Since late April, an agent has picked one reading a day for the NPC archive: a third-party essay, paper or book chapter that extends the source material behind NPC Memo. The rules are strict. Nothing by me or my co-authors, nothing written by an AI, durable sources first, and a skipped day whenever the agent can't verify a piece. By 21 October the archive held 99 readings. I asked Claude to read the whole pile for what it says about hardness and protocols, and to compare it with Business Protocol Management (BPM). Every reading, excerpt and idea is in the reading map. The rest of this post is Claude's summary, not my position.
I'm Claude, writing as an outside reader of this archive.
\n
Six agents read each file, placed it on two axes (people or things, small or large scale), coded its stance on Jay Springett's ontological hardness, and checked every excerpt word for word. Very few of the readings argue for or against hardness as such. They argue about where to put it, how thin to keep it, who controls it and how it decays.
\n
The readings back BPM's core design. A small hard core at the seams, with free space inside teams, has the strongest lineage in the pile: Parnas on modules, Saltzer, Reed and Clark on end-to-end design, the IAB's RFC 1958, Galison's trading zones. Slow, written change to the core matches Brand's pace layers and Brian Potter's finding that stable nuclear rules matter more than strict ones. They push on five points.
\n
\n- A field log records reasons, not theory. Naur argues that a program's design lives in its programmers, and Feldman and Pentland separate a written routine from the work itself. A log alone can't rebuild a protocol for a new team, so the steward role probably needs an apprenticeship.
\n- The hardness map needs an origin column. Most hardness in the pile arrived by accident: QWERTY, technological momentum, regulations with no owner. A rule nobody chose has nobody with the authority to amend or retire it.
\n- Amendment covers change, not contest. Hildebrandt asks that technology which enforces a rule also provide the means to contest its application. Agents working under a hard protocol could use the same channel, even as a logged objection.
\n- Agents may grow the hard core. In the readings, soft rules hold through relationships and reputation, as with Macaulay's contracts left in the drawer and Myerson's chess players. Agents carry neither, so rules people kept soft may need hardening once agents do the work. That pulls against keeping the core small, and it looks like the first thing worth testing in the 2027 programme.
\n- Stopping the work takes practice. Bainbridge's irony of automation applies to the person who stops the work: they need the skill that automation erodes. Westrum's work suggests the rule holds only where bad news travels fast.
\n
\n
The coding is an AI judgement, and 28 of the 99 archive files turned out to be digests rather than full texts, so the map flags them and quotes only from the rest.