Partner-Fit Engine Spec

Version: v0.2 Status: Review Owners: Foundation Last Updated: 2026-03-25 Change Type: additive Migration Required: yes Depends On: Hermophylax profiles, OpportunityCandidate, Semantic Airgap Supersedes: none Superseded By: none

Purpose

Rank the best startup or partner candidates for enterprise needs, meeting action items, and strategic opportunities while keeping the confidential join local.

Problem Statement

Foundation-V2 already has partner profiling, opportunity candidates, contradictions, and governed briefs, but it does not yet have a first-class engine that turns those into scored partner recommendations.

Ownership

• Foundation: scoring engine, fit objects, recommendation workflow, final join
• FMIE: upstream meeting/media primitives and provenance where needed
• Cortex: tuning recommendations and quality monitoring only
• Frontend: ranking, rationale, confidence, and approval UX

Inputs

• CorporationProfile
• startup and entity KB records
• OpportunityCandidate
• ClaimCluster
• meeting and action-item objects
• IntelligenceBrief
• external public trend imports
• abstracted external reasoning on surrogate IDs

Core Objects

NeedVector

Represents the enterprise need state. Fields: - need_id - corp_id - domain - priority_score - capability_gaps - sector_targets - geo_constraints - regulatory_constraints - timeline_constraints - evidence_refs

CapabilityVector

Represents startup/partner capability state. Fields: - candidate_id - entity_id - sector_strengths - capability_strengths - stage - deployment_regions - integration_patterns - traction_signals - risk_flags - evidence_refs

ConstraintVector

Represents exclusions and friction. Fields: - constraint_id - must_have - must_not_have - commercial_constraints - regulatory_constraints - disclosure_constraints

PartnerCandidateScore

Fields: - candidate_id - need_id - strategic_alignment_score - technical_fit_score - maturity_fit_score - deployment_fit_score - geo_regulatory_fit_score - relationship_access_score - timing_score - contradiction_penalty - confidence_score - disclosure_risk_score - final_score

PartnerRecommendation

Fields: - recommendation_id - corp_id - top_candidates - decision_summary - key_assumptions - key_blockers - diligence_next_steps - evidence_trace_refs

FitRationale

Human-readable explanation per candidate.

EvidenceTrace

Links scores back to claims, opportunities, meetings, and profile evidence.

Scoring Dimensions

• strategic alignment
• technical fit
• maturity fit
• deployment fit
• geographic / regulatory fit
• relationship access
• timing / why-now fit
• contradiction penalty
• evidence confidence
• disclosure risk

Execution Model

1. build NeedVector from corp profile, meetings, opportunities, and briefs
2. build CapabilityVector for candidate startups/partners
3. compute local candidate ranking
4. optionally invoke external abstract reasoning on surrogate vectors only
5. fuse local score and trusted external reasoning
6. resolve real candidate identities locally
7. emit governed recommendation set

Use Cases

• KDDI meeting action items to portco/startup partner shortlist
• Tokyo Gas style external-collaboration strategy
• enterprise domain brief to startup ecosystem mapping
• strategic partner discovery with confidentiality preserved

Telemetry

• recommendation count
• score distribution by domain
• contradiction penalty rate
• top candidate churn over time
• external reasoning usage rate
• trust score distribution

Foundation-V2 Gap Closed

Turns existing profiles, claims, opportunities, and watchlists into a native recommendation system instead of an ad hoc prompt workflow.

Open Questions

• whether candidate ranking is persisted per account as part of the Strategy Twin
• how much of fit rationale can be externalized under abstract_reasoning_only
• whether relationship access is manual CRM data, inferred data, or both

Change Log

• v0.2: expanded to implementation-grade contract
• v0.1: initial seed