Skip to content

AI uses for records

Freight quotation constraint dataset: label what each quote assumed

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

A freight quotation constraint dataset should label the desk's decision, not only the price: quote as requested, quote with a written assumption, add the service, hold for one named fact, substitute equipment or service, refer to a specialist, or decline. In NMFTA's open LTL standards, 21 of the 49 codes a carrier uses to report bill changes after pickup match nothing a shipper can declare on the bill of lading.

Agent run

Freight quotation constraint

This page covers freight quotation constraint dataset: label what each quote assumed.Illustrative

Key takeaways

  • Label the quote decision, not just the rate: seven labels, tested in a fixed order that puts safety and feasibility before margin.
  • Hold when an unknown fact could change the carrier, the equipment or feasibility; when it only adds a tariff charge, quote and write the assumption down.
  • NMFTA's charge notice standard defines 49 bill change codes; 28 match bill of lading services and 21 have no code a shipper can declare.
  • The open full-truckload quote schema has no field for what a price assumes, so assumption text needs its own place in the dataset.
  • Label from the evidence at quote time, store the bill's outcome separately, and keep lost quotes with an empty outcome.

Which decision does a freight quote request force?

A quote request forces one decision before any price: are the facts on the request enough to commit a carrier, equipment and a delivery date, and if not, what should the desk do about the gap? The constraint that matters is the one that changes who can haul the freight, which truck arrives, whether delivery can happen on the day asked, or which tariff charges apply. The rate follows from that answer.

Most requests are incomplete in ordinary ways. A shipper types a destination ZIP code, a weight and a pallet count, and leaves out that the consignee is a school with no dock, that the freight is aerosol paint, or that the receiver books appointments two days ahead. The desk can quote anyway, ask first, add the service it suspects, change the equipment, pass the request to someone qualified, or say no. Several of those answers are partial or deferred, and a model trained only on quotes that went out never sees them.

The evidence for a useful label already sits in records a broker or carrier keeps: the request as received, the customer's location history, the quote with any assumption it stated, and, after pickup, the freight bill, its change notices, the delivery receipt and any dispute. This page builds the label set, counts what an open industry standard lets a shipper declare against what a carrier can bill afterwards, and turns both into a casebook and a reviewer's checklist.

Seven labels for a quote request, tested in a fixed order

Test seven labels in order and stop at the first that fits: decline, refer to a specialist, substitute equipment or service, hold for one named fact, add the service and quote, quote with a written assumption, and quote as requested. The order puts the answers that protect safety and feasibility ahead of the answers that protect margin.

The line between holding and quoting with an assumption carries most of the weight. If the unresolved fact could change who can carry the load, what equipment arrives or whether the delivery can happen, hold the quote and ask one named question. If it would only add a tariff charge, quote now and write the assumption into the quote, so a later charge is a disclosed condition rather than a surprise. A liftgate shows the boundary: where there may be no dock, the wrong truck means a failed attempt, so an unknown dock is a hold, while a residential consignee whose last three bills carried a liftgate is an add.

Each label, the evidence it needs on file when the quote goes out, and the mistake it exists to prevent:

Seven labels for a quote request, tested in a fixed order
LabelUse whenEvidence on file at quote timeWrong call it prevents
DeclineNo equipment, service or mode the desk sells can lawfully meet the request, and the customer refuses alternativesThe request, a drive-time check, the customer's stated limitsA promised delivery nobody can make
Refer to a specialistHazardous material, over-dimension permits, in-bond or cross-border freightCommodity description and identification number, dimensions, customs statusA load tendered without the papers or permits it needs
Substitute equipment or serviceThe freight or the date cannot move as requested but can on other equipment or serviceTemperature needs, drive time, linear feet, commodityProduce in a dry van; a solo driver on a team schedule
Hold for one named factAn unknown fact could change the carrier, the equipment or feasibilityGaps in the location record, site type, receiving rulesA failed delivery and a redelivery charge
Add the service and quoteThe request omits a constraint the desk's own records show with confidencePrior bills, location profile, customer notesA charge added after pickup to a silent quote
Quote with a written assumptionAn unknown fact would change only a tariff chargeThe assumption text, the tariff item, quote validityA dispute over a charge nobody disclosed
Quote as requestedFacts are complete and match the recordsRequest, location history, clean prior billsNeedless questions that slow a routine quote

What 49 freight bill change codes reveal about missing constraints

They show that the industry already expects quotes to miss things. The Preliminary Freight Charges standard from NMFTA's Digital LTL Council lets a carrier notify the paying party when a freight bill changes between pickup and invoice; NMFTA presents it as a way to flag reweighs, reclassifications and accessorial fees before final invoicing. Version 1.0.4 of its OpenAPI file, read from the council's public repository at commit b9371ac (September 24, 2026), defines 49 event codes for those changes.

Twenty-eight of the 49 are the codes a shipper can enter on the council's electronic bill of lading, version 2.1.0: appointments, liftgates, residential and limited access, inside delivery, hazardous material, protective service, morning, noon and end-of-day delivery windows, time-critical service and the rest. So every service a shipper could have declared can still arrive as a change notice, and the specification says notices fire when accessorial charges are added or removed. The other 21 have no code on the bill of lading at all. Five come from the carrier's own measurement or inspection, five from a changed plan or paperwork, two from waiting and storage, two from dock labor, four from location surcharges and one from customs; two are process codes rather than charges.

Two smaller lists sharpen the point. When a shipper declares limited access, the bill of lading offers an optional field for the site type, drawn from a list of 15 categories and 86 subtypes that runs from schools, hospitals and churches to military bases, construction sites, farms and self-storage. And of the council's 37 dispute reason codes, 16 are rate disputes; two name a quote (incorrect quote, incorrect volume quote applied) and three an accessorial charge. The grouping below is this article's; the codes and their meanings are the standard's.

What 49 freight bill change codes reveal about missing constraints
GroupCodes in the standardCountWhat the quote desk could have done
Declarable on the bill of ladingAPTD, LFTD, RES, LTDAD, IDL, HAZ, PSC, TCS and 20 more28Ask, add the service or state an assumption before pricing
Measured by the carrierRCL, RWE, CBC, DMIN, CMIN5Require dimensions and the class basis, then state them
Plan or paperwork changedRCC, CBL, ACTCHG, VOR, RDL5Confirm consignee, bill-to party and site before tender
Waiting and storageDET, STG2Check appointment and receiving rules
Dock laborLUM, UNL2Ask who unloads, and whether a lumper is used
Location surchargesSEC, HIPU, HIDL, CCS4Check the site against the carrier's surcharge rules
CustomsINBC1Refer in-bond moves to a specialist
Process codes, not chargesFINL, OTHR2Nothing; they close the notices or flag another event
A strip of 49 cells, one per freight bill change code in NMFTA's charge notice standard: 28 in blue for codes a shipper can declare on the bill of lading and 21 that it cannot, listed below by cause as carrier measured 5, plan or paperwork 5, waiting and storage 2, dock labor 2, location surcharges 4, customs 1 and process codes 2, the last in gray.
The 49 event codes in the Preliminary Freight Charges standard, version 1.0.4, counted at commit b9371ac. Grouping by this article. Source: NMFTA Digital LTL Council, 2026.

Where does a quote write down what it assumed?

In the open full-truckload quote standard, nowhere structured. The council's Rate Quote API, a public preview added to its full-truckload repository in June 2026 and read here at commit 294ded5, lets a requester send 21 known additional-service keys, among them LIFTGATE, LIMITED_ACCESS, APPOINTMENT, HAZMAT, REFRIGERATION and TEAM_DRIVER. A carrier's response can return 25, including four that are not on the request list: CROSSBORDER, CALIFORNIA_COMPLIANCE, DOCUMENTATION and MISC.

The response carries a quote type (spot or contract), validity dates that the specification says typically span 24 to 72 hours for a spot quote, a yes-or-no allInclusiveQuote flag, and a message status whose WARNING value means the request succeeded with some exception. None of the 68 field names in the schema file mentions an exclusion, an assumption, a condition or a caveat. A price that assumes dock delivery has to say so in free text or in a field a carrier invents, which is where datasets usually lose it.

The council's less-than-truckload repository, at the commit read, publishes bill of lading, charge notice, pickup and in-transit specifications but no rate quote specification, although rate quotes are on the roadmap NMFTA announced for the council. Until one exists, the quote and its assumptions have to come from each company's own system, and the assumption needs fields of its own:

  • Assumption text: the exact words the quote used, or empty when it stated none.
  • Assumed value: what was assumed, such as dock delivery, a class basis or no appointment, from a controlled list.
  • Basis: the customer said so, the location record, a prior bill, or a desk judgment.
  • Consequence if wrong: a tariff charge, an equipment change, a failed delivery or a refusal.
  • Disclosure: whether the customer saw the assumption, with the time it was sent.

When the delivery date is the constraint, not the price

A quote can be priced correctly and still be undeliverable. Under the hours-of-service rules that FMCSA summarizes for property-carrying drivers, a driver may drive at most 11 hours after 10 consecutive hours off duty and may not drive beyond the 14th hour after coming on duty. A requested delivery time that needs more driving than one driver can lawfully do is a feasibility problem, and the right label changes the service or the date before anyone discusses the rate.

Illustrative arithmetic shows how quickly that bites. At an assumed average of 50 miles an hour, a 1,000-mile move needs 20 hours of driving. A solo driver needs at least 30 hours from pickup, because a 10-hour break must fall between two driving shifts, so a 28-hour window fails; a two-driver team, which the quote standard lists as a service, fits. A 900-mile move promised within seven hours fails for any truck. Both figures are lower bounds, since rest breaks, traffic and loading time only add hours.

Feasibility also hides at the destination. The bill of lading standard requires appointment details when appointment at delivery is declared and delivery-type details when time-critical service is declared, while detention and storage sit among the 21 change codes no shipper can declare. A dataset that stores only the rate and a transit estimate cannot teach a model that an appointment-only receiver is a scheduling constraint.

How is the right label known? Reading the bill backward

Check the label against what happened after pickup: the freight bill, its change notices, the delivery receipt and any dispute, read beside the quote's own words. A charge that arrives after a silent quote is evidence of a missed constraint; the same charge after a quote that stated the assumption is a disclosed condition doing its job.

Outcome alone is not the label. A desk that quoted silently and got lucky made a worse call than one that held for a fact the bill later showed was harmless, so reviewers with freight pricing experience should label each case from the evidence on file at quote time, without seeing the bill, and store the outcome in a separate field. Quotes that never became shipments have no bill at all. They can carry a decision label but never an outcome, and a dataset that drops them learns only from the quotes customers accepted.

Five outcome classes cover most bills:

  • Billed as quoted: no change notice before the invoice; this supports the label but does not prove it.
  • Declarable service added after pickup: a missed constraint, unless the quote stated it as an assumption.
  • Carrier measurement changed the charge: a reweigh, reclassification, density or cubic minimum, meaning the dimensions or class basis were wrong or missing.
  • Plan changed or delivery failed: redelivery, reconsignment, an unused vehicle order, detention or storage, meaning feasibility or site facts were missed.
  • Disputed: a dispute coded incorrect quote or accessorial charge not applicable, to be read with the quote text before deciding who erred.

Eight quote requests and the label each one earns

The casebook applies the fixed order to eight invented requests. A short script assigns each label from the listed facts, so every verdict uses the article's seven labels and no others; the two drive-time cases use the illustrative arithmetic above.

Eight quote requests and the label each one earns
Request (illustrative)Gap or conflictLabelDeciding fact
Four pallets of shelving, no services requestedLocation record and three prior bills show a residential delivery with a liftgateAdd the service and quoteThe desk's own records confirm both services
Two pallets of lab benches for a public schoolNo dock or receiving information on fileHold for one named factWithout a dock, the wrong truck means a failed delivery
Six pallets of furniture parts, weight given, no dimensionsDensity, and so the class, is unknownQuote with a written assumptionMissing dimensions change the charge, not the truck
Dry van requested for fresh produceThe commodity needs temperature controlSubstitute equipment or serviceA refrigerated trailer, not a cheaper dry van
1,000 miles with 28 hours from pickup to deliveryA solo driver needs at least 30 hoursSubstitute equipment or serviceA team fits the window; a solo driver cannot
Pallets of aerosol paint, no hazardous flagThe description indicates hazardous materialRefer to a specialistHazardous paperwork and carrier authority come first
Three pallets dock to dock, dimensions and class givenNone; the history matchesQuote as requestedComplete facts that agree with the records
900 miles with delivery promised within seven hoursNo truck can drive 18 hours in sevenDeclineInfeasible for every service the desk sells

Hard negatives that pull a quoting model the wrong way

A useful evaluation set includes requests that look risky and are not, and routine-looking requests that hide the constraint. Each case below punishes a model that reacts to words instead of records:

  • A business park unit on a residential-sounding street, with a dock on file: adding residential delivery overcharges the customer.
  • Hazard warning cones: the word is in the product name, not the commodity's hazard class, and a referral wastes a specialist's time.
  • An email saying an appointment is preferred, for a receiver whose record says walk-ins are accepted: holding the quote costs a day for nothing.
  • A liftgate already priced on the quote: adding it again charges the customer twice.
  • A hospital with a dock and receiving hours on file: a limited access charge may still apply, but the answer is add or assume, not hold.
  • A warehouse address that turns out to be a self-storage unit: self-storage is a limited access subtype in the standard, so the routine-looking request needs the add or the hold.
  • A complete set of dimensions that implies very low density: nothing is missing, yet the class basis still belongs in the quote as a written assumption.

The records a quoting desk keeps, and the reviewer's checklist

Seven record types carry the decision and its outcome, and a dataset needs them joined by quote and shipment identifiers rather than delivered as separate tables. Before a case enters a training or evaluation set, a reviewer with freight pricing experience should confirm every item on the checklist, which can double as the acceptance test a buyer attaches to a data request. The checklist, then the records and their limits:

Checklist · 8 items

  1. The request is stored exactly as received, with its channel and timestamp, before anyone edited it.
  2. The quote text, including any assumption, is linked to the request and to the shipment if one followed.
  3. The location record carries the date it was last confirmed, so the reviewer can tell what the desk could have known.
  4. The label was assigned from evidence at quote time by a reviewer who had not seen the bill.
  5. The outcome class comes from change notices, the invoice and the delivery receipt, not from the quote status.
  6. Lost and expired quotes stay in the set with a decision label and an empty outcome.
  7. Customer names, consignee addresses, contacts and any rate fields the task does not need are removed or generalized; a residential consignee is personal data.
  8. Carrier tariff text is referenced by item rather than copied, unless the carrier's terms allow reuse.

0 of 8 ticked. Tick each item as you go.

The records a quoting desk keeps, and the reviewer's checklist
RecordUsually lives inWhat it provesWhat it cannot show
Quote requestEmail, customer portal or EDIWhat the customer said, and whenWhat the customer knew but left out
Quote with lines and validityRating engine or TMSPrice, services included, validity, any assumption textWhy the desk chose that answer, unless noted
Location and customer profileTMS or CRMDock, hours, site type, liftgate historyWhether the record was current on the quote date
Electronic bill of ladingCarrier or broker APIServices declared at tender; appointment, time-critical and hazardous detailsCharges that have no code on the bill of lading
Freight bill and change noticesCarrier billing and charge-notice feedEach change, its event code and time, and whether a weight and research document existsFinal terms: the standard says a notice is not an invoice
Delivery receiptProof of delivery or driver appWhat happened at the dock: refusal, no liftgate, wrong addressWhether the exception traced back to the quote
Dispute recordFreight audit or payables systemReason code and resolutionWhether the disputed charge was disclosed in the quote

What a broker's or carrier's records can and cannot show

A broker's or carrier's records can show what was asked, what was quoted, what the bill said afterwards and how disputes ended. They cannot show what the customer knew and left unsaid, what a carrier would have charged on a quote nobody accepted, or why a desk chose its answer unless someone wrote it down. A sector label settles none of the rights questions either: the shipper's data, a carrier's tariff and a broker's confidentiality promises to carriers each need their own review before records leave the company.

Rate fields deserve extra care. Lane prices shared among competitors raise competition questions, so a constraint dataset should keep price fields only where the task needs them. The strongest use is evaluation, testing whether a model picks the right label and asks the right question, and that depends on the constraint evidence far more than on the dollars.

The counts come from a short script that read the council's YAML and text files as data on 28 October 2026 without running any repository code. Event codes are the items of the EventCodeType list in version 1.0.4; accessorial codes are the starred lines of the Accessorial_Codes list in bill of lading version 2.1.0, and versions 2.0.0 and 2.0.1 carry the same 28, while version 1.0 has INS where later versions have FVC. A code counts as declarable when it appears in both lists; codes were compared as exact strings, and a case-insensitive comparison gave the same sets. A limited access entry with no spaced hyphen counts as a category and one with a spaced hyphen as a subtype, and the script stopped unless the shared list matched the one in the bill of lading specification exactly. A dispute reason's prefix is the text before its first spaced hyphen, and quote and accessorial mentions are whole-word matches in any case. The quote's additional-service keys are the key names listed in its request and response schemas; field names are the keys directly under each properties block, and none contains the strings exclu, assum, condition or caveat. Not measured: how often any code appears on real freight bills, which carriers have implemented the standards, and how often quotes state their assumptions in practice.

SourceX sources freight records like these when a buyer asks for them rather than keeping them in stock. A request that names the task, the label set (these seven or your own), the record types above, the modes and years needed, and an acceptance test such as agreement with adjudicated labels on a held-out casebook gives a candidate supplier something concrete to check its systems against.

Frequently asked questions

What is a freight quotation constraint dataset?

It is a set of quote requests, each joined to the quote that went out, the constraint evidence the desk had on file and what the freight bill later showed, with a label for the desk's decision. The decision label says whether to quote, assume, add a service, hold for a fact, substitute equipment, refer or decline. The outcome is stored separately so luck does not become the label.

Why not train a quoting model on booked loads alone?

Booked loads are the quotes customers accepted, so they leave out held, declined and lost requests, which are exactly the partial and deferred answers a desk has to learn. A model trained on them learns to price whatever arrives. Keep lost and expired quotes with a decision label and an empty outcome.

Which accessorials should a quote ask about first?

Ask first about the ones that change the truck or the delivery: a liftgate where a dock is uncertain, limited access sites, delivery appointments, temperature control and hazardous material. Items that only add a tariff charge can go into the quote as written assumptions. That split is this article's label rule, not an industry standard.

Does the NMFTA standard say which accessorial charges apply?

No. The Digital LTL Council's specifications define codes for declaring services on a bill of lading and for reporting charge changes, but the charges themselves come from each carrier's tariff and contract. The standards make a missed constraint visible after pickup; they do not price it.

Can AI-generated quote requests replace real ones?

They can test whether a model reads formats and fields, but the hard cases come from a conflict between a request and the desk's own records, such as a location history or a prior bill. Invented requests have no records behind them, so they cannot show whether a model checks the evidence it would have in practice.

Sources

  • NMFTA Digital LTL Council Preliminary Freight Charges OpenAPI file, version 1.0.4 (PRD December 2025), read at commit b9371ac (24 September 2026): 49 event codes for freight bill changes; notifications fire on changes that affect charges, including the addition or removal of accessorial charges; a PFC is not an invoice. Source
  • The council's electronic bill of lading specification, version 2.1.0: 28 accessorial codes; an optional limited access type when LTDAP or LTDAD is declared; appointment details required with APTD, delivery-type details required with TCS, hazardous details required with HAZ. Source
  • The council's less-than-truckload repository at commit b9371ac: shared code lists with 15 limited access categories and 86 subtypes and 37 dispute reason codes, and specifications for the bill of lading, preliminary freight charges, pickup requests and in-transit visibility, with no rate quote specification. Source
  • The council's full-truckload Rate Quote API schemas, version 1.0.0 public preview, at commit 294ded5: 21 request and 25 response additional-service keys, quote validity dates (spot typically 24-72 hours), an allInclusiveQuote flag and a WARNING status for success with some exception. Source
  • NMFTA's DSDC page presents the Preliminary Freight Charges API as alerting stakeholders to billing discrepancies such as reweighs, reclassifications or accessorial fees before final invoicing. Source
  • NMFTA announced a Digital LTL Council API roadmap that includes rate quotes alongside the bill of lading, pickup, visibility and freight charge standards. Source
  • FMCSA's summary of hours-of-service rules: property-carrying drivers may drive 11 hours after 10 consecutive hours off duty and may not drive beyond the 14th hour after coming on duty. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies