Personas¶

When to read: You want to know who this repo is for before picking a doc. Visual tour: TOUR.md.

This page is the canonical audience model. Wiki “who this is for” tables elsewhere are short pointers back here. Named composites are not real people.

What this product is for¶

Operator ETL is a reference architecture and local proof for audited intake (FOIA / public comments and similar CSVs): medallion warehouse, PII policy, LangGraph orchestration, critic-gated insights, MCP allowlist.

It is not a production FOIA product today. The Streamlit Gov tab is a run inspector (PARTIAL HITL), not an officer approval product. Production officer UX is SPECIFIED in PRODUCT-UX.md.

Terms (do not conflate)¶

Term	Means here
Adopter	Someone proving the demo locally (`./scripts/verify.sh`) — usually Sam or Riley. Not “agency buyer.”
Officer	Priya-class reviewer: quarantine, PII flags, critic-checked insight. Today = inspector; later = PRODUCT-UX.
Buyer / decision-maker	Alex: budget, ATO, roadmap honesty — white paper audience, not a day-to-day repo user.
Data subject	Citizen commenters in the sample — not product users.

At a glance¶

Persona	Role	Job to be done	Start here
Sam	New engineer	Clone, prove the demo, then look around	QUICKSTART → TOUR
Priya	FOIA / program officer	See intake volume, quarantine, PII flags, a critic-checked insight	FOIA guide → DASHBOARD
Riley	Data engineer	Run pipelines, add a CSV, optional local Ollama wording	CLI → ADD-A-SOURCE → LLM
Jordan	Architect / reviewer	Honest scope, proof gate, scale path	WHY → FINAL-REVIEW
Casey (secondary)	AI agent / MCP client	Query allowlisted gold KPIs; never vault or raw SQL	AGENTS.md → operator-verify
Alex (secondary)	Decision-maker / FOIA director / CDO	Trust claims, budget/ATO framing, what is proven vs specified	WHY → white paper → ROADMAP

flowchart LR
  Sam[Sam engineer] --> Verify[verify.sh]
  Priya[Priya FOIA officer] --> GovTab[Gov dashboard]
  Riley[Riley data engineer] --> CLI[etl-graph / etl run]
  Jordan[Jordan reviewer] --> Wiki[wiki home]
  Casey[Casey MCP agent] --> MCP[allowlisted gold tools]
  Alex[Alex decision-maker] --> Share[white paper / ROADMAP]

Primary personas (builders and operators of the demo)¶

Sam — new engineer¶

Context: Joining the repo or evaluating it in a clone-and-run session. No agency FOIA duty yet.

Job to be done: Prove the invariant on a laptop without API keys or cloud spend.


Today	`./scripts/verify.sh` → `OPERATOR_ETL_VERIFY=PASS`, 76 pytest, FOIA demo `silver=10` / `quarantined=2`. Template insights; no Ollama required.
Later	Optional local Ollama wording (LLM.md); not required for CI.
Start here	QUICKSTART → TOUR → CONCEPTS
Do not expect	An API key, live GCP, or Presidio. Template insights are the CI path.

Priya — FOIA / program officer¶

Context: Program or FOIA staff who must see what arrived, what was rejected, and what a critic-checked summary claims — before any public release.

Job to be done: Inspect intake volume, quarantine reasons, PII flags, and a grounded insight.


Today	Streamlit Gov tab after a local FOIA run: KPI cards, quarantine expander, latest insight. HITL is PARTIAL (`needs_human` is not success; no approve/reject queue).
Later	Production officer UI — responsive queue, streaming progress, approve/reject + audit (SPECIFIED, PRODUCT-UX.md). ROADMAP L4 Track B.
Start here	FOIA guide → DASHBOARD → TOUR
Do not expect	A production approval UI or auto-publish. Agents never publish FOIA releases.

Riley — data engineer¶

Context: Owns pipelines, schemas, and new feeds (FOIA CSV today; 311 / grants / orders via APPLY).

Job to be done: Run graphs, add sources, optionally improve insight wording without inventing numbers.


Today	CLI / graph runs, ADD-A-SOURCE, optional Ollama; critic still rejects uncited digits.
Later	File inbox (L1), container staging (L2), BigQuery (L3), Presidio / Reg.gov adapter (L4) — see ROADMAP.
Start here	CLI → ADD-A-SOURCE → LLM · extend: APPLY
Do not expect	That optional LLM bypasses the critic, or that regex PII equals Presidio.

Jordan — architect / reviewer¶

Context: Design review, hiring loop, or scope honesty before claiming production readiness. Includes skeptical “show me proof” readers.

Job to be done: Separate proven / partial / specified; know the scale ladder and residual risk.


Today	Wiki + FINAL-REVIEW + RISKS + NIST + proof matrix in FOUNDATIONS. Local e2e is the gate.
Later	Staging / ATO evidence as L2–L4 exit criteria land — still not FedRAMP from this demo alone.
Start here	WHY → CONCEPTS → FINAL-REVIEW
Do not expect	Live GCP or Presidio from a green local verify. Do not overclaim.

Secondary audiences¶

Not day-to-day operators of this git repo, but they show up in docs and share packs.

Casey — AI agent / MCP client¶

Job: Call allowlisted tools (get_gold_metrics, etc.). No vault access, no raw SQL, no PII in prompts beyond gold KPI JSON policy.

Start here: AGENTS.md → operator-verify · runtime: HOW-IT-WORKS MCP section.

Do not expect: Unconstrained warehouse chat or publish rights.

Alex — decision-maker / FOIA director / CDO¶

Job: Decide whether the pattern is worth budget, staging, or ATO conversation. Needs honest proven-vs-specified language.

Start here: WHY → Operator-ETL-White-Paper → ROADMAP → share pack.

Security / ATO reviewer (same reading path): NIST · RISKS · SECURITY-HARDENING — alignment language, not certification.

Not product users¶

Who	Why
Citizen commenters	Data subjects in synthetic samples; not interactive users of this software.
Leadership reading a released memo	Downstream of officer review. Agents never auto-publish (decision).

What each persona should not expect (summary)¶

Priya does not get a production officer-approval UI today. The Gov tab is a run inspector (PARTIAL HITL).
Sam does not need an API key or Ollama. Template insights are the CI path.
Riley’s local Ollama path is optional. The critic still rejects invented numbers.
Jordan should not claim live GCP or Presidio from this demo.
Casey never sees the PII vault or arbitrary SQL.
Alex should not treat the local MVP as FedRAMP or agency-ready HITL.