Keeping the research current
*AI-generated recommendations for discussion. These are not Nick’s recommendations or approved JLL/PRISM guidance.
Current operating guide, verified September 29, 2026. This replaces the earlier monitoring proposal. Full coverage and current system · Collection status.
What runs today
Cloudflare hosts the website, database, evidence storage, MCP and collection jobs. Scheduled work runs without a laptop. The system uses the collection providers directly; it does not run a separate Firecrawl Monitor, Exa Monitor or Competitor Watch service.
| Component | Current job |
|---|---|
| Cloudflare Workers and Workflows | Serve the library and execute scheduled, batched collection |
| D1 and R2 | Store searchable records, captures, history, coverage state and downloads |
| Public help-center APIs and Firecrawl | Capture known documentation and public pages; inventories discover new official help articles |
| Exa | Discover relevant public product, company and CRE market evidence |
| Apify and public RSS | Retrieve configured company posts, selected review data and feeds |
| OpenRouter / Gemini 3.1 Pro Preview | Analyze topics and sentiment in captured posts and reviews with source quotes |
| Workers AI / Gemma 4 26B-A4B | Draft the weekly GTM digest from collected evidence |
| Read-only MCP | Retrieve that evidence for an agent using its own model and authorized internal context |
When it updates
Customer-review checks run daily at 13:00 UTC. On Mondays they are included in the broader collection, which also checks documentation, product/release pages, company posts, discovery and macro context. The digest follows that run. A minute-by-minute dispatcher resumes queued batches; it is not a minute-by-minute source monitor.
Audience comments are a historical, unscheduled pilot. Reports, screenshots and flow maps remain dated artifacts until revised. The schedule is configured in Cloudflare, not in the historical watchlist's per-entry metadata. Current source coverage reports successes and failures.
What colleagues can read
- Product evidence: current successful source captures, full normalized text and media references.
- Customer reviews: review samples, rating snapshots, themes, dates and coverage limits.
- Market signals: company posts and discovery evidence; company tone is not customer satisfaction.
- Weekly GTM digest: competitor activity, market context and suggested PRISM responses, with reasons and citations.
- Published updates: accepted source changes. New captures and digests are not automatically reviewed conclusions.
- AI setup: 15 tools for reports, workflows, visual evidence, feedback, changes and economic histories.
What the automation establishes
The first capture is a baseline. A newly found article may be old. A removed URL does not prove a feature was retired. Source publication dates, first-observed dates, collection times and review decisions remain distinct.
Official help-center inventories add new public articles. Wider discovery produces candidates; it does not automatically enroll every returned URL. Failures retain the previous successful capture rather than becoming an empty result. A proposed change may link to a relevant workflow, but the diagram and strategic recommendation still need editorial revision.
Visual evidence and comments
Existing screenshots, GIFs and sampled training frames remain available with provenance. New document captures retain ordered image/video references. The collector does not compare image bytes at an unchanged URL, rebuild task flows, obtain private product access or automatically create transcripts. Audience-comment examples at the pilot page are not a recurring cross-platform feed.
Cost and operation
The registry currently has 403 sources, 401 enabled. Collection uses bounded batches, provider operation caps and per-run request limits. These reduce uncontrolled usage but do not establish that every source can refresh within every month's allowance. See the current budget guide for the daily-review calculation, inference budget and provider gaps.
The September 29 daily review run completed successfully for all eight listings. Eleven documentation URLs remain inaccessible at the source. The site and MCP expose collection failures; an external email or incident-alert channel is not configured.
Transfer and Confluence
The deployed service is a Worker with Static Assets, D1, R2 and Workflows. Moving to a company account means provisioning those resources, restoring the database and objects, setting provider secrets, redeploying and verifying the result. The private GitHub repository contains the code and research files; it is not a live D1/R2 backup. A backup/export script is supplied, but a restoration exercise has not been certified.
The Confluence and Rovo guide remains an import plan for the dated research package. A Confluence space, Jira workflow or Rovo connection has not been deployed by this project. Stable source, claim, flow and asset IDs support cross-linking; no separate graph database or automatic Confluence sync is running.