# v2.3: what the outside read requires before merge, proposed as a second pass

**Type:** proposal (level 2 for the spend; level 3 for the wording, the author's at merge) · **Date:** 30 September 2026 · **Status:** awaiting the author · **Why:** GPT-6 Astra held the release (`20_read_astra.md`). The first pass used USD 9.42 of the USD 10 cap; this pass is new scope and needs a new cap · **Proposed cap:** USD 8 in Claude Sonnet 5.5 tokens, one Codex call for the re-read

## Already fixed by the editor on the branch, no further cost

- The claims file's note for objection 17 carried the sentence the amended spec had withdrawn; replaced with the amended reply's content.
- Objection 17's source note said the owner edited and adopted it; it now says the editor drafted it and the author adopts or refuses it at the merge.
- The published review now opens with an editorial provenance notice: written by the author's personal agent, commissioned by him, independent of the editor and the outside reader and not of the project.
- The candidate-v3 test's version check is pinned to 2.3 again, so the next release must change it on purpose.
- The author's untracked documents that an editor's commit had swept onto the branch (`md/Archive.zip` and thirteen Triangulation files) were removed from the branch history before anything was pushed.

## What the second pass would change (exact wording drafted by the editor before the implementer runs)

1. **R1 reconciled across the paper.** The corrected Part VI paragraph is right locally, but six other places still treat the two-orientation pattern as established: Part IV ("recur around two orientations"; "evidence that demands some explanation"), Part V ("robust enough, and concordant enough"), Part VI ("non-coincidental"), Part VII ("the pedagogically relevant fact"), M1's served text ("contemplative traditions document"), and objection 7's status ("Good in current version"). Each would be qualified to match the Academic Edition: an interpretive pattern the proposal reads in the reports, not yet established by systematic collection.
2. **S4 narrowed to what its falsifier can test.** Reports shaped by doctrine are still data; what the new falsifier defeats is one use of them. S4's claim would become that contemplative reports carry information about experience beyond doctrine, expectation and translation, which the rewritten falsifier then tests. M3's moved clause gets a measurement bridge, since two orientations need not show as two clusters.
3. **S3's falsifier matched to its role.** S3 is now a method that stands on its own, but its falsifier still targets a separation of structure from phenomenality, which would not defeat a method. It would be rewritten to say what would show structural investigation unproductive.
4. **Objection 17's reply names every place it bears on**, adding Part VI's complementary-appearances conjecture, Part VIII's process account, the cohering page, the home page, and the Academic Edition's "ontologically prior to the container".
5. **The paper's edition label.** The paper's HTML, Markdown and PDF still say v2.1, August 2026, under a v2.3 colophon; either update them or state that the edition label and the revision date differ.
6. **The changelog** states the scope of each repair and links to the outside read and what was done with it.
7. **Regression checks** for objection 17's note and the R1 passages.

Items 2 and 3 change claim text; the wording is the author's to accept or refuse at the merge.

## What would count against this

- The second pass growing the release into a paper-wide revision: items 1 and 4 touch several parts of the paper, which is what consistency requires once one paragraph is corrected; the alternative is to ship the four original repairs with a changelog that says plainly which passages remain to be reconciled.
- The cost estimate: the first implementer spent most of its tokens re-reading large files; the second pass gives it a named list of passages.