A rooftop air handler fails in the second winter after the owner moved in. The facility manager pulls up the asset in the CMMS and finds a warranty field that says the unit is covered for five years by the manufacturer, so they file a claim. The manufacturer rejects it: the five-year term covered the compressor, not the unit, the labor was never covered at all, and the general one-year contractor warranty that would have covered the install defect expired four months ago because nobody tracked the start date against the certificate of substantial completion. The owner eats a repair two different warranties should have split, and the reason is a single wrong cell in the handover dataset, a warranty attribute the closeout team copied from a template and never reconciled against the actual submittal. At handover, the owner is owed three linked things: a warranty matrix that says who covers what and until when, an O&M binder index that points to the operating and maintenance documents for every asset, and a commissioning (Cx) checklist that proves the systems were verified to perform. AI can assemble all three fast from the federated model and the submittal record. But AI invents asset attributes, the failure mode established two levels ago, so the data must be verified against the submittal record and reconciled against the commissioning agent's punch before it becomes the record the building is operated, warrantied, and litigated from. This lesson shows you how to produce the warranty and O&M COBie handover for an MEP scope and reconcile it against the Cx agent's punch, aligned to the owner's BIM Execution Plan and the Owner's Project Requirements.

The Handover Triangle: Warranty, O&M, and Cx

Three deliverables come due at the same moment, substantial completion, and they are not independent documents but three views of the same asset population. The warranty matrix states, for every installed asset, who warrants it, what it covers, and when the term starts and expires: the general one-year contractor warranty under AIA A201 Section 3.5, layered with the manufacturer's extended terms, which on MEP equipment run two, five, or ten years by component (a compressor at five years, a heat exchanger at ten, a controls board at two). The O&M binder index is the table of contents to the operations and maintenance documents: for every asset, where the operating instructions, maintenance procedures, parts lists, and manufacturer cut sheets live, so the facility manager can find them when the unit fails at 2am. The Cx checklist is the commissioning agent's verification that the systems were tested and perform to design intent: the functional performance tests, prefunctional checklists, and issues log the Cx agent works down to closure.

These three share one spine: the asset register. Every row in the warranty matrix, every entry in the O&M index, and every line in the Cx checklist points to the same physical equipment, the same VAV box, the same chiller, the same fire pump. If the asset register is wrong, all three are wrong in lockstep: a missing asset has no warranty, no O&M document, and no commissioning record; an asset with a wrong attribute carries the wrong warranty term, maintenance procedure, and Cx result. The handover is therefore not three separate assembly jobs but one asset register expressed three ways, which is why AI is good at producing it and why a single data error propagates into all three at once.

COBie Is the Container That Carries All Three

The structured format that carries this asset register is COBie, the Construction Operations Building Information Exchange, which you assembled in Level 2 as the facility manager's operating dataset: the linked worksheets of Facility, Floor, Space, Type, Component, System, Spare, Resource, and Job. All three deliverables live in or hang off it. The warranty data populates the warranty columns on the Type and Component worksheets and the linked Document and Job rows; the O&M index is the Document worksheet pointing to manuals and cut sheets; the Cx results attach to Components and Systems. This lesson extends the Level 2 COBie work into the closeout deliverables it was always meant to feed, so the verification discipline carries directly: the AI drafts the COBie fast, but a schema-valid COBie can be substantively false because the AI sourced or guessed the values.

The container enforces the shared spine. When the warranty, O&M reference, and Cx result all attach to the same Component row keyed to the same asset tag, an inconsistency surfaces structurally: an asset with a warranty but no O&M document, or a Cx result with no matching Component, is a broken link the schema check catches. Loose spreadsheets hide those inconsistencies because nothing forces them to share keys, so COBie makes the reconciliation checkable rather than a manual cross-walk no one has time for at closeout.

Generative Assembly From the Model and Submittal Record

The labor savings is the same generative-assembly move you have seen throughout the level. AI reads the federated model (the IFC asset geometry and parameters) and the submittal record (the approved equipment submittals, O&M manuals, manufacturer data sheets, and warranty certificates), then assembles the three deliverables: it populates the asset register from the model, pulls warranty terms and start triggers from the certificates into the warranty matrix, indexes the O&M documents into the binder index, and generates the Cx checklist skeleton from the systems that require functional testing. Assembling this by hand is the brutal, weeks-long closeout grind always done last under deadline pressure and therefore done badly: hundreds of assets, each with a warranty term, document set, and test requirement, cross-referenced across a model and submittal log in different systems. AI compresses that to a first-pass deliverable in a fraction of the time.

But the spine hook from Level 2 governs here without softening: AI invents asset attributes. When the AI populates a warranty term, a model number, a capacity, a maintenance interval, or a Cx pass result, it is sourcing or guessing, and that is where it fabricates a plausible-but-wrong entry: a five-year warranty where the certificate says two, a model number transposed from a similar unit, a maintenance interval invented from a generic template, a Cx result marked complete that the agent never closed. The assembly produces a complete, schema-valid handover that can be substantively false, the falseness invisible because the deliverable looks finished. Assembly is not verification, and a fast handover that embeds invented attributes is worse than a slow one, because the owner operates, warranties, and litigates from it for decades.

The handover is one asset register expressed three ways: the warranty matrix, the O&M binder index, and the Cx checklist, all carried by COBie. AI assembles all three fast from the federated model and the submittal record, which is real labor savings, but AI invents asset attributes, so every warranty term, document link, and Cx result must be verified against the submittal record and reconciled against the commissioning agent's punch before the handover becomes the record the building is operated and warrantied from.

Verify Against the Submittal Record

The submittal record is the authority for the warranty and O&M data; verification confirms the AI sourced each value from it rather than inventing one. The warranty matrix is the highest-stakes verification because a wrong cell is a missed claim or misallocated cost, exactly the rooftop-unit failure that opened this lesson. For each asset, the verifier confirms three things against the warranty certificate: the coverage (what component the term actually covers, since manufacturer warranties are layered and a five-year term often covers only the compressor, not the unit or the labor), the term (two, five, or ten years, matching the certificate, not a template default), and the start trigger (typically substantial completion or beneficial use, where being wrong by months is how a one-year contractor warranty silently expires before anyone files). The O&M index verification is lighter but real: confirm every indexed document exists and points to the right asset, since a link to the wrong cut sheet sends a technician the wrong procedure.

This is the same gate-not-mood discipline applied to closeout data: the warranty matrix sits behind a verification gate as serious as a pay application because it allocates dollars and contractual obligation, so rigor is proportioned to the stakes. The critical, high-dollar MEP assets, the chillers, air handlers, fire pumps, and switchgear, get rigorous per-asset verification; commodity assets get a lighter check. The verifier adopts the AI's term only after confirming it against the certificate, because the warranty matrix is what the owner hands a manufacturer to demand a covered repair, and a wrong term means the owner pays.

Reconcile Against the Cx Agent's Punch

The commissioning agent's punch is the second authority, the one that keeps the Cx checklist honest. The Cx agent runs the functional performance tests and prefunctional checklists and logs the deficiencies as commissioning issues (the Cx punch), which the contractor works down to closure before acceptance. The AI-generated Cx checklist describes what should have been tested and may optimistically mark items complete, but the punch records what actually failed and whether it was fixed. Reconciliation cross-checks the two: every open item on the punch must be reflected as an unresolved or conditional entry, and any asset the AI marked commissioned-complete must match a closed item on the Cx record, not an invented pass. An asset the AI shows as verified-performing while the punch shows an open functional-test failure is a false handover, the optimistic invention the spine warns about.

The asymmetry mirrors the false-negative weighting you applied to the register and the submittal review. A Cx item the AI shows as open but the agent closed is a visible, low-cost error: someone re-checks and clears it. A Cx item the AI shows as closed but the punch shows open is the dangerous error, because it hands the owner a system marked verified that was never made to perform, and the failure surfaces later as a comfort complaint, an energy overrun, or a life-safety gap on a system like smoke control or a fire pump. So reconciliation concentrates on confirming that nothing the agent flagged as open is buried under an AI-generated complete; the handover is reconciled when the two agree asset by asset.

Align to the BEP and the OPR

The handover does not float free; it is governed by two owner documents you have met earlier in the program. The BIM Execution Plan (BEP), authored under ISO 19650, specifies the information-delivery requirements: the COBie schema version, required attributes, naming conventions, level of development (LOD 500 for the as-built record), and asset classification (MasterFormat 2024 or Uniformat II per ASTM E1557) the owner's systems expect. A handover that ignores the BEP produces data the owner cannot load, the schema-valid-but-unusable failure from Level 2. The Owner's Project Requirements (OPR), which the owner's rep authors and which the basis-of-design and Cx Plan flow down from, defines what the owner needs the building to do and therefore what the commissioning verified and what the FM team must maintain. The Cx checklist exists because the OPR set performance requirements; reconciling the handover to the OPR closes the loop the owner opened.

Aligning to both makes the handover usable rather than merely complete. The BEP tells you the shape the data must take to load into the owner's CMMS and asset systems; the OPR tells you which assets and results actually matter (the critical clinical systems on a healthcare project, the cooling and power on a data center, the life-safety systems everywhere), so verification concentrates where the OPR placed the stakes. As the Level 2 COBie work closed on, the handover is a relationship to the receiving organization, not a file. The handover that aligns to both is the one the owner can load on day one and operate from for decades; the one that ignores them fails its purpose as surely as one full of invented attributes.

The Applied Problem: Produce and Reconcile the MEP Handover

Take the federated model and the submittal record for the MEP equipment (chillers, air handlers, pumps, fans, VAV boxes, switchgear, and controls), use AI to assemble the COBie asset register and generate the three deliverables, then run the verification: confirm each warranty term, coverage, and start trigger against the warranty certificates; confirm each O&M document link resolves to the right asset; and reconcile the Cx checklist against the commissioning agent's punch so every open item is reflected and every commissioned-complete claim traces to a closed punch item. Align the whole deliverable to the owner's BEP (schema, attributes, naming, classification) and the OPR (which assets and results matter).

Produce two things. First, the verified handover package: the COBie-carried warranty matrix, O&M binder index, and Cx checklist, corrected against the submittal record and reconciled against the Cx punch, in the shape the BEP requires so it loads. Second, the reconciliation record: the warranty attributes you corrected against the certificates (wrong terms, misstated coverage, wrong start triggers, invented intervals), the O&M links you repaired, and the Cx discrepancies you found, especially any asset the AI marked complete that the punch showed open. Concentrate verification on the critical, high-dollar MEP assets and the OPR-flagged performance systems, where a wrong warranty term, missing maintenance procedure, or false Cx pass has real operational, financial, and life-safety consequence over the building's life.

This is the warranty and FM-handover core of the BIM chapter, and it integrates the level's spine: generative assembly is real labor savings, but AI invents attributes, so verification against the authoritative source and reconciliation against the independent verifier are non-negotiable, and the deliverable must conform to the owner documents that govern it. The professional who masters this hands the owner a handover produced in a fraction of the old closeout time but true where decades of operations depend on it.

Key Takeaways

• The handover is three linked deliverables sharing one asset register: the warranty matrix (who covers what, and until when), the O&M binder index (where the operating and maintenance documents live), and the Cx checklist (the commissioning agent's verification that systems perform). One asset error propagates into all three at once.
• COBie is the container that carries all three, extending the Level 2 COBie lesson: the warranty data, O&M document links, and Cx results all attach to the same Component rows, so the schema enforces the shared spine and surfaces broken links that loose spreadsheets would hide.
• AI assembles the asset register and all three deliverables fast from the federated model and submittal record, real labor savings on work the industry chronically under-resources under deadline pressure.
• But AI invents asset attributes (the Level 2 spine hook): a fabricated warranty term, a transposed model number, an invented maintenance interval, an optimistic Cx-complete, producing a schema-valid handover that can be substantively false, the falseness invisible because the deliverable looks finished.
• Verify against the submittal record, concentrating on the warranty matrix: confirm coverage (warranties are layered, often five years on the compressor only), term (two, five, or ten years matching the certificate), and start trigger (typically substantial completion, where being wrong by months silently expires the one-year contractor warranty under AIA A201 Section 3.5).
• Reconcile against the Cx agent's punch: every open commissioning item must be reflected, and every commissioned-complete claim must trace to a closed punch item. The dangerous error is an AI-marked complete that the punch shows open, falsely certifying a system (including life-safety systems like smoke control and fire pumps).
• Align to the owner's BEP (under ISO 19650: COBie schema, required attributes, naming, LOD 500 as-built, MasterFormat 2024 or Uniformat II classification) so the data loads, and to the OPR (which assets and results matter) so verification concentrates on the stakes.
• The artifact: AI-assemble the warranty matrix, O&M binder index, and Cx checklist for an MEP scope in COBie, verify the attributes against the submittal record, reconcile the Cx checklist against the commissioning agent's punch, align to the BEP and OPR, and document the reconciliation record, because a schema-valid handover that embeds invented attributes or a false Cx pass is worse than none when the owner operates from it for decades.