Tree G: theory. Plan

Status 2026-10-30: sections 2 and 4 drafted, reading list and checklist written. Runs in parallel with Tree A from day one. Context: ../../research-plan.md section 2 (Tree G).

Goal

One position paper, written early, so that nobody can say the group does not know what it is claiming. It states what a positive residual would and would not show, fixes the vocabulary, and supplies the checklist every later claim has to pass.

Deliverable: position paper, ~6,000 words

Working title: "Pricing the free lunch: what an unpaid competency would show".

The sections, with the concrete work behind each:

1. The claim, stripped. Restate Levin's three hypotheses as one measurable sentence (the residual) and one metaphysical sentence (ingression), and say that only the first is tested here. Source: ../../Review.md sections 2 and 7.
2. The accounting problem. Effort, substrate bias, competency and residual, using the v0.1 definitions from Tree A. Relate effort-as-description-length to Kolmogorov complexity and to the Wolpert-Macready no-free-lunch theorems. An advantage on some tasks is paid for on others, so a residual is always a statement about a task family and never about a system alone.
3. Constructor theory. Deutsch and Marletto take possible and impossible transformations as primary. The 2026-09-30 draft called this Levin's "enablements, not just constraints" with a formalism attached; the reading (reading.md) does not support that, since constructor theory states its laws as possible and impossible tasks and Dodig-Crnkovic treats enabling and constraining as complementary. It also prices nothing: a task counts as possible if it can be done to any accuracy short of perfection. So it gives no effort measure, and this section is one paragraph (Q2).
4. Causal exclusion in ledger terms. Kim's argument first. Then the point that a positive residual does not show non-physical causation. It shows that the accounting left out a term, and finding the term is the science. We say this before any data exists.
5. The three discriminators from the Sheldrake test (Review.md section 5), rewritten as questions any emergent-mind claim must answer:
   • Does the effect require prior instances of the pattern (history-dependence)?
   • Is there a manipulable variable that switches the effect?
   • Is agency attributed to the pattern, or to the system?
6. Where the programme sits. Reconcile it with Dodig-Crnkovic's Cognitive Platonism, which keeps the ordered pattern space and argues the realm is not needed (it stays agnostic on whether it exists; II, p. 10), and with Birch's centrist programme. Corrected 2026-10-30: version 1 of Birch's manifesto tests biological naturalism comparatively across animals and waits on interpretability for the AI side; the "biology-independent theory with tests for non-biological systems" came from a secondary post. Read the 2026 text before relying on either reading. State that the ledger is an instance of the CMEP/Eleos "developmental evidence" class.
7. The checklist, on one page: metric substitution, matched-effort baseline, random-substrate null, pre-registration, adversarial replication, no self-report as primary evidence, and the three discriminators. It is published as a standalone page and also inside the paper.

Steps

1. Read Wolpert and Macready 1997; Marletto, "Constructor theory of life" (2015), the paper that deals with design; Kim, "Mind in a Physical World" ch. 2; Dodig-Crnkovic, Cognitive Platonism I and II; Birch, "AI Consciousness: A Centrist Manifesto"; CMEP/Eleos "Studying AI Welfare Empirically" (2026); and Chis-Ciure and Levin 2025 for their search-efficiency metric. Notes go in reading.md in this folder, one section per source, with page references.
2. Draft sections 2 and 4 first, because they constrain Tree A's definitions.
3. Draft the checklist (section 7) and circulate it to Tree A and Tree B as the pre-registration template.
4. Write the full draft and review it internally against style.md. Before posting, send it to one sympathetic critic (Dodig-Crnkovic) and one hostile one (Sweet, who will reply).
5. Post it as a preprint (PhilArchive or arXiv cs.AI) and as a page.

Decision gates

• Sections 2 and 4 are done before Tree A freezes ledger v0.1.
• The checklist is done before any Tree B pre-registration.

What this tree does not do

• It neither defends nor attacks the Platonic space. It states what would and would not count.
• It proposes no theory of consciousness. The paper is about accounting for competency.