SGA Reporting System — Planning Hub

The four buckets

Reshuffled from the original Headway/Daybreak/Lookout framework. Karen + Scott own the categorization session to assign every live deliverable to exactly one bucket.

The 8 Daybreak gate dimensions

Architecture inheritance — do NOT redesign

3-Week Sprint

Window: 2026-05-15 → 2026-06-05 (heads-down, per Brooke's approval)

The 5 gates that lock first (Phase 2)

Locked thresholds, per research stream 4. The other 3 stay under-review until benchmarks confirm.

Three-tier time-clock routing (research stream 1, adopted verbatim)

Each tier has 60 minutes to acknowledge before auto-escalation. Forces accountability.

4 parallel research streams — 12,700 words, 80+ sources

Ran in background while planning + scaffolding proceeded. Output synthesized into research/SUMMARY.md.

Cross-cutting conclusions (all 4 streams agree)

1. Composite-AND is universal. Banned single-metric alerts at alert-tab layer.
2. Hybrid floor + cohort thresholds. Both must fail to fire.
3. Causal attribution is non-negotiable from day one. Layer 4 directive store ships in Week 2, even though Signal Engine deferred to Phase 4.
4. Design against documented anti-patterns: alert fatigue, single-metric trap, gaming, causal theater, cadence mismatch, vendor bias.

Dashboard Audit — Four Families, Not Three

Answers the question from Meeting 2026-05-14: "labor analysis has one type, RIS has different type, DAX queries different type."

30+ live HTML dashboards fingerprinted by CSS variable scheme, font choice, Action Briefing presence, threat-bar presence, and drillable-dashboard markers.

Family A — Drillable-Dashboard canonical WINNER for ops/alert dashboards

CSS signature: --accent:#1e3a8a slate. Reference impl: .tmp/zoho-sentiment.preflight/index.html. Pairs with action-briefing skill.

Family B — People System scorecard WINNER for people-tier scorecards

CSS signature: --amber:#F59E0B; --amber-bg:#FFFBEB. In-house design system, not in .claude/skills/. Cross-nav across the 5 dashboards already works.

Family C — Hand-coded power-analytics MIGRATE

Pre-skill era. Chart.js + annotation plugin, custom colors. Works fine; not urgent. Refactor before adding features.

Family D — Inter-fonted standalone TRIAGE

Inter font, no skill signature. Quiet Ledger aesthetic adjacent.

Anti-pattern findings

1. Inconsistent CSS vars. A=slate (#1e3a8a), B=amber (#F59E0B), C=navy (#1a1a2e). Ad-hoc picks create a fifth family. Lock var sets in skill files.
2. kpi-metrics-dashboard is 1MB. Too much inline data. Should pull from snapshot.json.
3. procurement-analysis is 6MB. Needs investigation — embedded images or massive inline data dump.

Peer-Cohort Definitions

Required for every gate using peer-logic threshold (Call Volume, Conversion, Labor, Doctor Supply, Hygiene Reappt). Closes Phase 1 §1.4 prereq.

Tier 1 cohort: size × specialty × PMS

Expected: 12–24 cohorts of 8–25 practices each at SGA scale.

Cohort ID format: {size}-{specialty}-{pms} (kebab-case). Example: 2k4k-general-eaglesoft.

Empty / single-practice cohort fallback

If cohort has fewer than 4 practices, drop most-specific dimension in order: PMS → specialty → size. Log the fallback in gate output so brief discloses it.

Per-gate cohort assignment

Tier 2 dimensions (deferred until Tier 1 stable)

Validation requirements before Phase 2

1. Cohort assignment dry-run — Verify no cohort has fewer than 4 members after fallback
2. Cohort stability test — 4 consecutive days; flag any practice flipping cohorts ≥2× per week (likely master-data error)
3. Peer-median sanity check — 30 days historical; flag cohorts whose median is worse than industry floor

GSD Scaffold for Daybreak

Project structure at personal/SGA/daybreak/.planning/. Five files written by gsd-roadmapper agent.

5-Phase Roadmap

Top 3 requirements surfaced

1. REQ-028 (Phase 2): Rules engine with composite-AND + streak-day persistence. Architectural keystone — without composite-AND logic, labor headcount-reduction flag cannot fire correctly.
2. REQ-018 + REQ-019 (Phase 1): Registry stubs for all 8 gate dimensions + DAX validation of ~7 OpenClaw queries. If not done in Week 1, Phase 2 stalls day 1.
3. REQ-031 (Phase 2): Task Scheduler chain with PushNotification on failure. Direct mitigation of 6-month "collapse" risk.

Top 3 risks surfaced

1. R-02 (high): 6-month collapse risk — entire reason sprint exists. Mitigation: Phase 2 automation + Phase 3 hardening.
2. R-10 (high): Call tracking blocker (Dakota + Amy) — uniquely affects 2 of 8 gates. Mitigation: stub YAMLs status: pending-data-source, ship other 6 gates, chase in parallel.
3. R-40 (medium): Signal Effectiveness Engine scope creep — most exciting piece is highest-risk distraction. Hard rule in ROADMAP: Phase 4 is own project, own folder, mid-June.

Signal Effectiveness Engine

The IP — Phase 2 / own GSD projectThis is the self-improvement layer. NOT compliance-as-paperwork. A learning system that gets smarter every day.

The loop

Directive issued
  → Signal patterns observed
  → Execution inferred (not self-reported)
  → Target metric movement measured
  → Causal effect estimated
  → Gate priority re-weighted
  → Tomorrow's directives smarter than today's

Five jobs

1. Did the action happen? Infer execution from punch data + confirmation timestamps + chart audit patterns. Not "did OM say yes" — "did signals shift."
2. Did the practice improve? Target metric movement + externalities.
3. Which signals cause improvement? Per directive type × cohort × context: effect size, confidence interval, causal inference (DiD, synthetic control, propensity matching).
4. Re-weight gate priorities. Gates whose directives drive proven improvement get amplified. Weak causal signal gets demoted or retired.
5. Surface hypotheses for human ratification. "Morning directives to Brendan's region close 2.3x more often." Review → bake into routing OR reject.

Methodology stack (pre-committed from research stream 3)

Architecture

Schema

New signal_engine Postgres schema, 5 tables: directives, outcomes, attributions, weights, hypotheses.

Human ratification gate (Stitch Fix pattern)

Any proposed gate-weight re-weighting >30% w/w must be human-reviewed before going live.

Open Coordination Items

Triage decisions waiting for Scott

• Headway path: Family A migration vs Family B + briefing bolt-on. Recommendation: Path 2.
• om-morning-brief: Keep or supersede by Daybreak v2?
• om-dashboard: Keep or supersede by Daybreak v2?
• kpi-metrics-dashboard 1MB issue: Migrate now or defer?
• DAX Dictionary: Source folder unidentified; need to fingerprint when located

Open questions from research (6 deepest)

1. Call-tracking schema from Dakota + Amy — what fields? Call IDs? Booking outcomes? Recording transcription?
2. SAPS hygiene reappt taxonomy — Brittney + Incline review. Lock before §2 hygiene gate fires.
3. Donor pool construction for synthetic control — which practices are "comparable" for treated practice? Stream 3 #1 Signal Engine open question.
4. Override capture mechanism — UI for OMs to mark "intervention attempted" so Signal Engine distinguishes "didn't try" from "tried, didn't work."
5. Cosmetic-vs-restorative OSA split — 25-day OSA is bad-medical-good-cosmetic. Resolve via chair-type classification in master data.
6. Acquisition-cohort tagging — when did each practice join SGA? Signal Engine needs this for causal models.

Source-material gaps (GSD scaffold flagged)

1. Peer-cohort definition CLOSED — see Cohorts tab
2. Action-verb owner free-text in v2 — accept for sprint, defer to PWA
3. Treatment Acceptance + Hygiene Reappointment not in registry — TA deferred to Phase 3; Hyg Reappt uses Gen4 data in Phase 2
4. Active-patient-base bucketing for call-volume CLOSED — 1k/2k/4k/8k boundaries (Cohorts tab)
5. 3rd Next Available no registry entry — add stub in Phase 1 §1.4