Paid local software · v1.1.0 · Private delivery

Audit the time your join could not know.

Temporal Boundary Auditor inspects declared event, availability, period, grid, duplication, and freshness rules behind a backward as-of join, then can compare caller-declared timestamp interpretations against an explicit reference series—locally, deterministically, and without copying row values into its evidence packet.

Request the $499 tool Check v1 fit Inspect failing sample Compare leakage boundaries

No account is required to run the delivered package. The implementation and delivery archive are not public downloads and are not published to GitHub. Public materials are synthetic evaluation evidence only.

01 / Why this toolAvailability is not the same as event time.

A normal as-of join can select a later revision unless the boundary separately models when each value became knowable.

02 / What it is notNot a feature store or join engine.

It does not build, repair, or serve training data. It audits a caller-declared boundary and retains aggregate evidence.

03 / Do not buy whenYour current stack already covers the boundary.

Keep pandas, DuckDB, Feast, Timefence, Hopsworks, or an internal control if it already covers availability, revisions, periods, grids, freshness, and the evidence you need.

Supported decision

Does this declared backward as-of join have observable temporal-boundary violations?

Time-series and feature pipelines can join a context row whose event already happened but whose value was not yet available, use a period before it completed, silently accept off-grid timestamps or duplicate temporal keys, or carry context farther than the declared freshness window. The auditor makes those mechanical failure modes visible under a strict user-supplied contract.

Workflow

Declare, inspect, retain.

01 / Declare

Write the boundary contract

Name contained relative files, formats, entity keys, time columns, expected grids, maximum ages, and resource ceilings.

Inspect contract schema

02 / Inspect

Run local mechanical checks

Evaluate RFC 3339 timestamps, keys, ordering, grids, availability-aware matching, revision ambiguity, period completion, and freshness.

Open adversarial contract

03 / Retain

Keep aggregate evidence

Store byte identities, counts, join summaries, rule findings, and explicit limitations without copying row values or source paths.

Inspect report schema

Best for: Data engineers, ML engineers, quantitative researchers, validation teams, and reviewers of point-in-time pipelines
Input: A strict JSON contract plus local UTF-8 CSV or JSON Lines datasets
Output: Aggregate audit or clock-probe JSON, streaming JSON Lines, and human-readable Markdown
License: Perpetual internal use by one purchasing legal entity; no redistribution, resale, publication, or service-bureau use
Distribution: Private delivery after accepted request and confirmed payment; no public package URL

Boundary

An audit is not causal proof. A ranking is not clock truth.

Checks are limited to the supplied contract and bytes. The tool does not establish factual correctness, complete lineage, the true timezone, absence of every leakage channel, statistical validity, compliance, production safety, or fitness for a decision. SHA-256 values are byte identifiers, not signatures or attestations.

Clock-interpretation probeOpen the optional v1.1 comparison of caller-declared timestamp interpretations.

New in v1.1.0

Compare candidate clocks without declaring a winner true.

When a timestamp column may be UTC, a fixed offset, or a named local zone, the read-only probe measures backward-match coverage against an explicit RFC 3339 reference series. Candidate modes are supplied by the operator; the tool never invents a timezone or rewrites source data.

01 / Declare

Name every candidate interpretation

Compare explicit RFC 3339, fixed offsets from −840 to +840 minutes, and available IANA zones under a declared backward tolerance.

Inspect probe contract

02 / Separate

Keep entities and DST boundaries visible

Matches stay within equal-length entity keys. Ambiguous and nonexistent local times are counted and excluded rather than silently folded or shifted.

Inspect DST-adversarial report

03 / Classify

Retain dominance, ambiguity, or insufficiency

Coverage and the runner-up gap produce DOMINANT, AMBIGUOUS, INSUFFICIENT_COVERAGE, or REFERENCE_INVALID—not a claim about the true clock.

Inspect probe report

python -m temporal_boundary_auditor probe-clock contract.json --out clock-output

A completed classification exits 0; invalid input or operation exits 1. The ranking can change with reference completeness, record density, tolerance, entities, omitted modes, daylight-saving transitions, or the local timezone database.

Audit and clock-probe runtimeInspect contract rules, classifications, release files, test coverage, and support.

Licensed runtime

One contract in. Three evidence files out.

python -m temporal_boundary_auditor audit contract.json --out audit-output

Exit 0 means no error-severity findings, 2 means the audit completed with error-severity findings, and 1 means the contract, input, or operation was invalid. A zero exit is not proof that all temporal leakage is absent.

Licensed delivery

Implementation, contracts, tests, and synthetic evidence.

Source-available Python packageDependency-free CLI and library under the one-organization commercial license.
Wheel and exact checksumsInstallable v1.1.0 wheel plus SHA-256-bound commercial archive.
Four strict JSON SchemasDraft 2020-12 contracts for audit and clock-probe inputs and complete aggregate reports.
Fifty-eight testsPositive, adversarial, path, limit, timestamp, DST, entity-isolation, determinism, privacy, and CLI coverage; one Windows symlink case may skip without privilege.
Four reference casesSynthetic clean and adversarial audit and clock-probe inputs with deterministic expected outputs.
One support roundOne consolidated installation or contract-format email received within 30 calendar days after delivery.

Public evaluation recordsOpen the fit check, evaluator guide, synthetic reports, schemas, pack, checksum, and exact offer.

Request to buy

Confirm the licensee and runtime fit before payment.

Send the legal-entity name, billing contact, country, intended Python version, and whether your evaluation inputs are CSV or JSON Lines. Do not send datasets, credentials, private source, customer data, or regulated data. A request is not an accepted order, and availability is not guaranteed until accepted in writing.

Request the $499 tool Read product terms