Proof Card: Strategic Positioning (Wedge Statement + Claim Cleanup Gate)

Status: production Last verified: 2026-05-18 · git sha feat/pr-16-strategic-claim-cleanup

Claim

The wedge statement — "SuiteCentral 2.0 is an embedded ERP operations layer for NetSuite and Business Central — governance-grade middle-market integration delivered as workflows inside the ERP, not a separate iPaaS to operate alongside it" — is the canonical positioning across product-positioning surfaces (CLAUDE.md, docs/strategic/competitive-positioning.md). Every numerical claim in these surfaces carries an inline classification tag () or matches an allowlist pattern in scripts/strategic-claims.allowlist.json. CI fails (via scripts/check-strategic-claims.mjs) on any unclassified number, preventing the "unqualified 95-99%" drift class that motivated PR 8-OptB.

Source

Wedge statement (canonical): docs/strategic/competitive-positioning.md:7 (lede block)
Six-principle framing tracking the wedge: docs/01_VISION_DOCUMENT.md "Six Design Principles" (moved here from CLAUDE.md's Strategic Vision section, Phase 2 slimming, 2026-08-07)
Gate script: scripts/check-strategic-claims.mjs
Allowlist: scripts/strategic-claims.allowlist.json
Scan targets (PR 16 initial scope): CLAUDE.md, docs/strategic/competitive-positioning.md
Tier-B decision #11 (hybrid framing — category-split lede + earned TCO subsection): docs/plans/2026-05-01-a-grade-remediation-plan-merged.md PR 16 section

Tests

Unit: tests/scripts/check-strategic-claims.test.sh (9 scenarios A-I — synthetic fixtures covering tagged-pass, untagged-fail, allowlist-year/version, code-block-skip, URL-skip, HTML-comment-skip)
Integration: none — gate runs on real product-positioning files in CI; the regression test exercises the gate's logic against synthetic fixtures via --root.
Coverage: not yet measured — gate is a node script outside the Jest scope

Live vs Fixture

Real CI gate wired? Yes · evidence: .github/workflows/ci-minimal.yml (the check-strategic-claims step runs node scripts/check-strategic-claims.mjs on every PR)
Demo-mode toggle? No · gate runs identically in CI and locally; no environment branching
Production credential test on file? N/A · gate is a static-analysis script; no credentials involved

Known Gaps

Scan target scope is initial, not exhaustive. PR 16 ships with CLAUDE.md + docs/strategic/competitive-positioning.md as the scan targets. README.md, docs/01_VISION_DOCUMENT.md, and the Squire executive package (public/Squire-Executive-Package-v2/**) are scheduled for a follow-up sweep PR — the gate is live and tested here, the follow-up just appends those files to DEFAULT_SCAN_TARGETS and tags the existing numerical claims.
Regex is claim-shape targeted, not all-numbers. The gate flags percentages (95%), currency ($50), and comma-grouped counts (11,847) — the high-value drift surfaces. Bare integers (ordinals, section refs, port numbers, PR/issue refs) are intentionally not flagged because tagging every digit would create review noise without catching new claim drift. If a number warrants classification (e.g., "5 production connectors") the author can voluntarily add  — the gate doesn't punish over-classification.
TCO numbers in competitive-positioning.md are labeled-projection. No pilot has yet produced measured TCO data. The category-split lede earns the cheaper-than conclusion; specific cost ratios will move to pilot-result only after PR 15 ships a real pilot.
benchmark classification has no current artifact. The post-PR-8-OptA accuracy harness is Tier-C deferred, so no claim in scanned files currently uses . The tag is reserved for when that harness ships.
Reviewer-mirror exposes the gate scripts but not the production-enforcement targets. PR #828 added the gate's scripts + regression test + allowlist to scripts/reviewer-mirror.allowlist.json. The scan targets themselves — CLAUDE.md and docs/strategic/competitive-positioning.md — remain upstream-only by design (CLAUDE.md contains environment-specific paths that the mirror's forbidden-content scanner rejects; docs/strategic/** is broad-excluded since most files there are commercially sensitive). External reviewers running the gate against the mirror snapshot get OK — 0 file(s) scanned (the gate soft-skips missing targets). The verification recipe below has separate "Mirror" and "Upstream" stanzas so reviewers in either context can reproduce the relevant checks.

Verification (60-second AI-reviewer recipe)

Mirror snapshot (works in `KStratMD/Preston-Test-reviewer-snapshot` + upstream)

# Confirm the gate scripts are present:
ls scripts/check-strategic-claims.mjs scripts/strategic-claims.allowlist.json
# Expected: both files listed.

# Run the regression test (9 synthetic-fixture scenarios — exercises the
# gate's logic end-to-end against fixtures, mirror-resident):
bash tests/scripts/check-strategic-claims.test.sh
# Expected: "All 9 scenarios passed."

# Confirm the allowlist is bounded (not an escape hatch):
grep -c '"reason"' scripts/strategic-claims.allowlist.json
# Expected: 4 (year stamps, ISO date stamps, version numbers, HTTP status placeholders).

Upstream (private repo only — production-enforcement targets are upstream-only)

# Run the gate against the actual production-positioning targets:
node scripts/check-strategic-claims.mjs
# Expected: "[strategic-claims] OK — 2 file(s) scanned, all numerical claims classified."

# Confirm the wedge statement lede exists where claimed:
grep -n "embedded ERP operations layer" docs/strategic/competitive-positioning.md
# Expected: a line in the lede block around line 7.

If any check in either stanza fails for the relevant context, the wedge claim cleanup has regressed (upstream) or the gate's exposure has degraded (mirror).