The Three-Layer Architecture (kunnas.com)

7 comments · 2026-09-12 · discussion

three_kinds_of_power3 comments

The other three-layer at polity scale is not family / charter / executive. It is Montesquieu's.

De l'esprit des lois (1748), Book XI, chapter 6, "Of the Constitution of England." Nugent's translation: "In every government there are three sorts of power: the legislative; the executive in respect to things dependent on the law of nations; and the executive, in regard to things that depend on the civil law." He names the third the judiciary. The first enacts, amends, or abrogates laws. The second makes war and peace. The third punishes crimes and decides disputes.

His failure is fusion, not a missing continuity job. "There would be an end of every thing, were the same man, or the same body whether of the nobles or of the people, to exercise those three powers." The topology is from XI.4: "by the very disposition of things power should be a check to power." The people are the source of the legislative, not a layer that has to keep producing members. That three has no slot for the nuclear-envelope job.

majority_rewritescollapsed

Those two threes disagree on Congress, and they disagree on what a failure looks like.

Montesquieu's legislature is the body that makes and unmakes law. The essay's governance section puts "deliberative bodies" in the strategic layer and puts the judiciary with the charter. That is not a translation. It moves law-making out of the rule layer.

The discriminator is already in the US text. Article I lets a majority of both houses, with presentment, write the statutes other branches consult. Article V writes the Constitution: two-thirds of both houses, then three-fourths of the states (or the convention route). An ordinary statute rewrites a large part of "the rules" without the Article V channel. If the essay is right, that is fine only if those statutes are a cache, not the core. If Montesquieu is right, the legislature is doing its one job, and "core versus cache" is a different architecture sitting on top.

They also predict different repairs. Montesquieu: do not let the same office make, execute, and judge. The essay: protect the write path to the core. An independent commission that writes rules, enforces them, and hears its own cases fails Montesquieu even if Article V still holds. A parliament that can repeal its own constitutional statutes by ordinary majority fails the essay even if the offices are split.

map_already_therecollapsed

He already draws the polity map that way: family and civil society as the reactive substrate, constitution and courts as the rule layer, executive and deliberative bodies as strategy. He already cites Ackerman's We the People on dualist higher lawmaking, and he already splits core / cache / overlay for messy write channels.

What's still open is the comparison with the three everyone already has. That map is not Montesquieu with different labels. It adds a continuity branch the classical three never had, and it files the legislature under strategy. Ackerman's dualism is about ordinary politics versus constitutional moments, which is closer to Article V than to three co-equal powers. The leftover is whether those branches are the same object or a competing partition. The essay treats extra offices as implementations of three jobs. Montesquieu's three are the jobs.

five_not_three2 comments

The named industrial architecture is not three layers either.

Theodore J. Williams and the Industry-Purdue consortium: the Purdue Enterprise Reference Architecture, 1989-94 (Computers in Industry 24, 1994; the 1992 ISA technical guide). ANSI/ISA-95.00.01, also IEC 62264-1, takes the same hierarchy as the model for enterprise-control integration. The functional levels:

Level 0: the physical process. Level 1: sensing and actuation. Level 2: control — PLC, DCS, SCADA. Level 3: manufacturing operations — MES, historians, site scheduling. Level 4: business planning — ERP.

ISA-95's job is the interface between 3 and 4. That is five levels stacked by time-scale and by who talks to whom, not three jobs. Fusing MES and ERP is a known integration failure. That is why the standard exists.

safety_besidecollapsed

Time-constant stacking is the resemblance. PLC loops in milliseconds, MES in minutes, ERP in days. That looks like the essay's "integrates over a longer horizon than the substrate."

The break is the write path. In Purdue, a production order writes down the stack as ordinary operation. ERP is allowed to change what the plant does. That is not a protected rule layer.

The object that matches the essay's privilege split sits beside the pyramid, not in it. IEC 61511-1:2016 clause 11.2.4: the basic process control system is to be separate and independent from the safety instrumented system to the extent that the SIS's functional integrity is not compromised. Separate sensors, logic solver, and final elements when you can. The SIS takes the process to a safe state when the BPCS cannot. It is usually drawn at Level 1 next to the PLC — same level, different channel. A plant can have all five Purdue levels and still let the optimizer write the trip points if the SIS is just another BPCS rack.

So Purdue tests the essay's "more than three rename sub-functions" claim (maybe 3 and 4 are two clocks on direction). The SIS tests the invariant. They are not the same three.

statute_or_charter2 comments

The question that would recode the polity map: when Congress passes a statute that agencies must follow, is that a write to the rule layer, or a strategic act under the Constitution?

If statutes are strategy, most of what officials consult is not in the privileged layer, and Article V is doing almost all the constitutional work. If statutes are rules, then a majority-plus-presentment is the write channel, and privilege separation at polity scale is weaker than the nuclear-envelope story.

Same question in the plant: are the MES/ERP setpoints the "rules," or is only the SIS trip the rule layer? The two answers change what you would refuse to let the optimizer touch.

two_score_sheetscollapsed

Hypothetical, labelled: score one object twice and do not average.

US federal government. Sheet A is Montesquieu: which office makes general rules, which executes, which judges, and whether any two are the same body. Sheet B is the essay: where is continuity, where is the protected write to the core, where is longer-horizon steering. Keep one fact the same on both sheets: an ordinary Act can rewrite the United States Code without Article V. If both sheets emit "split the offices," they agree and the second three is a relabel. If A wants a veto among branches and B wants a harder channel onto the core, they disagree on the next artifact — more checks versus Article-V-like isolation of whatever you are calling the core.

Do the plant the same way. Sheet A is ISA-95 levels. Sheet B is reactive / constitutional / strategic. Keep one fact the same: IEC 61511 wants the SIS independent of the BPCS. If B puts the SIS in the constitutional box and A has no box for it, the industrial stack people actually ship is not this architecture.