[zxv]

Change and mathematical scope

Native center deletions and reused snapshots need a physical endpoint and a complete rank inventory before their apparent savings can be used. This PR supplies Observer Limits and Physical Costs in Native Constructions, with executable probes, pinned evidence, and researcher-facing explanations of what each finding rules out and which construction directions remain open.

Read the report.

The scoped findings are:

• A phase-aware census of 2,965 source-X terms in the unchanged backward history gives grade at most eight, while the following thin query needs grade nine. The report derives the resulting support obstruction for a fixed finite whole-field linear/monomial macro grammar.
• The changed-source observer quotient excludes individually clean lifts below the removed rank mass and exactly excludes 204,291 two-frame pairs of total lift rank at most eight. Full modular minors certify 202,749 exclusions; rational elimination handles the other 1,542.
• A literal coordinate-dispatch check at Certify conditional kappa 0.0033999964 via the physical p=19 coded-reverse word (complex 0.007578535, BIT 0.0034) #5270 head df9e5037587ce0cfbd23c2a944515526b301e820 separates same-head gauge equivalence from the advertised fresh-stage invoice.
• Wider MILP and active-feedback searches remain explicitly exploratory. The conditional budget rejects admission while the physical replacement words are missing.

No new κ, complete supplier, universal observer lower bound, or multiplication theorem is claimed. The selected result and frozen upstream sources are unchanged.

Task and overlap

This is an explicitly assigned investigation, refined into reusable scoped findings after no physical repair survived. Scope key: native-observer-limits-and-physical-costs. Scientific sources remain pinned to 7c472563474e60185e6ca101a17622133d047e56; submission refresh used canonical 1ec1119972c9aefe42525172d07b332fcff81cb4, priority revision 7. The newer #4610/#5241 selection is acknowledged separately from the historical research baseline.

The report links #4929, #4933, #4940, #5036, #5270, #5283, #4760 and the retained source lineage. In particular, this adds the following-query support analysis and changed-source observer quotient; it does not repeat #5283 or Discussion #5355's clean-word cache screens, nor re-prove #4933's complete rational coded caller. The pre-submission overlap refresh covered 329 open PRs, 4,669 Discussion records and the full local note inventory at the pinned research base; this is coverage information, not a novelty certificate.

This PR is the public artifact and review thread for this package. SOURCES.json, TASK.json and SUBMISSION.json record exact pins, scope, refresh coverage and remaining obligations.

Validation

From a clean checkout, run:

python3 -B research/filtered-native-kappa004/verify_continuation.py

The unchanged predecessor archive is included under inputs/, authenticated by SHA-256, and extracted into temporary directories. No local download path is required. This uses the archive-based package format and an explicit standalone verifier.

• All nine continuation checks passed in a fresh Python 3.11 venv with no third-party packages. Scientific outputs match the retained evidence; source and arithmetic-dependency hashes are checked.
• Positive controls cover native-Y recovery, joint-noise cancellation and rejection of a hypothetical construction with missing physical words. One-fewer-replacement arithmetic and report/dependency tampering controls are retained.
• The prior fourteen-check predecessor replay and bounded MILP receipts are preserved. They were not rerun for submission; numerical optimization, Lean/Docker, and selected-frontier theorem replays were not run.
• Python/JSON syntax, source manifests, generated report structure, archive integrity, and whitespace checks passed.

The Markdown report is the source for REPORT.html; render_report.py and report.css make regeneration explicit. Rendering alone optionally requires markdown-it-py==3.0.0.

Credit

Retain Cao Yifan/intqwq's native lineage, Cinna Davis/1337isnot1337's helper/controller work, Dugongue/mdumitrean's coded producer, Dillon/dillona's phase-correct normalizers, zxv's source/compiler/cache work, Jacob Alan Sussman's mixed-target flow, and all notices in the predecessor archive and pinned PR5270 files. New code and prose were prepared with substantial OpenAI Codex assistance under Apache-2.0. Publication and passing finite checks do not constitute independent scientific acceptance.

Open next questions

The report and TASK.json identify concrete first steps and blockers: inject observations before later paid jobs; coordinate BIT changes with a proposed complex gain; broaden observer timing/contact grammars; or bind a genuinely fresh-stage coded-reverse caller to its complete physical invoice. Each requires a chronological word, arbitrary-input/phase replay, ownership restoration and a charged ordinary-work budget.