← AI-Native VC & PE Funds·CASE 003·IMPACTMAYANK TALWAR·design partner slot
§ 00 ── MASAWA IMPACT FUND I · ARTICLE 9 · HEXFRAME + NURTURECAPITAL

AI Operating System for Masawa. So your methodology survives scale.

An operating substrate for the €40M Article 9 mental wellness mandate — Hexframe and NurtureCapital encoded so the alpha outlives the operator.

Prepared for · Sabine Fléchet · 2026-05-20

§ 01 ── THE OPPORTUNITY FOR MASAWA

Masawa compounds because Hexframe and NurtureCapital are the alpha. When the partner steps back, when the model churns, when Fund II opens — that alpha walks out unless it's encoded.

01
PARTNER DEPARTS

Hexframe weighting, Care-to-Dare patterns, NurtureCapital specifics — none of it is in the memos. It was in working memory.

02
MODEL CHURNS

The Anthropic API shipped six months ago is not the API shipping tomorrow. Carefully tuned prompts age. Fine-tunes go stale.

03
FUND CYCLE

2026 IC decisions produce outcomes legible in 2031–2033. By the time outcomes arrive, the original logic is five years old.

€40M FUND · 20–25 PORTCOS · TWO GPS · NOT A HIRING PROBLEM
CAPABILITY
HIRE
GENERIC AI
MASAWA AI OS
Inbound triage at €40M-fund pace
~
Hexframe (P × P × P) auto-scoring · κ ≥ 0.85
Mental wellness scientific evidence validation
Article 9 SFDR audit chain on impact KPIs
Founder wellbeing signal monitoring (privacy-bounded)
Care-to-Dare coaching pattern surfacing
Soft Landing trigger detection
LP-grade impact letters from sourced data
~
Knowledge persists when partner leaves
Code-path L8 floors (NEVER policy)
THE ANTHROPIC CLAUSE

The only AI category where defensibility increases with regulation. Article 9 SFDR · DNSH · PAI · 7-year forensic retention, indefinite for severity:critical · BaFin / ESMA inspection-ready. Tacit methodology falls over the fund cycle. Encoded substrate rises against the same clock.

§ 02 ── FIVE DISCIPLINES · ZERO EXCEPTIONS

Generic AI has none. Substrate-grade architecture has all five — architecturally, not declaratively — and all five apply to Masawa from Day 1.

  1. 01
    Audit chain by default

    Every action emitted via sole-emission Audit Curator. Append-only. Hash-chained. NEVER-deletable for severity:critical.

  2. 02
    L8 architectural floor

    Commitments encoded as code-path absence. L8-Enforcer continuously surveils five categories (six with Amendment 002). Self-test 5/5 PASS quarterly.

  3. 03
    Tether-pair discipline

    Every abstract claim ships paired with a concrete operational mechanism. Policy without architecture is refused at authoring time.

  4. 04
    3-layer ontology

    Methodology · Substrate · Model never blurred. L0 / L1 / L2 IP boundaries enforced contractually and architecturally.

  5. 05
    Substrate-mirroring · I15

    Every commitment Queryable Fund enforces on Masawa applies recursively to Queryable Fund's own operations. The substrate has no outside.

WITHOUT ALL FIVE · "SUBSTRATE" IS MARKETING. WITH ALL FIVE · IT IS WHAT THE WORD CLAIMS.

§ 03 ── THE ARCHITECTURE IS THE PRODUCT

Eleven agents · two classes · one audit chain — at three levels of substrate-mirroring nesting.

What Masawa gets on Day 1 of Phase 1 — eleven agents operating under one audit chain, with the Hexframe methodology overlay held as your L2 IP.

SYSTEM OF SYSTEMS · L0 / L1 / L2
L0
Universal Agentic Substrate Framework v1.0

Audit chain · L8 floor · Tether-pair · 3-layer ontology · Substrate-mirroring (I15). Operator-tier · open methodology.

L1
Queryable Fund · Specialist VC framework

Six customer-operational + five architectural agents. Operator IP · travels across customers in the segment.

L2
Masawa fund tenant

Methodology overlay v1.0 · Hexframe-class scorer · NurtureCapital framework · LP voice register · Article 9 DNSH/PAI · 11 agents · your namespace.

L2′
Portco tenants · per portco

Founder-consented · parallel 11-agent structure · founder namespace · 24h revocation halt · 72h aggregate purge · k ≥ 8 anonymity floor between tenants.

PATTERNS GRADUATE L2 → L1 → L0 · ATTRIBUTION + COMPENSATION · MSA § 8 · TRIPLE-SIGNED CONCURRENCE

ELEVEN AGENTS · TWO CLASSES
CLASS A · CUSTOMER-OPERATIONAL · L3 CAP
  1. 01
    Sourcing Analyst
    T3 Operator · inbound triage · founder discovery
  2. 02
    Diligence & Impact Lead
    T3 Operator · Hexframe scoring · evidence integration
  3. 03
    Investment Principal
    T2 Drafter · L8 CAP (F1) · IC memo synthesis · IAC commits
  4. 04
    Founder Support
    T2 Drafter · L8 CAP (F2) · wellbeing extraction · Care-to-Dare
  5. 05
    LP Reporter
    T3 Operator · per-tier voice · Article 9 audit chain
  6. 06
    Portfolio Strategist
    T3 Operator · cross-portfolio matching · k ≥ 8 federation
CLASS B · ARCHITECTURAL · T4 STEWARDS · do NOT count against L3 cap
  1. 01
    Audit Curator
    Sole-emission authority · chain integrity by construction
  2. 02
    L8-Enforcer
    Continuous surveillance · 5/5 PASS quarterly · 6/6 pending Amendment 002
  3. 03
    Registry Maintainer
    Weekly sweep · schema validation · 4-surface alignment · orphan detection
  4. 04
    Eval Suite Runner
    κ ≥ 0.85 verification · halt authority on κ drift · weekly + quarterly
  5. 05
    Version Controller
    MAJOR / MINOR / PATCH cycle · pre-snapshot · IAC concurrence routing

TIER CEILINGS ON INVESTMENT PRINCIPAL + FOUNDER SUPPORT ARE ARCHITECTURAL — THE AUTONOMOUS-ACTION CODE PATH DOES NOT EXIST.

SIX CAPABILITIES · MENTAL WELLNESS IMPACT · ENCODED
Hexframe (P × P × P) scorer
PPP××

Auto-score every portco against the rubric with evidence binding. Inter-rater κ ≥ 0.85 tracked weekly. Calibration drift flagged via Eval Suite Runner.

Mental wellness evidence validator

Citation-aware reasoning across PubMed · Cochrane · ClinicalTrials.gov · NICE. Replication-status flagging on every claim.

Founder wellbeing signal extractor

Privacy-bounded patterns from consented data. Surfaces Care-to-Dare opportunities. Never diagnoses. Tier-2 hard cap (F2).

Cross-portfolio NurtureCapital matcher

Have we seen this thesis before? What worked, what didn't? Cohort-level pattern recognition. Federation activates at k ≥ 8.

Impact pathway mapper

Theory of change → evidence chain → measurable outcome. Article 9 SFDR-grade traceability on every claim. DNSH + PAI tracking.

LP-grade audit chain

Every signal · every memo · every metric traceable to source. Plain-language explanations of why a signal fired. NEVER-deletable critical.

§ 04 ── INTERROGATE THE PROPOSITION

Sabine — ask it anything. Live.

Type a question. Get an answer with full audit chain.

Ask the substrate the questions Sabine would ask in a Monday IC. Affinity · email · every deck read · every memo written · every impact KPI logged — one model, one conversation, full provenance. The substrate answers in its own voice, the same primitive we'd install on Masawa's methodology.

STREAMING · κ ≥ 0.85 · ANONYMITY HOLDS · CITATIONS BY DEFAULT

↓ Type a question · streaming answer
MASAWA AI OS · LIVE
STREAMING
↵ to send · ⇧↵ for newline
four exemplars from the Tier-1 spec:

ONE MODEL · ONE CONVERSATION · FULL PROVENANCE · EVERY ANSWER LINKS BACK

§ 05 ── TWELVE WEEKS · EIGHT METRICS · FOUR PHASES

Where the partner's week actually goes — baselined W3, measured W12.

Baseline is Masawa today. Target is Week 12 of Phase 1 — Care-to-Dare and founder time preserved, triage and admin collapsed. 20% → 65% partner time on founder + Care-to-Dare work.

EIGHT METRICS · PRE-COMMITTED · ALL 8 PASS = PHASE 2 ENTRY
#
KPI
SOURCE
W12 TARGET
1
Inbound triaged per partner / week
CRM + email log
3 × baseline
2
Time inbound → first-pass decision
Operator log
≤ 30 min
3
Hexframe (P × P × P) scoring time
Operator log
≤ 5 min
4
Hexframe inter-rater reliability (Cohen's κ)
Independent rater pair
≥ 0.85
5
Mental wellness evidence-validation completeness
Operator log + review
≥ 95 %
6
Founder wellbeing signal lead time
Signal log + confirm
≤ 7 days
7
Impact KPI audit chain completeness
Audit log
100 %
8
Partner time on founder + Care-to-Dare
Calendar
+ 50 %

CONDITIONAL PASS = OPTIONAL REMEDIATION SPRINT · NO PENALTY · NO-GO = PHASE 1 ARTIFACTS RETAINED UNCONDITIONALLY

TWELVE WEEKS · EACH WITH A SHIPPED ARTIFACT
FOUR PHASES · ONE LADDER · KPI-GATED · NEVER AUTO-RENEWED
PHASE 0
10 days
FREE
GO / NO-GO Day 10
PHASE 1
10 weeks
KPI-GATED W12
8 metrics PASS
PHASE 2
Monthly retainer
12 mo min · internal owner
Tier 2 / 3 evolution
PHASE 3
Quarterly
RESIDENCY · OPTIONAL
New workflows · Fund II
INSIDE PHASE 2 · CAPABILITY EVOLUTION · KPI-GATED

EACH TIER-UP IS A CONTRACT AMENDMENT — NOT A HAND-WAVE. AUTHORITY ADDED THE WAY IT IS ADDED TO A HUMAN INVESTOR · BY DEMONSTRATED OUTCOMES.

THE MATH AT FUND SCALE · THREE COMPOUNDING VECTORS
01
LABOR LEVERAGE
60+
partner-equivalents

Two partners · three AI surfaces · 10× per-role output. Capacity that would otherwise require 6× current headcount, achieved at €40M-fund scale without expansion.

02
IMPACT-CARRY DEFENSIBILITY
50%
of GP carry, defensible

Impact-tied carry × 25 portcos × audit chains. Carry compensation backed by source-traceable evidence rather than narrative. LP-defensible at Toniic-tier review.

03
FOUNDER BURNOUT PREVENTION
€1–2M
per failure prevented

≤ 7-day signal lead-time × Care-to-Dare intervention. ~10–15% of portcos surfaced before crisis. Even one prevented failure recovers full position cost.

HYPOTHESES H1–H4 IN METHODOLOGY DOCUMENT · FALSIFICATION THRESHOLDS PRE-COMMITTED

§ 06 ── VERIFICATION + IP BOUNDARY

LPs and regulators have standing inspection authority. Trust by architecture, not by policy.

How Masawa's LPs and BaFin verify the substrate without taking our word for it.

D7 § 12 · 5-STEP VERIFICATION
  1. 01 · Adjacent-tier (operator)
    3–5 days notice

    Queryable Fund substrate inspected · anonymized cross-customer · I15 substrate-mirroring verified

  2. 02 · Own-tier (Masawa fund)
    Immediate

    Full visibility · 8 Tier-1 KPIs PASS · κ ≥ 0.85 · sole-emission 100% · critical = 0 unintended

  3. 03 · Sub-tier (portcos)
    Immediate

    Per-portco substrate state with founder participation · cross-tier composition operational

  4. 04 · Cross-tier composition
    3–5 days notice

    Hash continuity + anonymization end-to-end · 5 composition rules · L17/L18 watertight

  5. 05 · L8 floor pressure test
    Immediate

    N/N refusals · NEVER-deletable entries persist after deletion attempts · recursively audit-emitted

INSTITUTIONAL COMPLIANCE · STANDARDS POSTURE
  • GDPR
    EU residency · Frankfurt
  • Article 9 SFDR audit-grade
    DNSH + PAI + 7y + INDEFINITE critical
  • SOC 2-aligned
    trust services criteria mapped
  • ISO 27001:2022-aligned
    controls catalogued
  • EU AI Act
    obligations mapped
  • NIS2
    where applicable
  • DPA + SCCs
    third-country transfer
  • 24h consent revocation halt
    operationally enforced
  • 72h aggregate purge on revoke
    audit chain retains anonymized lineage
  • Right to erasure preserved
    without breaking chain integrity
IP BOUNDARY · L0 / L1 / L2 — CONTRACTUALLY + ARCHITECTURALLY ENFORCED
§ 07 ── THE ASK

Five things from Joshua + Sabine. Trust the discipline · then verify.

  1. 01
    Time commitments · across engagement

    Methodology workshop (Phase 0 W1 D3, 90–120 min) · weekly check-ins · IAC sessions · W12 gate review · partner training (~20 hours over Phase 1).

  2. 02
    IAC concurrence at Phase 1 kickoff

    v2.0 spine ratification · methodology overlay v1.0 baseline · 6 L8 floors + Amendment 002 (F6 cross-tenant isolation) ratification · Phase 1 scope confirmation.

  3. 03
    Methodology overlay v1.0 baseline · Phase 0 W1 D3

    Hexframe-class composite scorer + NurtureCapital framework + Article 9 SFDR mappings + LP voice register captured at the workshop.

  4. 04
    Founder concurrence · Phase 2 onwards

    Founders sign portco-tier consent agreements before per-portco onboarding begins. 24h revocation halt and 72h aggregate purge are architecturally enforced.

  5. 05
    Trust the discipline, then verify · continuous

    D7 § 12 5-step protocol available any time during the engagement · LP cadence quarterly · Tier 2 amendment at Phase 2 month 1 (internal owner trained).

§ END ── CLOSING

The substrate has no outside. Including the operator. Including you.

Article 9 SFDR · €40M · Hexframe + NurtureCapital · 3-portco · 11-agent fund tier · 11-agent per portco · cross-tier composition · NEVER-deletable.

If relevant, I can send a short async questionnaire — no call needed — and build a no-charge synthetic-data Phase 0 PoC "Masawa AI OS". Within two weeks, replay one AI-native portco's evidence chain against Article 6(1) before the notified bodies start citing it.

SMALL · SENIOR · SPECIALIST