Dashboard Legend

Column Definitions

Column	Meaning
Hour	Time slot (7am – 7pm, your business hours)
Pred	Predicted sessions — model's best guess based on 8 same-weekday samples
Range	Low–High — the min and max seen across those 8 historical samples
Act	Actual sessions — real count for completed hours (synced from BeyondTrust every 5 min)
✓ / ✗	Grade — green if accurate, red if off (see thresholds below)

Grading Thresholds

Each completed hour is graded by comparing Pred vs Act:

Grade	Condition	Example
✓ Green	Off by ≤ 1 session OR within 20%	Pred 5, Act 4 → diff 1 → ✓
✗ Red	Off by more than 1 AND more than 20%	Pred 5, Act 0 → diff 5 → ✗

Visual Indicators

Bar width — proportional to the predicted count (wider = busier hour)
Bar opacity — tighter range = more opaque (higher confidence), wider range = more transparent
Peak hours — top 3 busiest predicted hours are highlighted

Drift Alert

Appears above the hourly grid when viewing today. Compares cumulative actual vs cumulative predicted for all completed hours:

Alert	Condition	Meaning
Running Hot	Actual ≥ 40% above predicted	Busier day than expected
Running Cold	Actual ≥ 40% below predicted	Slower day than expected
(no alert)	Within ±40%	Tracking close to forecast

How Predictions Work

Model: dow_hour_baseline_v1.1

For each (weekday, hour) pair, the model looks at the 8 most recent same-weekday historical samples (excluding holidays)
Takes the median of those 8 samples as the base prediction
Applies a trend factor (linear regression over the 8 samples, capped at ±15%) to adjust for upward/downward trends
Range = median ± 1 standard deviation (Low/High bounds)

Prediction = median(8 samples) × trend_factor (±15% max)

Data Source

Historical data: BeyondTrust session reports (CSV downloads)
Live actuals: BeyondTrust CSV re-synced every 5 minutes (authoritative source)
Training window: ~168 days of history
Business hours: 7am – 7pm (13 hourly slots)

Score Composition

The SVC score (0–100) is a weighted composite of real-time metrics plus a pacing adjustment:

Component	Weight	How It's Calculated
RC Answer Rate	40%	RingCentral queue service level (% calls answered). Defaults to 100 if unavailable.
Queue Count	30%	100 − (customers waiting × 30). Each person in queue costs 30 points.
Wait Time	15%	100 − (longest wait in minutes × 10). Each minute costs 10 points.
Session Stress	15%	100 − (sessions per rep / 3.0 × 100). Maxes out at 3 sessions/rep.

Base Score = RC×0.40 + Queue×0.30 + Wait×0.15 + Stress×0.15

Final Score = Base Score + Pacing Adjustment (±3 to ±5)

Pacing Indicator

Compares today's cumulative actual sessions against cumulative predicted for all completed hours.

Pace Ratio = (Actual So Far − Predicted So Far) / Predicted So Far

Pacing	Condition	Score Effect	Meaning
+40%+	Way above forecast	−5	Much busier than expected — keep all staff
+25–39%	Above forecast	−3	Somewhat busier — be cautious releasing
−25% to +25%	On pace	0	Tracking as expected
−25 to −39%	Below forecast	+3	Slower than expected — OK to release
−40%+	Way below forecast	+5	Much slower — safe to release

Example: 35 actual vs 28 predicted = +25% → Pacing: Ahead → Score gets −3 adjustment

Forecast Outlook (Next 2 Hours)

The SVC system looks at the hourly forecast for the next 2 hours and classifies the outlook:

Outlook	Condition	Effect on Release
Low Volume	Both next hours have pred ≤ 2	Score threshold lowered from 80 → 75 (easier to release)
Normal Volume	Neither peak nor low	No change — standard threshold of 80
Peak Volume	Any next hour has pred ≥ 4 or is top 3 for the day	Release blocked — hold all reps

Early Release Recommendations

Only active during the 3:00 PM – 7:00 PM window on weekdays.

Recommendation	Conditions
Safe to Release	Score ≥ 80 (or 75 if low volume ahead) Queue empty (0 waiting) 3+ reps online Sessions/rep < 2.0 No peak hour coming No vacation understaffing risk
Release Blocked	Conditions for "safe to cut" met, but blocked by: • Peak hour predicted ahead, OR • Vacation schedule would leave < 3 reps
Keep All Reps	Score < 65 OR queue ≥ 2 OR sessions/rep ≥ 2.5
Monitoring	Score between 65–80 — watching but no action needed

Rep selection: When releasing, the system picks the idle rep (0 active sessions) with the fewest releases this month — ensuring releases are distributed fairly. John is excluded from release candidates.

Daily Card Fields

Field	Meaning
Predicted	Total sessions expected for the day
Range	Low–High bounds (sum of hourly ranges)
Actual	Real session count (for past days / today)
WoW Delta	Week-over-week change — compares to same weekday last week
Confidence	High / Med / Low — based on how tight the prediction range is
Reps	Suggested staffing — based on 12 sessions/rep/day ratio

The TODAY number in the status bar shows the total sessions for the current day. This is synced from BeyondTrust every 5 minutes via CSV download — it includes all completed and active sessions, not just currently active ones.

Hourly Outlook