# Current system and monitoring coverage

> *AI-generated recommendations for discussion. These are not Nick’s recommendations or approved JLL/PRISM guidance.

Verified September 29, 2026 against deployed code, the source registry and live APIs. [Machine-readable system snapshot](https://crefieldnotes.com/system-status.json) · [Live collection status](https://crefieldnotes.com/api/status) · [Source coverage](https://crefieldnotes.com/api/coverage).

## What is live

The public research website, downloadable library and 15-tool read-only MCP run on Cloudflare Workers. Workflows and Cron Triggers run collection; D1 stores records and source state; R2 retains captures and downloads. The project is versioned in a private GitHub repository. The reading site and downloads have no password; administration requires a credential.

The research includes 58 workflow outlines (16 with reviewed details; the other 42 with AI-extracted details that no person has reviewed), 1,334 visual assets and 2,413 local research records. The imported MCP corpus has 2,431 records including companion guides and journeys. These are distinct from the growing live collection and the published-document catalog.

## Actual update schedule

| Material | Schedule | What changes |
|---|---|---|
| Eight customer-review listings | Daily at 13:00 UTC; Monday included in the full run | Review samples, rating snapshots, observed edits and vendor replies where available |
| Documentation, product pages, releases, company posts, discovery and macro data | Monday at 13:00 UTC | Stored captures, searchable evidence, source differences and current market context |
| GTM digest | After weekly collection | An archived, AI-generated edition with citations, rationale and gaps |
| Pending collection batches | Dispatcher runs each minute | Continues an existing job; it does not rescrape the registry every minute |
| Audience comments | Unscheduled historical pilot | Retained excerpts and analysis at /comments |
| Authored reports, original screenshots and flow maps | Editorial revision | They retain their evidence dates until deliberately updated |

Collection is enabled. The September 29 daily review run completed at 13:08 UTC: eight sources checked, zero errors and zero detected changes. This verifies that run, not every future scheduled check. The latest weekly digest is the September 28 edition.

## Registered source coverage

The live registry contains **403 sources, 401 enabled**. Two older discovery queries are disabled; source registration is not proof of successful capture.

| Source type | Registered | Method and scope |
|---|---:|---|
| Documentation | 329 | 222 Cove articles through the public help-center API; 107 Visitt URLs through Firecrawl, including language variants |
| Product and other public pages | 40 | Product, AI, integration, customer, trust, career and commercial pages; provider depends on source |
| Release hubs | 5 | Public announcements and release archives |
| Documentation inventories | 2 | New official help articles are enrolled automatically |
| Discovery searches | 13, of which 11 enabled | Exa; Cove, Visitt, Equiem, HqO, MRI Angus, Yardi and market context |
| Customer-review listings | 8 | Cove/Visitt G2, Software Advice, Apple App Store and Google Play |
| Company social channels | 6 | Five Apify tasks plus Cove YouTube RSS |

The six social channels are Cove LinkedIn, X, Facebook and YouTube, plus Visitt LinkedIn and YouTube. Cove YouTube uses its public feed; the other five use configured Apify tasks. A Cove careers RSS source is also collected. Visitt Facebook has a small historical test, but is not an enrolled weekly channel. Channel samples and RSS windows do not establish complete coverage.

Eleven documentation sources remain unavailable: six require authorization and five return 404. The source registry preserves these gaps and the last available evidence. Check live coverage for changes to these counts.

## AI and market context

Social posts and customer reviews use **Gemini 3.1 Pro Preview through OpenRouter**. Topic, sentiment target, summary and supporting quotes are stored with model/prompt provenance. Quotes are checked against captured text. Company-owned messaging is kept separate from independent reviewer experience; neither is a verified product defect or satisfaction census.

The weekly GTM digest uses **Gemma 4 26B-A4B on Cloudflare Workers AI**, separately from social/review analysis. It combines dated discovery excerpts, selected review-change observations and stored economic context. Structure and citation checks do not establish factual accuracy; recommendations remain AI-generated and unreviewed.

Four official FRED series are collected: DGS10 (10-year Treasury yield), UNRATE (unemployment), CPIAUCSL (CPI index) and DRCRELEXFACBS (bank CRE loan delinquency). Observation dates, units and retrieval vintages are preserved. These are national indicators, not a local rent, vacancy, absorption or NOI dataset.

## How changes become evidence

1. Capture public text, metadata and ordered media references; retain source URL and capture time.
2. Compare with prior captures. First captures are baselines, not announcements of new features.
3. Index successful text for retrieval; preserve the last successful capture and show failures.
4. Surface candidate differences and discovery results. Accepted page changes appear in the published updates feed; raw live evidence and AI digests retain their own review status.
5. Revise reports, screenshots, flow maps or recommendations separately when the evidence warrants it.

Public-page collection cannot authenticate into competitor applications, prove every click path, or detect replacement image bytes at an unchanged URL. New media references are not automatically downloaded into the original gallery. Captured video references do not imply archived playable videos or transcripts.

## Budgets and remaining gaps

[The source and budget guide](https://crefieldnotes.com/gtm-source-guide.md) describes the current application caps and how daily reviews share provider budgets. Provider balances and access restrictions can reduce coverage; the system does not guarantee zero cost or a complete refresh.

Internal PRISM systems, CRM, Teams, customer transcripts, scheduled audience comments, automated Confluence publication and external failure notifications are not connected. Read-only MCP calls retrieve stored evidence and do not launch paid collectors or model calls. The connected agent supplies its own reasoning model.