<quick_start> Trigger: "SDR call coaching", "coaching cards", "score SDR calls", "review Edgar's/Vasil's calls" — or the Friday ~11 AM CST scheduled run.

Flow: Gates (Nooks + Clari read-only, technical-accuracy, privacy/scope) → Step 1 pull coachable calls per SDR (Nooks connects >=30s with recording; Clari external-only) → Step 2 score each deep-scored call on the 4-dimension rubric (cap 6 per SDR) → Step 3 build per-SDR coaching cards (2 wins, 2 fixes, spec correction, drill, best-call callout) → Step 4 week-over-week trend via outputs/sdr_coaching_prev.json → Step 5 team roll-up → deliver (HTML, Gmail draft to Tim, Slack DM to Tim). </quick_start>

<success_criteria>

Every deep-scored call has four 0-3 scores with one-line justifications + short verbatim quotes; composite normalized 0-100 with GREEN/YELLOW/RED badge; spec-accuracy 0-1 force-flags "SPEC FIX".
Spec claims judged only against live-verified facts (search_product_catalog / search_product_knowledge); unverifiable moments marked "unverified — flag for Tim," never called errors.
Each SDR card has: header stats + trend arrow, per-call rows with recording links, TOP 2 WINS + TOP 2 FIXES with quotes, one spec correction if any, ONE DRILL, best-call callout, live-review candidates.
Vasil zero-coachable weeks render "NEW · ramping" — never zeros or RED; holiday weeks post a short no-connects DM and skip the draft.
Privacy held: output goes only to Tim (Slack DM + Gmail draft); nothing sent to SDRs, Victor, HubSpot, or prospects. Only allowed writes: create_draft, slack_send_message, local HTML + sdr_coaching_prev.json. </success_criteria>

<core_content>

sdr-call-coaching