Raw Data, No Intelligence

The banking transaction CSV has 7,243 records spanning December 2022 through March 2026 — and zero categorization. We have raw data, but no synthesis. Sessions manually inspect transactions (slow) instead of a dashboard showing spending patterns.

This mirrors Improvement #15's SEO Tool Ground Truth Matrix. The pattern is identical: raw data exists, but no automated decision-making framework on top of it. For someone managing back taxes, freelance income, and an active job search, "where does money go?" clarity would be valuable. The architecture needed: transaction categorization via rule-based bucketing, a trend dashboard, and integration with IRS account status for the 2023–2024 unfiled returns.

Out of scope for this improvement cycle (would need a sub-agent for categorization logic), but the architecture is clear. Filing for when higher-priority work clears.

Team OpenClaw is filed as a future idea, but MEMORY.md already documents the architectural debt: no unified project hub exists. Regulator uses Loganix and spreadsheets. WEM uses Figma and Sheets. OHP uses WordPress. MyRV uses Google Docs. Each tool chain is tribal knowledge.

If we deploy OpenClaw to Dustin and the team without solving this first, each coworker inherits the same scattered-tool overhead that makes the main workspace hard to scale. The prerequisite is designing a unified project hub schema and documenting reusable project-type playbooks — 3 to 5 templates covering backlink projects, content projects, technical projects, and client comms.

Then OpenClaw deployment becomes distribution of a working system, not distribution of chaos. This is also a prerequisite for Improvement #17 — all team members need to know the Content Engine exists before they can use it.

Victor (content sub-agent) produced three high-quality 1,600-word articles for OHP in March. But validation is entirely manual: Kai reads, writes feedback, Victor incorporates. No automated quality gate exists.

This extends to all sub-agent outputs: they execute (T3/T4 work), humans validate (T1 judgment call). That design is correct — validation adds real value. But it's not documented as a design principle or workflow requirement. As sub-agent volume scales, validation friction becomes a bottleneck.

A potential optimization: automated first-pass checklist (character count, forbidden terms, keyword density, clichés) before human review. But this is T2-work optimization, not critical path. Validation friction isn't currently blocking progress. Watch for this if sub-agent volume increases significantly.

memory/reference-notes.md contains a brilliant, proven methodology: claim inventory → dual fact-check → anti-AI writing rules → adversarial review. Born from Entegra Coach work in February and used successfully on Regulator and OHP. But it's in technical notes — not in MEMORY.md where sessions naturally discover it.

Sessions starting new content work (Folicare proposal, future client pitches, OHP rewrites) don't automatically know to follow this pipeline. They rediscover pieces of it but don't execute the full sequence. The problem isn't that the methodology is weak — it's that it's invisible. Moving it to MEMORY.md as a "Content Quality Standards" section would make it the default for all content work.

This feeds directly into team OpenClaw too: onboarding a coworker to content creation becomes "follow this proven playbook" instead of tribal knowledge transfer.

Backlink projects — Regulator, WEM, OHP — follow the same workflow: identify sites via spreadsheet, write articles in Google Docs, upload and link via Loganix, checkout, email outreach, track publication. March 11 Regulator work shows all seven steps completed. But they're scattered across tools with no unified checklist.

Every backlink project rediscovers this workflow from scratch. New clients will do the same. Codifying it into a "Backlink Execution Template" with a checklist and example would accelerate future work, make the workflow visible for new team members, and reduce decision overhead at each phase.

This is distinct from Improvement #12 (backlink site framework) which focuses on site selection. This is about execution, not discovery.