DCE document consistency check with ChatGPT: method and checks
DCE consistency check with ChatGPT: checking cited standards (NF P03-001, CCAG-Travaux), five common discrepancy types and the method.
By Educasium
Install the ChatGPT skill: DCE document consistency check →
An architect is about to launch the tender process for renovating an office floor. Before sending out the file, they ask ChatGPT to cross-check the documents: drawing P02 revision B shows a 90 cm door where the CCTP, article 4.2, specifies 83 cm. A little further on, the CCAP (particular administrative conditions) lists the NF P03-001 standard among the contract documents, but the CCTP (particular technical specifications) cites no technical fascicule at all to govern execution of the relevant trade. Two discrepancies of a different nature, both invisible on a quick read-through, and both expensive to discover once the contract is signed.
This article details what a DCE (tender document set) is and why its documents must be consistent with one another, the check most often overlooked — the one covering standards and reference codes actually cited as contract documents —, the method for checking DCE consistency with ChatGPT without ever settling a discrepancy on the professional's behalf, and the situations where this check is not enough.
Summary
- What a DCE is, and why its documents must be consistent with one another
- The most overlooked check: cited standards, not just dimensions
- Five recurring types of discrepancy
- Checking DCE consistency with ChatGPT, step by step
- A discrepancy is never resolved by ChatGPT on your behalf
- When this check is not enough
- Training to make DCE files more reliable
- Frequently asked questions
What a DCE is, and why its documents must be consistent with one another
The dossier de consultation des entreprises (DCE) brings together the documents that define a works contract before signature: drawings, the CCTP (particular technical specifications), the DPGF (lump-sum price breakdown), the CCAP (particular administrative conditions) and, for a public contract, the consultation rules. These documents are drafted at different times, often by different people — architect, engineering consultant, quantity surveyor — which explains why a dimension changes between two versions with no one explicitly flagging it.
Why a DCE inconsistency is more costly after signature than before
A contradiction discovered during the tender process is fixed with a simple addendum or a written answer to bidders. The same contradiction discovered after the contract is signed opens a dispute over which document prevails, with the risk of extra work being invoiced or disagreement over what was actually ordered. Consistency checking therefore carries more value before the file is sent than during the site works.
The most overlooked check: cited standards, not just dimensions
A DCE consistency check almost always focuses on dimensions and quantities; it often forgets to check whether the standards and reference codes mentioned in one document also appear, or conversely are missing, in the other documents of the same file — even though that check directly changes what the contractor actually owes.
A standard or reference code never applies on its own
For a private contract, the NF P03-001 standard, the general administrative conditions for French private building contracts, only produces contractual effect if the contract expressly cites it among its documents. For a public contract, the principle is identical: the CCAG-Travaux, approved by the order of 30 March 2021, only applies if the contract expressly refers to it, and the CCAP may in addition depart from some of its clauses. A consistent DCE must cite the same normative reference in both the CCAP and the CCTP, or explicitly justify why they differ — never leave a reference code mentioned on one side and missing on the other without flagging it as a discrepancy.
| Document citing the reference | Document silent on the reference | Consequence if the gap is not corrected |
|---|---|---|
| CCAP cites NF P03-001 (private contract) | CCTP mentions no technical specification code | Ambiguity over which execution rules actually apply to the trade |
| CCAP cites the CCAG-Travaux (public contract) | Consultation rules stay silent on any departures | Risk of dispute over which clauses are actually in force |
| CCTP refers to a DTU for a given work item | DPGF has no line for the services that DTU requires | Service contractually owed but not priced in the bid |
What the Civil Code says about an uncited clause
The general principle is that a contract document's silence on a rule never amounts to tacit adherence to that rule, unless the law, usage, business relations or an express clause provide otherwise, as set out in Article 1120 of the French Civil Code. In other words, a contractor cannot be held to a standard the file does not explicitly cite among its contract documents, even if that standard is common practice in the profession.
Five recurring types of discrepancy
Beyond normative references, a consistency check systematically cross-checks five categories of documents against each other.
1. Dimension mismatch between drawing and CCTP. A dimension differs from one document to another for the same work item, such as a door width or a guardrail height.
2. Overlap or gap between two trades. A service appears in the CCTP of two different trades, or in neither, exposing the project to double billing or an omission.
3. Normative reference missing from one document while cited in another. The case detailed above, often invisible because it involves no quantity or dimension.
4. Inconsistent revision or version date between documents. The CCTP cites drawing revision A while the file sent to bidders contains revision B, with the change between the two versions never carried over into the text.
5. CCTP item with no matching DPGF line, or the reverse. A technically specified service is priced nowhere, or a price line corresponds to no identifiable requirement.
Checking DCE consistency with ChatGPT, step by step
Step 1: Inventory every document in the file with its revision and date. Without this prior inventory, ChatGPT cannot distinguish a current document from an outdated version mistakenly kept in the file.
Step 2: Import the documents into a ChatGPT project, explicitly identifying each one's role. State which document is the CCAP, which is the CCTP for each trade, and which is the DPGF, rather than letting ChatGPT infer it from the filename alone.
Step 3: Request a systematic cross-check trade by trade, work item by work item. The pack produces an Excel discrepancy register, each row carrying the two precise references in disagreement, never a single document judged on its own.
Step 4: Specifically check normative references cited in the CCAP against those missing or different in the CCTP. This often-overlooked check should appear as its own discrepancy category in the register produced.
Step 5: Settle every confirmed discrepancy with the design team, never by automatically picking the dimension that seems most likely. The correction itself remains a decision for the architect or the engineering consultant, not for the consistency check.
Fictional example for testing the method: drawing P02 revision B shows a 90 cm door, CCTP article 4.2 specifies 83 cm. The correct request to ChatGPT is to record a referenced discrepancy with both precise references, without automatically picking the correct dimension or modifying any source file.
A discrepancy is never resolved by ChatGPT on your behalf
The register produced by this pack lists differences, not corrections. For every row, ChatGPT must supply two precise references, a possible consequence if the gap persists, and a proposed correction to be validated — never a correction already applied silently to a modified file. The check finding no discrepancy does not prove the documents are consistent either: it only means the requested cross-check revealed nothing within the scope of the documents actually imported, which depends directly on how complete and accurate those documents are.
When this check is not enough
A well-built discrepancy register does not resolve every situation encountered when assembling a DCE.
A disagreement between designers over which document should prevail — the engineering consultant defending a technical dimension, the architect defending the drawing's intent — is not resolved with one more table: it calls for a design decision before the file is sent out. A public-contract DCE subject to formal publicity and competition rules requires a compliance check against the French public procurement code that goes beyond what a documentary consistency check can cover, particularly regarding the principle of equal treatment among bidders set out in Article L3 of that code. Finally, a discrepancy affecting the contract's overall price — a missing item that could significantly shift the total amount — should be reported to the quantity surveyor before any file is sent out, rather than silently fixed by adjusting a single register row.
Training to make DCE files more reliable
A well-structured discrepancy register is not enough on its own: it also needs adapting to every file, with documents that change in nature from one project to the next. For an architect practising as a self-employed professional (NAF code 7111Z in France), the FIFPL fund covers part of the cost of Qualiopi-certified training under 2026 criteria set at €300 per day and €900 per year, with e-learning capped at 50% of the daily rate — funding that must be arranged before training starts, not after.
Frequently asked questions
Can the consistency check decide which DCE document prevails over another?
No: ChatGPT flags a discrepancy with its two precise references, but the decision on which document prevails, or on the correction to make, remains with the design team. A check that automatically picked a dimension would amount to changing the project with no human validation. Until a decision has been communicated, the discrepancy must stay displayed as unresolved in the register, and never be removed just because one of the two values looks more likely.
Does the NF P03-001 standard apply automatically if the CCTP does not mention it?
No: a voluntary standard such as NF P03-001, or a reference code such as the CCAG-Travaux for a public contract, only produces contractual effect if the contract expressly cites it among its documents. A CCAP that cites it while the CCTP does not echo it is an inconsistency to correct, never an automatic application to assume. The check must then flag that missing cross-citation as a discrepancy in its own right, distinct from a simple dimension mismatch, and pass it to the design team before the file is sent out.
Does finding no discrepancy guarantee the DCE is complete?
No: finding no flagged discrepancy only means the requested cross-check revealed nothing within the scope of the documents actually imported. A document forgotten during import, or a trade not covered by the request, can hide a very real inconsistency the register will never see. Before treating the check as conclusive, it therefore remains necessary to explicitly verify that the inventory of imported documents actually covers the full DCE sent to contractors.
What should be done if a discrepancy affects the contract's overall price?
A discrepancy that could significantly change the total amount — a missing DPGF line for a service specified in the CCTP, for example — should be passed to the quantity surveyor before the file is sent out, and never resolved silently by adjusting a single register row. The register must keep a record of that discrepancy and of its referral, rather than making it disappear once flagged. A correction made to the DPGF or the CCTP following that decision should then be rechecked in a new consistency pass, to make sure it does not introduce a fresh inconsistency elsewhere in the file.
A consistent DCE rests on three elements that do not change from one project to the next: a precise inventory of documents with their revision and date, an explicit check of the standards and reference codes actually cited as contract documents, and a human decision on every confirmed discrepancy, never an automated one.
Once the DCE is secured, our ChatGPT method for the works CCAP details how to adapt the administrative clauses to that same file, and our ChatGPT method for tender comparison revisits these documents when comparing the contractors consulted.
Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To structure your tender files with AI as part of our AI for architectural management training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal. This pack is also available on its dedicated ChatGPT skill page.