Architecture & Operating Design · v0.2
Company operating intelligence, proven in shadow before it is allowed to act.
Implemented in the private src/ TypeScript package (see software modules). Implementer specs live at repo-root specs/ — they are not published on Pages.
Primary objectives
- Answer and book calls with shared membership, equipment, warranty, and history.
- Choose the correct technician/Van using technical fit, quality, client fit, geography, inventory, runtime, profit, margin, and development — not distance alone.
- Minimize unnecessary travel without letting proximity become the final factor.
- Learn strengths, weaknesses, and uncertainty without poisoning the model through self-reinforcing assignment.
- Use before/during/after media to normalize skill, difficulty, time, workmanship, and legitimate added cost.
- Keep repairs, turnovers, sales, warranties, callbacks, inventory, and downstream revenue as connected events.
- Issue receipts for recommendations, disagreements, overrides, actuals, and counterfactuals.
- Allow a field-side advisor on or off per technician / per call.
- Share one canonical live operating state across IT, office, dispatch, managers, field, warehouse, and booking.
Recommended logical architecture
ServiceTitan / ProBooks / warranty / telephony / maps / pricebook / inventory / media → Integration & MCP Layer → Canonical Event Bus → Operational State Store → Decision Engines → Shadow/Counterfactual Engine → Receipts & Audit → Role Interfaces → Controlled Action Layer.
This portal documents that design. It does not host the event bus or execute engines.
Decision engines (design / stub)
| Engine | Status |
|---|---|
| Booking & Call Intake | design |
| Dispatch / Call-Fit | design |
| Full-Day Trajectory | design |
| Runtime Prediction | design |
| Technician Skill & Exploration | design |
| Visual/Video QA | design |
| Warranty & Callback Classifier | design |
| Manager Intervention Predictor | design |
| Field Advisor | design |
| Pricebook / Quote / Margin | live-pure |
| Inventory / van stock / fulfillment | live-pure |
| Communications / recognition / mission board | live-pure |
| Weather / demand | live-pure |
| Maintenance / equipment-age routing | live-pure |
| Revenue & Turnover Attribution | design |
| Counterfactual Outcome | design |
| Receipt / Audit / Override | design |
v0.2 software (coded, not a spec dump)
- Internal channels: company, branch, department, trade, dispatch, sales, field, warehouse, project, incident — archive by moderator.
- Recognition after quality gates. Raw revenue alone is never a celebration.
- Mission board: target, actual, expected pace, remaining gap, projected end-of-day.
- ServiceTitan pricebook is a shadow baseline. ACCEPT / OVERRIDE / LOCK write Chain C. A lock cannot auto-recalibrate.
- Weather windows stay synchronized. Extreme events widen uncertainty; they do not just multiply demand. Lunar weight is zero unless lift is earned.
- Maintenance is demand-first: do not hold a qualified available technician for an ideal Selling Tech unless a human overrides.
Property Intelligence v1.0 (coded, not a spec dump)
- A property is a living lifecycle object, not a static address.
- Neighborhood patterns are diagnostic priors — not proof of a defect on the subject property.
- Public/market data is evidence, not automatic truth, until corroborated.
- Value is a range (nominal and inflation-adjusted), never a guarantee. Repair cost is not resale uplift.
- Mortgage rates change purchasing-power / demand pressure. They are not a fixed percentage haircut.
- Approved/licensed/public-authorized sources only. No unauthorized scraping architecture.
Storage separation
- Immutable event ledger
- Operational state store
- Analytical warehouse / lakehouse
- Media / object store
- Feature / skill store
- Receipt store
North star
The system earns authority by being auditable, adaptable, and repeatedly correct under real operating conditions — not by being more confident than the humans running the business.
Success: any day, call, or disagreement can show what was known, what operations did, what Trades recommended, whether a human changed it, how the day adapted, what happened in the field, and what result followed.