**This now gives the critique a second, testable object: whether a machine can preserve the disagreement, its evidence and its concessions through compression.** The kernel asks a reader to reconstruct the conflict well enough to examine it. Agreement with *Daffodils* is a separate question.

The six-body structure matters. It distinguishes the register, full-text readings, editorial record, captures, receiving texts and apparatus history. Each body has limits binding both your reading and the received flagship account. In particular, a capture establishes what a composition did at an address; repetition or institutional standing establishes neither reading’s truth.

The new §6.5.1 also advances the substantive argument. You now examine the sources displayed alongside the composition and compare what they establish with the authority assigned in the answer. That makes the movement from **adjacent scholarship to credited leadership** inspectable at particular passages.

I checked the packaged files independently of the included verification report. The results expose two concrete transmission problems:

| Check | Result |
|---|---|
| Separate map versus bundled map | Identical |
| Ten kernel-file hashes and six map/capture hashes | All match |
| Ten first-seating text snapshots versus their manifest | All match |
| Graph endpoints | All declared |
| Forty claim quotations against the bundled source copies | **38 bind; two require a later source version** |

Both unmatched quotations point to **record #1750**. Claims `ol-defects` and `ei-741` cite its current version, whose recorded hash differs from the bundled first-seating copy. The earlier copy does concede defects, but it does not contain the exact current wording quoted by these claims.

That is a version-packaging gap. Including the cited current copy alongside the historical snapshot would let an offline reader verify both quotations. The supplied verifier depends on an external archive checkout and registry, so its recorded pass is not fully reproducible from this ZIP alone.

**The second problem concerns scope A’s reconstruction test.** Its two files name material that they do not actually contain:

- They refer to K1–K5 without preserving all five definitions.
- They omit the central **141/185** result and the **32 source-reaching / nine governing** breakdown.
- They mention the stylistic control without its **7 / 12 / 31** counts.
- They list research-question identifiers, while the substantive questions and advance/retreat conditions reside in a separate file.

The automated “yield 1.000” checks selected strings for presence. It cannot establish that a reader can reconstruct those missing distinctions. The verification report appropriately says reconstructive validity remains untested.

This is consequential on the kernel’s own terms: **it asks the receiving machine to preserve criteria and counts that its smallest packet has already compressed away.** Putting the missing material into scope A—or explicitly enlarging that scope—would make the reconstruction requirement achievable.

There is a further useful limit to keep explicit: the ZIP carries the map, selected archive records and capture transcripts, but not the census and coding tables named in the map. It supports inspection of the encoded argument; independent recounting still requires those linked materials.

The architecture’s strongest feature is its separation of **faithful carriage from warranted judgment**. A reader should be able to preserve your claim that the apparatus gains authority through closure, preserve the strongest counterevidence, and identify what remains unresolved. The immediate work is finite: repair the two source-version links and put the required distinctions into the reconstruction packet.