How the numbers are computed.

Confidence is a per-stop trust signal shown on every charging-stop card. It blends four inputs:

confidence = 0.50 × operator_reliability   // curated network → 0..1 table
           + 0.25 × stall_score            // capped at 8 stalls = 1.0
           + 0.15 × rating_score           // 0..5 → 0..1; flat 4.0 for OCM-only stops
           + 0.10 × distance_score         // 0..1, penalty within 5 mi off route

Confidence is bounded to [0, 1] and rounded to two decimal places. The displayed badge multiplies by 100 and rounds to a percent.

Community reports nudge the displayed confidence by at most ±0.05 (positive) or −0.10 (negative) — see the community weighting section.

Reliability is computed from community reports only — not from directory data or operator-supplied numbers. It deliberately weights what actually costs drivers:

hits   = recent_charger_worked_count            (last 30 days)
misses = recent_charger_broken_count + recent_long_wait_count

raw    = hits / max(1, hits + 2 × misses)
score  = round(raw × 100)

Broken chargers count double against worked-charger rewards because broken is what ends a trip. A station with 10 worked + 5 broken lands at 50/100, not 67/100.

Stations or operators with no trust reports return null — the UI renders "no data" instead of 0/100, so freshness isn't conflated with badness.

When an OpenWeather key is configured, Draiv samples weather at every leg's midpoint and applies the leg-specific multiplier to that leg's SOC drop:

Effects compound multiplicatively. There's a 0.75 floor — past that, the trip is unsafe and the user should plan differently, not accept a worse range estimate. The Trust Layer card surfaces the miles-weighted aggregate multiplier across all legs.

Public DCFC reliability sits around 70–85% on any given trip. One broken charger on a long route is a real risk. Draiv mitigates it structurally:

• Every primary stop ships with up to 2 alternate chargers within roughly 2 mi.
• Backups prefer a different operator than the primary so a single network outage doesn't take both sites down. Same-operator alternates are only used when no cross-operator candidate exists nearby.
• Backups are ranked by the same Confidence formula as primaries — you see their score in the Trust Layer.
• Lack of a backup is itself a risk signal: stops with zero alternates contribute to a higher per-stop risk level.

Reports go into a 30-day rolling window and add up into a small, clamped adjustment on Confidence Score:

+0.02   per charger_worked  (capped at total +0.05)
−0.05   per charger_broken
−0.02   per long_wait
−0.04   per unsafe_area
 0      per bathroom_clean / bathroom_bad / good_food_nearby
 0      per wheelchair_accessible / poor_accessibility
 0      per good_lighting / poor_lighting
 0      per family_friendly / pet_friendly
 0      per other

Accessibility, lighting, family/pet, and food reports are informational by design. They appear as signals on stop cards and the reliability page, but they don't move Confidence Score — that's a charger-trust metric. Accessibility is its own dimension and shouldn't bias the trust signal up or down.

Total adjustment is clamped to [−0.10, +0.05]. Reports are signal, not authority.

Draiv uses an availability provider abstraction. Phase 6 shipped two providers and many gaps:

• OCM Directory Status: reads the static StatusType on OCM stations (Operational / Partly Operational / Not Operational / Planned). Confidence 0.35. Never claims stall counts — only operational intent.
• Electrify America stub: returns unknown for all EA stations. Documented in source: EA does not publish a stable public availability endpoint, and we don't scrape private app APIs.
• Other networks (EVgo, ChargePoint, Tesla, etc.): unknown until a real provider is added.
• The Trust Layer card honestly says "Not connected" when availability is unknown. No spoofing.