§ 01 · The Problem

AI-ready dataisn't the same as AI-ready.

Clean data is not ready data. A dataset can pass every hygiene check and still be the wrong shape for the model, silent on provenance, ambiguous to a machine, or out of compliance. Teams discover the gap in production — the most expensive place to find it.

Blind Deployment Risk

No way to know if agents will succeed

…until they fail in production.

Compliance Gaps

Regulators want audit trails

for autonomous and AI-assisted decisions.

Data Fragmentation

Critical data locked in silos

Databricks, Snowflake, Veeva — never assessed together.

Unquantified ROI

No line from data spend to AI outcome

Which fixes actually move readiness?

“Everyone says their data is AI-ready. Almost no one has measured it.”

§ 02 · What Compass Does

Evaluate readiness.Then close the gap.

DATA Compass evaluates datasets for AI readiness across four dimensions, grounded in FAIR — then turns every finding into a specific, executable recommendation and a quantified ROI, so AI investments succeed instead of stalling in a proof-of-concept.

01

Assess

Score the dataset across four dimensions of readiness, grounded in FAIR.

02

Recommend

Bridge AI turns findings into prioritized, causal recommendations.

03

Act

Every recommendation arrives as an executable remediation plan, ready to run.

04

Prove

Quantify the readiness lift and the ROI of each fix.

§ 03 · The Model

Four dimensions.One foundation.

Everyone sells FAIR and stops. Compass treats FAIR as the foundation, then measures the four dimensions of readiness FAIR enables but does not guarantee, with Bridge AI synthesizing it all into what to do next.

From enterprise evidence to defensible AI action

Compass connects heterogeneous evidence, reconciles it into a defensible baseline, measures readiness across four dimensions, identifies root causes, and turns findings into executable, measurable action.

01
Enterprise evidenceConnect data, knowledge, governance, and stakeholder evidence
Read-only · in place · zero egress

Data platformsDatabricks · Snowflake · Veeva · files & APIs

Knowledge systemsStardog · ontologies · metadata · vocabularies

Governance evidenceSOPs · policies · standards · regulatory documents

People & purposeInterviews · persona sentiment · the declared use case
02
The triangulated FAIR foundationReconcile what the data shows, what people experience, and what governance requires
16 FAIR rules · SHACL

Technicalwhat the data shows

Perceivedwhat people experience

Governedwhat policy requires

Defensible finding

Agreement builds confidence. Divergence reveals the problem — and names it:

Usability Gap Policy Theater Tribal Knowledge Unformalized Practice + 4 more

FAIR — findable · accessible · interoperable · reusable — is the substrate: necessary for readiness, not sufficient on its own. The engines below measure what it can't.
03
Four assessment enginesMeasure readiness across four interdependent dimensions
50+ algorithms · named tests
Semantic Clarity
- Ontology alignment
- Inferred column→class mapping
- Interoperability depth
Statistical Health
- Data quality — 8 algorithms
- ML readiness — 4 algorithms
- Bias & class separability
Contextual Validity
- Fit to the declared use
- Domain Pack rules
- Evidence-cited judgment
Regulatory Compliance
- ALCOA+ integrity
- FDA AI credibility
- Privacy & provenance
Statistical tests · deterministic rules · evidence extraction — every score traces to its source
04
Bridge AIFrom multidimensional signal to prioritized, evidence-grounded decisions

Cross-dimension synthesis

Causal analysis & root cause

Prioritized recommendations

Execution risk & stakeholder readiness

Value drivers & quantified ROI

Evidence traceability

No black-box conclusions — every claim cites the evidence that produced it.
05
Execution & proofExecute, reassess, and prove improvement

Executive briefing Prioritized action portfolio Remediation scripts Owners · effort · timeline Business case & ROI Regulatory evidence package Readiness trajectory Supported autonomy level

Reassess — every re-analysis feeds back through the engines and appends to the trajectory. Readiness is measured over time, never asserted once.

§ 04 · Where FAIR Sits

FAIR is the floor.Not the finish line.

FAIR makes data legible to machines and people — findable, accessible, interoperable, reusable. That is necessary, and it is not sufficient. A perfectly FAIR dataset can still be statistically broken, wrong for your use, non-compliant, or semantically ambiguous. Compass treats FAIR as the substrate and measures what comes next.

“We treat FAIR as table stakes — and measure the readiness it can't.”

FAIR guaranteesThe data can move.Findable · Accessible · Interoperable · Reusable — the plumbing is in place.

FAIR does not guaranteeThe data is ready.Statistically sound · fit for purpose · compliant · semantically unambiguous.

So Compass addsFour dimensions on the FAIR floor.Each measured, evidenced, and triangulated for defensibility.

§ 05 · Connect

Your data never moves.We query it in place.

Compass is an API overlay, not a data lake. It connects to your warehouse, runs read-only queries through your own compute, and analyzes the results in memory. No ETL, no copies, no new storage. Source data never leaves your environment — the answer to the first question every pharma security review asks.

Databricks · Snowflake · Stardog · Veeva

ConnectUnity Catalog / Snowflake, PAT or OAuth.Read-only. Credentials AES-256 in Key Vault / Secrets Manager.

Analyze in place50+ algorithms run on your warehouse.Results return as JSON — scores, recommendations, ROI, fix scripts.

Zero egressSource data never leaves.Most single-table analyses cost under $0.10 of warehouse compute.

§ 06 · The Method

Three reads on one domain.The disagreement is the finding.

Compass doesn't take the data's word for it. Every FAIR assessment triangulates three independent reads. Where they agree, you have a finding you can defend. Where they diverge, the divergence itself is the insight — and it's named, not hand-waved.

TechnicalWhat the data shows.SHACL + scoring across 16 FAIR rules.

PerceivedWhat people experience.Persona-weighted organizational interviews.

GovernedWhat the documents say.Governance documents scored against eight frameworks.

Agreement = Confidence · Divergence = The Finding

Confirmed Maturityall three legs agree, high

Aligned Maturitydata and people agree; the governed read is pending

Greenfieldlittle signal anywhere — start clean

Usability Gaptechnically strong, people struggle

Tribal Knowledgepeople know it, nothing is written

Unformalized Practicethe practice works; no SOP captures it

Abandoned Capabilitywas there, now decayed

Policy TheaterSOPs exist, practice doesn't follow

Eight named patterns — each one an actionable diagnosis.

A DATA Compass pillar card: the Accessible pillar confirmed as a Usability Gap, with Technical, Perceived, and Governed leg scores, coverage sufficiency, coherence, and what-to-do guidance — A live pillar card: three legs scored with coverage honesty, the pattern named, and the move spelled out.

§ 07 · The Four Dimensions

What we measure,dimension by dimension.

Dimension 1 — Semantic Clarity

Does the data meanwhat it says — to a machine?

Whether a machine can resolve the meaning of your data unambiguously, against the ontology you actually use. Compass reads your private ontologies, infers which columns map to which classes, and scores the alignment.

Ontology inferenceInfer column→class alignment.Credit the meaning that's there even before it's tagged (Tier-2).

Private ontologiesScore against your vocabulary.Your CDISC / IDMP / internal taxonomy — not a generic dictionary.

I-pillar depthInteroperability, made first-class.Declared · Inferred · None, at labelled confidence.

Dimension 2 — Statistical Health

Is the dataquantitatively sound?

Whether the numbers hold up before a model ever touches them. Compass runs two engines — data quality and ML readiness — across 50+ algorithms, from distribution and anomaly detection to bias, class separability, and feature sufficiency.

Data qualityEight algorithms.Information content · outliers · sparsity · anomalies · cleanliness · more.

ML readinessFour algorithms.Data sufficiency · bias detection · feature engineering · class separability.

EvidencedNamed statistical tests.Benford · Pearson / Chi-square · Cohen's Kappa · Shapley · LOO.

A DATA Compass Data Quality Assessment: overall quality 66 of 100 with an eight-component breakdown — information, statistical quality, independence, outliers, completeness, cleanliness, anomaly detection — plus dataset-size and sparse-data callouts — The Statistical Health read on a live warehouse table: eight components scored, warnings in plain language.

Dimension 3 — Contextual Validity

Is the data rightfor your specific use?

A dataset that's clean, sound, and FAIR can still be wrong for the job in front of you. Contextual validity scores fit-for-purpose against a declared use — using customer-authoritative Domain Packs that encode your sub-domains, archetypes, and rules.

Domain PacksCustomer-authoritative.Your sub-domains and record archetypes, versioned and yours.

Fit-for-purposeScored against a declared use.Validity is relative to the question you're asking of the data.

5-step pipelineSignal → rules → coverage → LLM → judgment.Deterministic where it can be, synthesized where it must be.

Dimension 4 — Regulatory Compliance

Does the data meetthe integrity bar?

In regulated work, readiness includes defensibility to a regulator. Compass scores governance evidence against the frameworks pharma actually answers to — line by line, with citations — so compliance is measured, not asserted.

ALCOA+Nine data-integrity attributes.Attributable · legible · contemporaneous · original · accurate · + four.

FDA credibilitySeven steps + Step 8.Context of use → fit-for-use, plus agent decision provenance.

PrivacyGDPR · HIPAA · CCPA.Cited line-by-line, mapped to FAIR pillars.

A DATA Compass ALCOA+ Compliance card scoring 68, partial: three evidence-backed principles scored, six indicators flagged as requiring source-system evidence, and twelve findings by severity — ALCOA+: evidence-backed principles scored; the rest honestly labeled as needing source-system evidence.

A DATA Compass FDA AI Credibility card: score 55, verdict Not Fit for Use, relevance and reliability sub-scores, and gaps found by severity — two critical, two major, one minor — FDA AI credibility: a verdict with its gaps attached — including when the answer is “not fit for use.”

§ 08 · Synthesis

Bridge AI reads it alland tells you what to do.

Four dimensions produce a lot of signal. Bridge AI is the synthesis layer: it runs causal analysis across the findings, separates root causes from symptoms, and writes prioritized recommendations in plain language — each one traced back to the evidence that produced it. No black-box conclusions; every claim cites its source.

Causal analysis

Root cause, not symptom.

Why the score is what it is — and what fixing it unlocks.

Prioritized recommendations

Ranked by impact and effort.

With an effort estimate on every action.

Grounded

Every recommendation traces to evidence.

From finding to source, auditable end to end.

A Bridge AI executive briefing in DATA Compass: a plain-language narrative connecting vocabulary fragmentation to missing-data rates, FAIR interoperability failures, and ML readiness, ending in a key-takeaway recommendation with a realistic timeline — grounded in the Pistoia Alliance business-value model — An executive briefing written by Bridge AI: the causal story across every dimension, ending in one keystone recommendation with an honest timeline.

“From four-dimensional signal to a short list of things worth doing.”

§ 09 · The Output

Every finding shipswith the fix attached.

Compass is not a scorecard that rots on SharePoint. Every recommendation routes to a structured, executable action plan: an auto-generated remediation script you can read, edit, and run.

Executable

Auto-generated remediation, ready to edit.

A Python script per finding — not a to-do line.

Routed

Findings connect to the plan automatically.

No re-keying scores into a separate tracker.

Traceable

Plan → recommendation → finding → source.

The whole chain survives an audit.

A DATA Compass smart recommendation: implement controlled-vocabulary master data management — impact, effort, and timeline up top with return, cost, ROI, and payback; below, the causal analysis chain from problem to root cause to fix to outcome, plus stakeholder-readiness context — One finding, fully dressed: the causal chain, the stakeholder read, the fix, and the ROI — ready to defend in a steering committee.

§ 10 · The Trajectory

A score is a photograph.A program needs the film.

Every Compass analysis appends to the dataset's history. Fix something, re-analyze, and the trend view shows each dimension moving — with the delta since your last checkpoint quantified in a progress report your steering committee can actually read.

Re-analyze in place

One click re-runs the full assessment.

On a connected dataset — no re-browsing the catalog, no re-upload.

Trend per dimension

Watch each score move over time.

Statistical health, semantic maturity, governance, ML readiness — with FAIR-pillar drill-down.

Progress since checkpoint

The delta, quantified.

What improved, what regressed, and which remediation produced it — evidence attached.

The DATA Compass technical trajectory view: Technical Asset Score and Statistical Health plotted across three readings from March to July, with the scoring-rubric version annotated on the timeline — A real trajectory: three readings across four months, with the scoring-rubric version pinned to the timeline.

“Readiness is a trajectory. One reading tells you where you stand; the trend tells you if the program works.”

§ 11 · The Value

Readiness, in dollars.A line item that defends itself.

Every gap carries a cost and every fix carries a return. Compass quantifies the business value of closing each gap — grounded in the Pistoia Alliance FAIR business-value model — so a data-readiness program competes for budget on the same terms as everything else: ROI.

Prioritize by return

Which fixes move readiness most, per dollar.

Rank the backlog by impact, not by who shouted loudest.

Quantified benefit

Trust · speed · cost · effectiveness.

Level-0 outcomes decomposed into measurable benefits.

A budget case

Defensible numbers for the steering committee.

The readiness program earns its line item.

“The first FAIR conversation a CFO will sit through.”

§ 12 · Know Before You Deploy

How much autonomycan your data actually support?

Readiness isn't a yes/no. Compass tells you which level of AI autonomy your data can support today — and exactly what to fix to climb to the next one.

Level 1AdvisoryAI suggests; a human decides everything.

Level 2CopilotAI drafts; a human reviews and approves.

Level 3SupervisedAI acts within bounds; a human monitors.

Level 4AutonomousAI acts; humans audit the trail.

Integrations

Connect your datawhere it lives.

Native connectors for enterprise data platforms. Assess data in place — without extraction or movement.

Databricks

Unity Catalog browser, SQL profiling, and MLflow model assessment via PAT or OAuth2.

Snowflake

Database and schema browser with SQL API profiling for warehouses and data shares.

Veeva Vault

QMS, CDMS, and RIM module adapters with OAuth2 SSO and Direct Data extraction.

Stardog

SPARQL querying, ICV constraint validation, and reasoning-powered ontology analysis.

File Upload & API

Drag-and-drop CSV, Excel, Parquet, JSON — or a REST API for CI/CD and pipelines.