Notes — Campaign intelligence POC

What we shipped.

A 30-day POC, closed 2026-05-12.

An honest write-up of a campaign-intelligence engagement that ran April 13 – May 12, 2026. Apr 13 kickoff, Mon/Wed/Fri briefs, 19-page handoff on day 30. Here's the shape, what we shipped, and what we learned doing it.

StatusClosed 2026-05-12 · paid in full
Engagement30-day fixed-scope POC · Apr 13 – May 12, 2026
CadenceMon · Wed · Fri
ComputeLocal, on our hardware
Attestationevery brief signed against pulse_id
Handoff19-page closing doc on day 30, plus a live Q&A engine over the corpus as leave-behind

What we did

We ran a 30-day campaign-intelligence POC for a regional client. They wanted a regular drumbeat of intel on a market they compete in, without burning an analyst's full week on manual scraping, and without sending their clients' material through a third-party AI vendor. We put a small pipeline in front of those three constraints and let the POC settle the question.

The brief shape

We landed on a Mon / Wed / Fri rhythm. Each day targets a different decision surface, so the week reads as a sequence rather than three copies of the same digest:

Monday — Arena Map / what's in the market this week / notable moments / a suggested play for the week Wednesday — Per-angle Deep-dive / longevity-winners by angle / variation winners / creative brief, ready to spec Friday — Verdict / HIT / PARTIAL / MISS header / cross-week arc / market snapshot + movement

Friday checks itself against Monday. We don't claim that loop is perfect — we claim it's the smallest honest feedback shape we could build, and that having one matters more than not having one.

How it actually ran

The pipeline lived on hardware we own. There were no public endpoints reachable from the open internet; everything stayed on a private fabric. Briefs were delivered to a private channel. The client's source material never crossed our wire to a third-party model.

Every brief was bound to a pulse_id from our live plasma physics beacon at the moment of delivery, signed by an Ed25519 key held on independent witness hardware at our edge. Months from now the client can re-derive the digest, fetch the witness public key, and verify the signature with the same one-liner we publish on every Rail release. That's the same primitive Ledatic uses for itself, not a separate marketing layer. See /entropy and /system.

What we learned

Two things honestly. First: the discipline of shipping a structured brief three times a week sharpened the questions we asked of the data more than we expected — the cadence forced clarity the analysis would not have produced on its own. Second: the attestation step turned out to be a smaller tax than we feared once the cron was wired, and the client never had to think about it. The cryptographic floor was there if anyone ever wanted to inspect it; nobody had to in the moment.

One quieter learning: the client consumed the briefs but didn't engage in real-time. Reactions and follow-ups happened on their own internal cadence. That's not a failure mode — it's a signal about who actually pays for delivered intelligence vs. who pays for interactive analysis. Useful to know.

If a shape like this is useful to you

This was one engagement. Most of what we ship is smaller — a weekend script, a one-day tool, a tiny watcher for one thing you care about. The cryptographic floor underneath is the same one Ledatic uses for itself, so anything we hand you can be verified the same way every Rail release can. If anything here resembles a thing you wish were easier, tell us about it and we'll figure out the smallest version that would actually help.