Scope3D
Get a demo
DesktopReview & automation wizards

Twenty-one checks
that write nothing.

Named, grouped, and honest about their limits. Every one of the twenty-one reads
your open Revit document and reports. None of them changes it.

PRE-ISSUE

The pre-issue checklist, pointed at the whole model: nine reviews run one after another and pooled into a single verdict, sorted into build order rather than the order the checks happened to finish in.

Twenty-one wizards, four groups, named

Every tile says what it will read before you press it, whether it answers for the whole model or one storey, and what it cannot see. All twenty-one are bridge reads: they talk to Revit, they spend no tokens, and they write nothing. Each finished run carries the milliseconds its read took, because a slow check should be visibly slow.

A read that fails is reported as a failure, not as a clean result — “the read failed” and “nothing is wrong” are different answers, and an empty list would make them look the same. Fixing anything a wizard found is a separate run with its own checkpoint, transcript and approval.

Review & QA — nine

Reads the model and tells you what is wrong with it. Nothing here writes.

Exterior review

envelope closure · wall joins · roof coverage · openings

Whole model or one storey. Reads only, no tokens.

Not checked

  • •the real wall outline — coverage is compared as rectangles, so an L-shaped plan is judged loosely
  • •insulation, air barrier and thermal bridging
  • •whether the elevation drawing shows a form this model does not
Floor review

slabs · openings · level offsets · stair arrivals

Whole model or one storey. Reads only, no tokens.

Not checked

  • •structure — whether a slab that exists can actually carry what is on it
  • •the slab build-up: thickness, layers and whether the finish floor level is where the drawing says
  • •a hole cut for a reason this cannot see, which is reported as deliberate on geometry alone
Stair & rail review

riser count vs floor-to-floor · riser height · tread · width

Whole model or one storey. Reads only, no tokens.

Not checked

  • •headroom, which needs a ray cast up from each tread rather than a bounding box
  • •handrail continuity and graspability
  • •landing dimensions where a stair turns
Door review

widths · hosts · duplicates · doors that foul each other

Whole model or one storey. Reads only, no tokens.

Not checked

  • •swing direction and the clearance a leaf actually needs — the geometry for that is not read here
  • •hardware, fire rating and acoustic performance
  • •whether a door is the right type — that is what Type match is for
Window review

head heights · sills · sizes · hosts · duplicates

Whole model or one storey. Reads only, no tokens.

Not checked

  • •which rooms are bedrooms, so egress is reported as worth checking rather than as a failure
  • •glazing performance, U-value and solar gain
  • •whether the elevation drawing shows a window this model does not
Room review

enclosure · placement · naming · area

Whole model. Reads only, no tokens.

Not checked

  • •whether a room is the right size for its use
  • •room-to-room adjacency and circulation
Model audit

Revit's warnings · duplicates · unused families

Whole model. Reads only, no tokens. The warning count comes off the state probe and the audit notes off a bridge workflow, and either is allowed to fail without costing the other. The notes are carried through as prose rather than parsed, because inventing a schema for someone else's report is how a check silently drops the half it did not expect.

Not checked

  • •anything the audit workflow itself does not look at
Clash review

hard intersections between walls, floors, doors and windows

Whole model. Reads only, no tokens.

Not checked

  • •soft clearances — a clash scan finds solids inside solids, not a door that cannot open
Pre-issue checklist

runs every review above and reports one verdict

Whole model. Reads only, no tokens. Nine reads run one after another — the bridge takes one client at a time — and a check that could not run counts against the gate rather than vanishing from it. The verdict is a state of the model, not permission to publish.

Not checked

  • •anything none of the individual checks looks at — this is their sum, not more than their sum

Model automation — six

Finds what is missing across the whole model. Finding it is a read; applying what it found is a separate run you approve.

Type match

placeholder families · wide openings built from swing doors

Whole model. Reads only, no tokens. Also one of the nine the pre-issue checklist runs.

Not checked

  • •whether a family is the right one aesthetically — only whether its size and its name disagree
  • •window types, which are harder to call from width alone
Tag sweep

untagged doors · windows · rooms · walls

Whole model or the active view. Reads only, no tokens.

Not checked

  • •whether a tag is in a sensible place, or overlapping another
Dimension openings

openings carrying no dimension string

The active view, or the whole model. Reads only, no tokens. An opening nobody has dimensioned cannot be built from the drawing.

Not checked

  • •whether an existing dimension is to the right face, or reads sensibly on the sheet
What is in the model

every counted category, with totals

Whole model. Reads only, no tokens. The census is also where “walls but no floors” comes from — counted in Revit, not inferred.

Not checked

  • •anything outside the fourteen categories the census counts
Naming normalisation

views and sheets off the naming standard

Whole model. Reads only, no tokens. If the connection cannot check names against a standard, the tile says so instead of reporting nothing found.

Not checked

  • •what your naming standard actually is — this reports what the bridge considers off-pattern
View cleanup

unused views · views on no sheet · duplicate crops

Whole model. Reads only, no tokens. It lists candidates; deleting any of them is a separate approved run.

Not checked

  • •whether a view is unused on purpose — a working view is not rubbish

Drawings & schedules — four

Turns the model into the things you actually issue. When a prerequisite is missing it says which one, rather than returning an empty schedule.

Takeoff

material quantities by assembly

Whole model. Reads only, no tokens. A model with nothing in it is reported as blocked — “no elements to count” — instead of coming back as a takeoff of zero.

Not checked

  • •prices — this counts quantities, it does not cost them
  • •waste factors and laps, which are an estimator decision rather than a model fact
Room schedule

rooms with their numbers, names and areas

Whole model. Reads only, no tokens. With no rooms placed it blocks and says why: place rooms first, and the schedule follows from them.

Not checked

  • •whether the field list matches your office standard
Room finishes

floor, base, wall and ceiling finishes per room

Whole model. Reads only, no tokens. With no rooms placed there is nothing to compute finishes for, and it says that.

Not checked

  • •whether the finishes are the specified ones — this reports what the model carries
Sheet set

sheets, the views on them, and what is missing

Whole model. Reads only, no tokens.

Not checked

  • •whether the sheet layout reads well — placement is checked for existence, not for composition

Knowledge & regulatory — two

Measures the model against documents you have loaded, and cites them. Both stop short of a verdict.

Code check

model measurements against the clauses you have loaded

Whole model. Reads only, no tokens. The answer is a measurement beside a clause — “measured 160 mm; the clause states a maximum of 200 mm” — with the clause number and the words the number came from.

Not checked

  • •compliance itself — this reports a measurement beside a clause and stops short of a verdict
  • •occupancy, exemptions and alternative solutions, all of which change what a clause requires
  • •any clause the corpus does not contain, or phrases in words the search did not match
Brief conformance

rooms and areas against the brief you have loaded

Whole model. Reads only, no tokens. A room the brief asks for and the model does not have is named, with the line of the brief it came from.

Not checked

  • •the brief's actual requirements — this matches room names, it does not read prose
  • •areas, adjacencies and anything expressed as a sentence rather than a name
KNOWLEDGE

The documents stay on your machine

1 Load a code, a zoning by-law, a brief, an office standard or a specification Text is extracted in the app and cut on clause numbering rather than on a fixed window of characters — 9.8.4.1 sits under 9.8.4 under 9.8 — so every piece already knows what to call itself and a citation names a clause instead of a page range.
2 Four million characters, and the cap is enforced out loud A few hundred pages of extracted text: enough for a project brief, an office standard and the parts of a code that matter, and not enough for a code in full. A document that will not fit is refused with the numbers, because a knowledge base that quietly forgot half its corpus would answer confidently from the half that survived.
3 Nothing is uploaded Extraction, chunking and search all run on your machine. The only thing that ever leaves is the handful of clauses that answer a question you asked, as part of that question — and there is no server on our side to send a document to.

The limit inside a clause is pulled out of the clause's own words — “not less than”, “shall not exceed”, “maximum of” — and a phrasing outside those shapes yields no limit rather than a guessed one. Then the measurement goes beside it and the app stops. It never returns a verdict of compliant, because that is a judgement about occupancy, exemptions and the parts of the code your corpus may not contain.

Twenty-one reads, zero writes.

A wizard gets three ways to reach Revit and no reference to the session, so it cannot start a run, open a dialog or save your document as a side effect of being read. That is what makes it safe to press one speculatively.

The rules themselves are pure functions of what was read, which is why the riser-count and slab-coverage tests are pinned by unit tests rather than hoped about. Fixing what a wizard found is a different thing entirely: a real run, with a checkpoint, a transcript and an approval.

The tile tells you the cost before the press

All twenty-one are bridge reads. None of them sends anything to a language model, so none of them shows up on your provider bill.

A failed read is reported as failed

Never as an empty result. “The read failed” and “nothing is wrong” are different answers, and one bug of exactly that shape has already shipped in this app once.

Every report names its own blind spots

Printed on the tile before you run it and on the report afterwards. A window review that goes quiet about never having looked at egress gets trusted for more than it can carry.

A clean run says so out loud

Findings include the ones that passed — “nothing counted is missing” — because a check that only ever lists problems stops being read.

Millisecond timing on every run

Shown, not hidden, because a slow wizard should be visibly slow. Verifying a fix is the same read again — a fix's own report of what it did is not evidence that it did it.

Your own named sequences.

The things people repeat are sequences, not single checks. “Room Schedule QA” is rooms, then schedule, then finishes — save it once and stop pressing three tiles in the right order from memory.

What a workflow is

An ordered list of wizards, each pointed at the whole model or at one storey. It runs them in the order you saved them.

Where it is kept

Per machine, not per project. A firm's pre-issue checklist belongs to the firm, not to the building, and re-creating it for every new .rvt is exactly the friction this removes.

How a storey is remembered

By name — “Level 2” — rather than by element id, because a saved workflow outlives the document it was made in. The runner resolves the name against whatever is open and falls back to the whole model when it finds no match.

Run count

Counted, so the list can put the ones you actually use first.

What it never does

Write. A workflow is a sequence of reads, so a saved one cannot become a saved change.

Questions about the checks.

If yours is not here, ask us and we will answer it properly.

No. A wizard is handed three ways to reach Revit and nothing else — run a program, call a named bridge tool, read the building's state — and no reference to the session at all, so it cannot start a run or save your document even by accident. Several of them end with a button that offers to fix what they found; pressing it starts a normal run, with a .rvt checkpoint written first and the change shown to you before it is applied. How a run is approved.

Nothing on your AI account. All twenty-one are bridge reads: they talk to Revit and to documents already on your machine, and they send nothing to a language model, so none of them moves the spend figure in the header. Asking the app to fix something it found does spend, at your provider's published rates — the model list and those rates are on models and cost.

No, and that is deliberate. It retrieves the clause, pulls the number out of the clause's own wording, puts your measurement next to it and stops — “measured 160 mm; the clause states a maximum of 200 mm”. Compliance depends on occupancy, exemptions, alternative solutions and the parts of the code your corpus may not contain, and software that says “compliant” on the strength of one retrieved paragraph is a liability rather than a feature. You read the pair and decide.

Into a store on your own machine, indexed there. Extraction and search run locally, the corpus holds four million characters, and a document that would push it past that is refused with the numbers rather than being half-saved. The only thing that ever leaves is the handful of clauses that answer a question you asked, travelling as part of that question to the AI provider you configured.

The panel is the unprompted half and the catalogue is the deliberate half. The panel populates the moment you connect and keeps a short list of what is wrong with the building in front of you; the catalogue is where you point a named check at a named storey because you are about to issue something. They share the same rules and the same “Not checked” discipline. What the panel finds on its own.

Yes. They read whatever document Revit has open — a model you built by hand years ago, one a consultant sent you, or one traced from a drawing set. Nothing about them assumes the model came from here. Each project does record which Revit document it was working on, so reopening it against a different building stops work rather than quietly checking the wrong one.

Run the checklist on a model you already have

Open any .rvt and press Pre-issue checklist. Nine reads, one verdict, nothing written and nothing spent — then decide whether the rest is worth your time.

Download for Windows Book a demo