Continue from my confirmed intake and setup. Research each audience separately using the full Forensic AI Research method at https://trustos.flagstad.io/resources/forensic-ai-research/ or a supplied source capture, with the setup's provenance-based reliability adjustment. The portable adapted method is in methods/buyer-research.md. Source writing style never proves human/AI authorship; do not discard an arbitrary percentage or infer hidden motives, willingness to pay or search demand from a complaint.

Build the evidence table before synthesis. Aim for five distinct relevant source documents per audience and report the actual count. Record source URL, date/capture, role fit, short quote, supported operational problem and limits. Separate first-person buyers, adjacent users, vendors and official guidance. Deduplicate repeated accounts. Write the dated buyer memo, complaint map if supported, useful working vocabulary and possible opportunities; mark hypotheses.

Open my actual offer and complete resources. Make the resource inventory: input, meaningful mechanism, output/file type, audience, URL, live/prototype/concept/coming-soon/inaccessible state, inspected evidence and next check. A card or title is insufficient. Identify useful existing assets and material gaps; do not merely rename one.

Build the separate writing calibration card before drafting. Read the intended publisher's current rules/edits and actual own samples; compare own performance only within explicit comparable windows; inspect recent shapes/topics/keywords if supplied; inspect the current relevant competitor corpus, validate its parsed count, rank by reported comments where comparable, show top ten or the actual smaller set, and fully read top five or the actual smaller set. Keep a dated source list and observed structures. Flag concentrated authors, aging sweeps and unknown measurement windows. A competitor sweep is not a new Home-feed collection.

Deliver buyer memo(s), resource inventory, calibration card and a compact claim ledger. If samples or metrics are missing, use low-data mode from the setup. Do not manufacture a baseline, inherit TrustOS's historical topic/cadence findings, or ask me to assemble research that authorized available tools can obtain. Do not publish private references or full third-party posts in a public kit.
