We thought churn was one growth lever; the deep dive showed it's three independent operational tiles with very different mechanisms — and after working the validation queue, all three earned a slot on the board.
T13 → "Your scheduling is leaking rework today." Pull last 20 broken/canceled, look for pattern by provider / day-of-week / appt-type / insurance. Verify recall list worked daily not weekly. (Trend tile — within-practice direction matters more than absolute level.)
T14 → "Your confirmation chain is breaking." Audit reminder cadence 7d/3d/1d/2hr. Confirm vendor firing. Pull last 100 booked + count touches. Check for bad numbers, opt-outs, language gaps. (Lever tile — the actually-modifiable front-desk workflow.)
T15 → "Your risk patients are leaving holes — protect the chair." STANDBY PARK COLUMN (5-10 patients on call). WAVE DOUBLE-BOOK (small backup proc for risk slots). HYPER-CONFIRM CHAIN (above standard). PRE-COLLECT (risk-flagged cash plans show better when prepaid). (Protocol tile — operational, not financial-prediction.)
Three tiles. Three sentences the OM can actually do something with. That's the product.
SAPS composite churn (T13) and risk_notcomp_rate (T15) are near-orthogonal — they share only 1.2% of variance. Different patient failure modes need different operational responses. The front-desk confirmation chain (T14) is the actually-modifiable workflow that lives behind both.
Each tile gets its own gate.yaml, its own evaluator in rules-engine.js, and its own drill-rail protocol. The OM doesn't see "your churn is high" — she sees "your confirmation chain is breaking" or "your risk patients are leaving holes" with a specific protocol to run.
Holcomb Family Dentistry, Coosa Dental Associates, Chattahoochee — Columbus, Waynesboro Family Dentistry, NHD New Horizons. All firing T13 + T14 + T15. Each one represents a different operational story — Holcomb's $518k/day estimated rework drag combines high broken%, slipping confirmation, AND lost risk-NC dollars.
Network pulse 58 → orange. Same drill spine, same Hopkins Tile contract, three new protocols at the OM huddle.
| # | Validation | Status | Result |
|---|---|---|---|
| 1 | risk_notcomp_dollars time window | ✅ RESOLVED | Window-respecting (1.8% per-day variance 7d↔365d). 365d direct = $53.9M, matches ×4.06 annualization within 4%. All-time = $65.8M. |
| 2 | Within-practice longitudinal | ✅ DONE | Confirmation slope: +$826/mo NP gap improving vs worsening. Churn slope: $397/mo gap. Rescues T14 from FE attenuation. |
| 3 | ROD fixed effects | ✅ DONE | T15 survives ROD FE (β=−0.38, p<0.001). T14 cross-section drops (p=0.083) — saved by within-practice. |
| 3b | PMS fixed effects (extension) | ✅ DONE | Combined ROD+PMS FE: β=−0.29, p=0.06 (borderline). |
| 3c | Codex Test #2 — same-PMS replication | ✅✅ CONFOUND FALLS | Every PMS cohort shows negative β. Dentrix Enterprise n=24: β=−0.98, p=0.03, R²=0.66 — strongest signal anywhere. Combined-FE attenuation was power loss, not signal loss. |
| 4 | Out-of-sample replication | ⚠ MIXED | T13 direction holds (r=−0.06), T15 direction holds but magnitude collapses (r=−0.12), non-monotone terciles. T14 untestable (PBI conf history shallow). Likely COVID-recovery dynamics. |
| 5 | Labor coverage expansion | ⏸ BLOCKED | sga/labor-analysis only has Accelerate brand punches. FO-staffing direction stays exploratory at n=43 until other-brand punches ingested. |
| Window | Total $ | Per-day $ |
|---|---|---|
| 7 days | $1.01M | $144,994 |
| 30 days | $4.23M | $141,025 |
| 90 days | $14.3M | $159,415 |
| 180 days | $26.9M | $149,712 |
| 365 days | $53.9M | $147,664 |
| ALL TIME | $65.8M | ~$140k |
| PMS vendor | n | Pearson r | OLS β | p |
|---|---|---|---|---|
| Dentrix | 96 | −0.142 | −0.215 | 0.30 |
| EagleSoft | 33 | −0.045 | −0.205 | 0.88 |
| Dentrix Enterprise | 24 | −0.717 | −0.985 | 0.03 |
| Open Dental | 22 | −0.183 | −0.135 | 0.44 |
| Challenge | Status |
|---|---|
| Risk-dollar magnitude | RESOLVED |
| Manager-quality (within-practice) | PARTIAL |
| ROD confounding | PARTIAL |
| OOS replication | PARTIAL |
| Labor / staffing inference | OPEN |
"We have enough evidence to deploy these as operating controls. T14 and T15 identify appointment-protection behaviors operators can act on immediately. The risk-dollar exposure is real at $53.9M over 365d, but recoverable dollars should be shown as scenario-based, not booked value. T13 is a rework/trend indicator, T14 is an operational discipline tile, and T15 is a protect-the-chair tile with medium-confidence financial upside."
"I would hold only the stronger claims: no 'T15 predicts growth robustly across periods,' no network staffing conclusion, and no guaranteed 40% recovery without pilot evidence."
| Tile | Operational confidence | Financial-prediction confidence | Ready to ship? |
|---|---|---|---|
| T13 — Patient Appointment Churn | Medium (within-practice trend) | Low (cross-section weak) | ✅ Ship as trend tile |
| T14 — Confirmation Discipline | High (within-practice +$826/mo gap) | Medium (cross-section borderline) | ✅ Ship as exploratory |
| T15 — Risk-Patient Non-Completion | High (protocol + PMS replication) | High ($53.9M verified, OOS instability noted) | ✅ Ship as active |
Three tiles. Three jobs. Three protocols at the OM huddle. Operational-grade today; financial-recovery framing as we extend the data coverage. The Codex floor is gone — the verified dollar exposure is $53.9M, recovery target is scenario-based not promised, and the protocol works regardless of which model you believe.