Closed-loop. Queryable. Replayable. The substrate that preserves institutional reasoning across model evolution, partner transitions, and fund cycles.
FIG. 01 ── REASONING DISPERSED ACROSS SUBSTRATES · MEMORY EXITS WITH PEOPLE
The IC made a decision in 2022. The reasoning lived in a partner's head, in a Slack thread, in a calendar invite, in three spreadsheet revisions, in a 90-minute meeting recording. Today, no one can reconstruct why. They see only what was decided.
And funds lose institutional memory the moment a partner leaves. Rationale, thesis lineage, weighting logic, failed-pattern memory — all of it walks out the door. Most funds rebuild from scratch every 5 years.
Irreproducible cognition and institutional memory decay are the same problem at different time scales.
FIG. 02 ── BOLT-ON FRAGMENTATION → QUERYABLE SUBSTRATE
SaaS / AI bolt-ons
Fragmented institutional reasoning into applications. Each tool owns a slice. AI bolted on top amplifies the fragmentation — non-deterministic outputs on non-replayable foundations.
Substrate
Recombines reasoning into queryable operational state. Closed-loop. Constrained. Provenance-tracked. Methodology becomes executable infrastructure.
Era 1
Pre-2010s
Decisions lived in partner heads.
Era 2
2010s — early 2020s
Decisions fragmented across applications.
Era 3
Now
Reasoning becomes operational, queryable, replayable.
Most firms will add AI to Era 2. Very few will cross into Era 3.
Generic AI has none of these. Substrate-grade architecture has all five — architecturally, not declaratively.
01
by default
Every action audit-emitted. Sole-emission per tier. Never-deletable critical events.
02
code-path absence
Architectural commitments encoded as the absence of code paths. Refusal events emit severity:critical.
03
claim ↔ verification
Every abstract claim ships paired with a concrete operational mechanism. Policy without architecture is rejected.
04
M · S · Model
Methodology · Substrate · Model never blurred. IP boundaries enforced. L0 universal · L1 per-domain · L2 per-customer.
05
recursive (I15)
Every commitment the substrate enforces on users applies recursively to our own operations. The substrate has no outside.
FIG. 05 ── ELEVEN AGENTS · ONE AUDIT CHAIN · OVERLAY THAT TRAVELS
Fund layer · 11 agents · two classes
Six customer-operational specialists (Sourcing · Diligence · Investment Principal · Founder Support · LP Reporter · Portfolio Strategist) + five architectural agents (Audit Curator · L8-Enforcer · Registry Maintainer · Eval Suite Runner · Version Controller).
Audit chain · append-only · hash-chained
Sole-emission per tier. Never-deletable critical retention. Replayable. Schema-migration-safe. Article 9 SFDR audit defensibility by architecture.
Methodology overlay · versioned YAML
Your scoring, weights, voice tiers, IAC composition — encoded. Your IP (L2). Travels.
Five questions only a queryable methodological substrate can answer. Every result carries lineage. Replayable. LP-defensible.
Which IC theses, replayed against today's evidence-base, would still pass thesis-fit at threshold?
Which sub-thesis areas correlate with Year-2 retention across the portfolio?
Which impact-KPI movements lack updated evidence-base citations >90 days old?
Which founder-coaching interventions correlated with positive Year-3 outcomes?
Which override patterns from the partners suggest the overlay should be amended?
Every answer carries lineage to the underlying portco substrate. Every claim re-verified on access. Every score evidence-mode labeled.
Time
≤30 min
Decision turnaround on initial screen. Partner time redirected from triage to founder work, fundraising, IC.
Defensibility
100%
Audit chain coverage on impact-KPI movements. Article 9 SFDR ready by architecture. Past decisions replay against past methodology.
Governance
L8 floors
Architectural commitments encoded as code-path absence. Tier caps · reversibility by design · refusal under low confidence. Constrains future decisions, not just explains past.
Intelligence
κ ≥ 0.85
Cohen's κ weighted agreement target between system and partners. Counter-evidence search at 0.85–0.90 confidence band. Cross-portfolio patterns above k ≥ 8.
Continuity
Yours
Methodology survives partner transitions, model upgrades, vendor migrations. Customer overlay (L2) is YAML. Versioned. Portable. Owned.
Leverage
Compounds
Every replay improves calibration. Every decision deepens the substrate. Substrate scales without adding partners.
Defensibility explains past decisions. Governance shapes future ones. That distinction is the moat.
§ 08.1 ── Impact / SFDR Article 9
LiveFirst production overlay live. Hexframe-class scoring, evidence-base integrations (PubMed · ClinicalTrials · NICE · Cochrane), four-tier LP voice calibration, Article 9 SFDR audit chain by default. Impact Advisory Committee workflows encoded.
§ 08.2 ── Climate / Energy Transition
In developmentOverlay in design. IPCC + ecoinvent + PCAF + SBTi integrations. TRL 1–9 scoring per technology family. EU CSRD / Taxonomy / IRA / CBAM drift tracking. Looking for our first Climate engagement to co-design v1.0.
§ 08.3 ── Healthtech / Biotech
In developmentOverlay in design. ClinicalTrials + PubMed + NICE + FDA pathway tracking. SaMD-specific regulatory tracking. CMS, AMNOG, HAS reimbursement modeling. Mechanism-of-action documentation rigor. Looking for our first Healthtech engagement.
The substrate architecture is genuinely domain-agnostic at the L0 layer. L1 derives per-domain. L2 per-customer. Specialist VC funds are our wedge — and the proof that the platform thesis works. The same substrate generalizes to specialist law firms, research labs, creative agencies, medical practices.
FIG. 10 ── PHASES · KPI GATES · TERMINABLE AT EACH BOUNDARY
Phase 0
10 working days
Free
Framework + PoC
Phase 1
10 weeks
Tier-1 production build · KPI-gated · Tier promotion
Phase 2
Monthly retainer
Operate + recalibrate + internal owner transition
Phase 3
Quarterly residency
Optional
Evolve · new workflows · new sub-thesis areas
§ 11 ── Continuity
The substrate travels.
Across partner transitions. Across model upgrades. Across vendor migrations. Across fund cycles. Across customers. Your methodology is YAML — versioned, portable, yours.
The model changes.
The substrate remembers.
── Fund AI OS · 2026
§ End ── Engage