docs(milestone): complete OpenYield v0.4 (Refinement — NFR)

Milestone v0.4 complete. NFR type, tags v0.3.x. Four NFR REQs shipped:
- REQ-029 lexicon.SyntheticBannedStrings() shared helper (G-014 closed)
- REQ-030 x/hub↔x/bond cross-const equality test (A-304 closed, G-015)
- REQ-031 council lifecycle divergence docs + regression guard (AUDIT §193)
- REQ-032 .gitea/workflows/docs-build.yml docs build CI (D-046 landed)

NFR purity gate GREEN: zero feat: commit subjects in the milestone.
go.mod unchanged (G-006). G-003 production firewall intact. Coverage:
x/hub 93.3%, x/council 96.4% (both above 80% target).

Phases: P0 -> v0.3.0, P1 -> v0.3.1, P2 -> v0.3.2, P3 -> v0.3.3,
P4 -> v0.3.4 (milestone release).

Requirements covered: REQ-029, REQ-030, REQ-031, REQ-032 (all complete).
Live-runtime promotions of v0.3 Bearers skeletons deferred to v0.5+
(feat:-class, rejected by D-001 refinement-only filter).

---ci---
project: oy
phase: 4
milestone: v0.4
status: complete
tag_base: v0.3.x
milestone_type: nfr
phase_role: final
milestone_complete: true
requirements:
  covered: [REQ-029, REQ-030, REQ-031, REQ-032]
  partial: []
---/ci---

.ciagent/CHECKPOINT.json

@@ -1,11 +1,14 @@
 {
-  "phase": 0,
-  "stage": "mvp_ux_check",
+  "phase": 3,
+  "stage": "complete",
   "milestone": "v0.4",
   "milestone_type": "nfr",
   "tag_base": "v0.3.x",
-  "phase_role": "pre_execution",
+  "phase_role": "execution",
   "project": "oy",
   "attempts": 0,
-  "updated_at": "2026-08-17T22:50:00Z"
+  "updated_at": "2026-08-17T23:45:00Z",
+  "phase_release_tag": "v0.3.3",
+  "release_id": 751,
+  "reqs_covered": ["REQ-029", "REQ-030", "REQ-031", "REQ-032"]
 }

.ciagent/oy/ARCHITECTURE.md

@@ -227,4 +227,36 @@ v0.4 touches NO new `x/*` modules. The touched files are:
 
 ### v0.4 Interface Contracts (unchanged from v0.3)
 
-v0.4 does not change any cross-component interface. The 6 cross-component interfaces (Standing, Forge/Fold, Mirror, Window, Fee Covenant, Voice/Council) are unchanged. REQ-031 documents a divergence in the Voice/Council interface surface (SignalKind shape) but does not change it.
+v0.4 does not change any cross-component interface. The 6 cross-component interfaces (Standing, Forge/Fold, Mirror, Window, Fee Covenant, Voice/Council) are unchanged. REQ-031 documents a divergence in the Voice/Council interface surface (SignalKind shape) but does not change it.
+
+### Council Voice/Council Interface — Lifecycle Type Divergence Decisions (v0.4, REQ-031)
+
+This section documents the lifecycle type shape-divergences flagged by AUDIT.md §193 for the Council/Voice interface surface. v0.4 is a refinement-only NFR milestone (D-047): the D-001 filter REJECTS `feat:`-class enum additions and locked-const shape changes, so these divergences are DOCUMENTED here, not fixed in code. A regression-guard test (`TestSignalKindShapeIntentional` in `x/council/types/types_test.go`) locks the current shape so a future agent does not silently "fix" a divergence by changing a locked const.
+
+**Divergence P1-1 (AUDIT §193): `Proposal`/`ProposalStatus`/`VoteOption` enums absent from `x/council/types`.**
+
+- **Spec source**: P3-01-01 deliverable recommended `Proposal`, `ProposalStatus` (5 states), `VoteOption` (3 options) enums mirroring OZ Governor / `x/gov`.
+- **Implemented**: `Council`, `CouncilMember`, `Voice`, `SignalKind`, `TallyResult` — no `Proposal`/`ProposalStatus`/`VoteOption` lifecycle types.
+- **Must-have impact**: NONE. The v0.2 P3 must-haves (3 councils, Mission Lock, `TallyResult` x/gov shape, no veto) are all met without the Proposal lifecycle.
+- **Decision (v0.4, D-050)**: ADDING `Proposal`/`ProposalStatus`/`VoteOption` is a `feat:`-class addition (new enum types). REJECTED by the D-001 refinement-only filter. **Deferred to v0.5+** when the council keeper is wired to a live governance runtime (the AUDIT's own recommendation: "add in v0.3 when wiring the council keeper to a live governance runtime"). The skeleton council keeper in v0.2 does not consume a Proposal lifecycle; adding the types without the runtime would be dead code.
+- **Severity (AUDIT)**: P1 (spec drift from deliverable text, not a must-have, not blocking).
+- **v0.4 action**: DOCUMENT only (this section). No code change.
+
+**Divergence P1-2 (AUDIT §193): `SignalKind` 4 sources vs spec `VoiceSource` 5 sources.**
+
+- **Spec source**: P3-01-01 deliverable specified `VoiceSource` with 5 sources (Stash/Standing/Vouch/Freeholder/Guild).
+- **Implemented**: `SignalKind` with 4 sources: `SignalStash`, `SignalStanding`, `SignalVouch`, `SignalCapital` (`SignalKindCount = 4`, locked const).
+- **Code rationale (AUDIT §193 P1-2)**: the 4-source shape is a defensible design refinement, not a defect:
+  - `Freeholder` is an ELIGIBILITY property (upstream in `x/standing`), not a voice signal. A Freeholder-eligible Reach is a precondition for voting, not a signal that feeds a vote's weight.
+  - `Guild` is a COUNCIL TIER (one of the three councils is the Guild Council), not a voice signal. Including Guild as a signal kind would conflate the council tier with the signal source.
+  - `Capital` is committed-capital (vision §9.1, one of the four Freeholder signals), which the spec's `VoiceSource` list omitted. Adding `Capital` corrects the spec list to match vision §9.1's four-signal definition (REQ-005: "Four Freeholder signals locked").
+- **Must-have impact**: NONE. The v0.2 P3 must-haves did not enumerate `VoiceSource` coverage; the 4-signal shape matches REQ-005's "Four Freeholder signals locked" exactly.
+- **Decision (v0.4, D-050)**: changing `SignalKindCount` 4→5 (to restore the spec's 5-source `VoiceSource`) is a LOCKED-CONST CHANGE. REJECTED by the D-001 refinement-only filter (changing a locked const is a behavioral change, not a refinement). The 4-source shape is the CORRECT shape per vision §9.1 and REQ-005; the spec deliverable text was wrong, not the implementation.
+- **Severity (AUDIT)**: P1 (design-choice divergence, tested and self-consistent, not blocking).
+- **v0.4 action**: DOCUMENT the rationale here + add `TestSignalKindShapeIntentional` (regression guard) so a future agent changing `SignalKindCount` from 4 to 5 must also update the intent-assertion test, surfacing the AUDIT rationale for review. No locked-const change.
+
+**Divergence P2 (AUDIT §193): `x/bearers/types` `ValidateGenesis` no-op.**
+
+- **Spec source**: P4-02-01 said "DefaultParams/GenesisState unchanged" (bearers is an EXTENSION in v0.2, not a new module; the A-212 `ValidateGenesis` upgrade was scoped to NEW modules only).
+- **Implemented**: `ValidateGenesis` remains a no-op (`x/bearers/types/types.go:108` returns `nil` unconditionally).
+- **Decision (v0.4)**: CORRECT per spec — no action (AUDIT explicitly notes "no action"). The A-212 upgrade applies to NEW modules (v0.2's `x/window`, `x/stand`, etc.), not to EXTENDED modules like `x/bearers`. Listed here for completeness; no code change, no test change.

.ciagent/oy/AUDIT.md

@@ -408,3 +408,108 @@ AUDIT.md appended:  /root/oy/.ciagent/oy/AUDIT.md  ✓ (v0.3 section appended; v
 ```
 
 AUDIT.md
+
+---
+
+# Audit: OpenYield (oy) — v0.4 (Refinement — NFR) Final Phase
+
+> **Auditor**: CIAgent security-auditor (read-only on source; writes only to `.ciagent/`)
+> **Date**: 2026-08-17
+> **Target**: v0.4 milestone (main..oy/milestone/v0.4-refinement) + `.ciagent/oy/` files
+> **Milestone**: v0.4 — Refinement (NFR)
+> **Autonomy**: full
+
+## 1. Reconstruction Test (git log ↔ `.ciagent/` files)
+
+### 1.1 Phase progression
+
+The v0.4 milestone progresses through 5 phases, each with a `---ci---` block and a tag:
+
+| Phase | Stage commits | Tag | Release ID | `.ciagent/` evidence |
+|-------|---------------|-----|-----------|----------------------|
+| P0 (pre-execution) | specify→clarify→research→plan→grill→mvp_ux_check→complete | v0.3.0 | 748 | PROJECT.md v0.4 section, REQUIREMENTS.md v0.4 table, ARCHITECTURE.md v0.4 section, PERSONAS.md v0.4, PLANS.md v0.4 plan, GRILL.md v0.4 section |
+| P1 (lexicon+const) | execute→verify→complete | v0.3.1 | 749 | REVIEW.md (P1), `lexicon/lexicon.go` helper, `x/hub/types/cross_const_test.go` |
+| P2 (lifecycle docs) | execute→verify→complete | v0.3.2 | 750 | ARCHITECTURE.md divergence section, `x/council/types/types_test.go` intent test |
+| P3 (docs CI) | execute→verify→complete | v0.3.3 | 751 | `.gitea/workflows/docs-build.yml`, `.gitignore` site/ |
+| P4 (final review+ship) | in progress | v0.3.4 (pending) | — | this AUDIT.md section + REVIEW.md v0.4 section |
+
+Reconstruction: the git log subject lines match the `.ciagent/` file state. Each phase's `---ci---` block is present in the commit messages. Tags v0.3.0..v0.3.3 exist. **PASS.**
+
+### 1.2 `.ciagent/` file discipline
+
+14 files in `.ciagent/oy/` (ARCHITECTURE, AUDIT, GRILL, P1-P4_SHIP_VERIFICATION [P1-P3 from v0.2/v0.3, P4 pending], PERSONAS, PLANS, PROJECT, REQUIREMENTS, RESEARCH, REVIEW, ROADMAP). All present and updated for v0.4. CHECKPOINT.json reflects the current phase. config.json has `milestone: v0.4`, `milestone_type: nfr`, `tag_base: v0.3.x`. **PASS.**
+
+### 1.3 Branch hygiene
+
+Local: `main`, `oy/milestone/v0.4-refinement`, `oy/phase/04-final-review-ship` (current). Remote: `origin/main`, `origin/oy/milestone/v0.4-refinement`. All merged execution phase branches (P1, P2, P3) deleted locally. No leftover phase branches. The P4 final phase branch will be deleted at milestone ship. **PASS.**
+
+### 1.4 Commit discipline
+
+All 20 milestone commits have `---ci---` blocks with `project: oy`, `phase: N`, `milestone: v0.4`, `status: <stage>`, `tag_base: v0.3.x`, `milestone_type: nfr`. Commit subjects use conventional prefixes (docs, refactor, test, chore, verify, decision, checkpoint, Merge). No `feat:` subjects. **PASS.**
+
+### 1.5 Build / test / coverage sanity
+
+- `go build ./...`: clean.
+- `go test ./...`: green (26 packages, all ok).
+- `go vet ./...`: clean on touched packages.
+- Coverage: `x/hub/types` 93.3% (v0.3 floor preserved; new cross-const test adds coverage). `x/council/types` 96.4% (improved from v0.3). Both above the 80% target (D-033). `go.mod` unchanged (zero deps, G-006). **PASS.**
+
+## 2. NFR Purity Gate (v0.4 mandatory)
+
+The NFR purity gate (D-047, D-001 filter) requires zero `feat:` commits in the v0.4 milestone range. The gate is checked on COMMIT SUBJECTS (not bodies), because `git log --grep` over-matches commit bodies that mention "feat:" in prose (e.g., the verify commits say "zero feat: commits in P2" in the body).
+
+**Gate command**: `git log --format="%s" main..HEAD | grep -E "^feat:"`
+
+**Result**: exit 1 (zero matches). The 20 commit subjects are: `docs(init)`, `docs(P00)` ×3, `docs(P00)` grill, `decision(P00)`, `Merge` ×3, `refactor(lexicon)`, `verify(P1)`, `checkpoint(p1)` ×3, `docs(arch)`, `verify(P2)`, `checkpoint(p2)`, `chore(ci)`, `verify(P3)`, `checkpoint(p3)`. None start with `feat:`.
+
+**NFR purity gate: GREEN.** ✓
+
+## 3. Non-Critical Observations (P1+ flags, not blocking)
+
+### P2-1: mkdocs build warnings (pre-existing v0.3 docs links)
+- **Files**: `docs/index.md`, `docs/shared/vision.md` (link to `../README.md` / `../../README.md`)
+- **Note**: `mkdocs build` produces 2 warnings about README.md not being in the docs tree. The build SUCCEEDS (warnings, not errors). These are pre-existing v0.3 docs content links, NOT introduced by v0.4. v0.4's REQ-032 is the CI workflow (which runs `mkdocs build` and succeeds despite the warnings), not the docs content.
+- **Recommendation**: post-hoc fix in v0.5+ (either include README.md in mkdocs nav, or fix the relative links). Not a v0.4 ship blocker.
+
+### P2-2: Gitea Actions `actions/upload-artifact@v4` compatibility
+- **File**: `.gitea/workflows/docs-build.yml`
+- **Note**: the workflow uses `actions/upload-artifact@v4`. Gitea Actions (which reimplements GitHub Actions) supports most `actions/*` but v4 of upload-artifact has had compatibility quirks. If the upload step fails, the `mkdocs build` step (the higher-priority check) would still have succeeded.
+- **Recommendation**: post-hoc verify on the first real CI run; downgrade to v3 or use Gitea-native upload if needed. Not a ship blocker (the build is the gate, the artifact is a bonus).
+
+### P2-3: NFR purity gate precision (documented in REVIEW.md P1+ #3)
+- **Note**: `git log --grep "^feat:"` over-matches commit bodies. The audit uses the subject-only gate (`--format="%s" | grep -E "^feat:"`). This is documented for future milestones.
+- **Recommendation**: bake the subject-only gate into the next milestone's PLAN. Not a ship blocker.
+
+## 4. Overall Audit Verdict
+
+### **PASS**
+
+The v0.4 (Refinement — NFR) milestone is **shippable**.
+
+**Per-check summary:**
+
+| # | Check | Verdict |
+|---|-------|---------|
+| 1.1 | Reconstruction (phase progression, tags, `.ciagent` evidence) | PASS |
+| 1.2 | `.ciagent` file discipline (14 files, CHECKPOINT, config.json) | PASS |
+| 1.3 | Branch hygiene (no leftover phase branches; P4 final pending) | PASS |
+| 1.4 | Commit discipline (`---ci---` blocks; conventional subjects; no feat:) | PASS |
+| 1.5 | Build / test / coverage sanity (build, test, ≥80%, lexicon, G-006) | PASS |
+| 2 | **NFR purity gate** (zero `feat:` commit subjects) | **GREEN** |
+
+**Critical issues: 0.**
+**Non-critical observations: 3** (all P2, post-hoc, non-blocking).
+**Confidence in overall verdict: 0.90.**
+
+## STRIDE security summary (per ci-auditor role, read-only)
+
+| Category | Finding | Severity | Disposition |
+|---|---|---|---|
+| Spoofing | No auth surface added (v0.4 is refactor+test+docs+CI; no new identity logic) | Low | Accept |
+| Tampering | The cross-const test (REQ-030) HARDENS tamper-resistance: a future change to `x/bond.CouponCapBps` or `x/hub.LendingCouponCapBps` is now detected by an automated test (was comment-only before v0.4). The absolute-value assertion (G-015) catches paired drift. Mission Lock (8%/0%) is MORE defended after v0.4. | Low (improved) | Accept |
+| Repudiation | No audit-log changes in v0.4 | Low | Accept |
+| Info Disclosure | The CI workflow (REQ-032) does NOT publish (build+artifact only, D-051); no secret exposure in the workflow YAML; `GITEA_TOKEN` is resolved via `resolveSecret()` from `.ciagent/.env.secrets` for release creation, never via shell-env `curl` | Low | Accept |
+| Denial of Service | No network surface added; the CI workflow runs on push but does not expose a service | Low | Accept |
+| Elevation of Privilege | No privilege surface added; the lexicon helper is a pure function; the regression guard only asserts existing consts | Low | Accept |
+
+No threat exceeds the low/accept threshold. No escalations. v0.4 hardens the mission-locked const firewall (REQ-030) and the lexicon firewall (REQ-029) without introducing any new attack surface.

.ciagent/oy/PROJECT.md

@@ -61,7 +61,7 @@ OpenYield (OY) is a durable, anti-greed, jurisdiction-light financial layer —
 - D-009: Rebased history to fix v1.0 → v0.1 in ---ci--- blocks
 
 ## Milestone
-v0.4 — Refinement (active milestone; NFR type; tags run on the v0.3.x patch line)
+v0.4 — Refinement (complete; NFR type; tags ran on the v0.3.x patch line)
 
 ### v0.4 Scope (Refinement-only NFR — v0.3 post-hoc forward-references)
 

.ciagent/oy/REQUIREMENTS.md

@@ -52,15 +52,27 @@ refinement-only filter applies to any IDEATE stage.
 
 | ID | Requirement | Source | Class | Priority | Status | Phase |
 |----|-------------|--------|-------|----------|--------|-------|
-| REQ-029 | Lexicon firewall: shared `lexicon.SyntheticBannedStrings()` helper — dedupe the synthetic self-test table between `lexicon_meta_test.go` and `lexicon_meta_docs_test.go`; both meta-tests derive count + strings from the single source so a future banned-term addition updates both firewalls from one place | GRILL G-014 | refactor/test | High | Pending | v0.4/P1 |
-| REQ-030 | Cross-package const-equality test: `x/hub.LendingCouponCapBps == x/bond.CouponCapBps` (and Floor) — test-only import (G-003 exempt), catches silent mission-lock drift between hub LOCAL consts and bond D-028 consts | REVIEW.md P2 / A-304 | test | High | Pending | v0.4/P1 |
-| REQ-031 | x/* lifecycle type shape-divergence review + alignment fixes — audit non-must-have lifecycle types across modules flagged by AUDIT §193; align shapes where divergent (no behavioral change) | AUDIT.md §193 | refactor/quality | Medium | Pending | v0.4/P2 |
-| REQ-032 | Docs build CI — Gitea Actions workflow that runs `go test ./...` (lexicon firewall) + `mkdocs build` on every push; upload the built `site/` as a CI artifact. Full Gitea Pages publishing deferred if no hosting target is configured (chore, not feat) | D-046 | chore/ci | Medium | Pending | v0.4/P3 |
+| REQ-029 | Lexicon firewall: shared `lexicon.SyntheticBannedStrings()` helper — dedupe the synthetic self-test table between `lexicon_meta_test.go` and `lexicon_meta_docs_test.go`; both meta-tests derive count + strings from the single source so a future banned-term addition updates both firewalls from one place | GRILL G-014 | refactor/test | High | Complete | v0.4/P1 |
+| REQ-030 | Cross-package const-equality test: `x/hub.LendingCouponCapBps == x/bond.CouponCapBps` (and Floor) — test-only import (G-003 exempt), catches silent mission-lock drift between hub LOCAL consts and bond D-028 consts | REVIEW.md P2 / A-304 | test | High | Complete | v0.4/P1 |
+| REQ-031 | x/* lifecycle type shape-divergence review + alignment fixes — audit non-must-have lifecycle types across modules flagged by AUDIT §193; align shapes where divergent (no behavioral change) | AUDIT.md §193 | refactor/quality | Medium | Complete | v0.4/P2 |
+| REQ-032 | Docs build CI — Gitea Actions workflow that runs `go test ./...` (lexicon firewall) + `mkdocs build` on every push; upload the built `site/` as a CI artifact. Full Gitea Pages publishing deferred if no hosting target is configured (chore, not feat) | D-046 | chore/ci | Medium | Complete | v0.4/P3 |
 
 > REQ-029..REQ-032 are NEW in v0.4. All are NFR classes (refactor/test/quality/
 > chore) — zero `feat:` phases by construction. The final-phase audit enforces
 > the NFR purity gate (zero `feat:` commits in the milestone).
 
+## Milestone v0.4 Summary (Refinement — NFR) — COMPLETE
+
+- 4 v0.4-scope REQs shipped as NFR (refactor/test/docs/chore): REQ-029, REQ-030, REQ-031, REQ-032
+- Closes 3 real v0.3 forward-references: GRILL G-014 (lexicon drift), REVIEW P2/A-304 (const drift), AUDIT §193 (council divergence docs)
+- Lands the D-046 docs-CI forward-reference (.gitea/workflows/docs-build.yml, build+artifact, no Pages publish per D-051)
+- NFR purity gate GREEN: zero `feat:` commit subjects in the milestone (20 commits, all docs/refactor/test/chore/verify/decision/checkpoint/Merge)
+- `go.mod` unchanged (G-006 — zero Go deps; Python deps isolated to CI docs-build job)
+- G-003 production firewall intact (no production import of `x/bond/types` in `x/hub/types`; cross-const test is test-only)
+- Coverage: x/hub/types 93.3% (v0.3 floor preserved), x/council/types 96.4% (improved); both above 80% target
+- Tags: v0.3.0 (P0) -> v0.3.1 (P1) -> v0.3.2 (P2) -> v0.3.3 (P3) -> v0.3.4 (P4 = v0.4 milestone release)
+- Tag-line note: v0.4 (NFR) ships on the v0.3.x patch line (config tag_base). The v0.3.4 milestone release IS the deliverable (D-008 — final phase patch IS the milestone release; no separate minor tag).
+
 ## IDEATE Traceability (Phase 0 — IDEATE stage, autonomy=full)
 
 The IDEATE stage ran the three ideation tiers (mechanical, backend-enriched,

.ciagent/oy/REVIEW.md

@@ -365,3 +365,59 @@ All verification commands pass. All locked-const invariants enforced and tested.
 **P0 fixes auto-applied: 0**
 **P1+ findings: 0 P1, 3 P2 (all nits, post-hoc, non-blocking)**
 **Confidence in overall verdict: 0.91**
+
+---
+
+# Review: OpenYield (oy) — v0.4 (Refinement — NFR) Final Phase
+
+> **Reviewer**: CIAgent multi-persona code review (lead-developer + backend-engineer lenses)
+> **Date**: 2026-08-17
+> **Target**: All v0.4 milestone commits (main..oy/milestone/v0.4-refinement) across P1..P3
+> **Milestone**: v0.4 — Refinement (NFR)
+> **Autonomy**: full
+
+## Scope Reviewed
+
+The v0.4 milestone ships 4 REQs across 3 execution phases (P1..P3) + phase 0 (pre-execution). 20 commits, 15 files changed (+764 / -157). The review covers the execution-phase deliverables:
+
+- **P1** (v0.3.1): REQ-029 lexicon shared helper, REQ-030 cross-const test
+- **P2** (v0.3.2): REQ-031 lifecycle divergence docs + regression guard
+- **P3** (v0.3.3): REQ-032 docs build CI
+
+## Adversarial Probes
+
+1. **Does the shared helper actually dedupe?** Probe: `grep -rn 'open a.*here\|make a.*now\|compounding.*rate' lexicon_meta_test.go lexicon_meta_docs/` — returns ZERO matches (the old duplicated table is gone; both meta-tests now call `lexicon.SyntheticBannedStrings()`). Verified at `lexicon_meta_test.go:93` and `lexicon_meta_docs/lexicon_meta_docs_test.go:155` (both consume the helper). ✓
+
+2. **Does the cross-const test fail closed on drift?** Probe: the test imports `bondtypes "github.com/oy/openyield/x/bond/types"` in `x/hub/types/cross_const_test.go:28` and asserts `LendingCouponCapBps != bondtypes.CouponCapBps` would fail the test. The absolute-value test `TestConstsAreMissionLocked800And0` (G-015) catches paired drift (both consts → 900). Both paths verified by reading the test. ✓
+
+3. **Does the regression guard lock the 4-signal shape?** Probe: `TestSignalKindShapeIntentional` at `x/council/types/types_test.go` asserts `SignalKindCount == 4` with a `t.Fatalf` (not `t.Errorf`) and checks `AllSignalKinds()` returns `[SignalStash, SignalStanding, SignalVouch, SignalCapital]` in order. The doc comment includes the AUDIT §193 P1-2 rationale. Changing `SignalKindCount` to 5 fails this test AND `TestSignalKindCountLockedConst`. ✓
+
+4. **Does the CI workflow YAML parse and enforce G-016?** Probe: `python3 -c "import yaml; doc=yaml.safe_load(open('.gitea/workflows/docs-build.yml')); assert doc['jobs']['docs-build']['needs'] == 'go-test'"` — passes. The `needs: go-test` line is present at `.gitea/workflows/docs-build.yml` in the `docs-build` job. ✓
+
+5. **Is `go.mod` really unchanged across the whole milestone?** Probe: `git diff main..HEAD -- go.mod` — empty. The CI workflow's Python deps are in a separate job; the lexicon helper adds no Go deps; the cross-const test adds no Go deps (test-only import of an internal package). G-006 intact. ✓
+
+6. **NFR purity gate — zero `feat:` commit SUBJECTS?** Probe: `git log --format="%s" main..HEAD | grep -E "^feat:"` — exit 1 (no matches). All 20 subjects are `docs(`, `refactor(`, `test(`, `chore(`, `verify(`, `decision(`, `checkpoint(`, or `Merge`. NOTE: `git log --grep "^feat:"` matches commit BODIES too (the v0.4 verify commits mention "feat:" in prose like "zero feat: commits in P2"); the correct gate uses subject-only check via `--format="%s"`. This is a documented refinement for the audit. ✓
+
+7. **Does G-003 hold — no production cross-module struct imports added?** Probe: `grep -rn "x/bond/types" x/hub/types/ --include="*.go" | grep -v "_test.go"` — zero matches. The only import of `x/bond/types` in `x/hub/types/` is in `cross_const_test.go` (a `_test.go` file, G-003 test-exempt). Production firewall intact. ✓
+
+## P0 fixes auto-applied
+
+**0.** No P0 (critical) issues found. The milestone ships clean.
+
+## P1+ flags (post-hoc review — do NOT block ship)
+
+**1.** [P2 nit, maintainability] The `mkdocs build` local run produced 2 warnings about README.md links (`docs/index.md` links to `../README.md`, `docs/shared/vision.md` links to `../../README.md`). These are pre-existing v0.3 docs links (not introduced by v0.4); the warnings are non-blocking (mkdocs builds successfully). Recommend post-hoc: either configure `mkdocs.yml` to include README.md in the nav, or fix the relative links. Not a v0.4 ship blocker (the docs site builds; v0.4's REQ-032 is the CI workflow, not the docs content).
+
+**2.** [P2 nit, CI] The `docs-build.yml` workflow uses `actions/upload-artifact@v4` with `retention-days: 14`. Gitea Actions may have a different artifact retention default; the explicit `retention-days: 14` is defensive. If Gitea Actions does not support v4 of the upload-artifact action, the workflow would fail at the upload step (the `mkdocs build` itself would have succeeded). Recommend post-hoc: verify Gitea Actions supports `actions/upload-artifact@v4`; if not, downgrade to v3 or use the Gitea-native artifact upload. Not a ship blocker (the build itself is the higher-priority check; the artifact upload is a bonus).
+
+**3.** [P2 nit, audit-gate precision] The NFR purity gate (`git log --grep "^feat:"`) over-matches commit bodies. The correct gate is subject-only (`git log --format="%s" | grep -E "^feat:"`). The P4 audit uses the subject-only check. Recommend post-hoc: document the subject-only gate in the next milestone's PLAN so future audits do not over-count. Not a ship blocker (the subject-only gate is green).
+
+## Overall verdict
+
+**SHIP.**
+
+All four REQs (REQ-029..REQ-032) delivered. The three real v0.3 forward-references (G-014 lexicon drift, A-304 const drift, AUDIT §193 council divergence) are closed; the D-046 docs-CI forward-reference is landed. `go test ./...` green across all 26 packages. `go.mod` unchanged (G-006). G-003 production firewall intact. NFR purity gate GREEN (zero `feat:` commit subjects). No P0 issues. Three P2 nits flagged for post-hoc review (none blocking).
+
+**P0 fixes auto-applied: 0**
+**P1+ findings: 0 P1, 3 P2 (all nits, post-hoc, non-blocking)**
+**Confidence in overall verdict: 0.90**

.ciagent/oy/ROADMAP.md

@@ -69,6 +69,47 @@ Target: Bearers skeleton (ROADMAP Phase 3 subset) + docs site for nomads and fre
 > (= the v0.3 milestone release, per D-008 — final phase patch IS the
 > milestone release; no separate minor tag).
 
+## Milestone v0.4 — Refinement (COMPLETE; NFR type; tags v0.3.x)
+
+Target: Close the v0.3 post-hoc forward-references (lexicon firewall drift,
+hub↔bond const drift, council lifecycle type divergence) + land the deferred
+docs build CI. Refinement-only NFR milestone: zero `feat:` phases.
+
+- [x] P0: Pre-Execution (spec/clarify/research/plan/grill/mvp-ux) → v0.3.0
+- [x] P1: Lexicon + const hardening (REQ-029, REQ-030) → v0.3.1
+- [x] P2: Lifecycle divergence docs + regression guard (REQ-031) → v0.3.2
+- [x] P3: Docs build CI (REQ-032) → v0.3.3
+- [x] P4: Final Review + Audit + Ship → v0.3.4 (milestone release)
+- Status: COMPLETE — 4 NFR REQs shipped; NFR purity gate GREEN (zero feat: commits); go.mod unchanged
+
+> v0.4 closes three real v0.3 forward-references (GRILL G-014 lexicon helper,
+> REVIEW P2/A-304 cross-const test, AUDIT §193 council divergence docs) and
+> lands the D-046 docs-CI forward-reference. Live-runtime promotions of the
+> v0.3 Bearers skeletons are deferred to v0.5+ (feat:-class, rejected by the
+> D-001 refinement-only filter).
+
+| Phase | Type | Scope | Patch |
+|---|---|---|---|
+| P0 | docs | Pre-Execution (spec/clarify/research/plan/grill/mvp-ux) | v0.3.0 |
+| P1 | refactor+test | Lexicon shared helper (REQ-029) + cross-const test (REQ-030) | v0.3.1 |
+| P2 | docs+test | Council lifecycle divergence docs (REQ-031) + regression guard | v0.3.2 |
+| P3 | chore+ci | Docs build CI workflow (REQ-032) | v0.3.3 |
+| P4 | final | REVIEW + AUDIT + milestone SHIP | v0.3.4 (milestone release) |
+
+### v0.4 Component mapping
+
+| Component | Deliverable | v0.4 Change | Phase |
+|---|---|---|---|
+| Lexicon firewall | Shared `SyntheticBannedStrings()` helper | `lexicon/lexicon.go` + both meta-tests refactored | v0.4/P1 |
+| Mission-locked const firewall | Cross-package const-equality test | `x/hub/types/cross_const_test.go` (NEW) | v0.4/P1 |
+| Council Voice/Council interface | Lifecycle divergence documentation + regression guard | ARCHITECTURE.md section + `x/council/types/types_test.go` intent test | v0.4/P2 |
+| Docs CI | Gitea Actions workflow (build + artifact) | `.gitea/workflows/docs-build.yml` (NEW) | v0.4/P3 |
+
+> **Tag-line note (G-010 continuation)**: v0.4 (NFR) ships on the `v0.3.x`
+> patch line (config.json `tag_base: v0.3.x`): P0 -> `v0.3.0`, P1..P3 ->
+> `v0.3.1..v0.3.3`, P4 -> `v0.3.4` (= the v0.4 milestone release, per D-008 —
+> final phase patch IS the milestone release; no separate minor tag).
+
 ## Phase 3 — The Bearers (Year 3) — v0.3 PARTIAL SKELETON
 **Target**: $10B annual volume → fee auto-declines to 0.07%
 

.gitea/workflows/docs-build.yml

@@ -0,0 +1,55 @@
+# OpenYield docs build CI (REQ-032, D-046 forward-reference, D-051, G-016).
+#
+# Runs the lexicon firewall (go test ./...) AND builds the MkDocs Material docs
+# site on every push. The docs-build job DEPENDS on go-test (G-016 binding:
+# firewall-gates-docs-build — a lexicon violation blocks the docs build so no
+# false-green docs artifact is produced from a repo with a firewall failure).
+#
+# Scope (chore, not feat: per D-001 refinement-only filter):
+#   - go-test job: setup Go 1.22, run `go test ./...` (lexicon firewall + all
+#     x/* tests + the v0.4 cross-const test). Zero external Go deps (G-006).
+#   - docs-build job: setup Python, pip install mkdocs + mkdocs-material
+#     (build-only Python deps, ISOLATED to this job — go.mod is NOT modified),
+#     run `mkdocs build` (produces site/), upload site/ as a CI artifact.
+#
+# Out of scope (deferred per D-051): full Gitea Pages publishing. v0.4 ships
+# build + artifact only; a hosting target is not configured.
+#
+# Triggers: on push (all branches) so the firewall + docs build are checked
+# on every change, not just on main.
+
+name: docs-build
+on:
+  push:
+
+jobs:
+  go-test:
+    name: go test ./... (lexicon firewall + all x/* tests)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version: '1.22'
+      - name: go test ./...
+        run: go test ./...
+
+  docs-build:
+    name: mkdocs build (docs site artifact)
+    runs-on: ubuntu-latest
+    needs: go-test  # G-016: firewall-gates-docs-build (no false-green docs build)
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+      - name: install mkdocs + mkdocs-material
+        run: pip install mkdocs mkdocs-material
+      - name: mkdocs build
+        run: mkdocs build
+      - name: upload site/ artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: docs-site
+          path: site/
+          retention-days: 14

.gitignore

@@ -2,3 +2,5 @@
 .env.secrets
 .env.*
 .ciagent/.env.secrets
+# MkDocs build output (REQ-032 CI produces site/ as an artifact; never commit it)
+site/

lexicon/lexicon.go

@@ -85,3 +85,41 @@ func FindBannedTerm(s string) (string, bool) {
 func ContainsBannedTerm(s string) (string, bool) {
 	return FindBannedTerm(s)
 }
+
+// SyntheticBannedStrings returns one synthetic string per banned term, each
+// embedding exactly one banned term in a plausible sentence context. This
+// is the single source of truth (REQ-029, GRILL G-014) for the synthetic
+// self-test table consumed by BOTH project-wide meta-tests:
+//
+//	lexicon_meta_test.go        :: TestLexiconMetaSelfTestTable        (package lexicon_meta, scans x/**/*.go)
+//	lexicon_meta_docs_test.go   :: TestLexiconMetaDocsSelfTestTable    (package lexicon_meta_docs, scans README.md + docs/**/*.md)
+//
+// Before REQ-029, both meta-tests DUPLICATED their own 10-string synthetic
+// table (byte-identical), creating a drift risk: a future banned-term
+// addition updating one table but not the other would silently drop coverage
+// in the unmaintained firewall. SyntheticBannedStrings() eliminates the
+// duplication — both meta-tests now consume this helper, so a future addition
+// updates both firewalls from one place. The strings are built from
+// BannedTerms() (already fragment-assembled), so this package's own source
+// stays lexicon-clean (the firewall's own code is allowed to name the terms
+// it bans, but only via the fragment-assembly bootstrapping pattern).
+//
+// The returned slice is indexed positionally against BannedTerms(): the i-th
+// synthetic string embeds the i-th banned term. Both meta-tests assert
+// len(SyntheticBannedStrings()) == len(BannedTerms()) and that each string
+// triggers FindBannedTerm with the matching term.
+func SyntheticBannedStrings() []string {
+	terms := BannedTerms()
+	return []string{
+		"open a " + terms[0] + " here",      // bank
+		"make a " + terms[1] + " now",       // deposit
+		"compounding " + terms[2] + " rate", // interest
+		"the " + terms[3] + " is 5pct",      // yield
+		"foreign " + terms[4] + " pair",     // currency
+		"price in " + terms[5],              // dollar
+		"price in " + terms[6],              // euro
+		"freeze the " + terms[7],            // account
+		"move to " + terms[8] + " now",      // savings
+		"the " + terms[9] + " lost money",   // depositor
+	}
+}

lexicon_meta_docs/lexicon_meta_docs_test.go

@@ -128,22 +128,14 @@ func TestLexiconMetaDocsNoBannedTermsInDocs(t *testing.T) {
 // breaks, this test fails before the firewall silently passes a real
 // violation in a docs page.
 //
-// G-014 self-test drift: this table is the docs mirror of the
-// TestLexiconMetaSelfTestTable in lexicon_meta_test.go (package lexicon_meta).
-// Both reuse lexicon.BannedTerms() as the single source for the 10 terms, so
-// a future addition updates both firewalls from one place. The synthetic
-// strings are assembled from lexicon.BannedTerms() fragments so this file
-// does not contain any banned term as a literal substring (it would otherwise
-// trip its own scan; the meta-test file is also excluded from its own scan,
-// but the self-test keeps the source clean for readability/searchability).
-//
-// CROSS-REFERENCE: keep this table aligned with
-//
-//	lexicon_meta_test.go :: TestLexiconMetaSelfTestTable
-//
-// Any change to the synthetic-string construction must be mirrored in both
-// files (or, preferably, add a shared helper in the lexicon package — see
-// G-014 minimum-viable: cross-reference comment + shared BannedTerms()).
+// REQ-029 (GRILL G-014): the synthetic strings are sourced from
+// lexicon.SyntheticBannedStrings(), the single source of truth shared with
+// lexicon_meta_test.go :: TestLexiconMetaSelfTestTable. Before REQ-029, this
+// file DUPLICATED its own 10-string table (byte-identical to the x/ meta-
+// test), creating a drift risk; the shared helper closes it. This file no
+// longer builds its own synthetic table — both meta-tests consume the same
+// helper, so a future banned-term addition updates both firewalls from one
+// place.
 func TestLexiconMetaDocsSelfTestTable(t *testing.T) {
 	terms := lexicon.BannedTerms()
 	// The spec lists 10 banned terms (plan docs say "9", counting dollar/euro
@@ -152,22 +144,10 @@ func TestLexiconMetaDocsSelfTestTable(t *testing.T) {
 	if len(terms) != 10 {
 		t.Fatalf("BannedTerms() len = %d, want 10", len(terms))
 	}
-	// Each synthetic string embeds exactly one banned term in a plausible
-	// sentence context. Each must be detected.
-	synthetic := []string{
-		"open a " + terms[0] + " here",      // bank
-		"make a " + terms[1] + " now",       // deposit
-		"compounding " + terms[2] + " rate", // interest
-		"the " + terms[3] + " is 5pct",      // yield
-		"foreign " + terms[4] + " pair",     // currency
-		"price in " + terms[5],              // dollar
-		"price in " + terms[6],              // euro
-		"freeze the " + terms[7],            // account
-		"move to " + terms[8] + " now",      // savings
-		"the " + terms[9] + " lost money",   // depositor
-	}
+	// REQ-029: consume the shared synthetic-string helper (G-014 single source).
+	synthetic := lexicon.SyntheticBannedStrings()
 	if len(synthetic) != len(terms) {
-		t.Fatalf("synthetic table len = %d, want %d", len(synthetic), len(terms))
+		t.Fatalf("SyntheticBannedStrings() len = %d, want %d (must match BannedTerms())", len(synthetic), len(terms))
 	}
 	for i, s := range synthetic {
 		found, ok := lexicon.FindBannedTerm(s)

lexicon_meta_test.go

@@ -76,10 +76,12 @@ func TestLexiconMetaNoBannedTermsInX(t *testing.T) {
 // firewall's detection logic is durably verified — if detection ever breaks,
 // this test fails before the firewall silently passes a real violation.
 //
-// The synthetic strings are assembled from fragments so this file does not
-// contain any banned term as a literal substring (it would otherwise trip
-// its own scan; the meta-test file is also excluded from the scan, but the
-// self-test keeps the source clean for readability/searchability).
+// REQ-029 (GRILL G-014): the synthetic strings are sourced from
+// lexicon.SyntheticBannedStrings(), the single source of truth shared with
+// lexicon_meta_docs_test.go :: TestLexiconMetaDocsSelfTestTable. Before
+// REQ-029, both meta-tests DUPLICATED their own 10-string table, creating a
+// drift risk; the shared helper closes it. This file no longer builds its
+// own synthetic table.
 func TestLexiconMetaSelfTestTable(t *testing.T) {
 	terms := lexicon.BannedTerms()
 	// The spec lists 10 banned terms (plan docs say "9", counting dollar/euro
@@ -88,22 +90,10 @@ func TestLexiconMetaSelfTestTable(t *testing.T) {
 	if len(terms) != 10 {
 		t.Fatalf("BannedTerms() len = %d, want 10", len(terms))
 	}
-	// Each synthetic string embeds exactly one banned term in a plausible
-	// sentence context. Each must be detected.
-	synthetic := []string{
-		"open a " + terms[0] + " here",      // bank
-		"make a " + terms[1] + " now",       // deposit
-		"compounding " + terms[2] + " rate", // interest
-		"the " + terms[3] + " is 5pct",      // yield
-		"foreign " + terms[4] + " pair",     // currency
-		"price in " + terms[5],              // dollar
-		"price in " + terms[6],              // euro
-		"freeze the " + terms[7],            // account
-		"move to " + terms[8] + " now",      // savings
-		"the " + terms[9] + " lost money",   // depositor
-	}
+	// REQ-029: consume the shared synthetic-string helper (G-014 single source).
+	synthetic := lexicon.SyntheticBannedStrings()
 	if len(synthetic) != len(terms) {
-		t.Fatalf("synthetic table len = %d, want %d", len(synthetic), len(terms))
+		t.Fatalf("SyntheticBannedStrings() len = %d, want %d (must match BannedTerms())", len(synthetic), len(terms))
 	}
 	for i, s := range synthetic {
 		found, ok := lexicon.FindBannedTerm(s)

x/council/types/types_test.go

@@ -109,6 +109,58 @@ func TestSignalKindCountLockedConst(t *testing.T) {
 	}
 }
 
+// TestSignalKindShapeIntentional (REQ-031, AUDIT §193 P1-2) is a regression
+// GUARD that documents and locks the 4-source SignalKind shape. It is NOT a
+// shape change — the existing TestSignalKindCountLockedConst already locks
+// the count. This test adds the INTENT documentation so a future agent who
+// changes SignalKindCount from 4 to 5 (e.g., to "restore" the spec's 5-source
+// VoiceSource list) must also update this test, surfacing the AUDIT rationale
+// for review.
+//
+// AUDIT §193 P1-2 rationale (why SignalKind is 4 sources, NOT the spec's 5):
+//
+//	The v0.2 P3-01-01 deliverable specified VoiceSource with 5 sources
+//	(Stash/Standing/Vouch/Freeholder/Guild). The implementation uses
+//	SignalKind with 4 sources (Stash/Standing/Vouch/Capital). The 4-source
+//	shape is a defensible design refinement:
+//	  - Freeholder is an ELIGIBILITY property (upstream in x/standing), not
+//	    a voice signal. A Freeholder-eligible Reach is a precondition for
+//	    voting, not a signal that feeds a vote's weight.
+//	  - Guild is a COUNCIL TIER (one of the three councils is the Guild
+//	    Council), not a voice signal. Including Guild as a signal kind
+//	    would conflate the council tier with the signal source.
+//	  - Capital is committed-capital (vision §9.1, one of the four
+//	    Freeholder signals per REQ-005), which the spec's VoiceSource list
+//	    omitted. Adding Capital corrects the spec to match vision §9.1's
+//	    four-signal definition (REQ-005: "Four Freeholder signals locked").
+//
+//	The 4-source shape matches REQ-005 exactly. The spec deliverable text
+//	was wrong, not the implementation. v0.4 (D-050) DOCUMENTS this and
+//	locks the 4-source shape; changing it to 5 is a locked-const change
+//	rejected by the D-001 refinement-only filter and deferred to a future
+//	milestone that re-litigates REQ-005's signal definition.
+//
+// See .ciagent/oy/ARCHITECTURE.md §"Council Voice/Council Interface —
+// Lifecycle Type Divergence Decisions (v0.4, REQ-031)" for the full rationale.
+func TestSignalKindShapeIntentional(t *testing.T) {
+	// LOCKED: 4 sources. Changing this to 5 requires updating this test's
+	// intent block AND re-litigating REQ-005's four-signal definition.
+	const expectedSignalCount = 4
+	if types.SignalKindCount != expectedSignalCount {
+		t.Fatalf("SignalKindCount = %d, want %d (REQ-031 intent guard: the 4-source shape is intentional per AUDIT §193 P1-2; see ARCHITECTURE.md v0.4 divergence section before changing this)", types.SignalKindCount, expectedSignalCount)
+	}
+	want := []types.SignalKind{types.SignalStash, types.SignalStanding, types.SignalVouch, types.SignalCapital}
+	all := types.AllSignalKinds()
+	if len(all) != len(want) {
+		t.Fatalf("AllSignalKinds() len = %d, want %d", len(all), len(want))
+	}
+	for i, s := range all {
+		if s != want[i] {
+			t.Errorf("AllSignalKinds()[%d] = %q, want %q (REQ-031 intent guard: the 4-source shape {Stash, Standing, Vouch, Capital} is intentional per AUDIT §193 P1-2; Freeholder and Guild are NOT signal kinds)", i, s, want[i])
+		}
+	}
+}
+
 // TestAllSignalKindsNames asserts the 4 signal names (Stash, Standing,
 // Vouch, Capital) cross-ref v0.1 x/standing FreeholderSignals (StashMaturity,
 // MultiDomainStanding, CommunityEndorsement, CommittedCapital).

x/hub/types/cross_const_test.go

@@ -0,0 +1,70 @@
+package types
+
+// cross_const_test.go (REQ-030, REVIEW.md P2 / A-304, GRILL G-015) is a
+// cross-package const-equality test that catches silent drift between the
+// x/hub LOCAL consts (LendingCouponCapBps / LendingCouponFloorBps) and the
+// x/bond mission-locked consts (CouponCapBps / CouponFloorBps, D-028).
+//
+// Before REQ-030, the two const pairs were cross-documented only by a comment
+// (x/hub/types/types.go:46-55) — no automated check existed. A future
+// mission-locked change to x/bond.CouponCapBps without a matching x/hub change
+// would silently drift. This test fails closed on either kind of drift:
+//
+//   - single-sided drift: hub stays 800, bond changes to 900 → the equality
+//     test fails.
+//   - paired drift: BOTH change to the same wrong value (e.g., both 900) → the
+//     equality test passes BUT the absolute-value test (G-015) fails, because
+//     the mission-locked value is 800, not 900.
+//
+// G-003 (no production cross-module struct imports): this is a TEST-ONLY
+// import of x/bond/types in a _test.go file. G-003's test-import exemption
+// (documented in v0.2 GRILL G-003 and already exercised by
+// x/bearers/types/types_test.go:7 importing x/processing/types) permits
+// cross-package test imports. NO production .go file in x/hub/types/ imports
+// x/bond/types (the P1-99-01 verification greps non-test .go files to confirm).
+
+import (
+	"testing"
+
+	bondtypes "github.com/oy/openyield/x/bond/types"
+)
+
+// TestLendingCouponCapMatchesBondCap asserts the x/hub LOCAL
+// LendingCouponCapBps equals the x/bond mission-locked CouponCapBps (A-304).
+// Fails on single-sided drift (one changes, the other does not).
+func TestLendingCouponCapMatchesBondCap(t *testing.T) {
+	if LendingCouponCapBps != bondtypes.CouponCapBps {
+		t.Errorf("A-304 drift: x/hub LendingCouponCapBps = %d, x/bond CouponCapBps = %d (must match)", LendingCouponCapBps, bondtypes.CouponCapBps)
+	}
+}
+
+// TestLendingCouponFloorMatchesBondFloor asserts the x/hub LOCAL
+// LendingCouponFloorBps equals the x/bond mission-locked CouponFloorBps
+// (A-304). Fails on single-sided drift.
+func TestLendingCouponFloorMatchesBondFloor(t *testing.T) {
+	if LendingCouponFloorBps != bondtypes.CouponFloorBps {
+		t.Errorf("A-304 drift: x/hub LendingCouponFloorBps = %d, x/bond CouponFloorBps = %d (must match)", LendingCouponFloorBps, bondtypes.CouponFloorBps)
+	}
+}
+
+// TestConstsAreMissionLocked800And0 (GRILL G-015) asserts the ABSOLUTE
+// mission-locked values: both caps are 800 (8pct, D-028) and both floors are 0
+// (0pct, D-028). This catches PAIRED drift — if both consts change to the same
+// wrong value (e.g., both 900), the equality tests above pass but this test
+// fails, because the mission-locked value is 800, not 900. The 8pct cap /
+// 0pct floor is the anti-greed covenant (vision §17, §18); defending the
+// absolute value is the highest-priority regression guard in v0.4.
+func TestConstsAreMissionLocked800And0(t *testing.T) {
+	if LendingCouponCapBps != 800 {
+		t.Errorf("G-015: x/hub LendingCouponCapBps = %d, want 800 (mission-locked 8pct, D-028)", LendingCouponCapBps)
+	}
+	if bondtypes.CouponCapBps != 800 {
+		t.Errorf("G-015: x/bond CouponCapBps = %d, want 800 (mission-locked 8pct, D-028)", bondtypes.CouponCapBps)
+	}
+	if LendingCouponFloorBps != 0 {
+		t.Errorf("G-015: x/hub LendingCouponFloorBps = %d, want 0 (mission-locked 0pct, D-028)", LendingCouponFloorBps)
+	}
+	if bondtypes.CouponFloorBps != 0 {
+		t.Errorf("G-015: x/bond CouponFloorBps = %d, want 0 (mission-locked 0pct, D-028)", bondtypes.CouponFloorBps)
+	}
+}