OverviewHow it works · in depth
The runtime path, not the pitch

See the data move.
Not just the claims.

Seven diagrams trace exactly what happens at the moment of access or share: where DataGuard sits, what the policy engine decides, how one source becomes a per-role view or per-recipient version, and how each decision turns into audit-ready evidence.

Read first

These flows are illustrative of the product's mechanics. Component names map to function, not to a literal service topology, and anything not yet shipped is marked planned.

DATAGUARD/How it works01 / 7
D1 · Control-plane architecture

A control plane.
Not another store.

DataGuard is a control plane, not a store. Your documents stay in your systems. At the moment of access or share, DataGuard decides what this subject may see, produces that view or version, and records the basis. Then it gets out of the way.

D1 · where DataGuard sits
Consumers
Employees

by role

AI / Copilot
Internal apps
External recipients
request access / share
DataGuard control plane
Enforcement points

intercept

Policy engine

decision

Classifier

data discovery

Transform

redact / scope

Evidence ledger

record

role / recipient view returned to consumer decision record peels off to the ledger
fetch source via connectors
Connectors
M365
Google Workspace
Clio
Email
Object storage
AI tools
Your existing storage · source of truth
Documents & data
DataGuard stores no source content
TakeawayDataGuard is a decision, transform, and evidence layer between your people and your existing storage. It holds no source content.
DATAGUARD/How it works02 / 7
D2 · Internal data flow

One file, two roles,
different content.

No pre-split copies, no separate redacted files to manage. The role-specific view is produced at request time from the single source, and the decision is logged before it is shown.

D2 · request-time sequence
  1. 1
    User · Billing Enforcement

    Open patient_record.pdf

  2. 2
    Enforcement Policy engine

    Decision request: subject, resource, action

  3. 3
    Policy engine Classifier

    What sensitive entities are in this doc?

  4. 4
    Classifier Policy engine

    SSN, diagnosis, billing codes

  5. 5
    Policy engine Enforcement

    Allow billing codes, redact SSN + diagnosis

  6. 6
    Enforcement Storage

    Fetch source

  7. 7
    Enforcement

    Apply per-role transform

  8. 8
    Enforcement User

    Billing-scoped view, diagnosis hidden

  9. 9
    Enforcement Evidence ledger

    Log who, what, basis, policy version, hash

TakeawayThe policy engine decides per field; the enforcement point fetches the source, applies the transform, and logs before returning.
D2 · same source, two views
Billing viewpatient_record.pdf
Patient
Jane R.
Billing codes
99213 · 80053
SSN
redacted
Diagnosis
redacted
Clinician viewpatient_record.pdf
Patient
Jane R.
Billing codes
99213 · 80053
SSN
•••-••-1782
Diagnosis
Type 2 diabetes
TakeawayThe single image that does more than a paragraph: identical file, redacted for Billing, full for the Clinician.
DATAGUARD/How it works03 / 7
D3 · External data flow

Each recipient
is a policy context.

The same source yields a different, scoped version per recipient: content, not just a watermark, and every version is recorded. Nothing leaves un-governed.

D3 · per-recipient sequence
  1. 1
    Staff Enforcement

    Share claim_file to Insurer + publish summary

  2. 2
    Enforcement Policy engine

    Evaluate per recipient: context, agreement

  3. 3
    Policy engine Transform

    Insurer = claim fields only; Public = de-identified

  4. 4
    Transform Insurer

    Insurer-scoped version

  5. 5
    Transform Public report

    De-identified public version, metadata stripped

  6. 6
    Enforcement Evidence ledger

    Log both decisions, delivery, access events

TakeawayOne share, evaluated per recipient against context and any agreement, transformed to each, both decisions logged.
D3 · one source, scoped versions
claim_file

single source

Transform engine

per-recipient

Insurer

claim fields only

Public report

de-identified · metadata stripped

TakeawayFan-out, not broadcast: one stored claim file becomes two different governed versions.
DATAGUARD/How it works04 / 7
D4 · Policy lifecycle

The policy isn't
written from a blank page.

Discovery drafts the policy, a human approves it, and it compiles to live rules. Time-to-value is the onboarding scan, not a six-month GRC project.

D4 · onboarding to enforcement
Connect tools

M365 · Google · Clio

Discover & classify

find sensitive data

Auto-draft policy

roles × data × actions

Admin reviews

adjust / approve

Compile

to enforcement rules

Enforce

every access & share

Log

every decision

continuous: log feeds discovery, the policy re-drafts as data changes
TakeawayDiscovery proposes, a human approves, it compiles to live enforcement, and every decision feeds the next discovery pass.
DATAGUARD/How it works05 / 7
D5 · Evidence to compliance

Evidence is a byproduct
of enforcement.

Every decision is hash-chained and mapped to the controls a regulator will ask about, exportable as an audit pack, not reconstructed after the fact.

D5 · log line to audit artifact
Access / share event
Decision record

who · what · basis · policy v · time

Hash-chained ledger

tamper-evident

map to controls
CCPA cyber-audit
SOC 2 · Privacy & Confidentiality
ISO 27701
HIPAA minimum-necessary
One-click audit export pack
TakeawayA decision record is hash-chained into a tamper-evident ledger, pre-mapped to the frameworks a regulated buyer answers to.
DATAGUARD/How it works06 / 7
D6 · Before and after

The same three leaks,
with the product in the path.

The three scenarios from the narrative, now as flows: the red path ends in exposure; the path with DataGuard ends in a governed view or version and a logged decision.

D6 · where DataGuard intercepts

The public report with hidden PII

Without DataGuard
Author exports report
Hidden columns, metadata, fake redactions remain
Published
Mass PII exposure
With DataGuard
Author shares for publication
Policy = 'public' recipient
Strip hidden data, flatten redactions, clear metadata
Clean public version, logged

The insider look

Without DataGuard
Opens full patient chart
Reads out-of-scope fields
HIPAA minimum-necessary breach
With DataGuard
Opens chart
Policy denies out-of-scope fields
Billing-scoped view, logged

The wrong recipient

Without DataGuard
Wrong attachment sent to outsider
Delivered before anyone notices
Privilege waived, client lost
With DataGuard
Send to recipient
Recipient / policy mismatch caught
Blocked or transformed, logged
TakeawayEach leak has a single point where a policy decision changes the outcome from exposure to a recorded, governed release.
DATAGUARD/How it works07 / 7
D7 · The policy model

Subject × resource × action.
Always, and always logged.

Every decision is attribute-based: who the subject is, what the resource is classified as, and which action they are attempting. There is no hardcoded rule list behind it.

D7 · attribute-based decision model
Subject

role + attributes: department, clearance, region, engagement

Resource

classification: PHI, PII, privileged, public + container

Action

view · share-internal · share-external · export · AI-read

policy engine resolves
Allow
Redact fields
Transform for recipient
Deny
every outcome is logged with its basis
TakeawaySubject, resource, and action attributes resolve to one of four outcomes, and the decision is always recorded.
The honest read

The per-role redaction mechanic exists in legacy ECM and vaults. What is unbuilt elsewhere is this bundle: everyday documents, internal and external, with a compliance-grade decision record, on the tools you already use.