Plan — concern-library expansion + make-concerns-bite + iterate-to-convergence (2026-05-27)

Plan-of-record for the current program (dogfoods the plan-as-file model). Captured as a file by default; under DDx this would be a bead. Status markers: [x] done · [~] in flight · [ ] queued.

Goal

Expand the HELIX concern library with architecture, platform, and cross-cutting concerns (operator: “do all tier 1-3” + databricks + MCP + usage-tracking).
Make concerns bite — a selected concern must change the ADR / FEAT / TD / SD / plan, not just exist as prose.
Operationalize “go check your work again” — a bounded evolve-until-converged loop.
Re-bench the 6 one-shots under it.39 + the new concerns, iterating until ACs met + specs align.

Concern inventory

Reviewed (codex SOUND-WITH-FIXES, minor):

domain-driven-design, onion-architecture, design-patterns-gof, enterprise-integration-patterns, event-sourcing
classic-layered, hexagonal-architecture, clean-architecture (fill architecture-style slot)
multi-tenancy, deployment-topology (guide), api-style (guide)
unity-catalog, databricks-apps (→ deploy-target), databricks-declarative-pipelines
mcp-server, usage-metering (+ api-style edited to add MCP as the agent-facing style)

In flight (Tier 2/3 — review pending):

[~] cqrs, resilience, enterprise-application-patterns, authorization-model
[~] relational-data-modeling, caching-strategy, frontend-architecture, twelve-factor, concurrency-model

Slotting decisions (reconciled)

New exclusive slot architecture-style (members: onion/classic-layered/hexagonal/clean) — no default, select-on-signal.
databricks-apps fills the existing deploy-target slot. All other new concerns are non-exclusive (no slot) — deployment-topology + api-style are decision-guides, not slots.

Consolidation (queued — after all land + reviewed)

Format fixes (from review): unity-catalog + databricks-declarative-pipelines DELETE their ## Slot block; databricks-apps → bare ## Slot: deploy-target; api-style MCP typing (only tools are JSON-Schema-typed).
concern-resolution.md §2: add selection bullets for every non-exclusive/guide concern + the architecture-style resolution rule (resolve only when an architecture-style member is selected). authorization-model now exists → resolves multi-tenancy’s forward reference.
Commit the whole batch on helix-self-improvement-2026-05-24.

Make-concerns-bite (methodology iteration — it.40)

Thread concern-resolution into the feature-specification, technical-design, and implementation-plan prompts (today only ADR/solution-design/design-system/principles reference concerns — verified gap).
Add a reconcile-alignment concern→artifact realization check: a selected concern must appear in its governing artifacts (codex gave the per-concern “selecting X must change artifact Y” map — use it). Empirically tested by the re-bench (diff artifacts with/without a concern).

Operationalize “check your work again” (methodology iteration — it.41)

A bounded evolve-until-converged loop: build → reconcile-alignment + it.39 verification gate → if findings, evolve against them → re-review → repeat (max-iterations cap) until all ACs SATISFIED (guard branches), zero blocking findings, specs aligned. it.39 set the EXIT criteria; this is the LOOP. File default; DDx → bead.

Re-bench (after the above)

6 one-shots = (claude,codex,ddx) × (invoicing,hubspot), each iterating to convergence under: the new concerns (DDD-minimum, both projects), the it.39 verification exit gate, and the converge loop. ddx via ddx plugin install helix --local /Users/erik/Projects/helix --force (symlink — verified; picks up the new concerns automatically).

Parked (operator-gated, unprompted = no)

push origin/main (local main at 38dffac; branch ahead with it.39 + this batch).
INT-CC release-gate slice (needs the ANTHROPIC_API_KEY repo secret).