GEN 86%
of GEN's flags were real trafficking cases
was 57%
✓ Approved
Client portal · fictional pilot client
Harbor Stays: does GEN work in your queue?
Results from the 60-day pilot on your Trust & Safety queue, measured against the 30 days before go-live. Every number shows its definition, period, denominator and review status. Values that have not cleared review are withheld, not rounded up.
At a glance · synthetic data
What the pilot did for your queue
Approved means GEN checked the number’s source and sample, so it can be shared. Held means there is not enough evidence yet, so the number is withheld, not rounded up. The detailed charts follow below.
69%
of cases found before anyone reported them
was 30%
✓ Approved
8 of 11
referrals that police or partners acted on
new in the pilot
✓ Approved
−31%
analyst time per case (24 → 16.5 min)
≈ $6,250 saved per 1,000 cases at $50/hour
✓ Approved
85%
of real cases caught
was 60%
✓ Approved
88%
of decisions ready for an audit
was 63%
✓ Approved
−56%
harmful content an analyst views per case
18.9 → 8.4 items
✓ Approved
—
hours from flag to report
Submission timestamps come from the client export and are not yet reconciled with CyberTipline receipts
‖ Held
Money saved assumes a $50/hour loaded analyst cost. Pilot: 60 days on a fictional booking platform, compared with the 30 days before.
Tier 1 · Detection results
Detection quality, speed and audit readiness
Lines show each 10-day pilot window; the dashed line is your pre-GEN baseline. Hover or focus a point for its value and sample size, or open the data table.
Tier 1 · HS-PREC
Alert precision and recall
ApprovedPrecision
86.2%
Baseline 57.1% · +29.1 pts
95% range 75.1–92.8% · n = 58
Data table
| Window | Dates | Value | n |
|---|---|---|---|
| Days 1–10 | 2026-07-03 – 2026-07-12 | 77.8% | 20 |
| Days 11–20 | 2026-07-13 – 2026-07-22 | 80.0% | 20 |
| Days 21–30 | 2026-07-23 – 2026-08-01 | 88.9% | 20 |
| Days 31–40 | 2026-08-02 – 2026-08-11 | 90.0% | 20 |
| Days 41–50 | 2026-08-12 – 2026-08-21 | 90.0% | 20 |
| Days 51–60 | 2026-08-22 – 2026-08-31 | 90.0% | 20 |
| Baseline | — | 57.1% | — |
Recall
84.7%
Baseline 60.0% · +24.7 pts
95% range 73.5–91.8% · n = 59
Data table
| Window | Dates | Value | n |
|---|---|---|---|
| Days 1–10 | 2026-07-03 – 2026-07-12 | 77.8% | 20 |
| Days 11–20 | 2026-07-13 – 2026-07-22 | 80.0% | 20 |
| Days 21–30 | 2026-07-23 – 2026-08-01 | 88.9% | 20 |
| Days 31–40 | 2026-08-02 – 2026-08-11 | 90.0% | 20 |
| Days 41–50 | 2026-08-12 – 2026-08-21 | 90.0% | 20 |
| Days 51–60 | 2026-08-22 – 2026-08-31 | 81.8% | 20 |
| Baseline | — | 60.0% | — |
- Definition
- Precision = adjudicated true positives ÷ high-priority alerts. Recall = adjudicated true positives ÷ all actual positives in the same adjudicated sample. Always shown together.
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 120 adjudicated items: 58 high-priority alerts, 59 actual positives (TP 50, FP 8, FN 9, TN 53)
- Claim limit
- Holds for this adjudicated sample only; not prevalence or field performance elsewhere. Higher is better.
Tier 1 · HS-MIN
Analyst minutes per decision
ApprovedMedian minutes per decision
16.5 min
Baseline 24 min · −7.5 min (−31%)
Data table
| Window | Dates | Value | n |
|---|---|---|---|
| Days 1–10 | 2026-07-03 – 2026-07-12 | 22.5 min | 10 |
| Days 11–20 | 2026-07-13 – 2026-07-22 | 18 min | 10 |
| Days 21–30 | 2026-07-23 – 2026-08-01 | 18 min | 10 |
| Days 31–40 | 2026-08-02 – 2026-08-11 | 15 min | 10 |
| Days 41–50 | 2026-08-12 – 2026-08-21 | 15 min | 10 |
| Days 51–60 | 2026-08-22 – 2026-08-31 | 14 min | 10 |
| Baseline | — | 24 min | — |
- Definition
- Median active analyst minutes from intake to documented disposition, with GEN (pilot) vs without GEN (baseline).
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 60 timed decisions with GEN; 30 without
- Claim limit
- Time-tracked sample of decisions; not a headcount saving until the client confirms its staffing model. Lower is better.
Tier 1 · HS-READY
Report-ready rate
ApprovedReport-ready
88.3%
Baseline 63.3% · +25.0 pts
95% range 77.8–94.2% · n = 60
Data table
| Window | Dates | Value | n |
|---|---|---|---|
| Days 1–10 | 2026-07-03 – 2026-07-12 | 70.0% | 10 |
| Days 11–20 | 2026-07-13 – 2026-07-22 | 80.0% | 10 |
| Days 21–30 | 2026-07-23 – 2026-08-01 | 90.0% | 10 |
| Days 31–40 | 2026-08-02 – 2026-08-11 | 90.0% | 10 |
| Days 41–50 | 2026-08-12 – 2026-08-21 | 100.0% | 10 |
| Days 51–60 | 2026-08-22 – 2026-08-31 | 100.0% | 10 |
| Baseline | — | 63.3% | — |
- Definition
- Decisions passing the audit checklist (provenance, rationale, reviewer, timestamp) ÷ sampled decisions.
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 60 audited decisions in the pilot
- Claim limit
- Checklist completeness, not a legal opinion on any filing. Higher is better.
Tier 1 · HS-NCMEC
Time to NCMEC report
HeldValue not released. Submission timestamps come from the client export and are not yet reconciled with CyberTipline receipts
- Definition
- Median hours from first detection to a submitted NCMEC CyberTipline report.
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 18 submitted reports with GEN; 10 without
- Claim limit
- Measures the platform’s own submission step; says nothing about what happens after NCMEC receives a report. Lower is better.
Tier 1 · HS-PROACT
Proactive-rate uplift
ApprovedProactive rate
69.0%
Baseline 30.0% · +39.0 pts
95% range 54.0–80.9% · n = 42
Data table
| Window | Dates | Value | n |
|---|---|---|---|
| Days 1–10 | 2026-07-03 – 2026-07-12 | 57.1% | 7 |
| Days 11–20 | 2026-07-13 – 2026-07-22 | 57.1% | 7 |
| Days 21–30 | 2026-07-23 – 2026-08-01 | 71.4% | 7 |
| Days 31–40 | 2026-08-02 – 2026-08-11 | 71.4% | 7 |
| Days 41–50 | 2026-08-12 – 2026-08-21 | 71.4% | 7 |
| Days 51–60 | 2026-08-22 – 2026-08-31 | 85.7% | 7 |
| Baseline | — | 30.0% | — |
- Definition
- Share of confirmed trafficking items found by GEN before any user report, pilot vs baseline (after Meta’s “proactive rate”).
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 42 confirmed items with GEN; 20 without
- Claim limit
- Detection share within this platform’s confirmed items; not a count of harm prevented. Higher is better.
Tier 2 · Mission and team
What happened next, and what your analysts had to see
Partners’ next steps mean a case was supported. GEN does not count rescues.
Tier 2 · HS-ACTION
Partner-confirmed next action
ApprovedPartner-confirmed next action
72.7%
95% range 43.4–90.3% · n = 11
- Definition
- Referrals with a documented, partner-acknowledged next step ÷ eligible referrals.
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 11 eligible referrals
- Claim limit
- Survivors and investigations are “supported”, never “rescued”: a next step is not an outcome. Higher is better.
Tier 2 · HS-EXPOSE
Analyst exposure
ApprovedItems viewed per confirmed case
8.4 per case
Baseline 18.9 per case · −10.5 per case (−56%)
Data table
| Window | Dates | Value | n |
|---|---|---|---|
| Days 1–10 | 2026-07-03 – 2026-07-12 | 13.8 per case | 10 |
| Days 11–20 | 2026-07-13 – 2026-07-22 | 9.4 per case | 10 |
| Days 21–30 | 2026-07-23 – 2026-08-01 | 11 per case | 10 |
| Days 31–40 | 2026-08-02 – 2026-08-11 | 7.8 per case | 10 |
| Days 41–50 | 2026-08-12 – 2026-08-21 | 4.8 per case | 10 |
| Days 51–60 | 2026-08-22 – 2026-08-31 | 5.2 per case | 10 |
| Baseline | — | 18.9 per case | — |
- Definition
- Items a human analyst had to view ÷ confirmed cases, pilot vs baseline.
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31; baseline 2026-06-03 to 2026-07-02
- Denominator
- 30 confirmed cases with GEN; 12 without
- Claim limit
- A workload proxy: fewer items viewed, not a mental-health result. Lower is better.
Tier 2 · HS-OUTCOME
Human outcomes
UnmeasuredNot measured. Unmeasured by design: no independently verified, consented outcome records exist, and partner actions are not converted into rescues.
- Definition
- Independently verified, consented outcomes for people. Not measured by this product.
- Period
- 60-day pilot, 2026-07-03 to 2026-08-31
- Denominator
- None: no verified outcome records exist
- Claim limit
- GEN does not convert alerts, reports or partner actions into rescue counts. Reach, not quality.
How to read a status
- Approved
- The evidence cleared every check and a named reviewer signed off for this audience.
- Held
- Calculated but not released: the reason is shown, and the value stays inside GEN until it is fixed.
- Unmeasured
- No events exist for it yet, or it is out of scope by design.