"PCT architecture in AI"? The authors' own criterion and their own paper, side by side

IAPCT
1

The session. The programme of the 36th IAPCT Conference (York, 7–10 October 2026) included a one-hour hybrid workshop on 9 October: “Clara (AI) demonstration (PCT architecture in AI)”. In the official Abstract Book, the entry reads: “No written abstract — this is a live demonstration. Demo link: https://chat.claranexus.com/”. The other talks and workshops carry written abstracts. On 8 September the company announced the session on LinkedIn as presenting Clara Nexus “as a PCT informed developmental platform”.

Our programme is the Association’s own document. When it describes a system as “PCT architecture”, it lends IAPCT’s name to a technical claim about PCT. That makes it fair, and I think useful, to check the claim against the record.

The authors’ own standard. The company’s research page publishes a seven-point PCT-Accuracy Criterion v2 for deciding when a claim that a system is “grounded in Perceptual Control Theory” is warranted. Among its commitments (verbatim, captured 15 Oct 2026):

  • Commitment 2: the reference value “must be identifiable as a distinct computational quantity, inspectable independently of the perception itself and independently of the output.”
  • Commitment 5: a hierarchy “must be demonstrated by showing that higher-level loops actually set the reference signals of lower-level loops […] regardless of the hierarchical language used in its documentation.”
  • Commitments 1 and 6: evidence of control is a demonstrated compensation of disturbances to the controlled variable.

What the authors’ peer-reviewed paper says about the same system. Front. Artif. Intell. 9:1936295, published 7 October 2026, two days before the session:

  • “Perceptual control theory primitives are reported as prompt-level Companion voice discipline rather than as a runtime-enforced mechanism.” (abstract)
  • “The controlled variable […] has no runtime representation.” (§3.2.3)
  • “The reference value is likewise absent as a runtime object […] the architecture holds no reference-value construct at all.” (§3.2.4)
  • Reorganisation and going up a level “would require a control-hierarchy representation that the architecture does not have before they could be reported as mechanism.” (§3.2.6)
  • No disturbance-compensation test is reported.

To the paper’s credit, it states these limits plainly. The gap lies between the paper and the description in our programme, not within the paper.

After the conference. The same research page now lists, under “Hierarchical perceptual control” (Powers, 1973), “The five-depth runtime: Each depth holds its own reference and its own error.” Among those it describes as “Expert-reviewed”, it lists “The PCT community — Reviews the Method of Levels and PCT claims”. A second paper (Front. Artif. Intell. 9:1985132, 15 October 2026) concerns the platform’s measurement harness and does not address PCT.

Questions

  1. Read against the authors’ own criterion, which of the PCT claims made about this system are supported by its published description?
  2. Should IAPCT require a written abstract for every programme item, and, where a session is described as “PCT architecture”, a short statement of which PCT constructs are implemented and how?
  3. How should the Association respond when “the PCT community” is presented as a reviewer of a commercial product’s PCT claims?
  4. Which perceptual variable would one test for, using the Test for the Controlled Variable, in a conversational system built on a hosted language model?

Sources

Luk