Summary
Fresh Opus session executing the policy-as-code handoff prompt v2.0. Produced the architecture spec, standalone spike brief, and Ch 3+ extension of reading notes. Recommendation is option B (first-class co-equal Apache policy layer alongside JSON Schema + TemporalRule). Sourcemeta partnership, MFI launch target, and four-tier commercial model all preserved under the recommendation.
Mid-session pivot — Rich expanded the mandate
While drafting, Rich added project_inherit_architecture_mandate.md: willingness to throw away JSON Schema entirely if a better paradigm (Semantic Web/SHACL, Akoma Ntoso, Datalog, logic programming, etc.) emerges. Eight candidate paradigms under consideration; 2–3 day broader evaluation pending.
Handled by reframing this spec as a sub-question within the broader evaluation rather than rescoping mid-flight or stopping to ask. Spec now carries an explicit “Scope note” in §1 stating that its A/B/C/D options all assume JSON Schema foundation; if broader evaluation picks a non-JSON-Schema paradigm, §3 decision test + §4 engine selection + §5 layer architecture + §11 spike brief remain reusable.
Outputs
| Artefact | Version | Location |
|---|---|---|
| Policy-as-code architecture spec | v1.0 | docs/superpowers/specs/2026-04-17-inherit-policy-as-code-architecture.md |
| Spike brief (paste-in) | v1.0 | docs/superpowers/scoping/policy-as-code-spike-brief.md |
| Reading notes (Ray 2024) | v0.2 | docs/superpowers/specs/policy-as-code-reading-notes.md |
| InheritKit design spec | v1.13 | docs/superpowers/specs/2026-04-16-inheritkit-design.md (v1.13 update + §2.16 extension) |
PDFs in Downloads: inherit-policy-as-code-architecture.pdf (v1.0), policy-as-code-spike-brief.pdf (v1.0), inheritkit-design.pdf (v1.13).
Key findings from book (via subagent on Ch 3+)
- OSCAL + Lula (Ch 15 pp 424–427) is the direct architectural analogue for INHERIT + InheritKit + policies. Open standard primary; Rego policies embedded. Strong evidence for option B, against C/D.
- OCI registries as policy distribution (OPCR, Ch 3) is the 2024 modern stance — free, signed, versioned, universal. Updated §5.4 to make OCI-first the default, OPA bundle service as Tier 1 managed offering.
- Cedar IS in the book (Ch 15 pp 436–438, corrected from v0.1 notes). Book explicitly pitches Cedar for “highly regulated industries” — strengthens the authorisation-slice case (PoA, executor powers).
- Effective-date routing is genuinely an INHERIT invention — no pattern in book. FW-PAC-03 confirmed as ours to solve.
- No legal-domain case study anywhere. INHERIT extends PaC into a domain the book doesn’t exemplify; spec acknowledges this plainly.
Decision summary
| Variable | Under option B | Rich sign-off needed? |
|---|---|---|
| Sourcemeta partnership | Continue; propose additive policy-bundle hosting conversation | No |
| MFI launch target | Unchanged (late 2028) | No |
| Four-tier commercial model | Tier 2 scope description revised; structure intact | No |
Option B was chosen partly because it was the ONLY option that delivered the interop improvement without triggering adjacent-arrangement changes.
Refined decision test (§3 of spec)
Apache policy if: output is yes/no or category; inputs schema-expressible; engine-agnostic; legal-fact status. InheritKit if: numeric computation; cross-layer integration; workflow; AI; performance-sensitive; close call (InheritKit-first principle).
Applied to SUCC-SCR-01–08 and TAX-SCR-01–29: ~30–50% rule-application surface becomes Apache policies.
Open items (FW-PAC-01–10)
Tagged in spec §10. Include: Cedar vs Rego for authorisation (confirmed by spike); Rego v0/v1 pin; TemporalRule helper pattern; bundle signing trust root; JSON Logic companion profile (probably deferred); whether Scoping Framework needs FW-PAC template field; conformance certification model (self vs central).
Interaction pattern worth preserving
Rich added a substantive new memory (project_inherit_architecture_mandate.md) mid-drafting without interrupting. Handled by finalising spec as-is with prominent reframing rather than stopping. This matches feedback memory preference for “take action, tell Rich what you did” over “shall I proceed.” End-of-session report flagged the positioning choice so Rich can redirect.
What’s next
- Rich reviews spec + spike brief PDFs
- Decide: kick off 1–2 week OPA/Cedar spike (using spike brief as paste-in prompt), OR first run the broader 8-paradigm evaluation, OR both in parallel
- If spike runs: decision gate at end of week 2 on option B commitment for v4.0.0
- Broader 8-paradigm evaluation is a separate workstream per
project_inherit_architecture_mandate.md