INIT
$ tabdot audit --strict·DuckDB Kernel·Zero Raw Retention

Data goes in.

The truth comes out.

Deterministic tabular data auditing with zero LLM hallucinations. Catch synthetic corruption, unmask hidden sentinels, and certify production datasets in milliseconds.

$pip install tabdot
Direct Benchmarks:
fin_ledger_2026_q3.parquet

Logarithmic First-Digit Distribution

Comparing empirical digit frequencies against Benford's Law (N = 500,000)

Digit 7 Anomaly Flagged (p < 0.001)
30.4%
1
17.2%
2
12.8%
3
9.5%
4
7.8%
5
6.4%
6
9.6%
7
4.8%
8
4.2%
9
Chi-Square Statistic42.18 (Critical: 15.51)
P-Value0.0003 (Rejection)
Sample Size500,000 Rows
Conformity VerdictNon-Conforming (Artifact)
ZERO-COPY TABULAR INTEGRATIONS:
Apache Arrow
Apache Parquet
DuckDB
Polars
dbt Core
Apache Airflow
Snowflake
FastAPI
PostgreSQL
Apache Arrow
Apache Parquet
DuckDB
Polars
dbt Core
Apache Airflow
Snowflake
FastAPI
PostgreSQL
Apache Arrow
Apache Parquet
DuckDB
Polars
dbt Core
Apache Airflow
Snowflake
FastAPI
PostgreSQL
CompletenessValidityConsistencyAuthenticity

Deterministic Forensics for High-Throughput Tabular Pipelines

Non-parametric statistical verification built directly into your streaming pipelines. Keep every audit reproducible, with zero raw data retention, sub-second latency, and cryptographic certificates built in.

Forensic Invariant Profiling

A deterministic forensic engine engineered for columnar streams. Scans tabular files in memory to detect disguised sentinels (-999, missing nulls), terminal float roundings, and Benford violations with mathematical p-values.

FORENSIC SIGNAL VIOLATION LOGN=25,000 ROWS
Benford Distribution Breach: Digit 1 OverrepresentedRow 1,482
Observed frequency 46.8% vs expected 30.1% (p < 0.0001, Nigrini MAD 0.0428).
Terminal Float Clumping DetectedRow 4,891
84.2% of continuous numerical values terminate exactly in .00 or .50 (synthetic rounding tell).

Cryptographic Schema & Invariant Enforcement

Manage schemas and enforce strict data contracts. Validates RFC 5322 email syntax, UUIDv4 uniqueness, monotonic timestamps, and logical price ranges without raw disk storage, keeping schemas and production contracts in sync.

INVARIANT VERIFICATION MATRIXSTATUS
RFC 5322 Email Syntax
Zero syntax violations in customer_email
100% PASS
Monotonic Timestamps (ISO-8601)
Chronological ordering across events
100% PASS

Continuous Kolmogorov-Smirnov & EMD Diff

Compare baseline vs target releases in milliseconds. Computes two-sample Kolmogorov-Smirnov continuous tests and 1D Earth Mover's Distance to catch silent distribution drift before models degrade.

KOLMOGOROV-SMIRNOV & WASSERSTEIN DRIFTDELTA SCORE
Baseline v1.2 vs Target IngestΔ -56.6 PTS
D-statistic: 0.4128 · p-value < 0.0001 · Significant statistical drift identified
Wasserstein: 4.82 EMDGrade A → Grade F

Automate CI/CD Tabular Gates

Take deterministic action on your datasets from anywhere. Trigger audits from GitHub Actions, fail dbt runs on invariant breaches, and drive pipelines directly from your Terminal or orchestrator.

Install CLI & API
tabdot — zsh — 80×24
user@macbook:~$tabdot audit --file transactions_2026.parquet --strict
[INFO] Initializing zero-copy Polars streaming kernel...
[INFO] Ephemeral memory allocated: 24.2 MB (zero raw row persistence)
[EXEC] Scanning 25,000 rows across 5 columns in 38ms:
› RFC 5322 Email: 100% conforming (0 invalid)
› Benford's Law First-Digit: MAD 0.0041, Chi² p=0.48 (Conforming)
› Terminal Float Clumping: 0.04% baseline (Natural continuous)
[PASS] SHA-256: 7f83b1657ff1... · Trust Grade A Certified94.8 / 100 PTS

Scientific Rigor & Ephemeral Architecture

High-throughput columnar speed, zero raw row retention, and cryptographic certification built in.

LATENCY<45MS
<0ms

Polars zero-copy stream processing 100 MB in under 800ms without RAM cloning.

RAW RETENTIONZERO
0.00%

Zero raw row retention. Ephemeral in-memory aggregation with 1-click panic wipeout.

ATTESTATIONFIPS 180-4
SHA-256

Tamper-evident standalone HTML export with cryptographic checksums.

DRIFT TEST2-SAMPLE
KS & EMD

Non-parametric Kolmogorov-Smirnov and 1D Wasserstein distance calculations.

RELIABILITYSLA
0.00%

Uptime SLA for enterprise private cloud and on-premise air-gapped clusters.

DATA PRIVACY100% LOCAL
Zero LLM API

Zero dataset rows are ever transmitted to third-party commercial LLM providers.

INTEGRATIONREST & CLI
CI/CD Gates

Block deployment in GitHub Actions or Airflow immediately upon invariant breach.

MATHEMATICAL RIGORN ≥ 30
Guardrails

Provisional seals issued when sample size bounds fail, preventing false certainty.

Transparent plans scaled for data teams

Start directly in your browser or deploy enterprise forensic verification in private clouds.

Community

For independent researchers and ad-hoc audits

$0/ forever free
  • Up to 100 MB per file (CSV, Parquet)
  • Full Benford, Clumping, & Entropy Engine
  • Interactive Dataset Diff & Drift Demo
  • Standalone HTML Certificate with SHA-256

Pro Team

RECOMMENDED

For data engineering teams and ML pipelines

$149/ month
  • Up to 1 GB per file (CSV, Parquet, JSONL)
  • Calibrated P(synthetic) + Isolation Forest
  • Unlimited KS-Test & Wasserstein Drift
  • 5 REST API keys (60 req/min for CI/CD)

Enterprise Air-Gapped

For defense, banking, and private cloud VPCs

$990/ month
  • Up to 10 GB+ streaming DuckDB ingestion
  • Air-Gapped VPC On-Premise (Docker/Helm)
  • Unlimited API keys & Dedicated Host
  • PGP Cryptographic Signature & 99.9% SLA
TECHNICAL SPECIFICATION COMPARISON MATRIXTECHNICAL SPECIFICATION →
ParameterCommunity ($0)Pro Team ($149)Enterprise ($990)
File Size LimitUp to 100 MBUp to 1 GBUp to 10 GB+
Forensic DetectionBenford, Clumping, EntropyFull + P(synthetic) + ForestCustom Domain Invariants
Diff & DriftInteractive DemoUnlimited KS & EMDCI/CD Continuous DAGs
API AccessNone5 Keys (60 req/min)Unlimited Keys

Frequently asked questions

Architecture guarantees, forensic mathematics, and enterprise integration details.

Never. tabdot operates on a strict Zero Raw Retention invariant. Raw tabular rows are streamed entirely in ephemeral memory via Polars and DuckDB, and immediately wiped after calculating mathematical aggregates. No individual row, column value, or numeric cell is ever persisted to relational disks or transmitted to commercial third-party LLM APIs (OpenAI, Anthropic, or Google).

Deterministic Data Trust for Mission-Critical Engineering

Audit Parquet filesVerify ML training datasetsCatch distribution driftStandalone SHA-256 certificate
DIRECT ENGINEERING & COMPLIANCE CHANNELS

Speak directly with data forensics engineers.

Whether you require air-gapped on-premise deployment, custom EU AI Act attestation pipelines, or specialized high-throughput data audit, our core engineering team responds within 4 hours.

Direct Engineering

Questions regarding Benford statistical bounds, zero-retention architecture, or custom rules.

engineering@tabdot.dev →

Security & Air-Gapped

Enterprise security audits, SOC2/FIPS cryptographic seals, or private cloud VPC deployments.

security@tabdot.dev →

Instant Interactive Studio

Test verified B2B transaction benchmarks, Benford analysis, and LLM synthetic data profiles right now.