Provifact™ by TMCO Consulting

From approved change to audit-ready proof.

Provifact turns approved endpoint changes into traceable audit evidence—repeatably, read-only, and without letting AI become the authority.

Regulated teams can manage thousands of endpoint settings and still spend weeks before an audit reconstructing what was intended, what was observed, who approved it, what changed, and which evidence supports the conclusion. Provifact joins those facts into one reviewable chain:

Git-approved intent → Microsoft Graph GET-only observation → deterministic drift → sanitized proof → bounded GPT-5.6 explanation → human judgment

Open the live Mission Control · Follow the judge path · Inspect the architecture · Run it locally

Why enterprises need this

Enterprise problem	Provifact response
Audit evidence is assembled after the fact from MDM consoles, tickets, screenshots, Git history, spreadsheets, and memory	Produce a versioned evidence chain as approved configuration and observed state change
A dashboard score can hide unsupported controls, incomplete collection, or guessed mappings	Keep aligned, drifting, unsupported, unreviewed, unevaluated, and collection-gap states distinct
Granting an evidence tool write authority expands operational and security risk	Use Microsoft Graph GET-only collection; no create, update, assign, remediate, or rollback method exists
Sending raw tenant context to a model creates privacy and authority problems	Publish through an allowlist, scan the sanitized package, and send GPT-5.6 only bounded evidence
AI-generated prose can sound authoritative even when it is wrong	Keep deterministic findings authoritative, verify typed claims and references, quarantine prose, and require human review

Provifact is designed for endpoint engineering, security, GRC, managed-service, and assessment-support teams that need to answer a simple question quickly: What proof do we have for this approved configuration state?

See the product in 90 seconds

1. Open Mission Control and confirm the evidence-mode banner, approved baseline, freshness, deterministic denominator, drift, backlog, and collection gaps.
2. Open a finding and follow the approved target, observed value, exact provider definition, assignment evidence, evidence IDs, fingerprint, and read-only operator guidance.
3. Open the Baseline Plan to see all 98 approved Level 1 rules. Four exact Intune joins are currently evaluated; the other 94 remain visible implementation work rather than fabricated failures.
4. Ask Provifact Assistant “What requires my attention?” The Worker selects a small sanitized context for fixed gpt-5.6-terra, attaches authoritative claims and references, rejects unsupported verdict language, and keeps generated prose subject to human review.
5. Finish with Evidence health and privacy: raw tenant responses are not public, production has no synthetic fallback, and Provifact has no Intune write capability.

What makes Provifact different

• Deterministic before generative. A model never selects finding status, calculates drift, approves an exception, accepts risk, publishes evidence, or decides compliance.
• Honest coverage. The current macOS plan shows all 98 approved rules, evaluates only four exact reviewed provider mappings, and exposes the 94-rule mapping and implementation backlog.
• Read-only by construction. The provider and transport expose collection and GET only. A Git revert changes reviewed desired-state history; it does not change Microsoft Intune.
• Privacy-safe publication. Public output is reconstructed from approved fields, fingerprinted, scanned, reviewed, and deployed only from an exact protected-main audit artifact.
• Evidence-bounded AI. Production pins gpt-5.6-terra, uses store: false, no tools, bounded context and output, rate limits, structured output, and deterministic post-verification.
• Vendor-neutral core. Microsoft Intune and managed Apple platforms are the first implemented provider slice; the evidence, publication, and verification contracts are designed for additional endpoint platforms.

The current implementation is portable at its core but not turnkey across providers. Domain, evidence, publication, and verification contracts are provider-neutral; Microsoft Intune is the only live provider. Adding Jamf, Omnissa Workspace ONE, or another MDM requires a new GET-only adapter, pinned reference data, exact reviewed field mappings and value transforms, least-privilege authentication, fixtures, contract tests, and deployment configuration. That same pattern can extend to Windows, Linux, or other mobile platforms, but effort depends on the target API and schema; it is not guaranteed to be configuration-only. The deterministic engine and public schema do not need to be rewritten when those contracts are met.

Current Build Week proof

The Phase 1 vertical slice connects the TMCO Consulting macOS CIS Level 1 Demo Baseline in Git to read-only Microsoft Intune collection, deterministic technical drift, a reviewed live sanitized Mission package, current/prior snapshot comparison, a Cloudflare-hosted operational dashboard, and Provifact Assistant.

The approved baseline is pinned to a reviewed NIST macOS Security Compliance Project revision and hashes. That approval supports technical drift detection; it is not CIS certification, a CMMC assessment result, a C3PAO conclusion, or an organizational compliance verdict. The approved-baseline runbook shows how reference selection, organization targets, reviewed provider mappings, observations, technical results, and human judgment remain separate.

The verified production record is in Judge-Readiness Validation. Production serves a scanned package labeled LIVE SANITIZED TENANT DATA; tenant, device, user, policy, group, assignment, credential, and raw Graph identities are excluded. One bounded live gpt-5.6-terra request passed typed-claim and evidence-reference verification, and no Intune mutation was performed.

What the slice proves

• Four macOS settings have reviewed, exact Microsoft Intune provider-definition mappings. A fifth desired setting remains visibly Provider mapping not reviewed instead of being guessed.
• Exact provider IDs—not policy display names or substrings—join observed settings to desired state.
• A setting is called missing only when its mapping is reviewed and collection evidence is complete. Unsupported values, parser gaps, unreviewed mappings, and unevaluated resources remain distinct.
• Current and immediately prior sanitized live snapshots can show new, resolved, and unchanged technical drift without retaining a tenant data lake.
• Production deployment accepts only a reviewed sanitized artifact from a successful protected-main live audit and verifies its exact snapshot before and after deployment.
• Synthetic data has no production fallback path.

Run the credential-free demonstration

Python 3.12 or later and Node.js are required.

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements-dev.txt
python -m pip install --no-build-isolation --no-deps .
npm ci --ignore-scripts --no-audit --no-fund
python -m provifact run-mission-demo --output-dir build/mission-demo
python -m provifact rebuild-static-demo
mkdocs build --strict
npm run validate:worker
npm run dev

Local and preview runtimes are explicitly fixture mode and need no tenant or OpenAI credential. The tracked package under docs/assets/data/ is deterministically generated synthetic data and is never accepted by the production deployment workflow.

Validate

python -m ruff format --check .
python -m ruff check .
python -m mypy
python -m pytest
python -m bandit -r evidenceops scripts -c pyproject.toml
python scripts/check_company_name.py
python scripts/check_secrets.py
python -m pip_audit -r requirements-dev.txt
mkdocs build --strict
python scripts/check_public_artifacts.py site
npm audit --audit-level=moderate
npm run validate:worker
git diff --check

Public CI uses no Microsoft or OpenAI credential. Live collection is a separately protected, environment-bound workflow on trusted main using GitHub OIDC and the documented read-only Graph permissions.

Safety and privacy boundaries

• Microsoft Graph access is GET-only and limited to the reviewed endpoint families.
• Raw tenant responses, Graph tokens, pseudonymization material, and private evidence are ephemeral and never enter the public artifact.
• Unknown public fields fail closed. Public and pre-model egress pass the shared credential and sensitive-value catalogs.
• Tenant, user, group, device, policy, assignment, and object identities are excluded or non-reversibly pseudonymized before publication.
• The deterministic evidence package is authoritative. AI cannot determine status, approve an exception, accept risk, publish, remediate, or decide compliance.
• Browser refresh checks only the current published snapshot; it never authenticates to Graph.

Scope and limitations

The slice begins with Microsoft Intune and managed Apple platforms while keeping the provider and evidence contracts vendor-neutral. The macOS inventory contains 98 pinned Level 1 rules with titles from the recorded mSCP revision; four exact provider mappings are currently reviewed. The remaining 94 are a visible onboarding and implementation backlog. iOS and iPadOS appear only as sanitized aggregate posture and are not scored against the macOS baseline. Long-term history, multi-tenancy, private policy-name presentation, broader exact provider mapping, Apple release intelligence, and all Intune write capabilities remain deferred.

External technical profiles—including CIS Level 2, DISA STIG, NIST, and CMMC-oriented mSCP profiles—are comparison and planning references only. Profile membership is not an approved company target, framework score, control-satisfaction result, completed assessment, or certification.

Build Week provenance

Provifact began during Build Week with repository safety foundations and a small synthetic schema proof. TJ Olnhausen retained the product, security, baseline-approval, external-system, review, and merge decisions. Codex implemented and tested the provider contracts, live collection boundary, versioned evidence, deterministic drift, sanitizer, Cloudflare runtime, dashboard, Assistant, and deployment controls. GPT-5.6 is a bounded runtime explainer, not the evidence engine. The private /feedback Session ID belongs only in submission metadata and must never be committed.

See Getting Started, Architecture, Security Model, and the Build Week demo package.

Brand and compatibility identifiers

The public product is Provifact™ by TMCO Consulting. Phase 1 retains the existing evidenceops Python import, evidenceops console command, schema/algorithm identifiers, EVIDENCEOPS_* environment variables, Worker resource name, OpenAI project/key labels, and artifact prefixes as compatibility identifiers. New operator documentation uses the provifact command. The public repository and hostname are tmcoconsulting/provifact and provifact.tmcoconsulting.com. The cutover was proven through the renamed immutable GitHub OIDC subject before the old repository-name trust was retired. The internal Worker name keeps its encrypted secrets and deployment history.

Copyright 2026 TMCO Consulting, LLC. Provifact™ is a trademark of TMCO Consulting, LLC. The software is licensed under the Apache License 2.0. The pinned baseline is derived from the NIST macOS Security Compliance Project under its upstream terms; see NOTICE and the recorded source hashes for attribution.