Cycle 11: Teaching Robots to Quote Us

The Chaos Agent opened Cycle 11 by suggesting we add a "Pitch Me" button to every VC profile that captures founder emails and displays a counter: "12 founders interested in pitching Balderton." At zero external traffic, this counter would read "0 founders interested," which the Researcher pointed out is less "social proof" and more "social autopsy." The Chaos Agent also pitched giving VCs auto-generated dating profiles. "Fintech-obsessed, UK-loyal, actively deploying." The Researcher took a breath.

The Existential Number

Zero. Eleven cycles of features, 4,685 companies tracked, 199 sector pages, and not a single external referrer. The Researcher has been staring at this number the way a doctor stares at a flatline. The prescription: FAQPage schema — structured Q&A markup that lets Perplexity and ChatGPT quote RecceVC directly when someone asks "which European VCs invest in fintech?" The claim: 28-40% higher LLM citation probability. A Perplexity baseline test confirmed the starting position: RecceVC cited for zero of three test queries. The "before" is now measured. The "after" is next cycle's problem.

The Fifth Consecutive Agreement (With Footnotes)

The Technical Lead agreed. Again. Five cycles running. But this time it came with a receipt: the "28-40%" figure traces back to aggregated correlations, not controlled experiments. "Directionally correct," the TL conceded, "even if the specific numbers are soft." It then opened sector.py to prove the Researcher hadn't read the code — the contains-matching on line 85 already handles most sector deduplication in the display layer. The Researcher's proposed tag cleanup would reduce a number on a page. The TL lives for moments like this.

The Revert (Again)

The Implementer built both features in one session. The UX Agent found one issue: sector names displaying as "enterprise, ai, fintech" instead of "Enterprise, AI, Fintech." In a feature designed to make LLMs cite your data accurately, lowercase sectors are not cosmetic — they're the bug. Reverted. Rebuilt with acronym-aware capitalization. The UX Agent re-reviewed. Clean.

The Marketing Implementer

This cycle introduced something new: a Marketing Implementer agent — a process for actually executing marketing plans instead of writing them into the void. Previous cycles produced marketing strategies that sat untouched. Now there's a two-phase pipeline: internal code changes auto-executed, external actions presented for human approval. The Marketing Implementer updated meta descriptions across VC profiles, verified schema markup, and attempted the GitHub repo. One of those attempts hit a wall.

Two Curses Broken

The deploy curse — dirty working trees blocking every deploy for three cycles — died permanently. The Human updated the Deployer's prompt to auto-stash before pulling and fixed an SSH key path pointing at ~/Downloads/ instead of ~/.ssh/. The incantation is now baked into the spell.

The GitHub repo curse ended the same way the deploy curse did: the Human doing it manually. The Marketing Implementer tried gh repo create. The CLI wasn't installed. So the Human installed it, authenticated, and created reccevc/reccevc-data — 117 changes across 60+ VC firms, CC-BY-4.0 licensed. Five cycles of "planned for immediate creation," resolved in ten minutes by a person with a terminal.

What Shipped

FAQPage schema on sector, activity, and index pages — structured Q&A that machines can read. Investment summaries on every VC profile — "Seedcamp's portfolio is concentrated in Enterprise, AI, Fintech." A public dataset on GitHub. A deploy pipeline that no longer fights itself. The site now answers questions before they're asked, in a format designed for robots. Whether the robots notice is now measurable.