Main site

Food & Beverage AI agent · Menu labelling

Menu-Labelling Substantiation AI Agent

Attach each standard menu item to the recipe version its nutrient values were calculated from, mark the basis that has drifted, and hold the substantiation file for the two individuals who sign it.

Typical delivery
4–6 weeks
Deployment
Your stack
On demand
Two signers
After launch
Agent Care

What this agent does

Assembles the file, never the certification

In

  1. 01

    An item is added, and § 101.11(a) reaches a chain of twenty or more locations.

  2. 02

    A basis is chosen, and § 101.11(c)(1) allows databases, cookbooks or laboratory analyses.

Reason

  1. 03

    A recipe is changed, and § 101.11(c)(2) wants preparation to adhere to the factors used.

  2. 04

    A file is assembled, and § 101.11(c)(3)(i) wants the database name, version and version date.

  3. 05

    An analysis closes, and § 101.11(c)(3)(i)(F) brings a responsible individual to sign.

Decide

  1. 06

    An outlet serves the item, and § 101.11(c)(3)(i)(G) wants a signer employed there.

  2. 07

    A guest asks for more, and § 101.11(b)(2)(ii)(A) sets the written nutrition information.

Out

  1. 08

    A request arrives, and § 101.11(c)(3) allows a reasonable period of time, not a date.

  2. 09

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

Product statement

The agent assembles the substantiation file and tracks what has drifted. A responsible individual signs and dates each statement under § 101.11(c)(3) — one for accuracy, one for adherence.

Example workflow

One menu item, evidence to certification

  1. 1Item evidence receivedRecipes, database printouts, laboratory worksheets or outlet preparation records
  2. 2Item context assembledThe item, its basis, the outlets offering it and the recipe version its values came from
  3. 3Substantiation evidence draftedThe items covered, the basis recorded, the gaps and completeness
  4. 4Controls appliedBasis-currency checks, outlet-coverage checks, written-information checks and completeness confidence

No human action required

Stages 1 to 4 run unaided, and nothing is certified at any of them — the agent is assembling, and the nutrition lane opens at the completeness gate.

5DecisionSplits at the completeness gate
Evidence sufficient

Goes to the responsible individuals to certify.

Anything thin

Adds a nutrition read first.

Nutrition review

The file is held with its items, its basis and the outlets behind them.

Certify · Append evidence · Send to nutrition review

Signed and dated — under § 101.11(c)(3)

  1. 6Recipe and menu records updatedOnly where write access and records policy allow it
  2. 7Outcome evaluatedBasis currency, outlet coverage, reviewer corrections and what the read found

Corrections

Each nutrition correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Signing the accuracy statement, § 101.11(c)(3)(i)(F).
    • Signing the adherence statement, § 101.11(c)(3)(i)(G).
    • Certifying that a nutrient analysis is complete.
    • Deciding a chain is not a covered establishment.

  • Automation boundary

    Agent acts unaided

    • ✓Hold every menu item against the recipe version used.
    • ✓Track the outlets where a standard menu item is offered for sale.
    • ✓Mark each item whose basis has drifted since signing.
    • ✓Assemble the evidence a request under § 101.11(c)(3) would reach.

    Nothing is signed or certified except by a responsible individual, inside agreed limits.

    • Judging what a reasonable period of time allows.
    • Telling FDA the substantiation file is complete.
    • Setting the basis a menu item is declared from.
  • Changes to recipes, menu boards or file records.

Example output

One menu item, annotated

Production runs within a reasonable period of time upon request under § 101.11(c)(3) — audit-readiness, not a clock; this record is what one standard menu item carries.

Item evidence · single menu item

Illustrative example
Menu item
Grilled chicken bowl, § 101.11(b)(1)
Recorded as
Declared from the recipe version on file, basis current
Item class
Standard item
Evidence of record
Database printout, 3 August 2026
Confidence
Held uncertified
Held for
A responsible individual at each desk

As received

Taken from the chain's own recipe files and database printouts — it reaches as far as those sources do.

What the record holds

  • Recipe of record
  • Database version
  • Outlet adherence

Why no certification here

Whether the file may be certified is a § 101.11(c)(3) act left to a person.

What the score decides

Below the configured threshold the file picks up a nutrition read before the signers see it.

Action

CertifyAppend evidenceSend to nutrition review

Value

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 menu item

From the recipe that sets it

03Evidence

Where the evidence is used

Our labelling and recipe helper owns what the pack declares and holds the artwork for the named label owner, while this page owns the menu board and the substantiation file behind it, with two different signers. Our drive-thru lane agent hands any allergen question to the shift team, and this page is what that shift team would have to have on file.

01Approved path

Two statements, two desks

§ 101.11(c)(3)(i)(F) may be signed at corporate headquarters or the parent entity; (G) only by a responsible individual employed at that covered establishment.

02Human review

What was checked, and not found

No Federal Register document amending § 101.11 in 2025 or 2026 was located, and the FDA menu labelling page still carries 13 December 2023; an eCFR marker reading last amended 8/19/2026 could not be reconciled against any published document, and Edition 2 of the supplemental guidance, announced 14 December 2023, remains draft.

04

Build an evidence trail

The item, the nutrient basis under it and the individual who dated it stay together.

Integrations

Typical integrations

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

  • Recipe and menu systems

    • Recipe management
    • POS
    • Menu boards and item files

  • Nutrient databases

    • Database name
    • version date
    • Calculation worksheets

  • Laboratory analyses

    • Analytical method
    • worksheets
    • Laboratory name and address

Agent

Menu substantiation

  • Reads the items
  • Assembles the file
  • Holds for the signers

  • Outlets and preparation

    • Store systems
    • training
    • Adherence records by outlet

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six reviews between the model and the signature

Six reviews, and not one of them repeats another. What clears every one is named in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to evidence assembly when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and basis rules; no Federal Register amendment to § 101.11 was located for 2025 or 2026.Track
L4Item trailRecord each item, its basis, the outlets it reaches and every read of the file.Record
L3Signer releaseHold the file for the two responsible individuals; the hold governs release, not whether the analysis behind it was right.Gate
L2Basis guardrailsTest the evidence against § 101.11(b)(2)(ii)(A), (c)(1), (c)(2) and (c)(3); the supplemental guidance of 14 December 2023 is still draft.Restrict
L1Confidence thresholdsRoute a thin file to a nutrition read before either signer sees it.Require review
Model coreEvidence assembled — the item, the basis, the outlets and completeness
L1 – L2
Test whether a file may stand
L3
Puts the certification in a person's hands
L4 – L5
Keep the item and the basis behind it
L6
Holds the statement undrafted when signals degrade

How Nestack evaluates it

Evaluate the whole assembly — not only the substantiation file that comes out.

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

Surface — the file a request reaches

  1. E1Final-output evaluationDid the evidence record what each menu item actually rests on?
  2. E2Step-level evaluationDid the agent read the right item, the right basis and the recipe version in force?
  3. E3Tool evaluationDid it read and write the correct item record and the correct basis?
  4. E4Confidence calibrationDo low-confidence files actually attract more nutrition corrections?
  5. E5Slice evaluationHow does performance change across specific menu items?
  6. E6Business outcomeHow many files needed a correction before the individuals signed?

Floor — the file the chain answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed at the stage where it first shows.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • JU-03

      Stale recipe read

      The recipe read is not the one the outlet now cooks.

    Stage gathers

    The items, the recipes, the basis and the dates

  2. 02 · Reasoning

    2 modes

    • JU-04

      Item asserted, not shown

      An item is called covered without its basis.

    • JU-06

      Draft guidance read as final

      The supplemental edition is worked as binding.

    Stage proposes

    The items, their basis and file completeness

  3. 03 · Tool / write

    2 modes

    • JU-02

      Thin file passed forward

      A file moves on without the nutrition read.

    • JU-05

      Evidence bound to wrong item

      A worksheet is filed against the wrong item.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • JU-01

      Certified, basis unrecorded

      The file shows a signature but not what supported it.

    Stage returns

    The file a signer dates and a request would reach

  5. 05 · Change / Version

    1 mode

    • JU-07

      Silent basis regression

      A recipe change moves the values, not the file.

    Stage tracks

    Model, prompt, basis rules and evidence fields

  • Sev-1 · a value posted on no basis
  • Sev-2 · wrong basis reaches the file
  • Sev-3 · degraded source, file uncertified
See the complete Food & Beverage failure-mode catalogue

Affected slices

Limited-time offers absorb the corrections

An item-level substantiation-coverage figure can read clean while limited-time offers carry most of the rework. Nestack reports the correction rate by menu item, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Limited-time offers6.8%3.6×Review
Items with supplier substitutions4.9%2.6×Review
Build-your-own combinations3.1%1.6×Watch
Core items unchanged for a year1.7%0.9×Normal

Bar: correction-rate lift vs. core-item baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

What an unsubstantiated value costs

The loop shuts when the unsubstantiated value is a regression case. That suite is what the next item posted is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Correction rate rises on limited-time offers.

  2. 02

    Diagnose

    The supplier substitution nobody told the desk that signed for the calorie count, and the board that kept posting the old number, are read back until one cause remains.

  3. 03

    Improve

    Any change goes out numbered, with the items that caused it attached.

  4. 04

    Verify

    One item case still failing is enough to hold the release back.

  5. 05

    Learn

    One case joins the suite, one line joins the substantiation file.

  6. Learn → Detect

    The return edge. The next file is measured 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, evidence assembly, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Menu-item discovery and automation-boundary work.Week 1
  2. 02Recipe, database and outlet sources.Week 1, Week 2
  3. 03Item-to-basis and outlet-adherence coverage mapping.Week 1, Week 2
  4. 04Item and basis evidence ingestion.Week 2
  5. 05Item, outlet and record binding.Week 2, Week 3
  6. 06Completeness scoring and review routing.Week 3
  7. 07Responsible-individual signing workflow.Week 3, Week 4
  8. 08Recipe and menu-system integration.Week 5
  9. 09Substantiation and outlet cases.Week 4, Week 5
  10. 10Guardrails and statement controls.Week 4
  11. 11Item-trail instrumentation.Week 5, Week 6
  12. 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 tierPilotOne chain, one menu cycleProductionProduction menu systemsAdvancedMultiple brands / menu sets
Introduced at Pilot
Evidence assembly to your items✓✓✓
Responsible-individual release✓✓✓
Menu-item inventory baseline✓✓✓
Introduced at Production
Reporting by outlet—✓✓
Signing workflow in your systems—✓✓
Approved write-back—✓✓
Recipe-system integration—✓✓
Introduced at Advanced
Multi-brand menu sets——✓
Cross-outlet evidence packs——✓
Large item registers——✓
Multi-outlet adherence controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, item register size, 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.

  1. 01Your covered establishments and the items they offer

    Item inventory mapping and evidence capture

    Week 1

  2. 02Representative recipes, database printouts and laboratory worksheets

    Record binding, basis logic and the evidence baseline

    Week 2

  3. 03Your reasonable basis under § 101.11(c)(1)

    Item mapping, basis binding and the automation boundary

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Recipe, database and outlet-source assessment, then integration setup

    Week 2

  5. 05Items you would not want requested

    Outlet cases and the evaluation run

    Week 4

  6. 06What no nutrient statement may settle

    Completeness scoring, review routing, guardrails and release controls

    Week 3

  7. 07Two responsible individuals to sign

    Signing 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

These widths measure the work and not the page, which is why a single week has to carry two phases.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Menu substantiation discovery, item mapping and the automation boundary

W2Source integration and the basis-coverage baseline

W3Evidence assembly, basis logic and release controls

W4Evaluation suite, outlet cases and failure-mode testing

W5Record integration, pilot items and targeted corrections

W6One menu cycle run under the responsible individual, then Agent Care handover

Reading the band
A band sits only on the weeks its own work is named in, and the fifth of them holds a pair.
At the end of W6
Once the substantiation file validates, Agent Care assumes the agent.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Food & Beverage AI agent

Build a menu substantiation agent around the file two individuals have to sign.

Show us one standard menu item and the recipe version its calorie count was calculated from. Send that item over and we will assemble the § 101.11(c)(3) file around it, accuracy routed to headquarters and adherence to the outlet. Pack declarations belong to a different agent.

Typical delivery
4–6 weeks
Deployment
Your menu items
Two signers
On demand
After launch
Agent Care available

Nestack Agents · Menu substantiation

AGT-FB-19 · Agent Care available after launch

Build a menu substantiation agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your menu items

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.