Fraud Protection

What it is

CeliaConnect provides two services on the same Flow pipeline:

1. Enrollment Intelligence — the Engagement, Readiness, and Yield scores, the Risk tier, and the recommended next action. Runs on every Flow, on every plan.
2. Fraud Protection — an additional analysis that looks at the same anonymized rows for patterns worth a human second look: inconsistent application data, unusual aid-seeking behavior, or a record that behaves very differently from its cohort. Opt-in per Flow, Enterprise plan only.

When Fraud Protection is enabled on a Flow, Celia runs it in parallel with the three Enrollment Intelligence analyses on the same run — no extra Slate queries, no extra setup beyond declaring four more ss_celia_fraud_* fields in your Slate Source Format (covered below). It works on the same no-PII data as everything else: anonymous IDs and behavioral signals only, never names, emails, or addresses.

The four signal categories

Every Fraud Protection finding is tagged with one or more of four categories. In plain English:

• Application Integrity (application_integrity) — the application's signals don't hang together the way genuine applications at your institution usually do. Example: an application completed at an implausible speed with no preceding engagement.
• Financial Aid Pattern (financial_aid_pattern) — aid-related behavior that matches patterns associated with aid-seeking abuse rather than genuine enrollment intent. Example: aid milestones completed while every enrollment-intent signal stays flat.
• Data Consistency (data_consistency) — fields in the record contradict each other. Example: a status date that precedes the activity that was supposed to produce it.
• Cohort Anomaly (cohort_anomaly) — the record's behavior is a statistical outlier against its own cohort. Example: a pattern of activity timing that no other student in the same program and cycle exhibits.

All four are pattern observations. A flagged record can have a completely innocent explanation — a transfer student, a counselor doing bulk data entry, a family sharing one device. That's exactly why the output is routed to a person.

What Celia produces per student

When Fraud Protection runs, each analyzed student gets:

• Fraud level — none, low, medium, high, or critical. How strongly the observed patterns warrant a look.
• Confidence — 0–1. How sure Celia is about its own read of the signals (see "Interpreting confidence" below).
• Categories — which of the four categories fired.
• Indicators — up to five specific signals, each with its category and a low / medium / high severity.
• Context — a short prose explanation (300 characters or fewer) a reviewer can read without decoding anything. PII-free by contract.
• Review recommended — a yes/no flag: does Celia think a person should look at this record now?

Inside CeliaConnect, this appears as a Fraud Level badge on the Students list and a collapsible Fraud Assessment card on each student detail page.

How to enable it on a Flow

1. Open the Flow (Create a Flow for a new one, or Flow detail → edit for an existing one).
2. Find the Fraud Protection Analysis section of the form and check Enable Fraud Protection Analysis.
3. Save the Flow. Fraud Protection runs starting with the next run.

The setting is per Flow — you can enable it on the Flow that covers aid-eligible applicants and leave it off everywhere else. Cloned Flows inherit the source Flow's setting.

Enterprise availability

Fraud Protection is an Enterprise plan feature. On other plans the toggle is visible but locked, with an "Enterprise feature" note — so you can see what's there without being able to switch it on. If you want it enabled for your institution, contact your CeliaConnect representative or open a ticket from Settings → Support; plan changes are handled via a signed agreement.

Reading the four fields in Slate

On every non-dry-run tick of a fraud-enabled Flow, Celia writes four additional fields to your Slate Source Format alongside the Enrollment Intelligence fields:

• ss_celia_fraud_level — None · Low · Medium · High · Critical. The headline signal. Most institutions build their Slate review query on Medium and above.
• ss_celia_fraud_confidence — decimal 0–1. Pair it with the level before escalating anything.
• ss_celia_fraud_categories — pipe-delimited category codes, e.g. financial_aid_pattern|data_consistency. Handy for Slate query filters (LIKE '%financial_aid_pattern%').
• ss_celia_fraud_context — the short prose explanation. Safe to show to reviewers; contains no PII by design.

Two Slate-side requirements:

1. Declare the four fields in your Source Format. Slate silently drops any POSTed field that has no matching prompt in the Source Format target — if the fields aren't declared, the writeback "succeeds" but the columns stay empty. The full 34-field contract (30 core + 4 fraud) is in the Slate starter kit doc.
2. Echo ss_celia_fraud_level back on your In Progress query. Your In Progress query already echoes the prior run's scores (risk, engagement, readiness, yield…) so Celia can compare run-over-run drift. Add ss_celia_fraud_level to that echo set on fraud-enabled Flows. This powers a stability safeguard: if the prior run said High or Critical and the current run suddenly says None or Low with no stage change to explain it, Celia raises a stability warning instead of letting the flag silently vanish. Without the echo, that safeguard never fires.

Interpreting confidence

The confidence value (0–1) is Celia's self-assessment of the signal quality behind the fraud level — it is not a probability that fraud occurred.

• High level + high confidence (e.g., High at 0.85) — the patterns are clear and mutually reinforcing. Review promptly.
• High level + low confidence (e.g., High at 0.4) — something looks off, but the signal is thin. Still worth a look; weigh it lightly.
• Low level at any confidence — routine. No action needed unless your own review policy says otherwise.

Whatever the numbers say, the decision belongs to a person. Treat level and confidence as triage inputs for where to look first, never as evidence in themselves.

Recommended review workflow

1. Scan the Students list. On the Students page, the Fraud Level column shows a badge on any student with a Low or higher level. Work Medium and above first.
2. Open the student detail. Expand the Fraud Assessment card to read the categories, the individual indicators with their severities, and the context sentence.
3. Open the Slate record. Use the external-link button to jump to the full, named record in Slate — that's where the actual evidence lives.
4. Follow your institution's review process. Route the record to whoever owns application or aid integrity review — admissions operations, financial aid, or compliance. Document the outcome in Slate per your own policy.

A few institutions also build a Slate query on ss_celia_fraud_level IN ('Medium','High','Critical') so the review queue lives inside Slate where the review actually happens. Either path works — the signal is the same.

What not to do

• Don't deny, rescind, or hold an application based on the flag alone.
• Don't describe a flagged student as fraudulent in notes or communications — the flag means "review this," not "this is fraud."
• Don't skip the human step because the confidence is high. High confidence means the pattern is clear, not that the explanation is.

Common gotchas

The ss_celia_fraud_* fields are empty in Slate

Three usual causes, in order of likelihood: the Flow's Fraud Protection toggle is off; your plan isn't Enterprise (the toggle can't be enabled); or the four fraud prompts aren't declared in your Slate Source Format target, so Slate silently drops them. Also check that dry run is off — dry-run ticks never write back anything.

A student's fraud level dropped and I don't know why

If your In Progress query echoes ss_celia_fraud_level, an unexplained drop from High/Critical to None/Low raises a stability warning you can filter on (ss_celia_stability_warning = 1). If you never see those warnings, verify the echo field is in your In Progress query — see the starter kit field roster.

No Fraud Level column on the Students page

The column only appears when at least one Flow in your organization has Fraud Protection enabled. No fraud-enabled Flows → no column — that's expected.