[aliromman]

The complete selected same-grid stochastic caller has a family of 2^(86*N) distinct public inputs with the same public output for every fixed retained seed, where N=512^f. Its PR #6546 native sampler makes that output exactly zero. Exact input recovery from this output and unchanged seed therefore needs at least 86*N bits of additional retained capacity: 44,032 bits, or 5,504 bytes, at width one. An explicit packed code attains the bound on the displayed family. This is a source-bound negative result and recovery construction, not a global minimum for all caller inputs.

No new complete kappa bound or selected-frontier change is claimed. The result identifies a concrete information obligation for owned_erasure and precision_tape; all five selected framework assumptions remain open. It does not invalidate the predecessor's numerical error theorem or its inverse, which keeps the required record.

Review the proof, exact receipt, source pins, causal checker and codec and verifier. Package head: 317e4a099488794c372f8a4ad1e70deef3626bbf. Base: canonical main 7c49fe32c55de02c2c873655c804b16e4c3fef0a. The unchanged selected word is the only external runtime input. The numerical contract and zero-native premise refer to the sampler at predecessor commit ff038070; that implementation is inspected, not copied or imported by this package.

Two independent first-access censuses agree for all 770 public fields. Exactly 43 fields first appear, as either a read or write, in a signed half-SCALE. For its sign p and fixed scalar seed u, choose each real or imaginary input component as x=p*(b-u) with independent b in {0,1}. Each component is in {-1,0,1} and its first half gives floor((p*x+u)/2)=0, while its literal retained residue is b. A full causal induction checks all 22,803 events. The last independent label disappears at cut 10448, after native calls have begun; this is not just a scalar-prefix count.

The whole public output is independent of all 43*2*N bits. Zero-preserving native functions make it zero. Even without that extra zero-value property, deterministic native functions with the same declared local read sets give the common output F_S(0), so the capacity count still holds. A finite arbitrary dirty bank cannot reduce the count while its initial information remains recoverable. This does not establish an initially clean hardware-storage minimum. Original input copies, changed seeds and other metadata that carry the distinction count as retained information.

The matching family codec packs the 43 half residues and uses the unchanged half seeds to recover the original input. Other division residues are seed-only constants on this family, and its saved COPY/CLEAR values are zero. Thus the bound concerns simultaneous variation in the actual complete record. No global maximum-fiber calculation or full-input minimum record is claimed. A fixed input pair, independent of the seed, also has identical complete output with probability at least one half under the predecessor's fresh uniform sampling law.

The input class is arbitrary public h*Z[i] fields in this fixed selected word and its tensor lift. An even-coordinate entrance promise removes this displayed family. The bound does not transfer automatically to actual multiplication input correlations, to the distinct exact entrance-grid contract in PR #6499 and PR #6506, or to a changed word/output/seed interface. Undeclared native reads can break the first-access proof. Compute/copy/uncompute with the original input retained remains permitted by the result.

Use CPython 3.12 and its standard library. These commands passed:

python3.12 -B scripts/research_package.py --package research/selected-caller-zero-fiber --check-only
python3.12 -B scripts/research_package.py --package research/selected-caller-zero-fiber
python3.12 -B scripts/check_research_commit.py --package research/selected-caller-zero-fiber --check-source-history

Expected markers:

PASS_PACKAGE_METADATA_AND_PINS (not a mathematical acceptance check)
PASS_SELECTED_CALLER_ZERO_FIBER
PASS_REGISTERED_RESEARCH_PACKAGE research/selected-caller-zero-fiber outcome=completed
PASS_RESEARCH_COMMIT_BYTES

Both normal and optimized registered runs passed. They check all first accesses and source events, eight signed-half endpoints, 86 individual label witnesses, 53 complete scalar/zero-contract replays totaling 1,208,559 events, nine semantic/domain controls and source-corruption rejection. The replays include two full width-one, 512-coordinate cases. At every MOVE they check the zero operand and apply only its proved zero contract; compiled native implementations are not run. The exponential family is established by the proof, not by exhaustive enumeration.

The committed-object check covered ten files including the declared source input and its main ancestry. The official export and input check passed:

python3.12 -B scripts/export_research_package.py export --package research/selected-caller-zero-fiber --ref HEAD --output /tmp/selected-caller-zero-fiber.tar
python3.12 -B scripts/export_research_package.py check /tmp/selected-caller-zero-fiber.tar --check-inputs

Expected marker:

PASS_RESEARCH_EXPORT

The official archive also passed an isolated optimized replay in 6.45 seconds with Git unavailable, all files read-only and all eleven exported file hashes and modes unchanged. Hosted CI is pending at publication. The complete predecessor native suite, full selected cold replay, independent human/proof-kernel acceptance, physical routing/randomness/cleanup bill and global fiber maximum are not supplied here.

Credit @DreamingOfClouds for the selected PR #6030 word and retained P104 lineage. Ali Romman's PR #6546 supplies the complete same-grid ABI and inverse record; PR #6532 and PR #6509 supply its pair-map predecessors. Credit @zxv for signed-half residues and matched inverse seeds in PR #6066, and Nick Petalas / @petalas for the reuse boundary in PR #6418. The earlier PR #5323 / Discussion #5303 inverse-code count and PR #6540 native period-lattice count are distinct local predecessors. NOTICE preserves their source authors and broader lineage. Pigeonhole counting, first-access dataflow and retained residues are classical; the new result is this complete caller fiber, independent census, joint bound and exact family code.

Canonical claim. Canonical main and current public Discussions/open/closed PR metadata were refreshed before submission. Related zero-fiber, scalar-record and native-codec work was inspected; a separate main-source search found no same first-access theorem. Arbitrary diffs and unpublished work remain overlap limits. Only nine files under research/selected-caller-zero-fiber/ are added. Ali Romman with substantial OpenAI Codex assistance; no subagents. Apache-2.0. The open next question is the actual entrance contract and maximum complete-map fiber, followed by a paid full-input recovery schedule.