Every month the contractor submits an application for payment, and on most projects that application is an AIA G702 application and certificate for payment with its G703 continuation sheet, the line-by-line breakdown of the schedule of values showing how much of each line is complete and how much is therefore due. And every month, on a contested project, the owner or the owner's representative disputes it: this line is not as complete as you claim, that stored material is not properly documented, this percent-complete is ahead of the actual installed work. The dispute is line item by line item, and resolving it is tedious, evidentiary, and high stakes, because the number at the bottom of the G702 is real money the contractor needs for cash flow and the owner does not want to pay ahead of value received. AI can reconcile the disputed pay application line by line, comparing the claimed percent-complete against the justified percent-complete for each G703 line, drawing on the schedule of values and the verified record of completed work, and surfacing the lines where the claim and the evidence diverge. But the percent-complete is a metric, a signal that must be verified against the actual installed work, and the number at the bottom is behind the dollars gate, so the human owns the reconciliation and the negotiation position. This lesson designs the pay-app reconciliation workflow, where AI surfaces the line-item divergences and the human verifies each against the work actually in place.

The G702 and G703: Where the Dispute Lives

The AIA G702 application and certificate for payment is the summary document: it carries the contract sum, the total completed and stored to date, the retainage, the previous payments, and the current payment due, the single number the owner certifies and pays. But the substance of the application, and the substance of any dispute, lives in the G703 continuation sheet, the line-by-line schedule of values that lists each item of work with its scheduled value, the work completed in prior periods, the work completed this period, the materials presently stored, and the resulting percentage of completion and balance to finish. The G702 number is just the sum of the G703 lines, so to dispute the G702 is to dispute one or more G703 lines, and to resolve the dispute is to reconcile those lines.

This structure is what makes the pay-app dispute tractable for line-by-line reconciliation: the dispute is not a single global disagreement about the total but a set of specific disagreements about individual lines, each with its own scheduled value, its own claimed percent-complete, and its own evidentiary basis. A dispute about a four-hundred-line G703 might involve a handful of contested lines and a great many uncontested ones, so the reconciliation is the work of identifying which lines are contested, why, and what the evidence supports for each, line by line. The schedule of values itself is the framework: it was agreed at the start of the project, allocating the contract sum across the lines, so the reconciliation works within that agreed framework, disputing not the scheduled values (those are fixed) but the percent-complete claimed against each value this period. This makes the G703 the natural object of the AI's analysis: a structured, line-item document with claimed values to be checked against an evidentiary basis, which is exactly the kind of structured reconciliation AI does well, surfacing the lines where the claim and the evidence diverge for the human to resolve.

How AI Reconciles the Pay Application Line by Line

AI reconciles the disputed pay application by working the G703 line by line, comparing for each line the claimed percent-complete (what the contractor's application asserts) against the justified percent-complete (what the evidence of completed work supports), and surfacing the lines where the two diverge, which are the lines the dispute turns on. For each line, the AI draws on the schedule of values (the line's scheduled value and prior completion) and the verified record of completed work (the installed quantities, the progress documentation, the stored-material records) to compute or estimate what percent-complete the evidence justifies, then compares that against the claimed percent-complete, flagging the lines where the claim exceeds the justification (the contractor claiming ahead of the evidence) or, less commonly but importantly, where the justification exceeds the claim (the contractor under-claiming, leaving money on the table).

The value is that the AI does this across the whole G703 fast, surfacing the contested lines and quantifying the divergence on each, which manually is a tedious line-by-line comparison of the claim against the records that consumes the days a monthly pay cycle does not have. The AI's reconciliation gives the human a structured starting point: here are the lines where the claim and the evidence diverge, here is the size of each divergence, here is the evidence the divergence rests on, so the human can focus the resolution on the contested lines rather than re-reviewing the whole application. But the AI's reconciliation is only as good as the completed-work data it draws on, and the percent-complete it computes is a metric, an estimate of completion from the available evidence, not a measured certainty, so the AI surfaces the divergences and the human verifies each against the actual installed work, which is where the reconciliation's authority comes from. The AI accelerates the line-by-line comparison and surfaces the contested lines; the human verifies the percent-complete against the work in place and owns the resolution.

AI reconciles the disputed pay application line by line, comparing the claimed percent-complete against the justified percent-complete for each G703 line and surfacing the divergences. But the percent-complete is a metric, a signal that must be verified against the actual installed work, and the number at the bottom of the G702 is behind the dollars gate, so the human verifies each contested line against the work in place and owns the negotiation position.

Percent-Complete Is a Metric to Verify Against Installed Work

The percent-complete on each G703 line is a metric, a number representing how much of that line's work is done, and like all metrics in this program it is a signal to interpret, not a verdict to accept, because the number is an abstraction of a physical reality (the actual work installed) and the abstraction can diverge from the reality. A line claimed at seventy percent complete asserts that seventy percent of that scope is in place, but whether that is true depends on the actual installed work, which the percent-complete number summarizes but does not itself prove, so the number must be verified against the work to know whether the claim is justified. This is the metric-as-signal discipline applied to percent-complete: the AI's computed justified percent-complete and the contractor's claimed percent-complete are both signals about the line's actual completion, and resolving the dispute means verifying which signal the actual installed work supports.

This is where the connection to computer-vision progress tracking enters, because the actual installed work is exactly what CV-based progress monitoring captures: the photographic and reality-capture record of what is physically in place, against which the claimed percent-complete can be checked. A line claimed at seventy percent can be checked against the progress record of what is actually installed for that scope, so the CV-progress link provides the evidentiary basis for verifying the percent-complete, turning the abstract metric into a claim checkable against the captured reality. The discipline is that the percent-complete is verified against the installed work, with the progress record (CV-based or otherwise) as the evidence, so the reconciliation is not a comparison of two numbers (claimed versus AI-computed) but a verification of the numbers against the physical work, which is what gives the resolution its authority. The percent-complete is a metric to be verified against the actual installed work, the metric-as-signal discipline made concrete by the progress record that captures what is physically in place, so the human resolves each contested line by checking the claim against the work, not by accepting either the contractor's claim or the AI's computation as a verdict.

The Number at the Bottom Is Behind the Dollars Gate

The G702's bottom-line number, the current payment due, is real money: the contractor needs it for cash flow to pay subcontractors, suppliers, and labor, and the owner is paying it out against work received, so the number is consequential, which puts the pay-app reconciliation behind the dollars gate. An over-certification, the owner paying for more than the value in place, exposes the owner to risk if the contractor later defaults with work paid for but not done, which is why owners and their lenders scrutinize the application; an under-certification, the owner paying less than the value in place, starves the contractor's cash flow and can itself become a dispute or a claim. So the number must reflect the actual value in place, neither ahead nor behind, which is the dollars-gate requirement: the consequential number must be verified before it is certified and paid.

This means the pay-app reconciliation cannot be left to the AI's computation, because the AI's justified percent-complete is an estimate from the available data, and the data may be incomplete or the AI's computation imperfect, so the consequential number requires the human's verification before it becomes the basis of a payment or a negotiation position. The human owns the reconciliation as the dollars-gate-responsible party, whether that human is the contractor's project manager defending the application or the owner's representative reviewing it, because the number each is responsible for, the contractor's claim or the owner's certification, is a consequential number behind the dollars gate. The AI's reconciliation accelerates and structures the analysis, surfacing the contested lines and quantifying the divergences, but the human verifies the contested lines against the actual work and owns the resulting number, the claim or the certification, as their responsible position. The dollars gate applies because the number is real money, so the human verifies the consequential number before it is certified or negotiated, the responsible-charge discipline applied to the pay-app reconciliation, ensuring the AI's accelerated analysis is verified against the work before it becomes a payment or a position.

Data Quality Drives the Reconciliation

The AI's pay-app reconciliation rests on two data sources, the schedule of values and the verified record of completed work, and the quality of the reconciliation is bounded by the quality of these data, so data quality drives the analysis. The schedule of values is usually sound, an agreed document, but its allocation can affect the reconciliation: a front-loaded schedule of values, where early lines are valued high to improve early cash flow, distorts the percent-complete economics, and the reconciliation must work within whatever allocation was agreed, so the human's understanding of the schedule of values' structure informs the interpretation of the line-item divergences. The completed-work record is the more variable data: if the progress documentation is current, detailed, and accurate, the AI's justified percent-complete is well grounded, but if the progress record is sparse, outdated, or inconsistent, the AI's computation rests on weak evidence and the justified percent-complete is unreliable, so the reconciliation's authority depends on the completed-work data being good.

This means the pay-app reconciliation is not just an AI-analysis problem but a data-quality discipline: the value of the AI's line-by-line reconciliation depends on the completed-work record being current and accurate, which depends on the progress-tracking process that feeds it, the field reports, the reality-capture, the installed-quantity tracking. A project with rigorous progress tracking gives the AI good evidence to compute the justified percent-complete, so the reconciliation is well grounded and the contested lines are clearly identified; a project with poor progress tracking gives the AI weak evidence, so the reconciliation is uncertain and the human has less to verify against. So the reconciliation's quality is a function of the completed-work data's quality, which makes the progress-tracking discipline a precondition for a strong reconciliation, and the human must interpret the AI's reconciliation aware of the data it rests on, weighting the divergences by the strength of the evidence behind them. The data quality drives the reconciliation, so the pay-app analysis is a reconciliation-plus-data-quality discipline, with the completed-work record's currency and accuracy determining how well grounded the AI's justified percent-complete is, and the human interpreting the divergences in light of the evidence's strength, which is the structured-data and metric-as-signal disciplines combined, applied to the consequential pay-app number.

The Applied Problem: Design the Pay-App Reconciliation Workflow

Here is the exercise. Design the pay-app reconciliation workflow: specify the AI's line-by-line reconciliation (comparing the claimed percent-complete against the justified percent-complete for each G703 line, drawing on the schedule of values and the completed-work record, surfacing the divergences), the metric-as-signal verification of the percent-complete against the actual installed work (the CV-progress link), the dollars-gate ownership of the consequential number, and the data-quality discipline the reconciliation depends on. Produce the workflow design that reconciles a disputed pay application line by line while the human verifies the contested lines against the work in place and owns the number.

Produce two things. First, the pay-app reconciliation workflow design: the AI's line-by-line reconciliation and the human's verification and ownership, in the form that would let a contractor defend or an owner review a pay application with the contested lines surfaced and verified against the work. Second, the signal-and-data analysis: why the percent-complete is a metric to be verified against the actual installed work (not a number to accept), why the number at the bottom is behind the dollars gate so the human owns it, and why the reconciliation's quality depends on the completed-work data's quality, with the reasoning. Pay particular attention to the verification of the percent-complete against the installed work, the CV-progress link, because that is where the reconciliation gets its authority, turning the abstract claimed percent-complete into a claim checkable against the captured physical reality, which is what resolves the dispute on the evidence rather than on assertion.

The deliverable is the pay-app reconciliation workflow design and the signal-and-data analysis, and the lasting product is a designed reconciliation workflow that uses AI to surface the contested G703 lines and quantify the divergences while the human verifies each against the actual installed work and owns the consequential number, on good completed-work data. This is the second step of the change-management and project-controls chapter, where the metric-as-signal discipline meets the dollars gate on the most directly financial of the project's monthly documents, and it applies the percent-complete-as-signal verification, the CV-progress evidentiary link, the dollars-gate ownership, and the data-quality discipline to the pay-app dispute. The professional who masters this reconciles the disputed pay application on the evidence, with AI surfacing the contested lines and the human verifying each against the work in place and owning the number, which is the only way an AI-accelerated reconciliation can become a defensible claim or a sound certification, because the number is real money and the human's verification of the percent-complete against the work cannot be replaced by the AI's computation, however fast and well structured, since the authority of the reconciliation comes from the verification against the work, not from the computation itself.

Key Takeaways

• The AIA G702 application and certificate for payment carries the bottom-line number, but the substance of any dispute lives in the G703 continuation sheet, the line-by-line schedule of values, so to dispute the G702 is to dispute G703 lines and to resolve it is to reconcile those lines.
• AI reconciles the disputed pay application line by line, comparing the claimed percent-complete (the application's assertion) against the justified percent-complete (what the completed-work evidence supports) for each G703 line, surfacing the lines where the claim and the evidence diverge, which are the lines the dispute turns on.
• The percent-complete is a metric, a signal to interpret, not a verdict to accept: it is an abstraction of the actual installed work that can diverge from the reality, so it must be verified against the work, the metric-as-signal discipline applied to percent-complete.
• The verification connects to computer-vision progress tracking: the actual installed work is what CV-based progress monitoring captures, so the claimed percent-complete can be checked against the progress record of what is physically in place, which gives the reconciliation its evidentiary authority.
• The number at the bottom of the G702 is real money (the contractor's cash flow, the owner's payment against value received), so it is behind the dollars gate: an over-certification exposes the owner, an under-certification starves the contractor, so the human verifies the consequential number before it is certified or negotiated.
• The human owns the reconciliation as the dollars-gate-responsible party (the contractor's PM defending the claim or the owner's rep reviewing it), verifying the contested lines against the actual work and owning the resulting number, because the AI's justified percent-complete is an estimate, not a verified certainty.
• Data quality drives the reconciliation: the AI's justified percent-complete is only as good as the completed-work record (the progress documentation, the reality-capture, the installed-quantity tracking), so a project with rigorous progress tracking gets a well-grounded reconciliation and one with poor tracking gets an uncertain one.
• The artifact: design the pay-app reconciliation workflow (AI line-by-line reconciliation, metric-as-signal verification against installed work via the CV-progress link, dollars-gate ownership, data-quality discipline), and analyze why the percent-complete is a signal to verify, why the number is behind the dollars gate, and why the reconciliation depends on the completed-work data's quality.