Canonical Metrics and Wording (style guide)

The definitive reference for how SuiteCentral 2.0’s numerical claims should be phrased in executive-facing material. Any future wiki page that quotes a test count, coverage percentage, connector count, or AI provider list should check this page first.

What this page is

A style-guide page. 26-canonical-metrics-and-wording is explicitly authored as a wording style guide — its governance section says “Any document added to NotebookLM should conform to this file before packaging.” This wiki page captures the same rules so future ingests can reference them without re-reading the raw source.

Why it matters

Consistency. The corpus has at least five reconciled test-count vintages (slide vintage, Talking-Points vintage, June rebaseline, July 17 vintage, current vintage; a separate Start Here ingest is internally inconsistent — all documented on production-proof). The canonical metrics file tells us which vintage to use in current communications and how to phrase it. Future Brain1 pages that cite numbers should cite the canonical form to avoid introducing another ungoverned vintage.

Canonical test wording (3-sentence sequence) — SEPTEMBER 2026 REBASELINE

Per the September 14, 2026 full rebaseline at Preston-Test source 8f12323e8d25bc80d2ee0f4f57add18bbb9720f4, the approved sequence is now:

1. “All 825 executed suites passed; one suite was skipped.”
2. “100% of executed tests passed (16,571 tests across unit, integration, and portal E2E; 26 skipped).”
3. “Full breakdown: 15,616 unit passed (10 skipped), 935 integration passed (16 skipped), 20 portal E2E passed.”

Each sentence’s historical progression — kept in its own scope below, because these are point-in-time snapshots that must NOT be re-baselined when the current numbers move:

• Suites: up from 630 on July 17, 607 on July 6, 604 on June 12, 416 on April 30, 412 on April 18, 406 on April 8.
• Executed tests: up from 12,734 on July 17, 12,320 on July 6, 12,214 on June 12, 9,514 on April 30, 9,410 on April 18, 9,364 on April 8.
• Breakdown: up from 12,154 unit + 560 integration + 20 E2E on July 17 (SuiteCentral control-plane security redesign, PRs #1006–#1013), 11,770 unit + 530 integration + 20 E2E on July 6, and 9,347 unit + 147 integration + 20 E2E on April 30. Skip count dropped from 30 → 6 via PR #694/#695 skip-discipline cleanup, then rose to 16 documented integration skips when the June 2026 re-measure picked up the full 60-suite integration profile; unit skips went 0 → 10 with the July 2026 security-redesign suites.

Do NOT shorten to “100% pass rate” without the executed/skipped context.

Space-limited version: “16,571 executed tests passed; 26 skipped.”

Note: the prior July 17, July 6, June 12, April 30, and earlier March 2026 canonical baselines are superseded (full progression archived in docs/CHANGELOG.md of the upstream Preston-Test repo). Some earlier wiki pages still cite those numbers in their vintage comparison tables — those are historically accurate snapshots. For new content, use the September 2026 numbers above.

Canonical coverage wording

• Statements: 72.47%
• Branches: 62.16%
• Functions: 74.74%
• Lines: 72.97%

Short form: “72.97% line coverage in the measured unit coverage profile.”

The current short form deliberately avoids a production-LOC denominator because this rebaseline measured Jest coverage and did not remeasure the production-TypeScript line total.

Canonical module wording

“16-module demo library (12 core + 4 extension/platform).”

If listing only core modules, explicitly say “12 core modules”. See module-library for the full list.

Canonical connector wording

“18 connector implementations: 5 production-ready, 1 beta, 11 demo-only, and 1 stub.”

The Preston-Test docs/connectors/CONNECTOR_STATUS.md inventory, connector registry, and status audit are the authority for this partition; see production-vs-demo for the full inventory. Production-ready means implementation readiness. The current connector proof cards record no live-credential verification; older technical evidence separately describes a NetSuite sandbox run, so do not treat the proof-card statement as a claim that no live system has ever been exercised.

Avoid: “4 connectors verified” (outdated).

The 6 connectors span ERP (NetSuite, Business Central, Oracle), CRM (Salesforce, HubSpot), and logistics/shipping (ShipStation). Oracle appears here as a connector even though it also appears as a competitor in the Oracle comparison page — these are different Oracle products.

Canonical AI provider wording

“4 production-ready AI providers (OpenAI, Claude, OpenRouter, LMStudio).”

With cost context:

• OpenAI GPT-5.4 mini — ~$0.0007/mapping, benchmark-measured (primary inference default)
• Claude Haiku 4.5 — ~$0.0011/mapping, benchmark-measured (Sonnet 4.6 upgrade tier)
• OpenRouter (multi-model, free tier available) — routing/fallback
• LMStudio (local, free) — on-premise/fallback

Both per-mapping figures are measured by the Phase-B accuracy benchmark matrix (2026-06-10 live run — Claude Haiku 4.5 measured 96.7% top-1 on the NetSuite pair, edging out GPT-5.4 mini’s 95.1%).

Avoid: “3 providers” (outdated); omitting OpenRouter from the list; the legacy “$0.02/mapping” / “85% cheaper” framing and the older “GPT-4o / Claude 3.5 Sonnet” model names (all stale).

External comparison wording (the “avoid” list)

Preferred phrasings (use these):

• “Comparable in complexity dimensions”
• “Positioned against”
• “Based on publicly available sources (date-stamped)”

Avoid unless source-locked and dated:

• “No competitor has this”
• “Uncontested 6-12 month window” (note: ai-governance-layer-video at 04:05 uses exactly this phrasing — the hook video is out of compliance with the current style guide; flagged)
• Any absolute timing claim without a refresh date

Competitive pricing wording (March 2026 refresh)

Celigo: “Celigo ~$10K–$50K+/yr, roughly $1K–$4K/mo equivalent, scope-based/per-endpoint pricing.”

Avoid: “$3K-$10K/integration”, “$4-8K/mo” (older phrasings, no longer fresh).

Celigo dual-ERP framing: describe as “Generic / not dual-ERP-specific.” Avoid positive-match framing that implies Celigo is purpose-built for dual-ERP; also avoid a flat “No” when the measurement is breadth rather than total connector absence.

Reviewer-facing proof routing

• Prefer reviewer-friendly proof pages over raw API/JSON payloads in executive-facing materials.
• Raw JSON endpoints should be labeled as “technical proof only” — not the primary reviewer path.

Date and decision wording

Avoid hardcoded past dates in reusable documents.

Preferred:

• “Decision requested within 10 business days”
• “Quarterly refresh required for external claims”

How to use this page

1. When adding a numerical claim to a new wiki page, look it up here first.
2. When updating an existing wiki page with a new source’s numbers, check whether the new phrasing conflicts with this page.
3. When a future ingest surfaces a claim using avoid-list phrasing (e.g., “Uncontested 6-12 month window”), flag it as a style-guide violation in the source-summary and use the preferred phrasing on the wiki page.
4. When this page itself needs updating (e.g., test counts change, new connectors ship), the update should come from a new ingest of 26-CANONICAL-METRICS-AND-WORDING.md — the source file is versioned and refreshed periodically.

Sources

• 26-canonical-metrics-and-wording — primary and only source; this is a style guide, not a claim document.