evidence-score

name	evidence-score
version	1.0.0
description	Ranks a prospect list or account list by evidence-backed outbound readiness. Scores each record 0-100 across five dimensions — fit (ICP match), timing (recent triggers), influence (decision-maker proximity), engagement (public signal on the offer's topic), and reachability (contact data confidence). Produces a ranked list plus one-line justifications tied to specific tool calls. Use when the user has a large output from build-prospect-list / account-research and wants to prioritize who to contact first. Proactively invoke when a list is > 50 leads and the user is about to start outreach.
benefits-from	["richapi-gtm","build-prospect-list","account-research","enrich-and-verify","pre-meeting-briefing","sequence-builder"]
allowed-tools	["Bash","Read","Write","Edit","AskUserQuestion"]
triggers	["score these leads","rank by signal","who should I contact first","hottest leads","prioritize this list","evidence-based ranking"]

evidence-score

A scoring engine. Takes whatever upstream produced — a prospect list, an account map, an enriched CSV — and ranks it by evidence-backed outbound readiness. Every score point is justifiable with a pointer to the tool call that produced it.

Preamble

~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-preflight

Phase 0 — what the user is scoring

Ask once:

• Records to score: prospects (people) or accounts (companies)?
• The offer: one sentence. The score weights depend on what "fit" means for this offer.
• Target volume: "give me the top 25" or "rank everything"? Affects what to surface.

If the upstream data is missing the signals evidence-score needs (e.g. no profile activity pulled), say so upfront and offer to enrich first.

Phase 1 — the five dimensions

Each record gets five sub-scores, 0-20 each, summing to 0-100.

Dimension	What it measures	Signals consumed
Fit (0-20)	Does this record match the stated ICP?	seniority, function, industry, companySize, geography from enrichment
Timing (0-20)	Is there a recent trigger that makes now a better moment than last month?	recentlyChangedJobs, recent posts via profile_activities, funding news, hiring surge, tech-stack additions
Influence (0-20)	Can this person actually buy or block?	Title + seniority + function; for accounts, the ratio of present decision-makers
Engagement (0-20)	Have they publicly engaged with the offer's topic?	post_keyword_search hits, profile_activities comments on related posts, content they've posted themselves
Reachability (0-20)	Do we have the data to actually contact them?	Email status (valid > risky > unknown > not-found), phone, LinkedIn URL present

Total = Fit + Timing + Influence + Engagement + Reachability, capped at 100.

Phase 2 — per-dimension scoring rules

Fit (0-20)

Signal	Points
Seniority exactly matches target band	6
Seniority adjacent to target	3
Function matches target	5
Industry matches target	4
Company size in target band	3
Geography match	2

Cap at 20. Missing signals → 0 for that sub-component (not penalized).

Timing (0-20)

Signal	Points	How to detect
Job change in last 90 days	10	enrich_profile.currentPositions[0].startedOn or recentlyChangedJobs flag
Company hiring surge (5+ open roles in target function)	6	Manual or via web_scrape on careers page (user tells us)
Funding event in last 180 days	6	From enrich_company headlines / post_keyword_search
Exec post referencing a pain word in the offer's domain	5	profile_activities + keyword match
Company posted on a relevant topic in last 30 days	3	post_keyword_search scoped to fromCompany

Influence (0-20)

Title/seniority band	Points
CXO / Founder / Owner	20
VP / SVP	16
Director	12
Senior IC	8
IC / Junior	4

Adjustment: +4 if function exactly matches where the offer is used (e.g. selling a sales tool → +4 for sales titles vs other VPs).

Engagement (0-20)

Signal	Points
Authored a post on the offer's topic in last 60 days	10
Commented on a related post	6
Reacted to a related post	2
Follows competitor / category account	2

Cap at 20.

Reachability (0-20)

Signal	Points
Valid work email confirmed (email_verifier = valid)	12
Risky / catch-all email	6
Personal email only	4
No email	0
Phone number present (normalized)	4
LinkedIn URL present (enables multi-channel)	4

Phase 3 — enrichment gap analysis

If the list doesn't have the data for full scoring, call it out. Offer to enrich the minimum required fields first:

Missing signal	Fill with	Cost per record
Email status	verify_emails batch	0.5
Current position + seniority	enrich_profiles_bulk	1
Recent activity (posts / comments)	profile_activities	2
Company hiring / posts	enrich_company + post_keyword_search	1 + 6 (per company, not per person)

A full-fidelity score for a 100-lead list typically costs 150-300 credits of enrichment on top of whatever produced the list. Worth it for high-ACV segments; skip for mass-market lists.

Phase 4 — score the records

Process the list. For each record:

1. Collect the signals. If a dimension has zero signals available, set that dimension to 0 (don't guess).
2. Compute the five sub-scores.
3. Sum to total.
4. Record, per dimension, the single most load-bearing signal that drove that sub-score — this is the "evidence" the user sees.

Phase 5 — output

For ≤ 25 records: full inline table

Rank  Score  Name              Company          Why
   1     91  Jane Smith        Acme Corp        Director; job change 45d ago; posted about our exact pain topic last week
   2     87  Alex Chen         Beta Industries  VP Eng; company hired 8 engs in 30d; valid work email
   3     84  Maria Kim         Gamma SaaS       CXO; commented on related post; HQ in target geo
  ...

For > 25 records: top 25 inline + CSV

Write the full scored list to ./evidence-scored-<YYYYMMDD>.csv with columns:

rank, score, fit, timing, influence, engagement, reachability,
firstName, lastName, company, title, linkedinUrl, email, email_status,
why_fit, why_timing, why_influence, why_engagement, why_reachability

The why_* columns are one-line evidence pointers. This is what makes the score auditable — the user can see why #3 beat #4.

Segmentation bands

Band	Range	Recommended action
Priority 1 — Hot	80-100	Contact this week. Multi-channel (email + LinkedIn). Custom opener.
Priority 2 — Warm	60-79	Next sprint. Default sequence.
Priority 3 — Watch	40-59	Add to a nurture / content drip. Don't hard-sell.
Below threshold	< 40	Drop or revisit next quarter.

Report band counts at the top so the user knows where to focus:

Scored 247 records
  Priority 1 (80+):   19 records
  Priority 2 (60-79): 88 records
  Priority 3 (40-59): 104 records
  Below threshold:    36 records

Phase 6 — handoff

Once scored, the user almost always needs one of:

• "Give me a sequence for the Priority 1 batch" → sequence-builder
• "Brief me on the top 3 before my meetings" → pre-meeting-briefing
• "Push Priority 1 into my CRM as a segment" → crm-sync-expert
• "Re-score this list weekly" → scheduled-workflow

Always offer the first two — they're the natural next actions.

Anti-patterns

• Scoring a list you haven't enriched. You'll get 0s on three dimensions and the ranking is useless.
• Hiding the why_* evidence columns. Users should be able to challenge a score; that requires seeing the signal.
• Double-counting: if recentlyChangedJobs gave Timing points, don't also award it to Influence.
• Scoring LinkedIn Sales Nav outputs with a one-size ICP. Weight dimensions to the offer — e.g. Influence matters more when selling to exec buyers than when selling to end-user ICs.
• Acting on score without a reachability floor — the #1 scored lead with email_status: not_found is unactionable. Filter for reachability > 8 before surfacing as "Priority 1".
• Re-scoring without new enrichment. Scores are a function of signals; if signals didn't change, the score doesn't change — save the compute.

Customizing the weights

The default weights assume a typical B2B SaaS offer. Users selling into specific segments can override:

• High-ACV enterprise → Influence × 1.5, Fit × 1.2. Reachability matters less (AEs can dig).
• Velocity SMB outbound → Reachability × 1.5, Timing × 1.2. Influence matters less.
• Developer tool to ICs → Influence × 0.5 (ICs buy themselves), Engagement × 1.5.
• Trigger-driven outbound (e.g. just-raised startups) → Timing × 2.0.

If the user hasn't said, ask once: "Any weight adjustments, or use the default weighting?"

Cost

This skill reads upstream data; it doesn't call paid tools itself. If data gaps require enrichment to score accurately, this skill surfaces the gaps but hands off to enrich-and-verify for the actual spending.