# Governance-as-Code — Research Notebook

## Question

[2026-10-10] Can the position paper make a testable contribution without treating its architecture metaphors or Base120 axis interpretation as empirical findings? This is a retrospective editorial question, not a preregistered experiment.

## Method

[2026-10-10] Read the manuscript, bibliography, and unexecuted orthogonality protocol at repository base `0b6cd1fa9865605499d0ad11bf96adaf1f26ed68`. Compare claims with supplied artifacts and check selected sources through publisher, author, or institutional pages. Revise living text without claiming a release decision or study results.

## Predictions

[2026-10-10] No pre-observation predictions were registered. Future axis or enforcement studies must freeze their own plans before collecting confirmatory outcomes.

## Observations

[2026-10-10] The original abstract calls the axes empirical, while §7 and the protocol say the reproducible study has not run. The protocol treats cosine distance as opposition and compares dependent family pairs. Six families yield 15 perfect matchings. A finite grammar alone supplies no termination guarantee.

[2026-10-10] Publisher records attribute the good-regulator theorem to Conant and Ashby (1970). Catala is published in PACMPL (2021), DOI `10.1145/3473582`. WMA records a 2024 Helsinki revision, contradicting the former “latest 2013” note. Sources are linked in the revised manuscript and bibliography.

## Interpretation

[2026-10-10] A narrower contribution is defensible: distinguish authority, executable policy, effect mediation, and outcome evidence, and state enforcement assumptions. Geometry, semantic judgment, and utility require different studies. These are methodological judgments, not implementation validation.

## Corrections and dead ends

[2026-10-10] Withdraw empirical confirmation of axes and promotion/rejection claims based only on private ledger references. Replace a binary pass/fail battery with a prospective diagnostic and evaluation protocol. Preserve the original dated changelog entry and append a correction.

## Open questions

- Which action boundaries are completely mediated at a given implementation commit?
- Which missing observations or alternate credentials invalidate an enforcement claim?
- Can independent semantic judgments support the proposed pairings?
- Do pairings improve held-out tasks compared with an unpaired taxonomy?

## Cross-references

- [Living manuscript](../papers/governance-as-code-normative-cybernetic-institution/README.md)
- [Prospective axis protocol](../papers/governance-as-code-normative-cybernetic-institution/orthogonality_test_methodology.md)
- [Revision record](../papers/governance-as-code-normative-cybernetic-institution/CHANGELOG.md)

## Exploratory diagnostic follow-up — 2026-10-10

The [separate plan](../papers/governance-as-code-normative-cybernetic-institution/analysis/exploratory-plan.md) was locally frozen at 13:18:40Z, after input-shape and one-record inspection and before numerical analysis. Its hash is `1e13ebe784cea0eaa9b9085947c04677fb876226ff740a03740fb1a02799bb06`. This is exploratory provenance, not external preregistration or a retroactive prediction.

All 120 definitions at Base120 `f27d603f401751a772fcade069038b45c248fe42` were retained. Under definition-only TF–IDF, the proposed P–IN/CO–DE/RE–SY pairing ranks 13/15 by mean paired distance (0.328995090258) and 5/15 by smallest maximum absolute contrast cosine (0.027593465578). Every matching has maximum absolute cosine below 0.079. The [package](../papers/governance-as-code-normative-cybernetic-institution/analysis/README.md) documents representation limitations, including sparse short texts, removed negation, and unnormalized family means. No scientific score or representation was changed after observing results.

Twelve synthetic tests and eleven reproduction checks passed. A fresh CLI run reproduced scientific JSON/Markdown bytes. A second native worker used a separate sparse-TF–IDF/Gram-matrix calculation without importing the analysis function, enumerated the alternatives independently, and matched all scores within 4.85e-13 and both ranks exactly. These are internal same-family checks, not independent scientific review. No human-rating, model, task-utility, or deployment experiment was run.