Rules, not machine learning. Every point comes from text already in the record's About Company field, and every row shows the points it earned. Enterprise size lowers the score, it never removes a company. Gate 6 is shown as a discovery question and costs nothing.
| Signal | Points | Read from |
|---|---|---|
| Verdict | YES 35 / MAYBE 15 | Prefix of About Company. A MAYBE is capped at 69. |
| Screening depth | 0 to 10 | Six-gate corridor sweep scores 10. The July Google Maps sweep, before the gate audit, scores 0 and is flagged for re-check. |
| Sector fit | 3 to 15 | Plastics, automotive, food and beverage, metals, electronics, heavy manufacturing score 15. |
| Demand | 0 to 15 | Several live reqs at one site 15, one live req 10. |
| TN read | 4 to 10 | Clean Gate 5 read 10. A Gate 5 issue scores 5 and is flagged, since the JD rewrite is the opening hook. |
| Contact | 0 or 10 | Contact linked and no "contact still needed" in the notes. |
| Freshness | 0 to 5 | Record touched in the last 3 weeks 5, last 2 months 2. |
| Large enterprise | −5 | Notes mark a long sell or a 20,000+ enterprise. |
The page reads every company record from RecruitCRM through the A&B Worker each time it opens, then scores them in this browser. The list call carries About Company, contact status and the Prospect Score field, so a full load is three calls and takes seconds. If RecruitCRM ever stops including those in the list, the page falls back to reading each record on its own, about 4 to 5 minutes cold, and says so in the header.
Write scores to RecruitCRM fills the company field Prospect Score: the score for Prospects, blank for every other lane. It writes only values that changed. Writing moves a record's updated date in RecruitCRM, so the Worker remembers each record's last human change and the freshness points keep reading that one. The page never reads its own field back as a scoring input.
The score is still read from prose, so a sentence can be misread. Open Why this score on anything that looks wrong; the points and the text they came from are both there.
The model needs outcomes, and today there are none to learn from. Each time a prospect replies, books a discovery call, signs or goes cold, that result goes on its RecruitCRM record. At roughly 150 to 200 recorded outcomes, the same signals above become the training columns, and a model can learn the weights instead of guessing them.