Education AI agent · Clery reporting
Clery Annual Security Report AI Agent
Hold the Clery reporting record whole — the authority register and who must never be asked, each report with geography applied, the log clock — while a named survey administrator submits.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- CSSA submits
- Never unfounds
- After launch
- Agent Care
What this agent does
Compiles the record, never the classification
In
- 01
A report reaches a person on the authority register, and it enters the record with what that person was asked.
- 02
A counsellor acting as one is not a campus security authority under 34 CFR 668.46(a), and is never asked.
Reason
- 03
A log entry falls due within two business days of the report, and the clock is run in business days.
- 04
A report is placed against Clery geography property by property, and a recognised fraternity house is a property.
- 05
A prescribed policy statement missing from the draft report is named before 1 October, not after.
Decide
- 06
A decision not to warn is filed with the time of confirmation and the decider, exactly as a decision to warn is.
- 07
An unfounding enters as a determination by sworn officers with its basis, and the removal becomes a disclosure.
Out
- 08
A hazing statistic is compiled on a stated reading, and 34 CFR 668.46 does not yet carry the word.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent compiles the record and runs the clocks; a named campus safety survey administrator submits, and sworn officers alone unfound.
Example workflow
One reporting year, report to submission
- 1Report reaches an authorityCampus police, a registered security authority, a local agency or a published reporting channel
- 2Record assembledThe report as received, the geography applied to it, the log time and the authority who took it
- 3File compiledAuthority register, log clock, geography, warning file and confidence
- 4Controls appliedExclusion checks, geography checks, the two-business-day log clock and confidence threshold
No human action required
Stages 1 to 4 run unaided, and none of them classifies an incident — the agent is compiling, and the administrator lane opens at the confidence gate.
Goes to the survey administrator to review.
Adds a sworn-officer read first.
Administrator review
The record is held with the geography applied, the clocks and the confidence.
Review · Correct · Send to sworn review
Reviewed — the administrator submits
- 6Reporting record updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedRegister completeness, log-clock accuracy, geography coverage and findings raised in review
Corrections
Every administrator correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Classifying an incident as final against the Clery definitions.
- Unfounding a report — reserved to sworn officers by 34 CFR 668.46(c)(2).
- Deciding that a timely warning is required, or that one is not.
- Asking a pastoral or professional counsellor to report what was told to them.
Automation boundary
Agent acts unaided
- ✓Register each authority by function, with the basis for including them.
- ✓Run the two-business-day log clock and flag an entry that is overdue.
- ✓Request statistics from the agency with jurisdiction over each noncampus property.
- ✓Show what the report is missing against the prescribed policy statements, and hold it.
Wrong after distribution, and the fix is a restated report and an amended federal submission.
- Submitting or locking the federal campus safety survey.
- Determining whether a property sits inside Clery geography.
- Choosing which reading of the hazing effective date the institution takes.
- Changes to register, geography or disclosure-control configuration.
Example output
One reported incident, annotated
Everything the agent records is attached to the report or the register entry it came from.
Reporting-record output · single incident
Illustrative example- Incident
- Residence hall report
- Recorded finding
- A report taken by a residence life coordinator has sat in the log with nobody asked to classify it
- Log clock
- 41 hours open
- Geography applied
- 34 CFR 668.46(f)
- Confidence
- 88%
- Classification status
- Unclassified — not yet put
As received
Taken from the report as received and the register as built — an alert our safety-monitoring agent routes to a counsellor is not a report here, and is never wired in as one.
Record used
- Report as received
- Authority register
- Daily crime log
Why this flag
The clock runs from the report, not from the classification — the overdue log entry is the finding, not the crime.
What the score decides
Below the configured threshold the record picks up a sworn-officer read before the administrator sees it.
Action
ReviewCorrectSend to sworn reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every report received
From the report as received
Compile against the definitions
Draw on the reports as received, the register and the geography applied. The 2016 Handbook was rescinded on 9 October 2020 as outside the scope of the statutory and regulatory authority — and the sector still cites its page numbers.
Some people must not be asked
Routine register upkeep and log-clock arithmetic arrive already done.
Point people at what a federal reviewer will demand
Unasked authorities, overdue log entries and geography gaps are marked. Hazing enters the annual security report for the first time, and the statute reads on one view as the report published in 2026 and on another as the report covering 2026 and published in 2027; the record carries which reading was taken, because no departmental guidance has been located.
Build an evidence trail
The report, the geography applied to it and the administrator who filed stay on the record.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Clery systems
- Daily crime log
- Campus safety survey portal
Incident records
- Dispatch and records systems
- Report management systems
Institution systems
- Conduct and Title IX systems
- Human resources and org data
Agent
Clery reporting and disclosure
- Reads the record
- Compiles the file
- Holds for the administrator
Notification and property
- Mass-notification systems
- Facilities and property records
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the report
Six passes, each narrower than the one before. What outlasts them is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an entry may stand
- L3
- Puts the record in the hands of the administrator
- L4 – L5
- Keep the report and the geography behind it
- L6
- Narrows to authority-register assembly when signals degrade
How Nestack evaluates it
Evaluate the whole record — not only the statistics table at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the report the campus reads
- E1Final-output evaluationWas every report received in the record, with its geography applied?
- E2Step-level evaluationDid the agent use the right register, geography map and clock configuration?
- E3Tool evaluationDid it read the correct incident and write the correct record?
- E4Confidence calibrationDo low-confidence entries actually attract more administrator corrections?
- E5Slice evaluationHow far apart do capture rates sit across specific crime-category types?
- E6Business outcomeHow many gaps were corrected, and how many surfaced only in a federal review?
Floor — the disclosure the institution answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each set where in the lifecycle it first arises.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- HJ-03
Report never reached the log
A report taken by a trip leader entered no register or log.
Stage gathers
Reports received, the register, geography and clocks
02 · Reasoning
2 modes
- HJ-04
Counsellor in the register
A confidential role is swept in from the org chart and chased.
- HJ-06
Hazing years overreached
A category is disclosed for years no obligation covered.
Stage proposes
Register, clocks, geography applied and confidence
03 · Tool / write
2 modes
- HJ-02
Record wired to a classifier
A completeness flag is piped into a classification tool.
- HJ-05
Log clock on calendar days
A holiday weekend is counted as two business days.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- HJ-01
Property never registered
A recognised fraternity house sits outside the geography register.
Stage returns
The record the administrator submits from
05 · Change / Version
1 mode
- HJ-07
Silent geography drift
A configuration change moves a property and open years do not recompute.
Stage tracks
Model, prompt, geography rules and register config
- Sev-1 · a classification is made
- Sev-2 · a report is missing at submission
- Sev-3 · source degrades, record goes to review
Affected slices
One crime category can carry most of the gaps
A campus-level authority-register figure can read sound while a single crime category holds most of the uncounted reports and the unapplied geography. Nestack reports the gap rate by crime category, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| VAWA offence categories | 8.6% | 3.7× | Review | |
| Hate-crime bias categories | 6.2% | 2.6× | Review | |
| Disciplinary referral categories | 3.9% | 1.7× | Watch | |
| Burglary and motor vehicle theft | 1.5% | 0.6× | Normal |
Bar: gap-rate lift vs. the burglary-and-vehicle-theft baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
An uncounted report ends as a test
A cycle is done when the uncounted report has become a case the next release must pass. That suite is what the next year compiled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Gap rate rises in one crime category.
- 02
Diagnose
The coach nobody had told, and the counsellor who must not be told, are read back through the register until the cause narrows to one.
- 03
Improve
Number the change; the incidents that drove it are filed under it.
- 04
Verify
Each touched incident case is run once more, and one red holds it back.
- 05
Learn
It stays as a standing test, and the geography rules move with it.
Learn → Detect
The return edge. The next year is compiled against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, reporting workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Reporting workflow discovery and refusal boundary.Week 1
- 02Incident and log-source assessment.Week 1, Week 2
- 03Authority register, geography and clock mapping.Week 1, Week 2
- 04Report and register ingestion.Week 2
- 05Clock arithmetic and geography binding.Week 2, Week 3
- 06Confidence scoring and finding routing.Week 3
- 07Survey administrator review workflow.Week 3, Week 4
- 08Incident and survey-system integration.Week 5
- 09Classification and geography cases.Week 4, Week 5
- 10Guardrails and disclosure controls.Week 4
- 11Incident-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne campus, one reporting year | ProductionProduction incident and survey systems | AdvancedMultiple campuses / systems |
|---|---|---|---|
| Introduced at Pilot | |||
| Record compiling to your register | ✓ | ✓ | ✓ |
| Administrator review | ✓ | ✓ | ✓ |
| Authority-register baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by category | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved record write-back | — | ✓ | ✓ |
| Incident-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-campus and multi-property rules | — | — | ✓ |
| Multi-stage institutional approvals | — | — | ✓ |
| High incident volume | — | — | ✓ |
| Multi-campus statistics controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, incident volume, review controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your reporting channels and where reports land
→Report and register ingestion, and geography binding
Week 102Representative files from a past reporting year
→Register baseline, clock normalisation and rule binding
Week 203Your geography register and your exclusion basis
→Authority register, geography and clock mapping
Week 104Access to relevant APIs, feeds or exports
→Incident, log and survey-source review, then integration setup
Week 205Statistics you would not want re-counted
→Log cases and failure-mode testing
Week 406What no security report may omit
→Confidence scoring, finding routing, guardrails and disclosure controls
Week 307The named campus safety survey administrator
→Administrator review workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
A band is exactly as wide as the weeks that one phase eats, which is why week five must carry two.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Reporting workflow discovery, register mapping and the refusal boundary
W2Record integration and the authority-register baseline
W3Clock arithmetic, geography binding and review controls
W4Evaluation suite, log cases and failure-mode testing
W5Incident and survey integration, a pilot year and targeted corrections
W6One reporting year run under the survey administrator, then Agent Care handover
- Reading the band
- Each bar sits only on the weeks its work is named in. The fifth-week overlap is real, not padding.
- At the end of W6
- When the report validates, Agent Care takes the agent on.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Education AI agent
Build a Clery agent around the person who submits.
Show us your reporting channels, your geography register and who holds the designation. Not a report generator. A reconstruction engine — because the evidence lists in the federal review letters of late 2025 are the specification.
- Typical delivery
- 4–6 weeks
- Deployment
- Your incident and survey stack
- CSSA submits
- Never unfounds
- After launch
- Agent Care available
Nestack Agents · Clery reporting and disclosure
AGT-ED-11 · Agent Care available after launch