The Trade SchoolThe honest spine.
A measure aimed at the underserved carries the highest duty of care. This page says what the measure is, what it is not, and what has to be true before it touches a real decision.
Regulatory positioning
The Trade School produces an analytical competence measure. It is a commercial scoring tool. It does not make consumer credit decisions and is not a consumer report under the Fair Credit Reporting Act. It does not make employment decisions and is not offered as a selection procedure. Any application that would influence a live hiring, promotion, or credit decision is a separate, counsel-reviewed step that has not been taken.
The model (tsc-v0.1.0) is six weighted, deterministic metrics. Unlike a thousand-variable network, it can explain itself, and it does: every score carries plain-language reasons mapped to the metrics. That explainability is a feature, not a courtesy.
Discipline 1 — calibrate against outcomes
The weights are expert hypotheses. Status: hypothesis, not calibrated. Every response says so in its calibration_status. The measure earns the right to matter only when tested against real hires and real income, and that test has not yet been run.
Discipline 2 — test for disparate impact
Inputs that describe a person — name, contact, age, gender, race, address, photo and the like — are stripped before scoring and never logged. That blocks direct input; it does not prove the absence of proxies. Before any live deployment, outputs are tested against a held-out set with protected-class labels the model never sees, with the four-fifths rule as the first screen. The test harness is published in the repository (compliance/disparate_impact_test.js) so the check is repeatable, not remembered.
Discipline 3 — protect the signal
Employer demand can be astroturfed; evidence can be inflated. Submissions arrive unverified and stay that way until reviewed; nothing is auto-verified. Volume limits, evidence-strength weighting (self-report is the floor), and verification tiers are the current floor. Signal integrity is the product.
Reasons on every score
| Code | Metric | Plain-language reason |
|---|---|---|
| VER-LOW | verification | Most evidence is self-reported or unverified. |
| DEP-LOW | demonstrated_depth | No competency is yet demonstrated at an independent level or higher with strong evidence. |
| REC-STALE | recency | The strongest evidence is more than a year old. |
| BRD-NARROW | breadth | Evidence covers fewer than two domains at an assisted level or above. |
| DEM-GAP | demand_alignment | The competencies held do not yet cover what the named demand asks for. |
| DEM-NONE | demand_alignment | No demand signal matches the sector/roles supplied. |
| TRJ-DOWN | trajectory | Older evidence is stronger than recent evidence. |
What is never collected
No names, no contact details, no government identifiers, no photos, no addresses finer than an ISO country-region. Paid machine queries are logged as shapes (domains, levels, evidence types, sector) with the paying wallet as the receipt key — never as descriptions of a person. We may classify a pattern in that log; we never act on a query's contents for our own account.
Questions on positioning: login@weckett.com. This page is a statement of design and current status, not legal advice.