Use case · Architecture and engineering

Approved buildings show what passed. Corrections show why.

Models trained on permitted buildings learn from approvals. The reviewer’s corrections, the prototype re-checked city by city and the calcs behind each beam are the part that is missing. Lasker licenses them from the firms that hold them, mapped into your schema.

For teams whose models produce permit-ready drawing sets, not concept sketches.

What reviewers rejectedNegative examples, with the reason
One prototype, many codesRe-checked city by city
Calcs behind the beamStructural and MEP packages
Rights on recordCopyright and owner release
// Where models break

Where each stage needs a signal approvals do not give.

Design models go from layout to systems to a buildable set, then check it against local code. Each stage has its own corrections, and they sit in firm archives.

Layout: site and zoning

Setbacks, parking counts and accessible routes change by city. Site-plan corrections show where a layout broke a local rule.

Site plan: add one accessible van stall.

Layout: egress and occupancy

Travel distance, exit count and corridor width follow occupant load. A tenant improvement that changes the use can break all of them.

Corridor C2: show 44 in clear width.

Systems: load path

Beam and column sizes the calcs support, and the corrections when a new opening or rooftop unit changed the load.

Provide calcs for beam B4 at new opening.

Systems: MEP fit

Ducts, plumbing and conduit have to fit the structure and the ceiling. Each discipline looks right on its own sheet.

RFI: duct conflicts with beam at grid C/4.

Building: a set a contractor can build from

References, schedules and details that agree across sheets, so the contractor does not have to ask.

Detail 5/A-501 referenced, not found.

Code check: per jurisdiction

Each city adopts its own edition and amendments. A code check learns them from that jurisdiction’s corrections.

Note the adopted code edition on the cover sheet.

An approval tells a model the set passed. It does not say what the reviewer made the architect change first.

Multi-site rollouts are re-checked for every location. Firms that adapt prototypes hold the same design with a dozen sets of corrections.

Under the standard AIA agreement the architect owns the drawings, so the rights have to come from the firm, project by project.

// What a production model needs

Six needs, and where the data comes from.

A model that issues drawings someone will build from, and signs for, needs more than examples of approved sets.

Know what plan check rejects

A verifier trained on real corrections, so the model catches what a reviewer would before the set goes out.

Source: plan-check pairs from firm archives

Adapt one prototype to many cities

The same store or restaurant prototype permitted in different jurisdictions, with each one’s corrections. The design stays fixed and the code changes.

Source: prototype rollout sets from site-adapt firms

Go down to the beam and the duct

Layouts are the easy part. Member sizes, load paths and duct routing have to agree with the calculations.

Source: structural and MEP packages tied to sheets

Cover more than houses

Retail, tenant improvements, offices, data centers, warehouses and manufacturing each have their own codes, systems and failure modes.

Source: completed projects from commercial firms

Price what it draws

A cost forecast is only as good as the estimate tied to the exact set it priced.

Source: estimates and bid tabulations

Learn where approved sets still cost money

RFIs and change orders show what passed review and still failed on site.

Source: construction-phase records

// What we deliver

Seven datasets, mapped to the stages.

Each record ties back to the sheets and, where the firm has one, the model element it changed. We map records into your schema in the pilot, not ours.

DatasetWhat is in a recordUsed forFormat
Plan-check pairsReviewer comment, code basis, the sheet before and after, and how it was resolvedCode check, verifierJSONL + sheet crops
Prototype rollout setsOne prototype across several jurisdictions: each site’s set, code basis and correctionsMulti-site adaptationJSONL + sets per site
Engineering packagesStructural and MEP calculations tied to the sheets they supportSystems: beam sizing, load auditsPDF + extracted values
Coordination recordsClash or RFI, the disciplines involved and the design change that closed itSystems: MEP fitJSONL + linked sheets
Native design setsRevit or IFC model with the matching construction set and schedulesMapping into your schemaIFC, RVT, DWG, PDF
Cost recordsEstimate or bid tabulation tied to the exact drawing set it pricedCost forecastsCSV + linked set
Expert reviewLicensed architects and engineers score generated sets against the adopted codeEvals before a set is stampedScores + rationale, JSONL
// Sample record

What records look like.

A plan-check correction tied to the sheets and model element it changed, and a prototype permitted in two jurisdictions with different corrections.

  • categoryFailure mode, so you can sample by what your model gets wrong
  • code_basisAdopted code edition for that permit, not the latest one
  • resolutionWhat the firm changed to clear the comment
  • prototype.idLinks every site of one prototype, so the code is the variable
  • rightsCopyright holder, owner release and license scope, per record

Illustrative. Field names are our draft schema, and we adapt it to your pipeline in the pilot.

pc-0184.jsonIllustrative
{
  "record_id": "pc-0184",
  "jurisdiction": {
    "state": "CA",
    "code_basis": "2022 CBC",
    "permit_year": 2024
  },
  "building": {
    "occupancy": ["B", "M"],
    "type": "retail prototype",
    "area_ft2": 8400
  },
  "discipline": "architectural",
  "sheet": { "id": "A-201", "rev": "1 -> 2" },
  "category": "egress.corridor_width",
  "comment": "Corridor C2: show 44 in clear width.",
  "resolution": "Wall moved 6 in, width dimensioned.",
  "artifacts": [
    "A-201_r1_crop.png",
    "A-201_r2_crop.png",
    "model.ifc#wall-2231"
  ],
  "rights": {
    "copyright": "firm-14",
    "owner_release": true,
    "personal_data_removed": true,
    "license": "non-exclusive, training + eval"
  }
}
pr-0031.jsonIllustrative
{
  "record_id": "pr-0031",
  "prototype": {
    "id": "proto-qsr-a",
    "type": "restaurant prototype",
    "area_ft2": 2850
  },
  "sites": [
    {
      "state": "CA",
      "code_basis": "2022 CBC",
      "corrections": [
        "11B: accessible stall slope at parking",
        "Title 24: lighting control schedule"
      ]
    },
    {
      "state": "TX",
      "code_basis": "2021 IBC + local amendments",
      "corrections": [
        "Grease interceptor sizing"
      ]
    }
  ]
}
// Expert review

How reviewers score a generated set.

The same questions a plan reviewer and a general contractor would ask, written down so two reviewers can agree.

DimensionWhat the reviewer checksEvidence per finding
Code complianceEgress, accessibility, fire separation against the adopted code editionCode section cited
CoordinationStructure, MEP and architecture agree across sheetsSheet and grid location
CompletenessReferences, schedules and details a contractor would needMissing items listed
ConstructabilityCould it be built without an RFIReviewer rationale
ApprovabilityWould it clear plan check in the stated jurisdictionPass, corrections or reject

Reviewers are licensed in the state whose code they score. Items with disagreement go to a second reviewer and are adjudicated.

// Pilot

A pilot on one building type.

Pick the stage and building type where your model is weakest. The pilot answers whether this data moves it.

Scope

One building type, one jurisdiction and the stage to improve: layout, systems or the code check.

Week one

Sample

Prototype rollouts and plan-check pairs for that type, mapped into your schema.

Your team audits it

Decide

Full set, ongoing feed or stop. Price is set per dataset after the sample.

No commitment before

We are scoping this vertical with firms now. No architecture dataset has been delivered yet, and we will not quote volumes until firms have signed.
// FAQ

Questions labs ask.

Data handling and our commitments are in the Trust center.

Isn’t approval already the signal?

It says a set passed. It does not show what failed on the way, and those failures, with the reviewer’s reason, are what a code check or verifier learns from.

Where does the data come from?

From the architecture and engineering firms that prepared the drawings and own the copyright, under a non-exclusive license. Each project also needs a written release from the building owner before anything is delivered.

Is it cleared for training?

Each record carries a rights record: copyright holder, owner release, a confidentiality check and the license scope, training, evaluation or both. We do not use scraped permit files.

Which jurisdictions and building types?

Tell us yours and they shape sourcing. Code basis is recorded per project, because the adopted edition and local amendments change the right answer.

Can you deliver native BIM?

Where firms hold it. Older projects are often PDF only, so we report the format mix before you commit to a full set.

Who are the reviewers?

Licensed architects and professional engineers, credential-checked and licensed in the state whose code they score. We are recruiting the bench now.

Can models trained on this data ship in a commercial product?

The license scope is written per dataset: training, evaluation or both, and whether a model trained on it can be used commercially. We negotiate with the firm for commercial use from the start, because that is what design teams need.

Can you deliver in our schema?

Yes. We map records into your schema during the pilot, and keep the source sheets linked so your team can check every field.

How is it priced?

Per dataset, after the sample. We do not publish ranges until we have closed deals to base them on.

Hold a drawing archive? Your firm keeps the copyright. License completed projects on your terms.

License your archive →
// Scope a pilot

Tell us which stage to start with.

Layout, systems or the code check. We reply with a sourcing plan for that stage and what a pilot would include.