101
applications, June 1–30
(all via the Long-form app)
(all via the Long-form app)
Every completed application graded 1–4 against the current rubric — recalibrated against real GHL-pipeline outcomes (front-end DQ rate and show rate), replacing the original lenient-bootstrap version that proved non-predictive.
The rubric is now meaningfully stricter than the lenient first pass — the majority of applicants land in DQ or setter-triage territory, and "perfect ICP" (grade 4) is a genuinely small minority, matching what a first-principles read of this offer's fit rate should look like.
Free-text answer length, not content, is what scores here — a terse answer is the single strongest red flag in the whole rubric (weighted +3, the heaviest single input). A substantive answer (6+ words) is required for a grade-4.
Free text, bucketed. Only an unambiguous "No" scores a flag (+1) — a supportive partner, a single person, or an ambiguous answer don't count against the applicant.
Volume is almost entirely VSL — the other channels don't yet carry enough applications to read grade mix by channel with any confidence.
| Funnel | n | Grade 1 | Grade 2 | Grade 3 | Grade 4 | MQL rate (3–4) |
|---|---|---|---|---|---|---|
| VSL (paid) | 74 | 18 | 25 | 25 | 6 | 42% |
| IG (organic) | 5 | 1 | 2 | 2 | 0 | 40% |
| Other / Organic | 1 | 0 | 0 | 1 | 0 | 100% |
| 1 | 1 | 0 | 0 | 0 | 0% |
Show rate climbs cleanly from the bottom two grades (23% each) to grade 3 (59%) to grade 4 (83%) — the rubric now sorts real outcomes, not just self-reported answers. Grades 1 and 2 read the same on show rate by design (both route away from a closer's calendar); the split between them is about front-end-DQ severity, not show-rate granularity — grade 1 front-end-DQs at a materially higher rate than grade 2 even though their show rates tie.
The first version of this rubric used three self-reported multiple-choice questions (Q8 living situation, Q10 capital, Q12 commitment) and counted red flags. Checked against real outcomes, it wasn't predictive: show rate by grade went 50% → 15% → 58% → 47% — grade 2 showed worse than grade 1, and grade 3 beat grade 4. The current version replaces Q12 (found to carry no real signal) with two free-text signals — Q7 answer length (the single strongest predictor found) and a bucketed read of Q9 — and switches from counting flags to a weighted score. Show rate now reads 23% → 23% → 59% → 83%, and front-end-DQ rate is 32% → 24% → 19% → 0%, both cleanly ordered.
| Grade 1 | Grade 2 | Grade 3 | Grade 4 | |
|---|---|---|---|---|
| Show rate — v1 (retired) | 50% | 15% | 58% | 47% |
| Show rate — v2 (current) | 23% | 23% | 59% | 83% |
| Front-end DQ rate — v1 | 45% | 29% | 17% | 6% |
| Front-end DQ rate — v2 | 32% | 24% | 19% | 0% |
Grade 1+2 share of applicants rose from 40% to 58%; grade 4 shrank from 20% to 7%. This is a deliberate, dated cutover (2026-07-01) — CPMQL/CPTQO reads from before this date aren't directly comparable to reads after it, since what counts as an MQL changed. Calibrated on June's own outcome data (some grade cells as thin as n=6) — not out-of-sample tested; scheduled for re-validation once July's outcomes resolve. Full detail: the wiki page typeform-grading-mql-handoff.md.