Retail AI agent · Substance disclosure
Chemical & Substance Disclosure AI Agent
Chase the supplier declarations behind each component, keep the per-state filing record and its clocks in one place, and leave verification and submission to the product steward.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Person verifies
- Per state
- After launch
- Agent Care
What this agent does
The chase is the deliverable, not the filing
In
- 01
A declaration is requested, and the request is logged with its channel, its date and the component it covers.
- 02
A supplier answers, and the reply is filed against the component and the state definition it was asked under.
Reason
- 03
A sub-tier goes silent, and the escalation runs on and is recorded rather than rolled up as a clean no.
- 04
A component stays open, and it is carried as an allowed unknown value, never as an absence.
- 05
A window opens: the Minnesota duty dates from 1 January 2026, the submission from 15 September 2026.
Decide
- 06
A federal window opens 31 January 2027 or sixty days after EPA announces, whichever is earlier.
- 07
A record is assembled, and it goes to the product steward who verifies and submits where a programme asks it.
Out
- 08
A product leaves the supply chain, and the Minnesota five-year retention clock starts there, not at the filing.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the record; a named person verifies and submits. In 40 CFR part 705 the only certification covers the confidentiality claims, not the chemistry.
Example workflow
One component, request to filing
- 1Component record receivedBill of materials, supplier declaration, test report or product record
- 2Evidence gatheredDeclarations, questionnaires, test data and the chase history behind each one
- 3Position draftedPer-state obligation, the evidence under it, open unknowns and confidence
- 4Controls appliedEvidence-binding checks, per-state definition checks, deadline checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is filed at any of them — the agent is assembling, and the steward's lane opens at the confidence gate.
Goes to the product steward to review.
Adds a regulatory counsel read first.
Steward approval
The record is held with its unknowns, its chase history and the confidence.
Approve · Amend · Send to counsel
Approved — released to file
- 6Product and filing records updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedDeclaration coverage, unknowns closed, filing acceptance and post-filing corrections
Amendments
Steward amendments are counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Verifying a state report as accurate and complete.
- Determining that a substance is present or absent.
- Deciding that a Proposition 65 warning is required.
- Drafting, placing or clearing on-pack warning copy.
Automation boundary
Agent acts unaided
- ✓Request declarations per component and log each request.
- ✓Carry each position with the evidence and the date behind it.
- ✓Track registration, reporting and renewal clocks per state.
- ✓Report the components still open, and why each one is open.
The verification is a person's. If the report was wrong, that file is what the state reads back.
- Signing the EPA confidentiality certification.
- Deciding that a filing obligation bites in a state.
- Judging whether a declaration supports a position.
- Changes to state, definition or approval rules.
Example output
One component, annotated
What the agent records is attached to the declaration it came from.
Substance output · single component
Illustrative example- Component
- Coated zip pull
- What the evidence shows
- Supplier declares no intentionally added PFAS; the coating sub-tier has not answered
- States in scope
- Minnesota and Connecticut
- Evidence on file
- Supplier declaration
- Confidence
- 74%
- Verification
- Held for the product steward
As received
Taken from the declaration and the chase log as filed — nothing on this side is asserted by the agent.
Declarations used
- Supplier declaration
- Third-party test report
- Sub-tier request open
Why this stays unknown
Minnesota contemplates allowed unknown values; the agent may not close one.
What the score decides
Below the configured threshold the record adds a counsel read before it reaches the steward.
Action
ApproveAmendSend 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 component
From the declarations on file
Build the position from the evidence
Draw on the declarations and test data on file. The Minn. Stat. §116.943 duty dates from 1 January 2026; MPCA has extended the initial submission to 15 September 2026 — an administrative date sitting on a statutory one.
Intentionally added is a finding
Minn. Stat. §116.943 puts it as a function test: deliberately added so the continued presence performs a specific function. Other states word it their own way, so the finding is made per state.
Not the conformity file
The catalogue agent leaves warning fields to your specialist; this agent assembles what is in the product and what the supply chain said, and a named person decides whether that requires a warning.
Build an evidence trail
The component, the declaration under it and the person who submitted stay on the filing.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Product and bills of materials
- Akeneo
- Salsify
- Component and BOM records
Supplier declarations
- Assent
- Sphera
- 3E
- Supplier portals
Test data and laboratories
- SGS
- Eurofins
- Intertek
- Bureau Veritas
Agent
Chemical & substance disclosure
- Reads the evidence
- Chases the gaps
- Holds for the steward
Filing portals
- Minnesota PRISM
- EPA CDX
- state 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 filing
The six narrow as they run. Whatever gets through them all is drawn in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a position may stand
- L3
- Puts the release in a steward's hands
- L4 – L5
- Keep the component and the declaration behind it
- L6
- Narrows to declaration chasing when signals degrade
How Nestack evaluates it
Evaluate the whole chase — not only the finished filing.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the filing the state reads
- E1Final-output evaluationDid each position cite evidence that actually supports it?
- E2Step-level evaluationDid the agent use the right state definition, rules and bill of materials?
- E3Tool evaluationDid it read and write the correct component and the correct programme?
- E4Confidence calibrationDo low-confidence records actually attract more steward amendments?
- E5Slice evaluationHow does performance change across substance classes and component types?
- E6Business outcomeHow many records needed an amendment or a correction after the filing went in?
Floor — the outcome the steward answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, set at the lifecycle stage each one begins in.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- HE-03
Superseded agency date
A window is read off EPA's page, which the rule has superseded.
Stage gathers
Bills of materials, declarations and chase logs
02 · Reasoning
2 modes
- HE-04
Definition carried over
A finding made under one state test is applied in another.
- HE-06
Unanswered tier rolled up
A silent sub-tier becomes a clean supplier no.
Stage proposes
State positions, open unknowns and confidence
03 · Tool / write
2 modes
- HE-02
Renewal closed as sent
A renewed notice closes with no confirmation of receipt.
- HE-05
Retention clock misstarted
Records are aged from the filing date, not the exit date.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- HE-01
Position without evidence
A state position cites no declaration on file.
Stage returns
The record the steward verifies and submits
05 · Change / Version
1 mode
- HE-07
Silent definition regression
A model or rule change widens what counts as added.
Stage tracks
Model, prompt, definition rules and state config
- Sev-1 · agent acts outside the boundary
- Sev-2 · a wrong position reaches the filing
- Sev-3 · source degrades, record goes to review
Affected slices
One substance class can carry most of the gaps
A portfolio-level declaration-coverage figure can read as sound while one substance class holds most of the unanswered components. Nestack reports coverage by substance class, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| PFAS, coated textile parts | 7.1% | 3.6× | Review | |
| PFAS, sub-assembly hardware | 5.0% | 2.6× | Review | |
| Listed chemicals, food contact | 3.5% | 1.8× | Watch | |
| Single-material moulded parts | 2.3% | 1.2× | Normal |
Bar: unanswered-declaration lift vs. moulded-part baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The cycle shuts on a standing test
A cycle is done when the unanswered supplier has become a case the next release must pass. That suite is what the next filing.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
The unanswered-declaration rate rises in one substance class.
- 02
Diagnose
The zip pull nobody had a declaration for is read back through its chase log until the cause narrows to one.
- 03
Improve
Stamp the change; the components behind it are filed against that number.
- 04
Verify
Each touched component case is run again, and one red stops the release.
- 05
Learn
It stays a standing test, and the substance rules move with it.
Learn → Detect
The return edge. The next round of detection meets a longer suite than this one.
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, record assembly, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Substance obligation discovery and boundary work.Week 1
- 02Supplier and test-data source assessment.Week 1, Week 2
- 03Per-state definition and deadline rule mapping and confirmation.Week 1, Week 2
- 04Bill-of-materials and declaration intake.Week 2
- 05Position assembly and evidence binding.Week 2, Week 3
- 06Confidence scoring and unknown routing.Week 3
- 07Steward verification workflow.Week 3, Week 4
- 08PIM and supplier-portal integration.Week 5
- 09Declaration and scope cases.Week 4, Week 5
- 10Guardrails and disclosure controls.Week 4
- 11Component-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 line, one state | ProductionProduction filing systems | AdvancedMultiple states / brands |
|---|---|---|---|
| Introduced at Pilot | |||
| Assembly to your states and rules | ✓ | ✓ | ✓ |
| Steward approval | ✓ | ✓ | ✓ |
| Declaration-coverage baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by substance class | — | ✓ | ✓ |
| Approval workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Bill-of-materials integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-programme definition rules | — | — | ✓ |
| Multi-stage disclosure approvals | — | — | ✓ |
| High component volume | — | — | ✓ |
| Multi-state disclosure controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval 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 bills of materials and state list
→Component ingestion and declaration filing
Week 102Representative supplier declarations
→Coverage baseline, evidence binding and unknown handling
Week 203Your per-state definition and deadline rules
→Per-state definition and deadline rule mapping
Week 104Access to relevant APIs, feeds or exports
→Supplier and test-data assessment, then integration setup
Week 205Filings you would not want reopened
→Supplier cases and the evaluation suite
Week 406What no filing may infer
→Confidence scoring, unknown routing, guardrails and approval controls
Week 307A named product steward to review records
→Steward verification 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 for the page; the weeks alone decide, and the fifth carries a pair of them.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Substance obligation discovery, definition mapping and the automation boundary
W2Supplier-declaration integration and the coverage baseline
W3Record assembly, chase logic and approval controls
W4Evaluation suite, unknown handling and failure-mode testing
W5Filing-portal integration, pilot components and targeted corrections
W6One reporting window run under the product steward, then Agent Care handover
- Reading the band
- Each bar runs across the weeks its work is named in and no further. The week 5 overlap is real.
- At the end of W6
- When the filing 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 · Retail AI agent
Build a substance-disclosure agent around the filings you owe.
It does not draft warnings, does not decide warning placement, does not maintain label artwork, and does not clear on-pack copy. Next step: bring one product family and we map which programmes reach it, chase what is missing, and hand your steward the record they verify from.
- Typical delivery
- 4–6 weeks
- Deployment
- Your bills of materials
- Person verifies
- Per state
- After launch
- Agent Care available
Nestack Agents · Chemical & substance disclosure
AGT-RT-21 · Agent Care available after launch