Architecture living owner: Azwaan reviewed: 2026-07-05

Architecture Risks

The ecosystem’s architectural risk register, produced by the Phase-D Architecture Review. Every risk is grounded in documented portal evidence; none asserts un-evidenced implementation detail. Priority is a function of impact × likelihood × strategic leverage.

Likelihood/impact scale: Low · Medium · High. Priority: P1 (address next) · P2 (plan) · P3 (watch).

Risk register

ID Risk Impact Likelihood Affected repos Recommended mitigation Priority
R1 Reusable assessment engine trapped in a venture repo — the domain-agnostic assessment API lives inside inexisstudios (a venture marketing site), so consumers couple to a venture. High Present (certain) inexisstudios, outreachagent, leadplatform Extract to a standalone Assessment Service in a new Platform Services layer (ADR 0005). P1
R2 Two divergent opportunity-scoring models — pack 0–100 domain score vs leadplatform’s own scoring; can disagree in front of a prospect. Medium-High Medium intelproducts, leadplatform, outreachagent Single source of truth (pack); have leadplatform consume the assessment/COP. P1
R3 Inference-derived scoring weights, no learning loop — weights are “review-gated defaults, not empirical”; Hermes feedback loop unbuilt. Medium High (documented) intelproducts, inexisstudios Instrument outcomes; build Hermes to replace defaults with outcome-derived weights. P2
R4 End-to-end flow mostly mock — assessment runs ASSESSMENT_MODE=mock by default in outreachagent; live run not operationally validated. Medium Present outreachagent, inexisstudios Operational live-run validation; track live-vs-mock in current-state. P2
R5 Pack drift via stale submodule pins — consumers vendor intelproducts; stale pins score against outdated intelligence. Medium Medium leadplatform (+ future consumers) Track pinned pack versions in the registry; CI pin-freshness check. P2
R6 Non-durable crawler — engine crawler jobs are in-memory / not durable across restarts (Phase 3d planned). Medium Medium inexisstudios Durable queue (Cloudflare Queues/Durable Objects); complete Phase 3d. P2
R7 Foundational loops unbuilt — FIP Phases 1–5 not started; lab core run/eval loop pending. Higher capabilities depend on these. High (strategic) High personalops, website-intelligence-lab Prioritise FIP capture→retrieve loop and the lab adapter→runs→eval loop. P1
R8 Documentation & knowledge duplication as the ecosystem grows — manual portal upkeep, near-duplicate skills, long hand-maintained intelligence chain. Medium High portfolio-portal, shared-skills, intelproducts Build the portfolio-portal-orchestrator; automate pack publication; consolidate skills. P2
R9 AI prompt surfaces spread across repos — engine narrative, pack report guidance, outreach skills → risk of divergent claims/tone, especially regulated verticals. Medium Medium inexisstudios, intelproducts, outreachagent Shared, fact-grounded claims/tone guardrail; health verticals already route to human review — extend that discipline. P2
R10 Two data platforms — Supabase (FIP) + Cloudflare D1/R2 (engine) fragment operations & backup/governance. Low-Medium Present personalops, inexisstudios A deliberate data-platform strategy (not necessarily consolidation); document the boundary. P3
R11 Key-person concentration — single founder owns all systems; private tools, no team. High Structural all The portal itself mitigates (captured knowledge, onboarding, AI-context). Continue capability documentation; keep digests current. P2
R12 Circular feedback risk (future) — the planned Hermes loop (outcomes → re-weight pack) could couple consumers back into the producer if designed as mutation rather than a new pack version. Medium Low (future) intelproducts, personalops, inexisstudios Design Hermes as a one-way feedback that publishes a new immutable pack version; never mutate a consumed pack. P3

Risk map (impact × likelihood)

quadrantChart
    title Architecture risk map
    x-axis Low Likelihood --> High Likelihood
    y-axis Low Impact --> High Impact
    quadrant-1 Address next (P1)
    quadrant-2 Plan mitigation
    quadrant-3 Watch
    quadrant-4 Monitor
    R1 engine trapped: [0.9, 0.85]
    R2 dual scoring: [0.55, 0.7]
    R3 no learning loop: [0.8, 0.55]
    R7 unbuilt loops: [0.8, 0.9]
    R8 doc duplication: [0.8, 0.5]
    R11 key person: [0.9, 0.8]
    R6 crawler durability: [0.5, 0.5]
    R10 two data stores: [0.6, 0.35]

Priority summary

  • P1 (address next): R1 (extract Assessment Service), R2 (single scoring source), R7 (build foundational loops).
  • P2 (plan): R3, R4, R5, R6, R8, R9, R11.
  • P3 (watch): R10, R12.

Risks are recommendations, not changes. Promote accepted mitigations to ADRs; the direction change for R1–R2 is captured in proposed ADR 0005.