The Church Chaos Index
For agents: The diagnostic methodology that scores a church’s communications health — entry point to the entire 2026 funnel. Canonical for anything about scoring, audit reports, or the funnel. See also 2026-chaos-era.
What it measures
The Church Chaos Index (CCI) is a 0-100 score that quantifies why people still say “I didn’t know that was happening” despite weeks of announcements, emails, and social posts. It measures the friction in a church’s communications.
Lower score = less chaos = healthier ministry.
Inputs
A church submits two things at churchchaos.com:
- Their website URL. We crawl homepage content and capture a full-page screenshot via Firecrawl for vision-based detection.
- Three full church services — YouTube URLs or audio uploads (MP3, M4A, WAV, AAC, OGG, FLAC). Full services, not just sermons. Includes announcements, welcome, transitions, calls to action.
Submission takes ~3 minutes. Audit delivery is within 24 hours (longer on weekends).
The four chaos dimensions
Each dimension is scored 0-25; the four sum to a composite 0-100.
1. The Announcement Avalanche
What it measures: how much promotional noise the congregation absorbs each Sunday. When everything is emphasized, nothing lands.
2. The “Where Do I Go?” Test
What it measures: can someone who decides to take action figure out where to go — or do they get sent on a wild goose chase across multiple destinations, people, and platforms.
3. The Sunday Savior Trap
What it measures: does the church have a clear promotional hierarchy, or is everything competing equally, forcing someone to play peacekeeper between ministry leaders all wanting stage time.
4. The First Impression Score
(Renamed from “Mixed Signals” 2026-05.) What it measures: can a newcomer evaluating the church online figure out when, where, and how to take their first step. Anchored on homepage entry-point legibility, not website-vs-stage mismatch.
The four severity tiers
| Tier | Range | Color | Description |
|---|---|---|---|
| Low Chaos | 0–25 | Blue | Strong systems. Announcements focused, destinations clear, newcomer pathway legible. |
| Moderate Chaos | 26–50 | Gold/amber | Some systems working, gaps remain. Targeted improvements would make a noticeable difference. |
| High Chaos | 51–75 | Orange | Significant friction. Congregation tuning out; newcomers face confusion. Fixable but requires intentional change. |
| Critical Chaos | 76–100 | Red | Communications infrastructure actively undermining the church. Urgent attention needed. Even small changes show immediate improvement. |
Tier colors are used dramatically across the YouTube show (flashed during pillar reveals) and operationally across the report and follow-up sequences.
The rubric — indicators per dimension
15 indicators across the 4 dimensions. Each indicator is bucketed 0 (Low) / 1 (Medium) / 2 (High); some binary indicators only use 0 or 2. Per dimension: (sum_of_buckets / max_possible) × 25.
D1 — The Announcement Avalanche (3 indicators)
| Indicator | Low (0) | Medium (1) | High (2) |
|---|---|---|---|
| Announcements per service | < 3.5 avg | < 6.5 avg | ≥ 6.5 avg |
| Announcement duration | < 3 min | ≤ 6 min | > 6 min |
| Unique promotions per week | bucketed by count of distinct topics |
Homepage CTAs indicator was dropped (false positives on seasonal banners).
D2 — The “Where Do I Go?” Test (4 indicators)
| Indicator | What it captures |
|---|---|
| Central Hub present | Does ONE destination clearly lead? (hierarchy) |
| Destinations mentioned | Total unique count across services. (volume) |
| Person-based CTAs | Count of “see Pastor X” style handoffs |
| Consistent destination | Same destination cited across all services? |
D3 — The Sunday Savior Trap (4 indicators)
| Indicator | What it captures |
|---|---|
| Clear #1 priority visible | Does ONE message dominate top-level? |
| Stage time variance | Within the service, does duration track priority? |
| Ministry prominence variance | Do top programs get more airtime? |
| Appeasement language | Apologetic / guilt phrasing count |
D4 — The First Impression Score (4 indicators)
| Indicator | What it captures |
|---|---|
| Service times on homepage | Binary (0 or 2). Vision-detected from screenshot. |
| Address on homepage | Binary (0 or 2). Vision-detected from screenshot. |
| Mission and values reinforcement | Does stated identity show up from stage? |
| New-visitor CTA on homepage | Clear “I’m new” CTA above the fold? |
Determinism map — what the model judges vs. what’s computed
The model’s job is bucketing, not prose. Most indicators are deterministic; the model is overridden if its bucket disagrees with the computed value.
Deterministic (from extraction or vision):
- D1: announcements per service, announcement duration, unique promotions per week
- D2: destinations mentioned, central hub present, consistent destination, person-based CTAs
- D3: clear #1 priority (when announcements ≥ 7), appeasement language
- D4: service times on homepage, address on homepage (via vision)
Model-judgment (genuinely interpretive):
- D3: stage time variance, ministry prominence variance
- D4: mission and values reinforcement, new-visitor CTA on homepage
Evidence prose is template-driven, not model-generated. Server-side templates indexed by indicator name and bucket interpolate canonical metrics from the extraction. The model stops at bucketing; templates handle voice. This eliminated months of prompt whack-a-mole on tone and number drift.
Qualitative flags
Pastoral Guilt Language (boolean + 2-3 example quotes): “I know we’ve mentioned this before, but…” / “Bear with me, I’ve got a lot to cover…” / “I promise this is the last time I’ll bring this up…”
Information Dump Patterns (boolean + examples): requires 3+ stacked logistics data points in one announcement (date AND time AND location AND cost AND signup). Single-fact mentions (“50 days until camp”) and financial transparency (“we’ve raised $9K of our $30K goal”) are explicitly excluded.
Sanity gates — auto-hold before delivery
Automatic flags block delivery and surface the audit for admin review:
| Flag | Triggers when |
|---|---|
noAnnouncements | Total announcements across all services = 0. Almost always sermon-only upload. |
shortServices | Avg service duration < 50 min. Sermon-only signal. |
extremeAvgTime | Avg announcement time outside [30s, 25min]. |
extremeAvgAnnouncements | Avg outside [1, 30] per service. |
extremeDestinations | Outside [1, 30] across services. |
missingLists / missingDimensions | Extraction or scoring returned malformed data. |
When a sermon-only flag fires, the system auto-creates a HelpScout draft asking the church to re-upload full services, pings Slack with a deeplink to the admin task, and holds delivery indefinitely.
Rejection workflow
A rejected audit can take one of two paths:
- Correctable (sermon-only, transcription failed, dead link): admin emails the church via the auto-created HelpScout draft, marks rejected. Church can resubmit; new church record is created.
- Terminal (non-English, not-a-church-service, gaming): admin clicks “Block from resubmitting” on the rejected banner. Sets
audit.blockResubmit = true. Intake’s duplicate check refuses any future submission from that email or domain. Reversible at any time.
The duplicate check on intake blocks: in-progress audits, audits delivered within the last 6 months, and any audit with audit.blockResubmit === true.
What the report contains
- Composite Chaos Score (0-100) and severity tier
- The four dimension scores (each /25) with description, what success looks like, and “What We Found” — template-rendered evidence interpolating canonical metrics
- Service Analysis — quantitative metrics: announcements per service, total announcement time, destinations mentioned, next step consistency
- Qualitative Flags — pastoral guilt language, information dump patterns
- Voice & Tone Analysis — formality and warmth (1-5 scale)
- The “So What Now?” prescription — pointing to the 7 Building Blocks. Tier-specific path-forward copy lives at chaos-index-path-forward-copy.
- Audit summary — capped at 350 chars / 2 sentences
Report delivered at churchchaos.com/audit/[church-slug], password-protected. PDF download available.
A real example (from beta — pre-2026-05 methodology, format reference only) lives at audit-example-beta.
Source of truth — runtime location
The runtime implementation of extraction, bucketing, vision detection, sanity gates, and evidence templates lives in the the-comms-dept repo. This doc is canonical for the methodology — names, dimensions, indicators, cutoffs, determinism boundaries. When runtime cutoffs change, update this doc to match.
Why it’s the funnel’s keystone
The audit collects proprietary data (full service audio + website, willingly given), produces source tape (every finding has a timestamp + audio of the moment), is the entry into the makeover funnel, and supplies an endless stream of churches for the YouTube show and reactive podcast.
Related
- 2026-chaos-era — why this exists
- 7-building-blocks — the prescribed solution every audit points toward
- chaos-index-path-forward-copy — tier-specific copy injected into every audit
- audit-evidence-pipeline — the technical pipeline (transcription → extraction → moments → briefs)
- makeover-funnel — what happens after a church receives their score