veridist

English | فارسی | Deutsch

Status

veridist 1.0.0 is an evidence-backed public contract release. It specifies and tests bounded delivery, replayability, pass budgets, transactional retry, checkpoint compatibility, typed failures, execution outcomes, and redacted provenance.

This build includes fixed-location Exponential, Weibull-minimum, and Lognormal MLE cells for exact and independently right-censored lifetimes. They provide point estimates when finite solutions exist and typed failures otherwise. The uncensored exponential cell also has refit Monte Carlo KS, AD, and CvM goodness-of-fit, AIC/BIC, calibration summaries, and adequacy-gated selection. Inference is restricted to that declared cell and requires a caller-owned NumPy generator. Its public CSV path is strict: UTF-8 CSV with exactly time,event_observed, event token 1, and right-censoring token 0. It executes one iterator pass with a declared logical retained-payload chunk budget and returns a closed, typed execution result. This is not a generic CSV reader or a portable RSS, throughput, cancellation, retry, checkpoint, or broad out-of-core claim. Retained evidence establishes bounded internal payload only for the measured 10k/100k/1m by 32KiB/64KiB/128KiB matrix; it does not establish a general big-data or high-throughput capability.

IterableDataSource is the reusable public stream adapter for caller-owned chunk iterables. Its immutable metadata explicitly declares one-pass or replayable acquisition: a single-pass source fails with a typed pass-budget error if acquired twice, while a replayable source requires an iterator factory. BoundedChunkBuffer charges queued and consumer-held chunks until BufferedChunk.release(); callers must release received chunks, normally in finally. The only shipped file adapter remains the strict CSV lifetime adapter—this does not add generic CSV, Parquet, Arrow, dataframe, database, or broad out-of-core adapters.

For a sequential pure reducer, veridist.engine.SQLiteCheckpointStore provides durable local checkpoint state with generation-based compare-and-swap and cross-process SQLite locking. It is limited to one host and a local filesystem; fit_exponential_checkpointed_csv supports explicit source-revision-bound resume for the strict lifetime CSV path.

The separate scalar surface exposes immutable FAMILY_REGISTRY metadata for normal, gamma, Weibull-minimum, lognormal, and right-Gumbel; scalar evaluate_log_density, CDF, survival, quantile, and caller-owned RNG sampling operations; and exact-state reduce_log_likelihood_chunks. These operations are scalar and the fit/inference capability matrix remains narrower than the operation registry. The reducer represents successful binary64 terms exactly and rounds the final total once; its unsigned-64 count cap implies a 2162-bit exact-total bound. Its retained 10k/100k/1m evidence is scoped to tested normal streams.

Candidate scale measurement is deliberately manual: the veridist-scale-evidence workflow first binds a clean, full candidate SHA and runs the evidence contracts, then measures the public iterable likelihood and strict CSV/exponential paths on Linux and Windows. It retains artifacts only after their own fail-closed SHA and schema validation. Merely having this workflow, or a historical artifact, is not evidence for a new candidate and is not a throughput or RSS claim.

Install an evaluation build

Install from the nested source project after cloning the repository:

git clone https://github.com/alisadeghiaghili/veridist.git
cd veridist/python
python -m pip install .

Or install a wheel that you built or obtained from a specific verified run:

python -m pip install /path/to/veridist-1.0.0-py3-none-any.whl

The project does not direct users to install an unreleased package name from a public index.

Try the experimental vertical

from pathlib import Path
from tempfile import TemporaryDirectory
from veridist import CsvLifetimeLimits, CsvLifetimeSchema, PublicSourceId, fit_exponential_csv
from veridist.families import ExponentialFitSuccess

with TemporaryDirectory() as directory:
    path = Path(directory) / "lifetimes.csv"
    path.write_text("time,event_observed\n1,1\n1,0\n", encoding="utf-8")
    fit = fit_exponential_csv(
        path, schema=CsvLifetimeSchema("time", "event_observed"),
        source_id=PublicSourceId("src_0123456789abcdef0123456789abcdef"),
        limits=CsvLifetimeLimits(32768, 32768),
    ).fit
assert isinstance(fit, ExponentialFitSuccess)
assert fit.rate == 0.5
assert fit.inference == "not_provided"
assert fit.censoring_assumption == "independent_right_censoring"

See the documentation toolchain and the repository's evidence ledger for implemented checks and explicit limits.

License

BUSL-1.1 with an Apache-2.0 additional-use grant for personal, non-commercial use; see LICENSE.