Grounded Abstraction

Source map and review scope

Review date: 9 October 2026. The first edition contains fourteen references. Entries 1-12 retain their recorded predecessor review scope unless explicitly updated below. The review was targeted, not a complete novelty audit or a reproof of the cited literatures. Every formal result used as a theorem in this manuscript has a self-contained argument in the manuscript. Contextual citations do not silently import stronger unexamined results.

Project predecessors

1. Machine Understanding v1.0.0, Komorebi P., 8 October 2026. The first-edition abstract and source relationships concerning behavioral state, acquisition, correction, and explanatory recipients were inspected. Its supplied source package's formatting was consulted. Related results retain their own proof boundaries; this new paper does not inherit a Lean certificate.
2. Self-Improving Intelligence v1.0.0, Komorebi P., 8 October 2026. Used as the immediate source context for finite diagnostic synthesis, persistent model tags, supplied semantics, and a guarded recursive research scope. No historical experiment is reinterpreted as evidence for grounded learning.
3. Sufficient Resolution, standardized traditional paper, Operation Komorebi, 2026. Library abstract, reading scope, supplied-operator behavioral results, calibration relationships, and source concordance were inspected. The known-operator behavioral quotient and the unknown-extension calibrated core solve different stated problems. No new Lean build of this predecessor was attempted.

Exact mounted predecessor PDF hashes are in history/v0.1.0/validation/predecessor_bindings.json. The Library-only Sufficient Resolution source was read as parsed text; it was not downloaded merely to invent a byte-binding claim. None of these predecessor files was edited.

Public foundations

1. Rothenberg (1971), Identification in Parametric Models. Primary bibliographic record checked, DOI 10.2307/1913267. The attempted JSTOR page did not provide readable full text. Citation is contextual, to classical identification from observational laws; no unexamined local-rank or information-matrix theorem is used.

2. Littman, Sutton, and Singh, Predictive Representations of State. Conference 2001, proceedings volume 14 (2002), pp. 1555–1561. The original proceedings PDF was obtained and its opening and definition context inspected. The web abstract's author list was incomplete; the original PDF supplied all three authors. An author-copy fetch failed and the BibTeX endpoint was not readable through the tool, so neither was treated as full-text verification.

Primary PDF: https://papers.nips.cc/paper_files/paper/2001/file/1e4d36177d71bbb3558e43af9577d70e-Paper.pdf

1. Angluin (1987), Learning Regular Sets from Queries and Counterexamples. DOI 10.1016/0890-5401(87)90052-6. Publisher metadata and an accessible copy of the author's paper were consulted, particularly its abstract and teacher/query assumptions. The current paper does not claim to implement L* or transfer its complexity guarantees to this setting.

Readable paper: https://people.eecs.berkeley.edu/~dawnsong/teaching/s10/papers/angluin87.pdf

1. van Waarde, Eising, Trentelman, and Camlibel (2020), Data Informativity: A New Perspective on Data-Driven Analysis and Control. IEEE TAC 65(11):4753–4768. Author publication metadata and the full arXiv paper were available. Review focused on the abstract, compatible-system sets, property informativity, and the common-controller quantifier. This is the closest acknowledged foundation for identifying a useful property without identifying a whole system. The paper does not claim that organizing principle as new.

https://arxiv.org/abs/1908.00468

1. Geiger et al. (2025), Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability. JMLR 26(83):1–64. Official abstract and publication metadata, including the author list, were checked. No complete proof audit of the 64-page paper was performed. Used to distinguish mechanism-level compatibility from an observer's acquisition and initialization problem.

https://jmlr.org/papers/v26/23-0058.html

1. van den Bos and Vaandrager (2019), State Identification for Labeled Transition Systems with Inputs and Outputs. Author arXiv record and readable article context were inspected for the testing framework and distinction between kinds of identification. It is not the source of the draft's explicit three-world destructive-test example.

https://arxiv.org/abs/1907.11034

1. Blackwell (1953), Equivalent Comparisons of Experiments. Annals of Mathematical Statistics 24(2):265–272, DOI 10.1214/aoms/1177729032. Primary publication metadata was available. The full-text endpoint did not yield readable text. The finite binary-testing and postprocessing arguments are proved directly, not attributed to an unread theorem.

Added primary sources for v0.5

1. W.-G. Tzeng (1992), A Polynomial-Time Algorithm for the Equivalence of Probabilistic Automata. SIAM Journal on Computing 21(2):216-227, DOI 10.1137/0221017. The primary publisher's bibliographic record and abstract were checked. The full article was not reviewed. Used only to acknowledge the established finite linear-algebra equivalence-testing tradition. The paired-kernel proof in this manuscript is self-contained and does not silently import the paper's complete algorithm or complexity theorem.

   https://epubs.siam.org/doi/10.1137/0221017

2. M. Petreczky, L. Bako, and J. H. van Schuppen (2013), Realization Theory of Discrete-Time Linear Switched Systems. Automatica 49(11):3337-3344, DOI 10.1016/j.automatica.2013.07.022. The author preprint, its readable HTML, and CWI publication record were inspected. Selected passages included reachability, observability, Theorem 3 on minimality, and the finite realization procedures. This was not a complete independent reproof of the cited article. The paper supplies the appropriate context for common linear dynamics and reachable/observable quotients. Our fixed-unknown-model comparison is kept separate from unrestricted switching.

   https://arxiv.org/abs/1103.1343 https://arxiv.org/html/1103.1343v2 https://ir.cwi.nl/pub/18901

The author surname uses Petreczky, as in the article and publication metadata; one surfaced preprint record used an inconsistent spelling. Entry 7's data-informativity metadata and abstract were also rechecked for the compatible-model interpretation. Its broader first-draft reading scope is retained, not represented as a new complete reading.

A related author paper, On the Complexity of the Equivalence Problem for Probabilistic Automata (https://arxiv.org/html/1112.4644v2), was consulted at its finite-span argument as an additional contextual check. It is not an added bibliography entry or an assumed theorem in the manuscript. Unrelated sections were not used.

Added primary sources for v1.0

1. F. Le Gland and N. Oudjane (2004), Stability and Uniform Approximation of Nonlinear Filters Using the Hilbert Metric and Application to Particle Filters. Annals of Applied Probability 14(1):144–187. DOI 10.1214/aoap/1075828050. The Project Euclid publication record, indexed primary-repository metadata, and an author publication list were checked. The full-text endpoint returned an iframe rather than readable article text, and HAL returned an anti-bot page. Those fetches are not full-text review. The citation only acknowledges established filtering-stability and approximation work. The finite retained-record theorem in this manuscript is proved directly and is not attributed to an unread theorem from that paper.

   Primary publication endpoint: https://projecteuclid.org/journals/annals-of-applied-probability/volume-14/issue-1/Stability-and-uniform-approximation-of-nonlinear-filters-using-the-Hilbert/10.1214/aoap/1075828050.full

   Author publication list: https://fime-lab.org/oudjane-nadia/

2. A. Girard and G. J. Pappas (2007), Approximate Bisimulation Relations for Constrained Linear Systems. Automatica 43(8):1307–1317. DOI 10.1016/j.automatica.2007.01.019. An author-hosted full PDF was obtained after the direct publisher endpoint returned HTTP 403. Review covered its title and bibliographic data, abstract, introduction, and selected approximate-relation definitions on the opening pages. No complete independent reproof is claimed. The paper is a primary-source comparison for observable system approximation. Our discrete visible/hidden block bounds and their composition with a probability record are derived explicitly rather than imported as an unstated result.

   Readable author copy: https://www.georgejpappas.org/wp-content/uploads/2024/04/GP-AUT07.pdf

   Author publication list: https://www.georgejpappas.org/publications/

Entry 7's arXiv metadata and abstract were revisited for the continuing distinction between identifying a system and justifying a property of all compatible systems. The larger predecessor review of that article was not repeated. The new precision, leakage, and decision arguments were checked from their stated proofs and small exact constructions; reference review is not a substitute for those arguments.

What the review does not establish

No comprehensive priority claim, citation-completeness claim, empirical performance claim, or proof-assistant result follows from this source map. Some foundations were inspected only through metadata or selected passages. No third-party full papers or standalone fonts are included in the package.