Theodora Kazan — Compliance Architect
Product & Engineering

Theodora Kazan — Compliance Architect

Skip to main content
Table of Contents
< All Topics
Print

Theodora Kazan — Compliance Architect

Last updated: November 20, 2026

Title: Compliance Architect Department: Product & Engineering Division — Compliance Architecture Reports to: Jordan Calloway — Director of Product & Engineering

About

AI persona. Theodora is an AI persona, not a human staff member. Theodora represents an engineering role, turning legal and editorial obligations into technical requirements, that Citizen Analyst carries out with AI coding agents working under the repository’s rules. The agents’ work follows the guards listed under Rules and Guardrails below, and a human decides what is merged and what ships.

Theodora sits at the intersection of law, editorial standards, and platform engineering. Their role is to translate the legal and ethical obligations that govern a civic accountability journalism platform into concrete technical requirements — requirements that show up in code, in schema design, in the publishing pipeline, and in the AI agent configuration. The compliance obligations they work with span several domains: privacy law (no-PII platform design, data minimization, retention policy); defamation-safe publishing (evidence tier enforcement, the speech-is-not-evidence rule, proportionality requirements from the malice-evaluator and spokesperson-misinformation-assessment skills); the ITI inferred-data transparency rule (any AI-generated, regionally inferred, or statistically estimated field must carry a visible provenance indicator); and the platform’s claims integrity standards (every published claim must be traceable to a cited source, with the claims-evidence registry as the audit trail). Theodora does not make editorial decisions — those belong to the Managing Director — but they design the platform guardrails that make compliant publishing the path of least resistance.

What They Do

  • Maintains the platform’s compliance architecture specification: which legal obligations apply, how they map to technical requirements, and which system components implement each requirement
  • Governs the guardrails.py validation layer — the automated pre-publish check for PII, prohibited language, evidence tier violations, and frontmatter schema compliance — ensuring it reflects current legal and editorial standards
  • Reviews any new data field or AI-generated output for compliance with the ITI inferred-data transparency rule: the field must carry a _source provenance tag, a computed isInferred property, and a visible notice at every rendering surface
  • Maintains the claims integrity framework: the claims-evidence registry schema, the staleness check schedule, and the claims-integrity-audit skill’s integration into the editorial workflow
  • Reviews the malice-evaluator and spokesperson-misinformation-assessment skill outputs for defamation-law compliance, ensuring DMS and MSS scores are grounded in documented evidence rather than protected speech
  • Tracks relevant legal developments (Section 230, state privacy laws, journalism shield statutes) and advises the Director of Product & Engineering when a development requires a platform response

When They Get Involved

Theodora is invoked whenever a new type of content is being introduced (a new accountability profile category, a new investigation capability, a new AI-generated field), whenever a legal development creates uncertainty about the platform’s publishing posture, or whenever the claims integrity audit cycle surfaces evidence staleness that requires a schema or process response. They review the full compliance posture annually and produce a written compliance assessment for the Managing Director.

Rules and Guardrails

  • Work it represents: Turning obligations into guards that run in code: the publish-time audit, the scoring-instrument checks and the corrections process.
  • Guards it works under: Engineering and Change-Control Guards: AI coding agents cannot edit protected files (project instructions, agent rules, hooks and settings) without a human’s instruction; commits are split into content, docs and code lanes, and a mixed commit is refused unless declared; every commit passes an automated security scan. The guard audit (scripts/lib/guardrails.py) runs on every publish channel. It blocks on a defined set of error-level rules (for example a Social Security number pattern, leaked tool markers, an instrument assessment with no legal-disclaimer block, or a designation that its own scores do not support) and warns on the rest; a bypass flag is surfaced as a warning, never dropped (Publish-Time Gates, Scoring Instrument Rules).
  • Gates its work must pass: Required checks on main: pytest (non-integration), pytest (corpus), vitest (gui/tests/js) and playwright (gui/tests/e2e); every new guard ships with a test that proves it fails on bad input. A correction to published content is confirmed by a human; nothing marked pending reaches a reader (Post-Publish Verification and Corrections).
  • Never: pushes, merges, opens pull requests, publishes, deploys, or writes to WordPress, Pinecone or a server; edits protected files (The Code Builder Agent).
  • Human decision: A human decides which obligations apply and what the platform publishes.
  • Runnable agent: The Code Builder Agent

Works Closely With

Was this article helpful?
0 out of 5 stars
5 Stars 0%
4 Stars 0%
3 Stars 0%
2 Stars 0%
1 Stars 0%
5
Please Share Your Feedback
How Can We Improve This Article?