Capability Documentation

Capabilities

1. Cross-domain strategic synthesis High confidence

   Reads investment portfolio context, operational status, governance commitments, and family-system signals together. Produces a single coherent strategic framing rather than siloed analyses.

   Evidence: Trained on the family fund's own institutional voice via the workbench synchronization layer. Synthesizes against the operator's stated strategic priorities, not generic management theory.

2. Family fund priority framing High confidence

   Translates strategic intent into operational priorities at family-fund granularity. Maintains alignment between principal's stated direction and day-to-day decision queue.

   Evidence: Operator's documented strategic priorities + ongoing workbench calibration provide the framing reference.

3. Investment-operational translation Medium confidence

   Connects investment-side language (sleeve allocation, manager selection, IRR targeting) with operational-side language (cash flow, vendor management, family services).

   Evidence: Pattern recognition across institutional family fund operating models; per-tenant calibration sharpens fidelity over the first 30-60 days.

4. Governance pattern recognition Medium confidence

   Identifies recurring governance patterns (decision queue concentration, advisor coordination overhead, succession pacing) and surfaces them for principal review.

   Evidence: Library of family-fund failure modes (see Engagement Blueprints) provides the pattern reference set.

Limitations

1. Autonomous decision execution By design

   Does not execute decisions on the operator's behalf.

   Alternative: Drafts decisions and queues them for principal review. The principal retains all execution authority.

2. External system access Fiduciary

   Does not access banking, brokerage, or external operating systems without explicit per-action operator authorization.

   Alternative: Surfaces required external action with full context; operator (or delegated team member) performs the action.

3. Material decisions without human review Fiduciary

   Cannot finalize material strategic decisions without human-in-the-loop review. Material thresholds defined per family fund (e.g., commitments above 1% of AUM).

   Alternative: Material decisions route through council pressure-test before principal sign-off.

4. Family-system mediation Capability boundary

   Does not mediate active family-system disputes or provide therapeutic guidance.

   Alternative: Family-system tensions surface to the Family Communications Coordinator (per Engagement Blueprints role definitions) or to qualified external family-systems facilitators.

Escalation Protocol

Material decision triggers

Capital commitments above the family fund's defined material threshold · Cross-generational decisions affecting succession or estate planning · Charitable giving decisions above defined annual limit per recipient · Strategic posture changes that contradict the documented quarterly priorities

Escalation path

High-stakes decisions surface to the standing Council for multi-agent pressure-test. The Council's synthesized recommendation routes to the family Principal for final review. Cross-generational items additionally surface to the Family Council Chair.

Timing

Material decisions enter the principal review queue within the same business day they are surfaced. Council pressure-test completes within 48 hours of escalation. Principal review SLA is set per family fund.

Override authority

The family Principal can override any agent recommendation. The Council Chair can override on cross-generational matters. The agent records overrides with reasoning to inform future calibration.

Training Data Notes

The agent's institutional voice is calibrated from the family fund's own operating context — documented priorities, prior decisions, voice patterns, and standing council positions. No cross-tenant training. The workbench cognitive synchronization layer captures and reinforces voice fidelity over the first 30-60 days of operator interaction.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every CEO Equivalent invocation logs: input context, synthesized output, confidence indicators, escalation triggers fired (if any), and operator response (acted on / overrode / declined). Per-tenant audit trail accessible to family fund counsel via consultant master console.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel; ad-hoc review on material capability or limitation change.

Authored by

Condifi Architecture Team

Applicability

Family fund clients deployed under any consultant. Activates once operator passes initial workbench calibration (typically 30-day onboarding).

Capabilities

1. Financial structure analysis High confidence

   Reads the family fund's entity structure, capital flows, and reporting obligations into a coherent financial picture. Identifies structural inefficiencies and compliance pressure points.

   Evidence: Family fund's own entity map + operating financial data. Quarterly cross-reference with external CPA observations.

2. IC memo drafting in institutional voice High confidence

   Drafts investment committee memos that read in the family fund's established institutional voice, integrating the CIO's investment thesis with the financial-structure implications.

   Evidence: Workbench-trained on the family fund's prior IC memos, decision memos, and LP communications.

3. LP reporting cadence tracking High confidence

   Maintains the LP-by-LP reporting calendar, drafts cadence-appropriate communications, and flags upcoming deadlines.

   Evidence: Family fund's LP commitment documents + reporting schedule + prior communication archive.

4. Capital call coordination support Medium confidence

   Drafts capital call letters, sequences LP payment expectations, and tracks call response timelines.

   Evidence: Family fund's standing LP commitment terms + prior capital call patterns.

5. Tax structure modeling Medium confidence

   Models tax implications of structural alternatives (entity restructuring, jurisdictional moves, charitable vehicle structuring). Surfaces material decisions for external tax counsel review.

   Evidence: General tax framework + family fund's existing entity context. Specific applicability requires external tax counsel sign-off.

Limitations

1. Transaction execution Fiduciary

   Does not execute capital transactions of any kind.

   Alternative: Drafts execution instructions; family fund's CFO, treasury team, or external operations partner performs actual transactions with proper dual-control authorization.

2. External banking system access Regulatory

   Does not access bank accounts, brokerage platforms, or treasury systems without explicit per-action operator authorization.

   Alternative: Surfaces required banking action with full context; authorized human performs the action via direct system access.

3. Specific tax advice Regulatory

   Does not provide jurisdictional tax advice that would constitute professional tax counsel.

   Alternative: Tax-significant outputs flag for qualified external tax counsel review before any client-impacting action.

4. Audit opinion substitution Regulatory

   Does not substitute for an independent audit opinion.

   Alternative: Quarterly and annual independent audits remain external. The agent supports audit preparation but cannot issue an opinion.

Escalation Protocol

Material decision triggers

Capital structure changes affecting LP rights or fund governance · Tax positions material to total tax liability (above threshold defined per family fund) · Multi-jurisdictional restructuring proposals · Audit findings of material weakness or significant deficiency

Escalation path

Tax-significant outputs flag for external CPA review before operator action. Capital structure changes surface to the family Principal before any execution. Audit-relevant findings route to the audit committee or family council per the family fund's governance framework.

Timing

Tax-significant items enter the CPA review queue within 24 hours. Capital structure escalations surface to the Principal within the same business day. Material audit findings escalate immediately.

Override authority

The family Principal can override any CFO Equivalent recommendation on strategic financial direction. Tax positions remain subject to external tax counsel sign-off regardless of agent recommendation.

Training Data Notes

Calibrated from the family fund's prior IC memos, LP communications, and financial reporting. Tax modeling references general framework knowledge but is always rechecked against jurisdiction-specific counsel before client-impacting action.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every CFO Equivalent invocation logs: financial data accessed, draft memo content, escalation triggers fired, external counsel routings initiated, and operator action taken. Audit trail preserves the chain from data input through draft output through human review.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + annual review by external CPA on tax-modeling boundaries.

Authored by

Condifi Architecture Team

Applicability

Family fund clients with formal financial reporting requirements. Activates once operator passes initial workbench calibration on financial voice patterns.

Capabilities

1. Investment thesis development High confidence

   Synthesizes a structured investment thesis from sourcing materials, due diligence inputs, and the family fund's standing portfolio context.

   Evidence: Family fund's own portfolio context + library of family-fund-relevant due diligence frameworks. Calibrated against operator's prior commitment decisions.

2. Portfolio drift detection High confidence

   Continuously monitors sleeve allocations against target ranges. Surfaces drift and rebalancing opportunities before they require reactive correction.

   Evidence: Live portfolio data + family fund's documented allocation framework + historical rebalancing patterns.

3. Private markets document synthesis Medium confidence

   Reads PPMs, side letters, and LPA documents to extract structural terms, fee implications, and governance impacts.

   Evidence: Document synthesis across institutional private markets format. Specific clauses always rechecked by external counsel for binding interpretation.

4. Sleeve and concentration analysis High confidence

   Calculates concentration exposure across sleeves, sectors, geographies, and manager relationships. Flags when concentration crosses operator-defined thresholds.

   Evidence: Family fund's portfolio data + operator's stated concentration discipline parameters.

5. Peer family fund signal monitoring Limited confidence

   Tracks publicly observable peer family fund positioning (deployment signals, sector emphasis, manager relationships) to surface coordination opportunities and divergence patterns.

   Evidence: Public signal aggregation only — no private LP-side or counterparty-confidential data. Useful as context, not as basis for action.

Limitations

1. Trade execution Fiduciary

   Does not execute trades, commitments, or capital movements.

   Alternative: Drafts the investment decision; the family fund's investment committee, CFO, or treasury team executes through proper authorized channels.

2. Specific investment advice Regulatory

   Does not provide individualized investment advice that would constitute regulated investment advisory services.

   Alternative: All outputs are analytical synthesis for the family fund's internal use. External investment advisory relationships remain with registered investment advisors.

3. Past performance prediction By design

   Past performance synthesis does not predict future investment returns.

   Alternative: Performance analysis frames historical context; forward-looking decisions are surfaced for human judgment.

4. Confidential counterparty data Fiduciary

   Does not access non-public LP-side or counterparty-confidential data.

   Alternative: Analysis based on family fund's own data + publicly observable signals. Counterparty-confidential context surfaces through direct relationship channels.

Escalation Protocol

Material decision triggers

Commitment amount above family fund's defined material threshold · Concentration crossing operator-defined sleeve, manager, or geography limits · Investment thesis with material divergence from family fund's standing strategic posture · Manager-level governance changes affecting fiduciary duty

Escalation path

Investment decisions surface to the family fund's investment committee for formal review. Concentration risk patterns flag for principal review. Manager-level governance changes route to General Counsel and the IC simultaneously.

Timing

Material concentration patterns surface within same business day. Standard investment thesis review enters IC queue per IC cadence (typically bi-weekly). Manager governance changes escalate immediately.

Override authority

The Investment Committee has primary authority on investment decisions. The family Principal can override IC recommendations. The agent records overrides for future calibration.

Training Data Notes

Calibrated from the family fund's prior commitments, IC memos, rebalancing decisions, and concentration framework. Document synthesis trained against generic institutional private markets format; specific binding interpretation always requires counsel.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every CIO Equivalent invocation logs: portfolio data accessed, source documents reviewed, draft thesis content, concentration calculations, escalation triggers fired, and IC routing initiated. Full chain auditable from data input to IC decision.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel; annual review by family fund's external investment advisor on capability boundaries.

Authored by

Condifi Architecture Team

Applicability

Family fund clients with active investment programs. Activates once operator passes investment-voice calibration in the workbench.

Capabilities

1. Risk pattern detection High confidence

   Continuously monitors portfolio and operational signals for emerging risk patterns. Surfaces patterns at the convergence of multiple signals before they require reactive response.

   Evidence: Pattern library from family-fund failure modes (see Engagement Blueprints) + per-tenant signal aggregation. Conservative confidence calibration to minimize false positives.

2. Compliance posture monitoring Medium confidence

   Tracks the family fund's posture against regulatory commitments and internal compliance framework. Flags drift between stated policy and observed operations.

   Evidence: Family fund's documented compliance policies + observed operational patterns. Specific compliance opinions remain external counsel territory.

3. Cybersecurity threat landscape synthesis Medium confidence

   Aggregates threat intelligence relevant to family fund operations. Surfaces threats in priority order with family-fund-specific impact framing.

   Evidence: Public threat intelligence + sector-specific reporting. Augments family fund's existing security infrastructure rather than replacing it.

4. Regulatory change tracking Medium confidence

   Monitors regulatory changes affecting family fund operations across applicable jurisdictions. Surfaces material changes for counsel review.

   Evidence: Regulatory monitoring sources + family fund's jurisdictional footprint. Final interpretation always requires counsel.

Limitations

1. Compliance counsel substitution Regulatory

   Does not replace qualified compliance counsel.

   Alternative: Surfaces compliance posture data for counsel review. Final compliance opinions and remediation decisions remain external counsel territory.

2. Legal opinions Regulatory

   Does not provide legal opinions on regulatory interpretation.

   Alternative: Regulatory change flags route to external counsel for binding interpretation before operational action.

3. Active security infrastructure Capability boundary

   Does not replace active cybersecurity infrastructure (firewalls, endpoint protection, SOC monitoring).

   Alternative: Threat synthesis augments existing security infrastructure. Active threat response remains with security operations team or external SOC.

4. Incident response leadership Capability boundary

   Does not lead active incident response operations.

   Alternative: Provides context and pattern intelligence during incident response. Operational leadership remains with the family fund's designated incident commander.

Escalation Protocol

Material decision triggers

Material risk patterns crossing operator-defined severity thresholds · Regulatory changes affecting active commitments or governance framework · Cybersecurity threat indicators of active targeting · Compliance drift exceeding operator-defined tolerance

Escalation path

Material risk patterns surface to the family Principal within 24 hours. Regulatory changes affecting fund operations flag for external counsel review. Cybersecurity active-targeting indicators escalate immediately to security operations.

Timing

Material risk: within 24 hours. Regulatory: within 24 hours. Active cyber targeting: immediate. Compliance drift: within next compliance review cycle (or immediate if drift exceeds threshold).

Override authority

The family Principal can defer non-immediate risk patterns. External counsel has override authority on regulatory interpretations. Security operations team has override authority on cyber threat response.

Training Data Notes

Calibrated against family-fund failure-mode library + family fund's stated compliance and risk-tolerance framework. Threat intelligence aggregated from public sources; private intelligence relationships remain with the family fund's security team.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every CRO Equivalent invocation logs: signals aggregated, patterns detected, severity assessments, escalations fired, and counsel routings initiated. Audit trail enables retrospective review of when patterns were first detectable vs. when action was taken.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + semi-annual review by family fund's security operations team on threat-synthesis boundaries.

Authored by

Condifi Architecture Team

Applicability

Family fund clients with active risk monitoring requirements. Activates once operator configures risk-tolerance framework in the workbench.

Capabilities

1. Relationship cadence tracking High confidence

   Maintains per-relationship contact cadence against the family fund's standing relationship discipline. Surfaces drift when an anchor LP or strategic counterparty falls outside expected cadence window.

   Evidence: Family fund's own contact log + operator-defined cadence framework per relationship tier.

2. Communication drafting in institutional voice High confidence

   Drafts outreach in the family fund's institutional voice — LP communications, anchor LP updates, capital call letters, condolence and acknowledgment communications.

   Evidence: Workbench-calibrated on the family fund's prior communication archive. Voice fidelity verified against the Identity Layer Equivalent's resonance score.

3. LP sentiment pattern detection Medium confidence

   Reads observable sentiment patterns across LP communications and interactions. Surfaces sentiment trajectory shifts before they reach quarterly LP reports.

   Evidence: Pattern analysis across communication archive + operator's documented LP relationship trajectory framework. Operator-defined sentiment vocabulary preferred over generic NLP sentiment scoring.

4. Co-invest and peer relationship coordination Medium confidence

   Tracks co-invest opportunities across the peer family fund network. Drafts coordination communications and tracks reciprocation patterns.

   Evidence: Family fund's documented peer relationships + observable co-invest patterns. Confidentiality boundaries enforced — no cross-tenant peer data leakage.

Limitations

1. Autonomous external communication By design

   Does not initiate external communications without explicit operator authorization.

   Alternative: Drafts are queued for principal or designated relationship lead review. Send authority remains with the human.

2. Sensitive family communications Fiduciary

   Does not send sensitive family-system communications without principal review.

   Alternative: Sensitive communications surface to principal review with full context before send. Family Communications Coordinator (per Engagement Blueprints) may also be the human-in-the-loop.

3. Private LP data access Fiduciary

   Reads patterns from publicly observable signals and shared documents; does not access private LP-side data.

   Alternative: Analysis based on family fund's own communication archive + publicly observable LP signals. Confidential LP context surfaces through direct relationship channels.

4. Family-system mediation Capability boundary

   Does not mediate active family-system disputes through communications.

   Alternative: Family-system tension surfaces to Family Communications Coordinator + qualified external family-systems facilitator.

Escalation Protocol

Material decision triggers

Anchor LP cadence drift exceeding operator-defined tolerance · Sensitive family communication requiring drafted content · Peer co-invest coordination affecting active commitment · Sentiment shift indicating relationship-trajectory risk

Escalation path

Sensitive communications queue for principal review. Relationship drift patterns surface in the weekly briefing. Anchor LP risks escalate to the principal within same business day.

Timing

Standard drafts queue within the operator's communication review cadence. Sensitive communications surface within 4 hours. Relationship drift escalates within 24 hours.

Override authority

The family Principal has final authority on all external communications. The Family Communications Coordinator (when defined per Engagement Blueprints role) holds intermediate authority on family-system communications.

Training Data Notes

Calibrated from the family fund's prior LP communications, anchor letters, and relationship archive. Voice fidelity continuously cross-checked against the Identity Layer Equivalent's resonance score. Cross-tenant peer relationship data never used for calibration.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every Relationship Officer Equivalent invocation logs: relationship context accessed, draft content produced, sentiment patterns detected, escalation triggers fired, and send-authority resolution (queued / approved / sent). Full chain auditable for any communication.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + annual review by family fund's relationship operations team.

Authored by

Condifi Architecture Team

Applicability

Family fund clients with active LP relationships. Activates once operator passes communication-voice calibration in the workbench.

Capabilities

1. Reputational risk assessment Medium confidence

   Assesses reputational risk implications of proposed external communications, charitable giving decisions, and public posture changes against the family fund's documented public values.

   Evidence: Family fund's documented public values + observable reputational pattern library. Specific reputational decisions remain principal authority.

2. Public posture consistency monitoring High confidence

   Monitors family fund's external posture across communications, charitable giving, and public participation for internal consistency.

   Evidence: Family fund's complete external communication archive + charitable giving record + public participation log.

3. Charitable giving alignment High confidence

   Cross-references proposed charitable giving decisions against the family fund's stated values framework and historical giving patterns.

   Evidence: Family fund's documented values + historical giving archive + observable recipient public posture.

4. Family communication drafting Medium confidence

   Drafts internal family communications maintaining the family's documented voice. Coordinates with the Relationship Officer Equivalent on external-facing drafts.

   Evidence: Family fund's prior internal communications + values-aligned voice pattern. Sensitive family-system communications always require principal review.

Limitations

1. Autonomous public statement approval By design

   Does not approve public statements autonomously.

   Alternative: All public statement drafts queue for principal review. Send authority remains with the human.

2. Strategic reputation decisions Fiduciary

   Does not make strategic reputation decisions requiring principal-level authority.

   Alternative: Strategic reputation framings surface to the principal for decision. The agent provides options and risk framing, not authority.

3. Family value interpretation By design

   Does not autonomously interpret what the family's values mean in novel situations.

   Alternative: Family value interpretation is guided by the family's stated values framework. Novel situations route through family council or principal for new interpretation.

4. Crisis communications leadership Capability boundary

   Does not lead crisis communications response.

   Alternative: Provides context and draft options during crisis. Operational leadership remains with the family fund's designated crisis-comms lead or external PR counsel.

Escalation Protocol

Material decision triggers

Material reputation risks identified in public communications drafts · Charitable giving decisions diverging from stated values framework · Public posture inconsistencies that could surface external scrutiny · Family-system communications affecting cross-generational alignment

Escalation path

Material reputation risks surface to the family Principal within 4 hours. Public statement drafts require principal review before publication. Family-system communications route to the Family Council Chair when defined.

Timing

Material reputation: within 4 hours. Public statement drafts: per principal review cadence. Charitable giving alignment: within next charitable giving review cycle.

Override authority

The family Principal has primary authority on reputation strategy. The Family Council Chair has override authority on cross-generational reputation decisions. External PR counsel has override on crisis communications.

Training Data Notes

Calibrated from the family fund's complete external communication archive + documented public values + historical charitable giving. Voice fidelity cross-checked against the Identity Layer Equivalent's resonance scoring.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every CBO Equivalent invocation logs: communication context accessed, risk assessments produced, values alignment checks, escalation triggers fired, and approval-authority resolution. Audit trail preserves the full review chain for any external-facing communication.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + annual review by family fund's PR counsel on reputational-risk framing.

Authored by

Condifi Architecture Team

Applicability

Family fund clients with documented public values and external communications activity. Activates once operator passes values-framework calibration in the workbench.

Capabilities

1. Estate planning synthesis Medium confidence

   Synthesizes the family fund's existing estate planning posture against current life events, regulatory environment, and multi-generational structure. Identifies update opportunities.

   Evidence: Family fund's existing trust documents + estate planning archive + general estate planning framework. Specific structural recommendations always require external trust & estate counsel.

2. Legal document review support Medium confidence

   Reads contracts, LPAs, side letters, and operating agreements to extract structural terms, material obligations, and governance implications.

   Evidence: General contract interpretation framework + family fund's existing contractual context. Specific binding interpretation always remains with external counsel.

3. Trust structure analysis Medium confidence

   Analyzes trust structures and family-fund entity relationships for governance clarity and tax-efficiency implications.

   Evidence: Family fund's existing trust framework + general trust law context. Restructuring recommendations always require external trust counsel.

4. Family governance documentation High confidence

   Maintains and updates family governance documentation: family constitution, family council charter, decision-rights matrix, and succession protocols.

   Evidence: Family fund's documented governance framework + observable operational patterns. Governance is documentation, not interpretation — agent maintains the document; family council interprets it.

Limitations

1. Licensed attorney substitution Regulatory

   Does not substitute for a licensed attorney's legal opinion in any jurisdiction.

   Alternative: All material legal questions route to qualified external counsel before any client-impacting action. The agent provides synthesis and surfacing, not legal advice.

2. Jurisdictional legal opinions Regulatory

   Does not provide jurisdiction-specific legal opinions.

   Alternative: Jurisdiction-specific questions route to external counsel admitted in the relevant jurisdiction. Multi-jurisdictional matters require multiple counsel relationships.

3. Document execution Fiduciary

   Does not execute, sign, or notarize legal documents.

   Alternative: Drafts documents for external counsel review. Execution remains with authorized signatories and notaries through proper legal process.

4. Litigation strategy Regulatory

   Does not develop litigation strategy.

   Alternative: Litigation matters route to qualified litigation counsel. The agent provides factual synthesis when requested by counsel, never strategic counsel itself.

Escalation Protocol

Material decision triggers

Trust structure changes affecting beneficiary rights · Estate planning gaps surfaced by life events · Material legal questions in contracts under negotiation · Governance framework changes requiring family constitution updates

Escalation path

Legal-significant items flag for external counsel review before any client-impacting action. Trust structure changes route to family fund's Trust & Estate Counsel. Governance framework changes route to family council for adoption.

Timing

Material legal questions enter counsel review queue within 24 hours. Trust structure changes follow counsel's review cadence. Governance framework changes route to next family council meeting.

Override authority

External legal counsel has primary authority on all legal interpretations. The family Principal has authority on strategic legal direction. The family council has authority on governance framework adoption.

Training Data Notes

Calibrated from the family fund's existing legal document archive + governance framework + general legal interpretation knowledge. Specific binding interpretation always defers to external counsel. No legal advice given; only synthesis and document maintenance.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every CLO Equivalent invocation logs: documents accessed, synthesis produced, counsel routings initiated, governance changes proposed, and family council action taken. Audit trail preserves the chain from legal question through counsel review to final decision.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + annual review by family fund's external trust & estate counsel on capability boundaries.

Authored by

Condifi Architecture Team

Applicability

Family fund clients with active legal documentation requirements. Activates once operator configures legal-document archive access in the workbench.

Capabilities

1. Multi-agent decision pressure-testing High confidence

   Convenes the seven crew agents around a specific decision or signal. Each agent reads the decision through its own lens (strategic / financial / investment / risk / relationship / brand / legal). Synthesizes the perspectives into a structured recommendation.

   Evidence: Each agent's individual safety card defines its capability boundaries. Council protocol synthesizes within those boundaries; conflicts between agents surface explicitly rather than being smoothed over.

2. Conflict resolution between calibrated agents High confidence

   When two or more crew agents reach contradictory framings, the council protocol surfaces the disagreement explicitly with reasoning from each side. Does not paper over disagreement.

   Evidence: Architectural intent: disagreement among calibrated agents is informational, not noise. The protocol records disagreements for human resolution rather than synthesizing them away.

3. Consensus synthesis High confidence

   When agents converge on framing, produces a synthesized recommendation in the family fund's institutional voice. Synthesis preserves the per-agent reasoning chain.

   Evidence: Calibrated multi-agent synthesis (Patent 3 substrate) — convergent agent framings produce higher-confidence recommendations than any single agent.

4. Decision rationale documentation High confidence

   Documents the reasoning chain from operator-submitted decision through each agent's framing to the synthesized recommendation. Audit-ready for counsel review.

   Evidence: Architectural requirement — every council pressure-test produces a documented reasoning chain that family fund counsel can audit.

Limitations

1. Autonomous decision execution By design

   Does not execute the decision under test, regardless of council consensus.

   Alternative: Council output is recommendation, not execution. The family Principal (or designated IC) executes through proper authorized channels.

2. Calibration bias Capability boundary

   Council pressure-test inherits calibration biases from the configured crew agents. If a crew agent's calibration is misconfigured, the council protocol amplifies rather than corrects.

   Alternative: Agent calibration is reviewed during the workbench onboarding and quarterly thereafter. Material miscalibration patterns surface through the Identity Layer Equivalent's resonance scoring.

3. Truly novel decision contexts By design

   Pressure-tests against the calibrated agents' known framings. Truly novel decision contexts (where no agent has applicable framing) surface as 'novel context' rather than producing a recommendation.

   Alternative: Novel contexts surface to the family Principal with the 'no agent has applicable framing' annotation. Human judgment then determines whether to defer the decision, engage external advisors, or extend the framing of one or more agents.

4. Disagreement smoothing By design

   Does not smooth over agent disagreement to produce a clean recommendation.

   Alternative: Disagreement is informational. Split decisions surface explicitly with the conflict pattern documented for human resolution.

Escalation Protocol

Material decision triggers

Split decisions where agents disagree at council protocol · Material decision threshold met (above family fund's defined material limit) · Conflict patterns repeating across multiple council sessions · Novel context flagged where no agent framing applies

Escalation path

Split decisions and novel contexts surface to the family Principal for direct review. Conflict patterns repeating across sessions log for family-fund-level calibration review. Material decisions route through standard IC or principal channels.

Timing

Council pressure-test completes within 48 hours of operator submission. Split decisions and novel contexts surface to principal review queue immediately upon council completion. Conflict pattern reviews are quarterly.

Override authority

The family Principal can override any council recommendation. The IC has primary authority on investment-related council outputs. The agent calibration framework can be adjusted by the operator with workbench-level changes.

Training Data Notes

The council protocol itself is not trained on operator data — it is a structural multi-agent orchestration. Each underlying crew agent is calibrated per its own safety card.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every council pressure-test logs: operator-submitted context, each agent's framing, conflicts identified, synthesis produced, escalation triggers fired, and final principal/IC action. Full reasoning chain auditable by counsel.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + annual architectural review on the multi-agent synthesis approach (Patent 3 substrate).

Authored by

Condifi Architecture Team

Applicability

All family fund clients with active council pressure-test workflow. Activates immediately upon workspace creation — does not require workbench calibration since each underlying agent's calibration is the relevant gating factor.

Capabilities

1. Institutional voice fidelity learning High confidence

   Captures voice patterns from the family fund's documents, communications, and decision archive. Builds a per-tenant voice fidelity reference that crew agents consult for institutional-voice drafting.

   Evidence: Workbench-fed by the operator's own document archive. Voice fidelity continuously cross-checked against the Identity Layer Equivalent's resonance scoring.

2. Decision pattern capture High confidence

   Records the family fund's decision-rights matrix, authority thresholds, and prior decision rationales. Crew agents reference this when surfacing recommendations.

   Evidence: Operator's documented decision-rights framework + observable decision patterns from prior decisions logged through council and decision room.

3. Doctrine evolution tracking High confidence

   Tracks how the family fund's documented doctrine evolves over time. When operator endorses a doctrine shift, the workbench updates the reference substrate for all crew agents.

   Evidence: Operator's explicit doctrine endorsements + observable doctrine application patterns. Doctrine shifts require operator endorsement — cannot drift implicitly.

4. Per-tenant cognitive synchronization High confidence

   Maintains the per-tenant cognitive substrate that crew agents reference. Architectural isolation enforces that one family fund's substrate never informs another's.

   Evidence: Single-tenant architecture enforced at the workspace level. Cross-tenant substrate access is architecturally impossible.

Limitations

1. Onboarding baseline period Capability boundary

   Requires sustained operator interaction to develop voice fidelity. Minimum 30-day baseline before crew agents can confidently invoke the institutional voice.

   Alternative: During the initial 30-day window, crew agents draft in a generic family-fund-appropriate institutional voice. Voice fidelity confidence indicators surface in every draft.

2. Voice drift on calibration changes By design

   Institutional voice can drift if calibration parameters change mid-engagement (e.g., new principal generation, major staff turnover, doctrine update).

   Alternative: Calibration changes surface to the operator with explicit acknowledgment. Voice fidelity drift indicators flag whenever observed outputs deviate from the documented voice baseline.

3. Implicit doctrine evolution By design

   Does not allow doctrine to drift implicitly. Doctrine shifts require explicit operator endorsement.

   Alternative: Doctrine update notifications surface to the operator for explicit endorsement. The substrate does not auto-update from observation alone.

4. Cross-tenant synchronization Regulatory

   Does not synchronize across family fund clients, even those under the same consultant.

   Alternative: Each family fund client has its own isolated cognitive substrate. Cross-client patterns can be aggregated at the consultant level only with anonymization (see Engagement Blueprints and refresh feed patterns).

Escalation Protocol

Material decision triggers

Voice fidelity confidence below operator-defined threshold · Significant doctrine shift queued for endorsement · Cross-tenant access attempt (architectural impossibility — but logged if detected) · Substrate drift indicators exceeding tolerance

Escalation path

Voice fidelity below threshold surfaces to operator for review. Significant doctrine shifts queue for explicit endorsement. Substrate drift surfaces to operator for calibration session. Cross-tenant access attempts surface to compliance counsel immediately.

Timing

Voice fidelity reviews: within next workbench session. Doctrine shifts: queued for next endorsement window. Substrate drift: weekly review. Cross-tenant access attempts: immediate.

Override authority

The operator has authority on all substrate calibration. Doctrine shifts require explicit operator endorsement. Compliance counsel has override on cross-tenant access investigations.

Training Data Notes

The workbench substrate is the per-tenant training surface itself. The substrate is not cross-trained across tenants. Each family fund's substrate reflects only that family fund's own documents and decisions.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every workbench substrate operation logs: documents ingested, voice patterns captured, doctrine endorsements recorded, calibration changes applied, and crew agent calibration version. Substrate evolution is fully auditable by counsel.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + annual architectural review on cognitive synchronization approach.

Authored by

Condifi Architecture Team

Applicability

All family fund clients. Activates immediately upon workspace creation with baseline generic voice; institutional voice fidelity matures over the first 30-60 days.

Capabilities

1. PII detection across configured categories High confidence

   Scans input text for configured PII categories (names, addresses, account numbers, custodian names, deal codenames, family member identities) before any AI provider call. Detection sensitivity tunable per family fund.

   Evidence: Recognizer library with both base (universal) and family-fund-specific recognizers. Per-tenant sensitivity configuration via the onboarding wizard.

2. Reversible tokenization High confidence

   Substitutes detected PII with reversible tokens that the provider can use as semantic placeholders. Tokens restored to original values locally inside the family fund's workspace before output is shown to operator.

   Evidence: Token vault stores reversible mappings within the family fund's isolated workspace. The provider never sees raw identifiers; the operator never sees tokens (substitution is invisible).

3. Per-tenant token vault isolation High confidence

   Each family fund client has its own isolated token vault. One tenant's tokens are architecturally invisible to another. Token reuse occurs only within the originating family fund.

   Evidence: Single-tenant architecture enforced at the database row-level security tier. Cross-tenant token lookup is architecturally impossible.

4. Audit trail of every tokenization event High confidence

   Logs every tokenization event: input fingerprint, detection categories triggered, tokens generated, provider call timestamp, and detokenization. Family fund counsel can audit any AI call's full tokenization chain.

   Evidence: Per-tenant ai_query_audit table records the full chain. Counsel can request the audit log for any query.

Limitations

1. Recognizer accuracy Capability boundary

   Relies on configured recognizer accuracy. Universal recognizers (names, emails, phone numbers) have high accuracy; family-fund-specific recognizers (custodian names, deal codenames) require per-tenant calibration to reach high accuracy.

   Alternative: Family-fund-specific recognizers default to conservative sensitivity (more false positives, fewer false negatives). Operator can tune sensitivity in the workbench.

2. Novel PII patterns Capability boundary

   False negatives possible for novel PII patterns not yet in the recognizer library.

   Alternative: Low-confidence detections flag for operator review. Operator can add custom recognizers per family fund. Recognizer library is reviewed and updated quarterly.

3. Conservative-bias overhead By design

   Conservative sensitivity setting produces slight latency overhead from over-tokenization.

   Alternative: Operator can tune sensitivity (conservative / standard / aggressive) per family fund in the workbench. Conservative is default for new family funds; can be tuned after 30-day calibration baseline.

4. Image and document-binary content Capability boundary

   Tokenization applies to text content only. Image and document-binary content (PDFs, scanned documents) require separate redaction processes.

   Alternative: Image-content redaction processes activate in a future batch. Until then, family fund counsel reviews image content before AI provider call.

Escalation Protocol

Material decision triggers

Low-confidence detections flagged for operator review · Recognizer accuracy below operator-defined threshold · Cross-tenant token access attempt (architectural impossibility — but logged if detected) · Novel PII pattern detected requiring recognizer library update

Escalation path

Low-confidence detections queue for operator review in the workbench. Recognizer accuracy issues surface to platform engineering for library update. Cross-tenant access attempts escalate to compliance counsel immediately.

Timing

Low-confidence reviews: queued for next workbench session. Recognizer accuracy reviews: weekly. Cross-tenant access: immediate.

Override authority

Operator can override individual recognizer decisions per query. Operator can adjust sensitivity setting per family fund. Compliance counsel has override on recognizer library updates affecting all family fund clients.

Training Data Notes

Universal recognizers (names, emails, phones, account numbers) trained on synthetic and aggregated public data. Family-fund-specific recognizers trained per-tenant from the operator's own documents. Cross-tenant recognizer training does not occur.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every tokenization event logs: input timestamp, detection categories triggered, token count generated, provider call timestamp, detokenization confirmation, and operator output timestamp. Per-tenant audit log accessible to family fund counsel.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + monthly recognizer-accuracy review by platform engineering.

Authored by

Condifi Architecture Team

Applicability

All family fund clients. Tokenization gateway activates as soon as TOKENIZATION_GATEWAY_ENABLED flag is set true for the tenant (or platform-wide). Pre-activation, the gateway is in non-blocking observe mode.

Capabilities

1. 8-surface continuous monitoring High confidence

   Monitors eight configured intelligence surfaces: regulatory updates, peer family fund signals, market context, sector-specific intelligence, counterparty news, jurisdictional changes, philanthropic sector activity, and family-system patterns.

   Evidence: Each surface has documented sources, quality benchmarks, and per-tenant relevance routing. Sources are configured (not unlimited) and reviewed quarterly for quality.

2. 4-tier refresh velocity High confidence

   Four refresh tiers calibrated to information criticality: real-time (regulatory + counterparty), daily (peer signals + market context), weekly (sector + jurisdictional), monthly (philanthropic + family-system). Each surface assigned to a tier.

   Evidence: Tier assignment reflects the operating criticality of timely intelligence on each surface. Real-time tier targets sub-15-minute latency from source publication to operator visibility.

3. Per-tenant relevance routing Medium confidence

   Filters surface signals through the family fund's own relevance framework before surfacing. One family fund's relevant signal may not be relevant to another, even within the same consultant's portfolio.

   Evidence: Per-tenant relevance framework from workbench calibration + operator-defined relevance thresholds. Cross-tenant aggregation occurs only with anonymization at the consultant level.

4. Signal attribution and source tracking High confidence

   Every surfaced signal includes attribution to its source. Family fund counsel can trace any signal to its origin for verification and confidence assessment.

   Evidence: Source attribution is non-optional in the refresh engine architecture. Every signal carries source URL, source confidence rating, and timestamp.

Limitations

1. Configured sources Capability boundary

   Sources are configured per surface (not unlimited). Coverage gaps exist for unconfigured sources.

   Alternative: Source coverage is reviewed quarterly. Operator can request additional sources through the consultant. Critical gaps trigger immediate source-library updates.

2. Source quality variance Capability boundary

   Source quality varies by surface. Regulatory and counterparty sources are high-confidence; peer family fund and family-system pattern sources are observational and lower-confidence.

   Alternative: Every signal carries source confidence rating. Operator can filter by confidence threshold. Low-confidence signals do not automatically trigger escalations.

3. Refresh latency by tier Capability boundary

   Refresh latency varies by tier. Real-time targets sub-15-minute latency but cannot guarantee it for all sources. Monthly tier may have multi-week lag from source publication.

   Alternative: Tier assignments reviewed quarterly. Latency degradation surfaces engineering review. Operators with stricter latency requirements can request tier escalation per surface.

4. Acting on signals By design

   Does not act on signals on the family fund's behalf. Surfaces — does not execute.

   Alternative: Signals route to relevant crew agents (e.g., regulatory signals route to CRO Equivalent + CLO Equivalent) for surfacing, recommendation drafting, or council pressure-test. Final action remains with humans.

Escalation Protocol

Material decision triggers

Critical regulatory changes affecting active commitments — escalate regardless of tier · Signal-quality degradation patterns on configured sources · Counterparty news indicating material impact on active commitments · Family-system pattern indicators flagged across multiple weeks

Escalation path

Critical regulatory changes flag immediately to CLO Equivalent + CRO Equivalent + family Principal. Counterparty news routes to CIO Equivalent + family Principal. Signal-quality degradation surfaces to platform engineering review. Family-system patterns route through Family Communications Coordinator when defined.

Timing

Critical regulatory: immediate (regardless of source-tier). Counterparty material news: within source-tier latency (typically real-time). Signal-quality degradation: weekly engineering review. Family-system patterns: monthly review.

Override authority

Operator can suppress sources or signals per relevance framework. Family fund counsel has override on regulatory signal interpretation. Platform engineering has override on source-quality issues.

Training Data Notes

Source library is configured at the platform level with per-tenant relevance routing. Per-tenant relevance frameworks calibrate from operator interaction. Cross-tenant signal aggregation occurs only at the consultant level with explicit anonymization.

Model Routing Disclosure

Primary provider category

frontier-tier AI provider

Fallback behavior

If the primary provider is unavailable, the agent fails over to a configured secondary provider. If no provider is healthy, the agent surfaces unavailability to the operator rather than degrading output quality.

Per-query audit trail

Every agent invocation is logged to the per-tenant ai_query_audit table with timestamp, query type, provider used, tokens consumed, duration, and outcome status. Family fund counsel can audit any agent invocation end-to-end via the consultant master console.

PII handling

The tokenization gateway substitutes detected PII with reversible tokens before any provider call. The provider never sees raw identifiers; tokens are restored locally inside the family fund's workspace before output is shown to the operator.

Audit Trail

Every refresh event logs: source polled, signals detected, relevance routing applied, signals surfaced, operator response (read / acted / dismissed). Counsel can audit any signal's full chain from source to operator response.

Metadata

Version

1.0.0

Effective date

2026-05-17

Review cadence

Quarterly review by compliance counsel + monthly source-library review by platform engineering.

Authored by

Condifi Architecture Team

Applicability

All family fund clients. Refresh engine activates with default surface configuration immediately on workspace creation. Operator can customize relevance routing and source preferences in the workbench.