Education AI agent · Substantive change
Accreditation Substantive-Change AI Agent
Catch a planned change while it is still planned, screen it against the accreditor's own categories, build the evidence pack and run the dates — and never classify, notify or certify.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Two may submit
- Never notifies
- After launch
- Agent Care
What this agent does
Assembles the notification, never the decision
In
- 01
A change is planned, and a row opens keyed on the change, the accreditor and the intended implementation date.
- 02
A lease is executed for space nobody screened, and the site is raised as an event before anyone calls it a location.
Reason
- 03
A change is screened against that accreditor's own categories, and the category it was tested against is recorded.
- 04
A programme is scheduled, and the 90-day notice under 34 CFR 600.20 runs to the first day of class.
- 05
A building is renamed, and 34 CFR 600.21 puts it on a 10-day federal clock, not an estate one.
Decide
- 06
A change is implemented before its approval lands, and that is raised as the violation, not as a late filing.
- 07
A submission date nears, and the pack goes to the chief executive or the accreditation liaison to decide on.
Out
- 08
A change closes, and the screening record, the evidence pack and the person who decided stay against the row.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the record; the chief executive or the accreditation liaison decides what is submitted, and nobody else may submit it.
Example workflow
One change, detection to submission
- 1Change detectedExecuted lease, curriculum vote, board resolution, HR change or scheduler entry
- 2Evidence gatheredThe implementation date from the operational system, the contract, the catalogue and the ECAR row
- 3Screening row assembledCategory tested against, evidence held, dates in play and confidence
- 4Controls appliedCategory-coverage checks, date checks, evidence completeness and confidence threshold
No human action required
Stages 1 to 4 run unaided, and none of them classifies the change or files anything — the agent is assembling, and the liaison lane opens at the confidence gate.
Goes to the accreditation liaison.
Adds a counsel read first.
Liaison review
The row is held with its evidence, the category it was tested against and the confidence.
Review · Correct · Send to counsel
Reviewed — the pack goes to the submitter
- 6Change register and inventory updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedScreening coverage, row corrections, submission timeliness and findings at review
Corrections
Every liaison 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 a change as substantive, or as not substantive.
- Submitting to the accreditor — two named roles only.
- Notifying or certifying anything to the Department.
- Deciding whether a leased suite is a location or a branch.
Automation boundary
Agent acts unaided
- ✓Hold a row per detected change, with the category it was tested against.
- ✓Run the accreditor dates and the federal clocks on one row.
- ✓Reconcile the location and programme inventory against the ECAR.
- ✓Show what the pack is still missing, and hold it for a named submitter.
Wrong after the fact means students already enrolled and money already moved at that address.
- Deciding what evidence is sufficient for a prospectus.
- State licensure certification to the Department — our licensure agent.
- Records requests, amendments and holds — our enrollment and records agent.
- Changes to screening, date or submission-control configuration.
Example output
One detected change, annotated
Everything the agent assembles is attached to the system it came from.
Screening-register output · single change
Illustrative example- Change
- Certificate, leased site
- Evidence line
- A lease was executed for a suite where a partner will teach most of a certificate
- Days to report
- Day 6 of 10
- Rule applied
- 34 CFR 600.21
- Confidence
- 88%
- Screening status
- Unscreened — no category recorded
As received
Taken from the executed lease and the room scheduler — nothing on this side is the agent's.
Source used
- Executed lease
- Room-scheduler entry
- Partner agreement
Why this flag
How much of it the partner teaches is the question; the liaison decides what it means.
What the score decides
Below the configured threshold the row picks up a counsel read before the liaison sees it.
Action
ReviewCorrectSend to counselValue
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 detected change
From the systems the change happens in
Screen against the accreditor
Draw on the accreditor's own published categories, the executed contract, the catalogue and the ECAR. 34 CFR 602.22 binds the accreditor to hold such a policy; the duty an institution owes runs from that policy. A departmental rulemaking on that section is open, and nothing in it is in force.
Approval is never backdated
Routine screening rows and date arithmetic arrive already done, keyed on the implementation date.
Point people at the changes nobody screened
One accreditor runs fixed dates, another runs submission cycles with fees, another publishes no advance deadline and says it cannot approve retroactively — so each body's calendar is data the product refreshes, not copy. The one national rule sits elsewhere: under 34 CFR 600.20(f)(5) the institution is liable for all Title IV funds disbursed to students enrolled at a location or programme it never applied for.
Build an evidence trail
The change, the classification it drew and the officer who submitted 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.
Student and academic systems
- Banner
- Colleague
- Workday Student
- Jenzabar
Curriculum and catalogue
- CourseLeaf CIM
- Acalog
- Coursedog
- Kuali
Contracts and real estate
- Lease and partner agreements
- Contract-management systems
Agent
Accreditation substantive change
- Detects the change
- Assembles the pack
- Holds for the submitter
Federal and accreditor records
- ECAR
- eligibility record
- Accreditor submission portals
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 submission
Six bands read from the outside in. What survives the innermost is listed in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a row may stand
- L3
- Puts the pack in the submitter's hands
- L4 – L5
- Keep the change and the classification behind it
- L6
- Reverts to notification assembly when signals degrade
How Nestack evaluates it
Evaluate the whole detection net — not only the pack on the desk.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the pack the submitter files from
- E1Final-output evaluationDid each row carry the category it was tested against?
- E2Step-level evaluationDid the agent use the right accreditor policy version, contract and calendar?
- E3Tool evaluationDid it read the correct source system and write the correct row?
- E4Confidence calibrationDo low-confidence rows actually attract more liaison corrections?
- E5Slice evaluationHow far apart do screening rates sit across specific change types?
- E6Business outcomeHow many changes were caught late, and how many surfaced only at review?
Floor — the submission the institution answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the lifecycle stage each one starts in.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- HP-03
Source system unwatched
A signed lease never reaches the screening net.
Stage gathers
Contracts, catalogue, scheduler and the ECAR
02 · Reasoning
2 modes
- HP-04
Category mis-keyed
A certificate is keyed to the degree it sits inside.
- HP-06
Stale accreditor policy
A row is built on a policy version the accreditor has replaced.
Stage proposes
Category tested, evidence held and confidence
03 · Tool / write
2 modes
- HP-02
Federal clock unrun
The accreditor date runs and the federal one does not.
- HP-05
Disbursement not tied
Money is scheduled at a site whose approval has not landed.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- HP-01
Change never screened
An implemented change has no screening row at all.
Stage returns
The pack the chief executive or liaison files
05 · Change / Version
1 mode
- HP-07
Category set moves
An accreditor retires categories and the net keeps the old ones.
Stage tracks
Model, prompt, category rules and accreditor config
- Sev-1 · a change classified without a person
- Sev-2 · a change is implemented unapproved
- Sev-3 · a source goes quiet, row to review
Affected slices
One change type can carry most of the misses
An institution-level classification-coverage figure can read sound while a few change types carry most of the events that reached a deadline unscreened. Nestack reports the unscreened rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Leased sites and partner space | 6.7% | 3.6× | Review | |
| Partner-delivered programmes | 4.9% | 2.6× | Review | |
| Programmes inside an existing CIP | 3.1% | 1.7× | Watch | |
| New degrees through committee | 1.4% | 0.8× | Normal |
Bar: unscreened-rate lift vs. the committee-degree baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A late notification ends as a test
A cycle closes when the late notification is a regression case. That suite is what the next change tracked is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Screening coverage falls in one change type.
- 02
Diagnose
The site that opened in September and was notified in March is read back through the register until the cause narrows to one.
- 03
Improve
The change goes out with a number, and the sites behind it travel with it.
- 04
Verify
Nothing submits while one touched change case is still red.
- 05
Learn
It is kept for good, and the classification rules move in the same commit.
Learn → Detect
The return edge. The next change tracked is checked 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, screening workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Submission workflow discovery and the boundary.Week 1
- 02Contract and catalogue source assessment.Week 1, Week 2
- 03Accreditor category and submission-date mapping.Week 1, Week 2
- 04Change detection and event ingestion.Week 2
- 05Screening rows and evidence binding.Week 2, Week 3
- 06Confidence scoring and flag routing.Week 3
- 07Liaison review workflow.Week 3, Week 4
- 08Programme-inventory and ECAR integration.Week 5
- 09Classification and date cases.Week 4, Week 5
- 10Guardrails and submission controls.Week 4
- 11Change-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 accreditor, one campus | ProductionProduction source systems | AdvancedMultiple accreditors / campuses |
|---|---|---|---|
| Introduced at Pilot | |||
| Rows and dates against your policy | ✓ | ✓ | ✓ |
| Liaison review | ✓ | ✓ | ✓ |
| Classification-coverage baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by accreditor | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved register write-back | — | ✓ | ✓ |
| Programme-inventory integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Complex multi-accreditor policy sets | — | — | ✓ |
| Multi-stage liaison review | — | — | ✓ |
| High change volume | — | — | ✓ |
| Multi-accreditor submission controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, change 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 accreditor's policy and category list
→Change detection and event ingestion, with evidence binding
Week 102Representative changes from a past cycle
→Classification-coverage baseline, screening capture and row keying
Week 203Your sources for leases, curriculum and scheduling
→Accreditor category and submission-date mapping
Week 104Access to relevant APIs, feeds or exports
→Contract, catalogue and scheduler review, then integration setup
Week 205Locations you would not want unapproved
→Date cases and failure-mode testing
Week 406What no notification may leave late
→Confidence scoring, flag routing, guardrails and submission controls
Week 307The chief executive and the accreditation liaison
→Liaison 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
Nothing here is sized to the page; elapsed weeks decide, and the fifth carries two of the bands.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Submission workflow discovery, category mapping and the boundary
W2Source integration and the classification-coverage baseline
W3Screening rows, date arithmetic and review controls
W4Evaluation suite, date cases and failure-mode testing
W5Inventory integration, a pilot change set and targeted corrections
W6One submission cycle run under the accreditation liaison, then Agent Care handover
- Reading the band
- A bar covers only the weeks its own work is named in. The fifth-week doubling is real, not padding.
- At the end of W6
- Validation closes on live changes, and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Education AI agent
Build a substantive-change agent around the two roles who may submit.
A register is worth building before the next change is planned, not after a reviewer finds it. Bring us your accreditor's category list, the systems your changes actually happen in and the two people who may file.
- Typical delivery
- 4–6 weeks
- Deployment
- Your source and student stack
- Two may submit
- Never notifies
- After launch
- Agent Care available
Nestack Agents · Accreditation substantive change
AGT-ED-15 · Agent Care available after launch