Prove the source-independent boundary obstruction for fixed R480 cleanup

[aliromman]

The selected R480 supplier cannot defer any of its 402 final helper climbs
through its unchanged physical FULL-frame cleanup while keeping canonical
data outputs and arbitrary source-independent helper boundary types.
The proof covers dense helper mixing, arbitrary fiber dimensions and all
batch lengths and finite recursive depths. Exact inverse-source witnesses
give the obstruction; the optimum is zero omitted climbs.

Review files: Proof and scope · Evidence · Source pins and credit.

The package also classifies the larger backward target-covariance interface:
19 free operator components on 22 helpers. It checks their native ranks and
mandatory output repairs. For each paired rank-one tail, a uniform inequality
excludes a moment saving with any host of rank at most eight, even with free
scalar adapters and perfect fusion. Changed native cleanup, source-history
protocols, joint encodings and changed owners remain open.

The selected local and global profiles are unchanged, so the conditional
kappa remains 0.001672660197076918. This is a scoped mathematical obstruction,
not a new exponent or a new native construction.

Validation: the registered Python 3.12 replay passes. It checks both complete
712-coordinate scalar maps, all 402 helper witnesses, all 78 isolated
nonfinal helpers, identical directional climb events, two independent exact
covariance classifications, native frames, target projectors, owner ranks,
and eight boundary controls. Direct optimized replays also passed on
Python 3.9.6 and 3.14.7. A read-only relocated fixture passed, and altered
source bytes and certificate coefficients were rejected. No full selected
cold replay, new hosted stream, or hosted CI run is claimed.

Reproduce from the repository root:

python3.12 -I -B scripts/research_package.py --package research/r480-source-independent-cleanup

Base: bce54fb3f895cbe5cd4c0eb477c3afeed100190f. Audited supplier source:
0fc329d0bec93ca7cd288eaea1295172e8789810. The new package contains the proof,
exact certificate, immutable source pins, original task card, and replay
manifest. No selected record, frozen input or upstream file changes.

Prepared by Ali Romman (@aliromman) with substantial OpenAI Codex assistance in algebra,
implementation, proof review, verification and writing. Credit the retained
PR #1094, PR #1042 supplier lineage; Sagar Thakrar (@thakrarsagar)'s PR #1012 and PR #1227; Ross Richardson (@RossRichardson)'s PR #1287; @Scratch-net's PR #1198; and the reader-contract and native
frame predecessors listed in the package. New work is Apache-2.0.

[Mister-Meeseeks]

100-PR batch review at 3bceac90599a02a96e9b744736b3feeb00171504.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Full R480 cleanup classification for source-independent helper outputs, with arbitrary operator mixing and all 712 columns. Source-history and backward target-covariance escapes are classified separately; transported-tail costs and shared-word restrictions are explicit, not a universal fusion obstruction.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.