, bandwidths, interpolation, and ECDF against R PHP smoothing tests passed (15639 R comparisons, missing policies and validation) |x − r| ≤ 1e-12 × |r|; at r = 0, x = 0 exactly 15,639 pass
optimize() and uniroot() against R PHP optimize/uniroot tests passed (91 bit-for-bit R comparisons, evaluation counts and validation) exact equality 91 pass
Seeded sampling against independent ChaCha streams and R quantiles PHP RNG tests passed (3764 independent R/ChaCha values, seed isolation, validation and moments; uniform, categorical and permutation samplers) |x − r| ≤ 1e-15 + 1e-12 × |r| 3,764 pass
Posterior-draw kernels against 80-digit Python references 25 independent draw fixtures, identities and strict input contracts passed |x − r| ≤ 1e-11 + 1e-11 × |r| pass
CSV reader: parsing rules, options, types, and errors csv: 190 checks passed exact equality 190 pass
CSV reader against phibble's reader on a fixed corpus csv-conformance: 120 cases matched phibble's reader, at every block size and in chunks exact equality 120 pass
CSV reader against phibble's PHP reader on generated files csv-phibble: 6278 comparisons against phibble's PHP reader passed exact equality 6,278 pass
Extreme but identified inputs against base R PHP extreme-input checks passed as stated in the record pass

Reproduction

From the repository root 
cargo build -p phit --release --locked
PHP=$(command -v php) python3 docs/verifications/run_test_suites.py
Raw record
site/verifications/test-suites.json
Full results
docs/verifications/test-suites-results.json