docs(spec): IHF specification v0.2 — Phases 9–12

Extends the framework beyond Phase 8 with four planned phases:
- Phase 9: External API Surface and Consumer SDKs (OAuth, OpenAPI, TS/Python SDKs, webhooks)
- Phase 10: Hub Registry and Widget Marketplace (pattern library, adoption tracking, governance templates)
- Phase 11: Advanced AI Federation (agent registry, model routing, delegation, collective proposals)
- Phase 12: Platform Memory and Continuous Learning (outcome correlation, adaptive thresholds, knowledge base)

Includes dependency graph, v0.2 risk register additions, and extended success criteria.
Status: draft — for review.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-03-29 21:58:54 +00:00
parent c9aa004bf8
commit 63fb0e8277

View File

@@ -0,0 +1,437 @@
InteractionHubFramework
*Inter*
# Interaction Hub Framework (IHF)
*Formal Specification with Phased Implementation Plan*
**Version:** 0.2
**Status:** Draft Extension Specification
**Supersedes:** `InteractionHubFrameworkSpecification_v0.1.md`
**Purpose:** Extend the IHF beyond Phase 8 into external platform maturity — public API surface, consumer SDKs, hub marketplace, and advanced AI federation.
---
## 1. Definition
The **Interaction Hub Framework (IHF)** is a platform framework for building **commentable, observable, governable, and evolvable user interaction systems**.
It provides a stable core for:
* addressing UI elements as governed platform assets
* capturing user feedback and interaction signals in context
* transforming observed friction and intent into structured requirements
* connecting requirements to decisions, implementations, deployments, and outcomes
* supporting multiple evolving UI technologies without losing semantic continuity
* federating governance across hub boundaries in multi-team AI factory deployments
The IHF is intended to serve as the **interaction substrate** of a hub-based AI factory, especially in environments where product development, operations, governance, finance, security, and AI-assisted workflows interact continuously.
---
## 2. Design Intent
The framework is designed to support the following long-term properties:
**Stable semantic core**
UI technologies may change, but interaction semantics, traceability, and governance remain stable.
**Antifragility**
The system should improve through systematically captured friction, confusion, unmet expectations, and emergent user needs.
**Observability by design**
Interaction, feedback, and governance events are first-class operational signals.
**Governance by design**
Requirements, decisions, policy concerns, and implementation consequences remain linked and auditable.
**Agent compatibility**
AI agents should be able to inspect, interpret, propose, implement, review, and monitor changes within explicit boundaries.
**Framework independence at the edge**
Any UI technology that can emit the widget envelope is a first-class participant. The IHF does not mandate a particular frontend stack.
**Platform composability**
The IHF should be consumable by external systems, third-party tooling, and downstream hubs via stable, versioned contracts. Internal governance state must not be a closed silo.
---
## 3. Scope
### In Scope
* Widget identity, lifecycle governance, and semantic addressability
* Interaction event capture and contextual enrichment
* Annotation and structured comment threads
* Requirements distillation from raw feedback clusters
* Governance ledger: decision records linked to requirements and implementations
* Outcome observation and pre/post change comparison
* AI-assisted distillation, bounded, attributable, reviewer-controlled
* Cross-framework UI adapter protocol
* Operational observability: friction scoring, bottleneck detection, health snapshots
* Federated governance: ownership delegation, requirement routing, policy overlays, stewardship, archival
* External API surface for third-party and inter-system consumption
* Consumer SDKs for common platform integrations
* Hub registry and marketplace for reusable widget and governance patterns
* Advanced AI federation for multi-model, multi-agent governance workflows
### Out of Scope
* A complete universal frontend framework or design system
* Pixel-level visual design or CSS conventions
* Full product management methodology
* Replacement for DevOps observability tooling
* Unrestricted autonomous AI decision-making on requirements
* Mandatory single UI technology for all hub surfaces
* General-purpose data warehouse or business intelligence platform
---
## 4. Core Concepts
*(Unchanged from v0.1 §4 — see `InteractionHubFrameworkSpecification_v0.1.md`)*
---
## 5. Phased Implementation Plan
The v0.1 specification defined Phases 08. Phases 07 are complete as reference implementation. Phase 8 (Federated Hub Maturity) is in progress.
This v0.2 specification extends the plan with Phases 912.
### Phase 0 — Concept Foundation *(complete)*
### Phase 1 — Minimal Interaction Core *(complete)*
### Phase 2 — Structured Feedback and Triage *(complete)*
### Phase 3 — Governance and Decision Linkage *(complete)*
### Phase 4 — Outcome Observation and Antifragility Loop *(complete)*
### Phase 5 — Agent-Assisted Distillation and Suggestion *(complete)*
### Phase 6 — Cross-Framework UI Adaptation Layer *(complete)*
### Phase 7 — Advanced Observability and Operational Integration *(complete)*
### Phase 8 — Federated Hub Maturity *(in progress)*
---
## Phase 9 — External API Surface and Consumer SDKs
### Goal
Make the IHF consumable by systems outside the reference IHP implementation.
Phase 8 established federated governance within a single deployment. Phase 9
exposes that governance state as a stable, versioned, authenticated REST API and
ships consumer SDKs that make integration a day's work rather than a project.
### Scope
* Versioned REST API (`/api/v2/`) for all core IHF artifact types
* OpenAPI 3.1 specification generated from the live schema
* Authentication: OAuth 2.0 client credentials flow (superseding per-hub Bearer tokens)
* API key management UI for external consumers
* Consumer SDKs: TypeScript/Node, Python
* Webhook delivery for interaction events, candidate creation, and decision records
* API usage dashboard: request counts, error rates, consumer identity
* Rate limiting and quota management per consumer
### Functional Capabilities
* External systems can read widget registry, interaction events, annotations,
requirement candidates, decisions, deployments, and outcome signals
* External systems can submit interaction events and annotations via the API
* Downstream hubs can subscribe to governance events via webhooks
* SDK consumers get type-safe access to IHF contracts without reading the spec
* API consumers are tracked, quotaed, and auditable
### Exit Criteria
* All core IHF artifact types are readable via `/api/v2/`
* Interaction events and annotations are writable via `/api/v2/`
* OpenAPI spec is generated and accurate
* TypeScript SDK and Python SDK published (as static files or packages)
* Webhook delivery confirmed for at least two event types
* API usage dashboard renders correctly
* OAuth token flow works end-to-end
### Data Artifacts Introduced
`ApiConsumer`, `ApiKey`, `WebhookSubscription`, `WebhookDelivery`
---
## Phase 10 — Hub Registry and Widget Marketplace
### Goal
Enable reuse of proven widget patterns, governance templates, and hub
configurations across deployments. Phase 9 made the IHF externally consumable.
Phase 10 makes it composable: hubs and widgets can be discovered, rated,
adopted, and evolved as shared platform assets.
### Scope
* Hub registry: a catalog of registered hubs with public metadata, capability
declarations, and health summaries
* Widget pattern library: reusable widget definitions that can be instantiated
into any hub
* Governance template library: requirement distillation and decision templates
that can be cloned across hubs
* Widget ratings and adoption tracking: which widgets are in use where, with
aggregated friction scores across deployments
* Pattern versioning: widget patterns have explicit versions; hubs can pin or
follow-latest
* Marketplace dashboard: browse, search, and adopt patterns
### Functional Capabilities
* Hub operators can publish a widget pattern to the shared library
* Hub operators can adopt a published pattern into their hub
* Governance templates (requirement categories, decision checklists) can be
cloned across hubs
* Widget adoption across hubs is tracked for aggregate friction and outcome
analysis
* Pattern authors receive friction and outcome feedback from all adopter hubs
(opt-in anonymised)
### Exit Criteria
* Hub registry renders all registered hubs with capability metadata
* Widget pattern library lists published patterns with version history
* A pattern can be published from one hub and adopted into another
* Adoption tracking shows which hubs use which patterns
* Governance template cloning works end-to-end
* Marketplace dashboard renders search and browse
### Data Artifacts Introduced
`WidgetPattern`, `WidgetPatternVersion`, `PatternAdoption`, `GovernanceTemplate`,
`GovernanceTemplateClone`
---
## Phase 11 — Advanced AI Federation
### Goal
Move AI assistance from single-model, single-agent operation to a federated,
multi-model, multi-agent architecture. Phase 5 introduced bounded AI support
for distillation and suggestion. Phase 11 introduces agent specialisation,
model routing, inter-agent delegation, and collective AI governance — while
keeping humans firmly in control of final decisions.
### Scope
* Agent registry: named, versioned AI agents with declared capabilities and
trust levels
* Model routing: select model per task type based on cost, capability, and
policy
* Inter-agent delegation: an agent can spawn or invoke a specialist sub-agent
with a bounded scope
* Agent collaboration records: when multiple agents contribute to a single
proposal, all contributions are attributed
* Collective proposal review: proposals that require multi-agent consensus
before human review
* AI governance policies: per-hub rules controlling which agents may operate,
on which artifact types, with what authority
* Agent performance tracking: acceptance rate, revision rate, confidence
calibration over time
### Functional Capabilities
* Hub operators can register and configure specialised agents (triage agent,
synthesis agent, policy agent, implementation agent)
* Agent tasks are routed to the lowest-cost model that meets capability and
policy requirements
* A delegating agent can hand off a subtask with a capped token budget and
scoped context
* All inter-agent delegations are recorded and auditable
* Humans can inspect the full provenance of any AI output: which agents
contributed, which models were used, in what order
* AI governance policies can restrict specific agents to read-only or
advisory-only roles per hub
### Exit Criteria
* Agent registry lists all registered agents with capability and trust metadata
* Model routing selects the correct model for at least three task types
* Inter-agent delegation creates an auditable delegation record
* Collective proposals show all contributing agents
* AI governance policy blocks an out-of-scope agent action
* Agent performance dashboard renders acceptance and calibration metrics
### Data Artifacts Introduced
`AgentRegistration`, `ModelRoutingPolicy`, `AgentDelegation`,
`CollectiveProposal`, `AiGovernancePolicy`, `AgentPerformanceRecord`
---
## Phase 12 — Platform Memory and Continuous Learning
### Goal
Close the longest feedback loop in the IHF: from deployed outcome signals and
accumulated governance history back to improved distillation, better routing,
and sharper AI proposals. Phase 12 makes the IHF a learning platform, not
merely a record-keeping one.
### Scope
* Long-term outcome correlation: connect deployment outcomes to the original
requirement and annotation signals that drove them, across quarters
* Pattern performance memory: track which widget patterns, governance templates,
and routing rules have historically produced good outcomes
* Adaptive friction thresholds: friction score weights and bottleneck thresholds
adjust based on observed hub-specific outcome data
* Institutional knowledge capture: significant decisions, their context, and
their outcomes are distilled into a queryable knowledge base
* Retroactive lineage enrichment: as new outcomes arrive, earlier records in
the traceability chain are enriched with outcome metadata
* Learning dashboard: shows what the platform has learned — best-performing
patterns, most-predictive signals, longest-value decisions
### Functional Capabilities
* The platform can answer "which types of annotation predicted a negative outcome
in this hub?"
* Widget patterns are ranked by historical outcome quality, not just adoption
* Friction thresholds for a hub can be calibrated from its own outcome history
* Governance decisions are enriched with outcome summaries when signals arrive
* A knowledge base of significant decisions is queryable by agents and operators
* The learning dashboard shows platform-level insights with evidence links
### Exit Criteria
* Long-term outcome correlation produces at least one actionable hub-specific
insight from test data
* Pattern performance ranking changes after seeding outcome data
* Adaptive thresholds update friction scores after calibration run
* Institutional knowledge base is queryable and returns relevant decisions
* Retroactive lineage enrichment updates at least one decision record with
outcome metadata
* Learning dashboard renders all panels with correct evidence links
### Data Artifacts Introduced
`OutcomeCorrelation`, `PatternPerformanceRecord`, `AdaptiveThresholdConfig`,
`InstitutionalKnowledgeEntry`, `LearningInsight`
---
## 6. Revised Phased Summary
| Phase | Title | Status |
|-------|-------|--------|
| 0 | Concept Foundation | complete |
| 1 | Minimal Interaction Core | complete |
| 2 | Structured Feedback and Triage | complete |
| 3 | Governance and Decision Linkage | complete |
| 4 | Outcome Observation and Antifragility | complete |
| 5 | Agent-Assisted Distillation | complete |
| 6 | Cross-Framework UI Adaptation | complete |
| 7 | Advanced Observability | complete |
| 8 | Federated Hub Maturity | in progress |
| 9 | External API Surface and Consumer SDKs | planned |
| 10 | Hub Registry and Widget Marketplace | planned |
| 11 | Advanced AI Federation | planned |
| 12 | Platform Memory and Continuous Learning | planned |
---
## 7. Dependency Graph (Phases 912)
```
Phase 8 (Federated) ──→ Phase 9 (External API)
Phase 10 (Marketplace)
Phase 7 (Observability) ──→ Phase 11 (AI Federation)
Phase 5 (Agent Assist) ──┘ │
Phase 12 (Platform Memory)
Phase 4 (Outcomes) ───────────┘
```
- Phase 9 requires Phase 8 (stable federated schema, OAuth replaces per-hub
Bearer tokens)
- Phase 10 requires Phase 9 (marketplace API is built on v2 API surface)
- Phase 11 requires Phase 5 (agent model) and Phase 7 (observability signals
needed for model routing and performance tracking)
- Phase 12 requires Phase 4 (outcome signals), Phase 7 (friction/health
history), and Phase 11 (agent performance as a learning input)
---
## 8. Risks and Failure Modes (v0.2 additions)
### 8.1 API Surface Becoming a Governance Bypass
Risk: external consumers write interaction events or annotations without
honouring the IHF envelope contract, polluting the governance record.
Mitigation: the `/api/v2/` write endpoints validate against the active
`InteractionReportingContract` and `EnvelopeEmissionContract`. Writes that
fail validation return 422 with contract-referenced errors.
### 8.2 Marketplace Pattern Divergence
Risk: adopted widget patterns diverge from their source over time, breaking
the aggregate friction and outcome correlation.
Mitigation: `PatternAdoption` records the version pinned at adoption time.
Pattern authors can issue new versions; adopters choose whether to migrate.
Aggregate metrics are version-scoped.
### 8.3 AI Federation Complexity Outpacing Auditability
Risk: multi-agent delegation chains become too deep to audit meaningfully.
Mitigation: `AgentDelegation` has a max depth field enforced at creation time
(default: 2). Chains deeper than the hub's `AiGovernancePolicy.max_depth` are
rejected with an audit record of the refusal.
### 8.4 Learning Feedback Loops Reinforcing Bias
Risk: outcome correlation training on biased historical data produces
systematically wrong friction weight adjustments.
Mitigation: `AdaptiveThresholdConfig` records the data window and sample size
used for each calibration. Calibrations below a minimum sample threshold are
flagged as `low_confidence` and do not apply automatically.
### 8.5 Knowledge Base Becoming Stale
Risk: `InstitutionalKnowledgeEntry` records reflect decisions that were later
superseded, misleading agents and operators.
Mitigation: entries link to their source `DecisionRecord`. When a decision is
superseded (a new decision is linked to the same requirement), the knowledge
entry is flagged `superseded` and a new entry is created for the successor.
---
## 9. Success Criteria (v0.2 additions)
The Interaction Hub Framework at v0.2 maturity is successful when:
* external systems can read and write IHF state via stable, versioned contracts
* widget and governance patterns are reusable without copy-paste
* multiple AI agents collaborate on proposals without reducing accountability
* the platform visibly learns from its own outcome history
* hub leaders can make evidence-based decisions about pattern adoption and
agent configuration without reading source code
* the IHF can be deployed across an organisation with clear ownership,
policies, and audit trails at every layer
---
## 10. Short Strategic Summary
The Interaction Hub Framework v0.2 extends the governed interaction substrate
into a **composable, federated, learning platform**.
Where v0.1 asked: *can we make interaction observable and governance auditable?*
v0.2 asks: *can the platform teach itself, share what it knows, and remain
accountable while doing so?*
The answer depends on making every new capability — API surface, marketplace
patterns, AI federation, outcome correlation — subject to the same contract
discipline and human oversight that governs the core.
That continuity of principle is what makes the framework trustworthy at scale.