gtmskills.com
SkillsCompaniesGitHubContribute
Warm intro intelligence

Reference file

Evidence Rules and Confidence Scoring

evidence-rules-and-confidence-scoring.md

Evidence Rules and Confidence Scoring

Sub-page of the Warm Intro Intelligence skill. Load when executing Step 5 - assessing edge confidence for each relationship path found.

Edge Confidence Tiers

Verified

Direct, dated, logged evidence of two-way interaction between the bridge and the target contact. Any one of the following is sufficient:

  • CRM meeting or call logged with the target contact - but ONLY if it passes the Meeting Substance Test below. A meeting record alone is not Verified.
  • CRM email thread where both the bridge and target contact appear, with at least one reply from the target contact's email address. A one-way send with no reply is NOT Verified (see Email Substance Test).
  • LinkedIn DM or InMail thread with at least one response from the target contact (accessed via the bridge's connected LinkedIn account).
  • Explicit note in the CRM or account memory naming both the bridge and the target contact with specific relationship context. Must reference the relationship directly - generic CRM notes and auto-generated placeholders do not qualify.
  • Warm Relationship Intake flag confirmed by the internal bridge, with intro_willingness of "strong" or "possible" and source of "intake" or "inferred+confirmed".

Always record the date of the most recent evidence item as the last interaction date.

Evidence Integrity Rules (apply before assigning ANY tier)

These rules exist because a record existing is not the same as a relationship existing. A meeting count, an email count, or a title is metadata - not evidence. Violating these produces false-Strong paths that waste rep trust.

Rule 1 - Score from content, never metadata

Before assigning any confidence tier, open the record and read its actual content. Never score from the existence of a record, a count of meetings/emails, or a title alone. If you have not read what the interaction actually contained, you cannot assign Verified.

Rule 2 - Meeting/Call Substance Test

A CRM meeting or call qualifies as Verified only if ALL of the following are true:

  • It has an assigned owner, AND
  • It has a substantive note, body, agenda, or outcome (NOT an auto-generated placeholder such as "left no additional comments," "no notes," or an empty body), AND
  • The title indicates a real interaction (NOT a bare "Event Meeting," "Badge Scan," "List Import," or similar capture label).

A meeting record that is title-only, has a placeholder/empty body, or has no owner is treated as Moderate at most, labeled: "event capture - interaction depth unconfirmed (likely badge scan or lead grab)." It is NOT Verified.

Rule 3 - Email Substance Test

  • A two-way thread with at least one genuine reply from the target contact -> may be Verified.
  • A one-way outbound send (sequence step, single email, no reply) -> Moderate at most, never Verified. A logged email existing is not a relationship.
  • Auto-replies, out-of-office, and bounce messages do NOT count as a reply.

Rule 4 - Strong-Tier Verification Gate

Before labeling any path Strong, run a final check: does the underlying record contain a confirmed two-way interaction that passed the substance tests above? If you cannot point to the specific content that confirms a real interaction, downgrade to Medium. When genuinely uncertain between two tiers, choose the LOWER one and state what would confirm the higher tier. Honest under-labeling protects rep trust; false-Strong destroys it.

Moderate

Credible signal of a relationship, but depth is unknown. Any one of the following is sufficient:

  • LinkedIn 1st-degree connection between bridge and target contact - confirmed via Sales Nav or connected LinkedIn account. No message history required, but none visible.
  • Shared employment at the same company during overlapping years, AND their roles are at seniority levels plausibly in contact. Rule: any overlap at a company under 100 employees qualifies; VP-level or above at a larger company qualifies. Junior roles at large companies do not qualify without additional evidence.
  • Both tagged in the same event in account memory (co-presence confirmed, interaction not confirmed).
  • One-way CRM email outreach from the bridge to the target contact with no reply logged.
  • Confirmed shared investor: a firm in {{INVESTOR_REGISTRY}} also invested in the prospect company. Verifiable overlap; VCs routinely facilitate portfolio-company intros. Firm-level - specific partner relationship not yet identified.
  • Unconfirmed Warm Relationship Intake flag where source is "inferred" and not yet confirmed by the bridge.

Inferred

A pattern suggests a possible relationship. No direct evidence of person-to-person interaction. Any one of the following is sufficient:

  • 2nd-degree LinkedIn connection (bridge is connected to someone who is connected to the target contact).
  • Company-level commercial relationship between one of your customers or partners and the prospect company - without any person-specific evidence linking specific individuals.
  • Public co-mention (both names appear in the same press article, conference panel recording, podcast, or content piece).

Champion Migration Edge Confidence

A champion-migration edge (a pool member who now works at the target account) inherits the confidence of the ORIGINAL relationship formed at the prior company. It never gets a confidence boost for having moved into the target.

  • Logged meeting or call, or a replied email thread, at the prior company -> Verified
  • 1st-degree connection, shared employment only, or other Moderate-tier signal -> Moderate
  • Pattern only, no logged interaction -> Inferred

Mandatory on every champion-migration edge:

  • Migration flag: "[Verified/Moderate] at [prior company] - [contact] moved to [target] as [current role]."
  • Recency governs, not the move date. Apply the standard recency weighting to the LAST INTERACTION date, not the date the contact joined the target. A relationship that has gone quiet for 18+ months is stale even if the migration is recent.

When multiple champion-migration hits exist at the same target, each is a separate candidate path and is ranked by the standard Path Ranking logic below - confidence first, then recency, then persona priority, then hops. There is no separate "which champion wins" rule.

Path Confidence

Path confidence is determined by the weakest edge confidence in the path:

  • Every edge is Verified -> path confidence: Strong
  • Weakest edge is Moderate, no edge below Moderate -> path confidence: Medium
  • Any edge is Inferred -> path confidence: Low

Edge labels (Verified / Moderate / Inferred) and path labels (Strong / Medium / Low) are always shown separately. Never combine them or use one label for both levels.

Required Fields Per Edge

Every edge in every output path must include all of the following. Do not omit any field - if a value is unknown, state that explicitly.

  • Confidence: Verified / Moderate / Inferred
  • Evidence: what was found, which system it came from, specific details (meeting count, email count, date range, etc.)
  • Last verified / last interaction date: the most recent date associated with the evidence. If unknown, state "date unknown."
  • Unknown: the specific gap that limits this edge's confidence. Be precise - "no direct communication found" is better than "relationship depth unknown."
  • Why worth checking: why the rep might still act on this path despite the limitation. Only include if there is a genuine reason - do not pad this field.
  • What would increase confidence: the specific signal or action that could upgrade this edge. Examples: "[the bridge rep] confirming the relationship directly," "a CRM meeting logged with this contact in the last 12 months," "a LinkedIn reply thread confirmed via Sales Nav."

Weak Link Rule

For every path, the weakest edge is the confidence-limiting edge. Always name it explicitly in the path summary so the rep can identify the gap without reading all edge details.

Format: "Weak link: [Edge description] - [one sentence explaining the gap]" Example: "Weak link: Customer Contact -> Prospect VP - company-level relationship only, no person-to-person connection confirmed."

Path Ranking

  1. Sort by path confidence: Strong -> Medium -> Low.
  2. Within the same confidence tier, apply recency weighting:
    • Last interaction within 6 months -> full weight
    • Last interaction 6-18 months ago -> reduced weight (rank lower within tier)
    • Last interaction 18+ months ago -> lowest weight (rank last within tier, flag as potentially stale)
    • Last interaction date unknown -> treat as 18+ months for ranking purposes
  3. Within the same confidence tier and recency band, sort by target persona priority (per {{PERSONA_MAP}}): a path reaching a higher-priority buyer persona ranks above one reaching a lower-priority or non-buying persona.
  4. Within the same confidence tier, recency band, and persona priority, sort by path length: 1-hop before 2-hop.
  5. Return the top 2 paths.
  6. Prefer paths targeting different personas. If both top paths target the same persona, note it explicitly.

Persona priority is a ranking axis, not a filter. A known contact who moved into the target in a role outside the buying committee still surfaces - they can intro internally - but ranks below an in-persona path. Never discard a valid path solely because the persona is non-ideal. Confidence and recency always outweigh persona: a high-confidence, recent relationship with a slightly-off persona beats a stale relationship with the perfect persona, because an intro that will actually happen is worth more than a cold path to the ideal contact.

No-Path Case

If no paths exist at any confidence level, do not return an empty output. State: "No relationship signals found between your company's network and this prospect. The paths below are the best available." Then present whatever was found, fully labeled.

If genuinely nothing was found - no bridges, no connections, no overlapping employment: state that clearly and include the gap-triggered intake prompt from the Warm Relationship Intake sub-page.