Nestack Agent Care
Industries / Education / Timeline agent

Education AI agent · IEP timelines

Special-Education Timeline AI Agent

Run each IEP clock in the unit your state actually uses, check who was on the team and which required fields are empty — and never draft, never propose accommodations, never summarise a file.

4–6 weeksTypical delivery
Your stackDeployment
Checks onlyTeam decides
Agent CareAfter launch

What this agent does

Checks the file, never writes in it

In
01

Consent arrives, and the evaluation clock starts on the day of receipt, in the unit that state uses.

02

A Texas clock runs in school days and a Georgia one in calendar days, and the exclusions ride with the count.

Reason
03

A New York consent starts two clocks in two different units, and the district has to hit both of them.

04

A review falls due, and 34 CFR 300.323(a) still asks whether an IEP is in effect on the first school day.

05

A meeting is convened, and 34 CFR 300.321(a) is read against the roster as it stands, before it opens.

Decide
06

An excusal is claimed, and the file is checked for the written consent and the written input it needs.

07

A notice falls due, and each content field 34 CFR 300.503(b) lists is tested for emptiness, nothing more.

Out
08

A re-evaluation date nears, and the agreement to forgo one is looked for before anything is chased.

09

Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent reports clocks and gaps; the team, which must include the parent, decides everything the plan says.

Example workflow

One case file, consent to review

AgentHuman
1Consent receivedSigned consent form, referral record, transfer file or eligibility determination
2Clocks computedEvaluation, eligibility, review and re-evaluation dates in the unit the state declares
3File checkedRoster as convened, excusal documents, empty required fields and confidence
4Controls appliedState-rule checks, roster tests against 34 CFR 300.321, notice-date checks and confidence threshold
No human action required

Stages 1 to 4 run unaided, and none of them writes a word into the plan — the agent is checking, and the case-manager lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the case manager to review.

Low confidence

Adds a compliance-office read first.

Case-manager review

The finding is held with the state rule it was computed under and the confidence.

Review · Correct · Send to compliance
Reviewed — the team convenes
6Case-management record updatedOnly where write access and approval policy allow it
7Outcome evaluatedClock accuracy, roster findings, notice dates and compliance findings after the meeting
Corrections

Every case-manager correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items
Drafting any part of the plan — goals, services, minutes or placement.
Proposing, ranking or scoring an accommodation for a child.
Writing the other-options-considered text a notice has to carry.
Summarising a case file, a meeting history or a child.
Automation boundaryAgent acts unaided
Compute each clock in the unit the named state rule uses.
Test the roster as convened against the members required.
Report which required fields are empty, and nothing about content.
Track whether the invitation and the notice went, and on what date.
Get the unit wrong and a child is owed compensatory services; the district answers, not a signer.
Determining eligibility, or that a re-evaluation is unnecessary.
Sending the meeting invitation or the prior written notice.
Deciding that an excusal is acceptable in the circumstances.
Changes to state-rule configuration, clocks or review thresholds.

Example output

One case file, annotated

Everything the agent reports is attached to the date or the field it came from.

Timeline output · single case fileIllustrative example
Student
Finding
Clock
Rule applied
Confidence
Team status
Initial evaluation
Consent reached the school office before the date the system logged against it
Recount required
Tex. Educ. Code 29.004
91%
Regular education teacher unconfirmed
As receivedTaken from the consent record and the state rule — nothing on this side is decided by the agent.
Record fields used Date-stamped consent Excusal document Meeting invitation log
Why this flagThis state counts school days, and the office log is not the receipt date.
ActionReviewCorrectSend to compliance
What the score decidesBelow the configured threshold the finding picks up a compliance read before.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04
Every case fileFrom the consent record
03Checking

Check against the state rule

Draw on the consent dates, the roster and the fields on file. Deal v. Hamilton County, 6th Cir., 16 December 2004: participation must be more than a mere form, it must be meaningful — a finished document is a decision that arrived before the discussion.

01Approved path

It runs clocks, it writes nothing

Routine clock arithmetic and roster checks arrive already done.

02Human review

Point people at what can void

Missed units, unconfirmed members and empty fields are marked. The second most common AI use here is choosing accommodations, nearly doubled in a year; the agent refuses it.

04Build an evidence trail

The clock, the state rule it runs under and the team who decided stay on the file.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Special-education systemsFrontline Special Education
PowerSchool Special Programs
Student information systemsPowerSchool SIS
Infinite Campus · Skyward
State reportingSEA data collections
Part B child-count extracts

Agent

IEP timeline tracking

Reads the file
Runs the clocks
Holds for review

Assessment and servicesEvaluation report stores
Related-service scheduling
Observability & evaluationOpenTelemetry · 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 file

Six layers, each drawn tighter than the last. What the stack misses is named in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeHold the case file at incomplete and stop reporting when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, state-rule and clock-configuration changes.Track
L4TraceabilityRecord the dates, the rule applied, the findings and the review time.Record
L3Case-manager reviewHolds findings for a named case manager; it governs what is reported, not whether the meeting was real.Gate
L2Policy guardrailsTest each finding against the configured state rule and the refusal list; a failure returns the finding.Restrict
L1Confidence thresholdsRoute low-confidence clock computations to a compliance read before the case manager.Require review
Model coreFinding produced — clocks, roster status, empty fields and confidence
L1 – L2Test whether a finding may stand
L3Puts the report in a case manager's hands
L4 – L5Keep the clock and the rule behind it
L6Holds the case file at incomplete when signals degrade

How Nestack evaluates it

Evaluate the whole check — not only the date it returns.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the finding the case manager reads
Depth of coverage ▼
E1Final-output evaluationDid each clock run in the unit the named state rule sets?
E2Step-level evaluationDid the agent use the right consent date, roster and state configuration?
E3Tool evaluationDid it read the correct case file and write the correct timeline record?
E4Confidence calibrationDo low-confidence findings actually attract more case-manager corrections?
E5Slice evaluationHow far apart do miss rates sit across specific timeline types?
E6Business outcomeHow many findings were corrected, and how many surfaced only after a meeting?
Floor — the timeline the district answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed where it starts in the lifecycle.

Agent lifecycleDirection of processing →
01 · Retrieval1 mode
HR-03

Log date read as receipt

The office logged the consent days after the parent handed it in.

Stage gathersConsent dates, roster, notices and the state rule
02 · Reasoning2 modes
HR-04

Wrong unit applied

A calendar-day count is used where the state counts school days.

HR-06

Second state clock missed

One of the two clocks a state runs is modelled and the other is not.

Stage proposesClocks, roster status, empty fields and confidence
03 · Tool / write2 modes
HR-02

Flag wired to a generator

An empty-field flag is piped into a text writer elsewhere.

HR-05

Checkbox read as evidence

A special factor is ticked considered and never discussed.

Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
HR-01

Unconfirmed member missed

A required team member never confirmed and the meeting went ahead.

Stage returnsThe finding the case manager acts on before a meeting
05 · Change / Version1 mode
HR-07

Silent state-rule drift

A rule change alters a clock and nothing recomputes the open files.

Stage tracksModel, prompt, state rules and clock config
Sev-1 · content written into a plan Sev-2 · a state timeline is already missed Sev-3 · date degrades, finding goes to review

Affected slices

One timeline type can carry most of the misses

A district-level field-completeness figure can read sound while a single timeline type holds most of the missed clocks and empty fields. Nestack reports the miss rate by timeline type, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example
SliceFailure rateLift Lift vs. thresholdStatus
Initial evaluation clocks6.4%3.4× Review
Transfers in from another state4.5%2.4× Review
Re-evaluation due dates3.5%1.8× Watch
Routine annual reviews2.1%1.1× Normal
Bar: miss-rate lift vs. the routine-annual-review baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

A missed clock ends as a test

A cycle is done when the missed timeline has become a case the next release must pass. That suite is what the next file checked is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect

Miss rate rises in one timeline type.

02Diagnose

The evaluation that started on the day the form was signed, not the day consent arrived, is read back through the case file until the cause narrows to one.

03Improve

Any change goes out with a number, and the timelines that caused it attached.

04Verify

One timeline case still failing is enough to hold the release.

05Learn

One case joins the suite, one line joins the compliance record.

Learn → DetectThe return edge. The next detection runs 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, timeline workflow, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6
01Timeline workflow discovery and refusal boundary.
02SIS and case-management system.
03State timeline, unit and exclusion-rule mapping.
04Consent-date ingestion and normalisation.
05Clock arithmetic and roster checks.
06Confidence scoring and finding routing.
07Case-manager review workflow.
08Case-management and SIS integration.
09Timeline and composition cases.
10Guardrails and refusal controls.
11Timeline-trail instrumentation.
12Deployment, documentation and Agent Care handover.
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 state, one district ProductionProduction case-management systems AdvancedMultiple states / districts
Introduced at Pilot
Clocks in your state's unit
Case-manager review
Field-completeness baseline
Introduced at Production
Reporting by timeline type
Review workflow in your systems
Approved date write-back
Case-management integration
Introduced at Advanced
Multi-state unit and exclusion rules
Multi-stage compliance review
High case-file volume
Multi-state timeline controls
Build price From $5,000 From $8,000 Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, case-file volume, review controls and deployment requirements.
Separate from buildBuild 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 bringWe build with it
01Your case files and where consent dates live Consent-date ingestion and clock mappingWeek 1
02Representative files from a past review cycle Timeline baseline, unit normalisation and rule bindingWeek 2
03Your state rule, its unit and its exclusions State timeline, unit and exclusion-rule mappingWeek 1
04Access to relevant APIs, feeds or exports SIS, case-management and assessment-store review, then integration setupWeek 2
05Files you would not want examined Team-composition cases and the eval suiteWeek 4
06What no agent may ever write Confidence scoring, finding routing, guardrails and refusal controlsWeek 3
07Named case managers to review findings Case-manager review workflow, then pilot and production validationWeeks 5–6
Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Each width follows the weeks that phase truly occupies, which is why two of them share week five.

Phase W1W2W3W4W5W6
Discovery W1
Build W2 – W3
Evaluate W4 – W5
Pilot & Launch W5 – W6
Week focus W1Timeline workflow discovery, state-rule mapping and the refusal boundary W2Case-file integration and the clock baseline W3Clock arithmetic, roster checks and review controls W4Evaluation suite, composition cases and failure-mode testing W5Case-management integration, a pilot cohort and targeted corrections W6One review cycle run under the case manager, then Agent Care handover
Reading the bandBands cover the weeks their work is named in and no more. The doubled fifth week is real, not padding.
At the end of W6Once the timeline record validates, Agent Care assumes the agent.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Education AI agent

Build a timeline agent around your state's own clocks.

Show us your case files, your state rule and who convenes. Your coordinator already knows which field gets filled in last — we tell her it is empty and stop there. Drafting belongs to our teacher copilot; our tutoring copilot will not move an accommodation.

Nestack Agents · IEP timelinesAGT-ED-16 · Agent Care available after launch