Evidence before claims
A public record of what Quantum Steel Design can demonstrate today, how steel automation will be evaluated, and where qualified human authorization remains mandatory.
Verified public baseline
These are inspectable capabilities and dated validation results, not estimates of client productivity, revenue, schedule improvement, or avoided rework.
The web reference exposes US customary dimensions and section properties, nominal profile previews, calculators, and downloadable formats. Maintainer checks validate all generated shape pages and their representative exports.
The offline Windows application documents 2D profiles, reports, structured data, and length-based 3D mesh outputs. It includes a command-line interface for repeatable local workflows.
PDF Rescue processes eligible vector drawing content locally in the browser. The Windows beta publishes release metadata and a SHA-256 checksum. Its page states the boundary between recoverable PDF content and original CAD authority.
The maintained video library pairs public YouTube videos with written context, local thumbnails, and a dedicated video sitemap. Long-form pages include transcripts and chapters. Release checks validate the current catalog and the steel-shape page embeds together.
The browser workbench can save tool snapshots into projects, maintain drawing and RFI records, and export a complete project package. A useful inspection is to save two different revisions, export the project, and import it as a separate copy before relying on the backup process.
Selected PDFs can be searched for embedded text, displayed side by side or overlaid, and annotated for review. Use a non-confidential pair of sample issues to inspect page selection, alignment, exported marks, and the limitations recorded on each tool page.
The offline workbench displays its included tools, release, and file size. It verifies downloaded files before replacing a previous pack. Users can check status and remove the download separately from their saved project records.
Published evaluation protocol
Each future automation result should be tied to a frozen input set, expected outcomes, a defined scorer, recorded exceptions, and a named human approval point. Failed and unsupported cases belong in the record.
State the exact input, output, intended user, supported file or data versions, prohibited uses, and consequence of a missed issue.
Version the test set and include ordinary examples, shape-family variation, incomplete inputs, ambiguous conditions, and known edge cases without exposing confidential project data.
Create the reference result independently of the automation where practical. Record the source, assumptions, units, reviewer, and any cases without a single defensible answer.
Record the software and ruleset version, runtime environment, input hash, warnings, raw output, elapsed time, and any manual intervention.
Classify correct detections, incorrect flags, misses, unsupported cases, and reviewer overrides. Do not remove difficult cases merely because they lower a headline result.
Report sample size, method, metric definitions, results, failures, version, date, reviewer role, conflicts of interest, and whether the result was internal or independently reproduced.
Evaluation metrics
A fast automation result is not useful if it hides uncertainty or moves an error downstream. QSD will match metrics to the task rather than select one flattering number.
True issues detected divided by all known issues in the frozen test set, reported with the sample size and issue classes.
Confirmed issues divided by all flags produced. Low precision creates review noise and can hide the checks that matter.
False negatives reported by severity and category. Consequential misses are not averaged away by numerous low-risk successes.
The share of the defined task set that receives a valid result, including explicit unsupported and indeterminate outcomes.
How often a qualified reviewer accepts, edits, rejects, or escalates the result, with the reason preserved for later analysis.
Median and high-percentile elapsed time measured against the same baseline scope. Setup, checking, correction, and approval time remain visible.
Human-accountable automation
Automation can assist collection, comparison, checking, and explanation. Project responsibility and release authority stay with the qualified people assigned by the contract and organization.
Confirm the governing issue, scope, units, and input provenance.
Run versioned rules and retain warnings, exceptions, and raw evidence.
Trace each proposed finding to its source rather than accepting a bare answer.
Accept, edit, reject, or escalate based on assigned responsibility and expertise.
Release through the project's document control process and preserve the decision record.
Known limitations
Open evidence register
These slots will be populated only after the work exists, the baseline is documented, the result is reviewed, and publication permission is clear.
Target: one bounded detailing or checking workflow with a pre-agreed baseline, success criteria, confidentiality terms, human review, and permissioned reporting.
Discuss a pilotTarget: compare the same defined task before and after an assisted workflow, including setup, checking, correction, approval, limitations, and exceptions.
Contact QSDTarget: publish a versioned set of non-confidential test cases with expected outputs, scoring definitions, misses, false alerts, and reproducibility notes.
Follow QSD ResearchQSD's present portfolio uses fictional quantities and conditions to demonstrate intended organization and deliverables without presenting them as client history.
Review illustrative samplesHelp create the first measured record
A useful pilot begins with a real problem, a controlled input, a current baseline, clear responsibility, and permission to publish only what both parties approve.