Exact Check

As uploaded

Measured

Zoom and compareWatch (0:16)

How it works#

1. Upload a .step or .stp file, or a ZIP of an assembly's STEP files. Uploads
2. Run one check: a job with recipe: "cad_check". It measures the model once.
3. Ask for an evaluation as often as you like: the stored measurements judged against one process and material, or all of them, at the scale and assembly intent you choose. Each answer is free.

What it accepts#

• A .step or .stp file, or a ZIP holding an assembly's STEP files. Any other format is refused with 422 validation_failed and a sentence that says why.
• A ZIP may unpack to at most 2,000 entries and 850 MiB, at a ratio of at most 200:1. Operating-system clutter such as __MACOSX/ and .DS_Store is skipped.
• In a ZIP, the files no other file references are opened, each with the files it references. Files that only reference each other are opened from the first of them by name. Every file is measured once. A ZIP with so many top-level files that one check cannot be given them all fails with invalid_input and a sentence saying so.
• Your account's upload limits apply as for any file. Rate limits & limits

Running a check#

curl -sS "$API/v1/jobs" -H "Authorization: Bearer $PRINTY3D_API_KEY" \
  -H "Content-Type: application/json" -H "Idempotency-Key: $(uuidgen)" \
  -d "{\"input_file_id\": \"$FILE_ID\", \"recipe\": \"cad_check\", \"params\": {\"max_credits\": 120}}"

Then poll the job until it is succeeded, failed or cancelled, exactly as for any other job (Jobs & polling), or let a webhook tell you.

Where it runs, and how long it may take#

• Checks run on their own CPU workers, not on the GPU fleet, so GPU capacity never refuses them, and queue.capacity is null.
• When no Exact Check worker is running (none has asked for work or held a check in the last 2 minutes), a live check is refused with 503 capacity_exhausted, reason: "cad_unavailable" and Retry-After: 300. Nothing is queued or held. A test key is never refused this way.
• A queued live check fails free with error_code: capacity_exhausted once it has waited 10 minutes with no worker running, or after 24 hours in the queue.
• The checker gets 290 s, or max_credits seconds when that is less, and stops itself there; one still running 10 s later is stopped. The check then fails with error_code: time_limit and is not charged.
• A check that dies because the file needs more memory than a check is allowed fails with error_code: memory_limit and is not charged either.
• A file the checker cannot read fails with invalid_input and costs the 10-credit minimum (When a job fails).
• A check the checker stops on a fault of its own, not the file's, fails with invalid_result and is not charged.
• Only a file the checker refuses is charged. Any other failure (the checker crashing or stopped, engine_failed) is ours and is not charged either.

What it costs#

• A live check costs 1 credit per second of run time, with a minimum of 10 credits, like every live job (Credits & billing).
• When the job is admitted it holds its expected run time: 120 credits by default, more for a larger file, or the measured run time of similar checks once enough have run; never more than max_credits when you set it. The charge never goes above max_credits either.
• time_limit and memory_limit failures cost nothing; a file the checker cannot read costs the 10-credit minimum.
• Evaluations are free: they start no job and touch no balance.
• A check made with a test key is free and simulated (see Test keys).

The result#

When the job has succeeded, job.result holds a count summary. The findings come from the evaluation, not from the job.

The whole arm

One part, six copies

Zoom and compareWatch (0:13)

{
  "schema_version": 1,
  "parts": 36,
  "placed_copies": 36,
  "pairs": 132,
  "open_parts": 1,
  "sheet_parts": 0,
  "accuracy_mm": 0.02,
  "sample_limit_per_shape": 3000000,
  "truncated": false,
  "unsupported_placements": 0
}

The demo part

The open top

Zoom and compareWatch (0:13)

How walls are read#

A wall is measured from each sample point straight into the material, to the surface facing it. Where two faces meet at a sharp edge, the reading near the edge measures the edge, not a wall, and is not reported. A blade that is genuinely thin is still reported.

Thinnest wall

Section at the wall

Zoom and compareWatch (0:12)

A wall or gap of 0.01 mm or less is not reported: at that size the reading measures how the curved surfaces were traced, not the model. Neither is a patch of surface too small to hold one sample point.

A gap inside a part is read the same way, from each sample point straight out of the material, across the space, to the surface facing it.

Demo part

Section at the slot

Zoom and compareWatch (0:17)

The limits table#

The limits that apply to your account, one row per process and material. Read it to offer your users the processes to choose from, and to show them the data notice.

{
  "version": 3,
  "gap_floor_mm": 0.2,
  "notice": { "processing_region": "...", "retention_days": 30 },
  "rows": [
    {
      "process": "fdm",
      "material": "pla_abs_petg",
      "label": "FDM, PLA / ABS / PETG",
      "wall_min_mm": 1.2,
      "gap_inside_mm": 0.5,
      "gap_assembled_mm": 0.5,
      "gap_moving_mm": 0.5,
      "interior": "closed_ok",
      "hole_count": null,
      "hole_min_diameter_mm": null,
      "source": "...",
      "wall_by_size": null
    }
  ]
}

Evaluation#

Judges a succeeded check's stored measurements against the limits table. It is free, so change the question and ask again: another material, a larger print, parts printed together or apart.

FDM

SLA / DLP

SLS

MJF

Zoom and compareWatch (0:17)

Printed separately

Printed together

Zoom and compareWatch (0:12)

Scale 1

Scale 1.3

Zoom and compareWatch (0:14)

curl -sS "$API/v1/jobs/$JOB_ID/evaluation?process=fdm&material=pla_abs_petg&assembly_intent=together" \
  -H "Authorization: Bearer $PRINTY3D_API_KEY"

{
  "table_version": 3,
  "items": [
    {
      "process": "fdm",
      "material": "pla_abs_petg",
      "summary": "blocked",
      "findings": [
        {
          "code": "NOT_CLOSED",
          "severity": "blocking",
          "message": "Planet Gear is not a closed solid, so it cannot be printed as it is.",
          "part": "Planet Gear",
          "value_mm": null,
          "limit_mm": null,
          "location": null,
          "frame": null
        },
        {
          "code": "WALL_BELOW_MIN",
          "severity": "warning",
          "message": "Sun Gear wall 0.2773 mm, under the 1.2 mm minimum.",
          "part": "Sun Gear",
          "value_mm": 0.277328,
          "limit_mm": 1.2,
          "location": [-0.4596, 6.1625, 9.5642],
          "frame": "part"
        },
        {
          "code": "PARTS_TOUCH",
          "severity": "warning",
          "message": "Actuator Housing and Bearing Retainer touch. Printed together they will fuse: print them separately, or open a gap in the design.",
          "part": null,
          "value_mm": 0.0,
          "limit_mm": 0.5,
          "location": [40.3617, 37.3564, 62.5],
          "frame": "assembly"
        }
      ]
    }
  ]
}

summary is blocked when any finding is blocking, warnings when any is a warning, and ok otherwise. A finding is {code, severity, message, part, value_mm, limit_mm, location, frame}; message is a sentence you can show as it is.

• Bodies of one part that touch or overlap are how multi-body parts are modelled and are not reported; only a real space between them is a gap inside the part.
• A part has at most one gap-inside finding, for its smallest gap, whether that gap lies inside one body or between two bodies, and however many times the part is placed. Findings name parts by their names in the file, so parts that share a name share that one finding, for the smallest gap among them.
• location is a point in millimetres, or null, and frame says which frame it is in. A wall, or a gap measured inside one body, is in that part's own frame (part), before it is placed in the assembly, and applies to every copy of the part. A finding between two parts, or between two bodies of one part, is in the frame of the whole assembly (assembly). frame is null when location is.
• 409 when the job is not a cad_check, has not succeeded, or its stored measurements have expired.

Artifacts#

In both JSON files, a part's volume is null unless every shape in the part is closed. The positions of a region (centroid, bbox_min, bbox_max) and of a part's thinnest wall and narrowest gap (wall_location, gap_location) are in the shape's own frame; the location of a pair is in the assembly's. Downloading works as for any artifact. Artifacts & exports

How long the measurements are kept#

A check's stored measurements are kept as long as its other outputs, the artifact retention of your plan, and no longer. After that the evaluation answers 409; run a new check to judge the file again. The region where files are processed and the retention period are in the limits table's notice.

Test keys#

A check made with a test key is free and runs in the sandbox. The file is not read: the result and the artifacts describe one simulated part named SIMULATED SANDBOX PART, and the preview is a placeholder. The evaluation judges that simulated part, and the webhook is sent as in live mode, so you can build your whole flow before your account is approved.