Government & public sector • customer story

State Digital Office — Oversight-safe public communications templates

A Type-III (CRSP-governed) deployment that accelerates public communication templates for agencies operating under scrutiny: legislative oversight, procurement review, FOIA risk, and administrative-law constraints. Outputs remain drafts for accountable review, recorded through Ask → Evidence → Release (AER).

Context

The state digital office needed repeatable, consistent public communications: citizen FAQs, press statements, service change notices, AI transparency summaries, and incident communications. The risk was not “typos”— it was language that could trigger avoidable oversight conflict, FOIA exposure, or legal misinterpretation.

Public communications Oversight scrutiny FOIA posture

Type-III wedge

Start with communications templates (safe to approve) and transparency documents. Expand into a standardized governance layer for AI and digital services: consistent AER gates, defeaters, evidence bundling, and audit logs across agencies and vendors.

Land → Expand Template-first Cross-agency consistency

Boundaries (non-negotiable)

  • No policy commitments. No binding commitments or legal promises unless explicitly authorized by policy owners.
  • No legal determinations. Drafts do not assert legal conclusions; they propose language for counsel review.
  • Draft-only posture. Outputs are proposed communications requiring approvals by named officials.
  • Explicit accountable roles. Every draft names the human owners and approval gates (comms, legal, agency lead).
  • Evidence-first transparency. Any factual claims must be anchored to approved sources; uncertainty must be stated.
Operational rule: AI may draft communications and transparency language, but may not announce, commit, or decide on behalf of the state.

CRSP architecture (defensibility by construction)

CRSP (Coherent Reflexive Super-Prompt) makes public-sector drafting defensible: role boundaries, required disclaimers, evidence rules, defeaters, and release criteria are versioned artifacts. This produces consistent templates that anticipate oversight questions: scope, authority, transparency, and review.

  • Context bundle (versioned): public communications posture, prohibited commitments, approved tone and accessibility, role definitions.
  • Policy layer: allowed artifact types (FAQ, press statement, incident notice, transparency summary), escalation and refusal rules.
  • Evidence rules: approved sources (agency policy, service metrics, incident logs, procurement references as permitted).
  • Defeaters: blocks for unauthorized commitments, legal conclusions, uncited claims, or misleading certainty.
  • Audit log: immutable record of request, policy version, evidence bundle IDs, and release outcome.

AER gate: Ask

Normalize the request into a bounded drafting task: artifact type (FAQ/press/notice), audience, required disclaimers, and prohibited commitments.

{ "ask": { "domain": "STATE_DIGITAL_OFFICE", "task": "Draft public communications template (draft-only)", "artifact_type": "citizen_faq", "audience": "public", "must_include": [ "scope and purpose", "what we know / what we don't know", "where to get official updates", "accessibility and plain-language constraints" ], "must_not_include": [ "binding policy commitments", "legal conclusions", "uncited factual assertions", "statements implying decisions are final" ], "accountable_roles": [ "Public Affairs Lead", "Agency Policy Owner", "Legal Counsel Reviewer" ], "policy_version": "crsp.gov.state.comms.v1" } }

AER gate: Evidence

Evidence is bundled from approved sources (service status dashboard, incident log, policy memos, procurement-approved statements). Drafts reference evidence IDs and explicitly flag gaps.

{ "evidence": { "bundle_id": "evb_2026_01_24_058", "sources": [ {"type":"service_status","id":"STATUS_DASHBOARD_2026_01_24"}, {"type":"incident_log","id":"INCIDENT_2026_01_23_LOGIN_OUTAGE"}, {"type":"policy_memo","id":"AI_TRANSPARENCY_GUIDE_v2"} ], "notes": [ "State uncertainty explicitly", "Avoid commitments; route commitments to policy owners" ] } }

AER gate: Release

Release is permitted only when the draft stays within authorized posture: plain-language, evidence-anchored, and ownership-named. Defeaters block or downgrade release if the draft implies commitments, legal determinations, or certainty without evidence.

{ "release": { "status": "APPROVE_FOR_REVIEW", "output_type": "citizen_faq_draft", "includes": [ "what we know / what we don't know section", "evidence-linked facts", "official update channels", "review gates and accountable roles", "non-commitment disclaimer" ], "defeaters_checked": [ "no_unauthorized_commitments", "no_legal_conclusions", "evidence_linked_claims_only", "plain_language_accessibility" ], "review_required_by": [ "Public Affairs Lead", "Agency Policy Owner" ], "audit": { "policy_version": "crsp.gov.state.comms.v1", "evidence_bundle_id": "evb_2026_01_24_058", "timestamp_utc": "2026-01-24T00:00:00Z" } } }

Defeaters (block or downgrade)

These conditions prevent release or force a safe rewrite:

  • Unauthorized commitments (“We will…” / “We guarantee…”) without policy-owner approval
  • Legal conclusions stated as final determinations
  • Uncited facts about incidents, timelines, or impacts without evidence references
  • Misleading certainty where the record is incomplete
  • FOIA risk posture errors (including sensitive details beyond permitted disclosure scope)

Mitigation language (release-safe)

Built-in phrasing patterns that reduce backlash and oversight friction:

  • Uncertainty framing: “Based on currently available information…” / “We are still assessing…”
  • Non-commitment posture: “This draft does not constitute a final policy commitment.”
  • Evidence anchoring: “Per the incident log…” / “As shown on the status page…”
  • Escalation hooks: “For official determinations, defer to the agency policy owner / counsel.”