Parked SC angle — "Put your agent on an information diet"

The hook phrase

Founder-coined, iMessage 2026-05-14 15:41 ET: "There will be other people that want to 'put their agent on an information diet'."

This is the durable frame for a Sanity Check piece about RDCO's content-ingestion discipline. Better than any phrase we'd surfaced in the parking lot.

Why this could be a Sanity Check piece (original re-frame test)

Everyone is writing about information overload + AI noise + context windows + RAG. Most of those takes are vibes. The ORIGINAL re-frame: most people are trying to scale UP the agent's information bandwidth. The actual play is scaling it DOWN — putting the agent on a curated diet.

Specifically what RDCO does that nobody else writes about cleanly:

Whitelist-only newsletter ingestion — process-newsletter K + F tables. Default skip everything not on the list.
Multi-layer skip filters — sender-pattern skip + subject-line triage + sales-CTA detection + list-serv-header signal. Several filters compound before any expensive read happens.
Sub-agent fan-out for budget bounding — full text never enters parent context; sub-agents read + return summary lines.
Post-condition audit invariants — deterministic structural checks on EVERY filed note. Catches schema drift before it metastasizes.
State files as the freshness boundary — each watch run is bounded by a state file, no infinite history scans.

The diet metaphor is sharp because it's testable: a diet succeeds when the body looks/feels different over time, not when the food is "tasty" in the moment. Same for agent context.

Why parked, not pitched

The angle is locked but no draft started
The piece needs concrete worked examples: what RDCO's diet actually looks like (16-sender whitelist, the K/F distinction, audit invariants count, etc.) — that's craft work
Doesn't need to ship immediately — better paired with the broader 4-tier model concept once that concept-cluster has a public surface
Per [[~/.claude/projects/-Users-ray/memory/feedback_no_derivative_sanity_check_pieces]] — needs the original-reframe test, which the "scale DOWN not UP" frame satisfies but the draft will need to defend specifically against the dozen-or-so "AI noise" pieces written badly each month

What would unlock pitching

Any of:

A trigger event — major news that makes "information diet" suddenly newsworthy (an LLM context-rot study going viral, an Every essay about agent diets, etc.)
A specific worked example reaches photogenic shape (e.g. /process-youtube watch flow surviving the May 2026 RSS deprecation cleanly because the diet-mindset made the migration painless — half-written, could be the lead)
Founder explicitly says "let's ship this"

Source attribution if it ever ships

Phrase coined by founder in iMessage 2026-05-14 15:41 ET
Connects to: process-newsletter skill, /sync-contacts whitelist filters, process-youtube watch flow, the 4-tier BUY/BUILD model (this is a tier-3 BUILD pattern that doubles as content)

[[concepts/2026-05-14-four-tier-buy-build-stack-soloproneur-tam-filter]] — parent thread that surfaced this phrase
[[concepts/2026-05-13-amble-is-software-losing-its-head-defensibility-migration]] — adjacent defensibility frame; agent diet IS a defensibility move
[[01-projects/sanity-check/parked-angles/2026-05-13-compounding-intelligence-parked]] — sibling parked angle in the same cluster
[[02-sops/2026-04-19-newsletter-output-invariants]] — the audit invariants spec is a load-bearing worked example for the draft
[[~/.claude/projects/-Users-ray/memory/feedback_no_derivative_sanity_check_pieces]] — why parking matters