Derive balanced CRT node selection and precision from the physical source before invoking the numerical producer (no new kappa)

research/encoded-balanced-crt-entry/CHECKPOINT.md

@@ -0,0 +1,51 @@
+# Author-frozen checkpoint; independent review pending
+
+Base: `e54523517d0f03e58352e647c403a5af3e525a57`.
+Branch: `research/continuous-20261110-crt-control`.
+Package: `research/encoded-balanced-crt-entry/`.
+Recorded on2026-11-10 at author freeze. The universal source-relative argument and
+author replay are complete; separate review is still a publication gate.
+This checkpoint is not independent acceptance or publication approval.
+
+## Executed
+
+Initial registered normal/optimized replay completed in3.813 seconds combined.
+After the scoped review corrections, the relabelled receipt was regenerated,
+metadata/source-pin checks passed, and both registered modes passed again,
+each with `PASS_ENCODED_BALANCED_CRT_ENTRY nodes=14 owned_roles=78 controls=8`.
+Four actual-source fixtures cover d=2,4,8; nonzero node starts; changing
+half-size; changing numerical precision; invalid left coordinates and
+padded right coordinates. Runtime geometry/decode functions in the new
+frontend are poisoned, and Port/Word/Arena constructors and Word.inspect
+are sealed after the fixed initial inventory is acquired. The unchanged
+Entry and numeric producer actually run. All source bytes/head restore,
+and all78 owned histories retire blank after complete traversal.
+
+The oracle independently solves the small CRT congruences by enumeration.
+Its decoded values are observer-only and never enter runtime control.
+The receipt is frozen and type-sensitive. Twelve scientific source inputs
+are authenticated; catalogue/selected provenance is historically pinned.
+The eight-axis fixture retains97 extra source-tail cells explicitly.
+
+## Remaining gates
+
+The independent source/mathematical review is in progress. Review required
+explicit linear-use/current-lease scope and removal of an unsupported
+numerical finite-control expansion constant. Those statements are now
+corrected: the raw Port-action envelope remains `2^16(M+2)^2`, while the
+complete frontend bill is `C_front(M+2)^2` for a fixed unspecified constant.
+The written proof covers the cursor invariant, that uniform bill, complete
+retained envelope, source input compatibility, and failure boundaries.
+Shape-rejection cleanup, early close, fault controls, source pins, metadata
+and optimized replay all passed. Committed-tree integrity is the next gate.
+No complete CRT row/ROM/payload permutation, binary admission, complete
+assumption discharge or kappa improvement is claimed.
+
+## Next
+
+Resolve independent review findings if any, preserve its exact receipt,
+then perform publication checks. The immediate existing replay:
+
+```sh
+python -B scripts/research_package.py --package research/encoded-balanced-crt-entry
+```

research/encoded-balanced-crt-entry/NOTICE.md

@@ -0,0 +1,11 @@
+# Attribution
+
+New tape-driven balanced-node frontend, quantified composition proof and
+verification prepared for Jerry Chen / CubeJerry with substantial OpenAI
+assistance,2026. Apache-2.0.
+
+The numerical Entry, producer, physical bank, canonical-word routine and
+Word/record semantics are imported unchanged from their authenticated
+predecessors. Preserve Ali Romman, JoyboyBrian and all predecessor credits
+listed in README.md and in the original source notices. The balanced CRT
+identity and general historical-capacity principle are not claimed anew.

research/encoded-balanced-crt-entry/OVERLAP.md

@@ -0,0 +1,38 @@
+# Scope and bounded overlap checkpoint
+
+Base e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8.
+The selected #7000/#7001 record retains all five assumptions at kappa
+.00363005817. The task catalogue has older contextual selected references;
+the actual selected record governs scientific status.
+
+The assigned Wave 2 question follows the actual numerical-entry boundary,
+after the preceding whole-tree source-state composition retained compiled
+topology. The parent approved this explicit narrower scope before code.
+The new program derives its axis count, depth, node cursor, half-size and
+precision from physical tapes and invokes the actual numerical provider.
+It does not merely price a host-generated schedule.
+
+Relevant predecessors:
+
+| Work | Retained result | Difference here |
+|---|---|---|
+| #5053 | Current-coordinate numerical arithmetic | Same exact producer |
+| #5219 | Guide-driven Entry and lifecycle | Guide and exact protected words now come from actual selection |
+| #5353 | Canonical records from a compiled node | Node selection and widths are tape-driven here |
+| #5505 |63-role varying-precision bank | Same exact owner; no new allocator claim |
+| #5496/#5625/#6100 | Compiled row query/update/reuse | Row/ROM/source-update topology stays open |
+| #5486 | Binary refinement of one fixed u2/t2/y8 repair | No variable geometry or numerical frontend supplied there |
+| #6629 | Compact generic routing ROM | Different source and consumer interface |
+| Wave 1 whole-tree state | Actual state traversal with compiled topology | No runtime dependency; this package treats numerical node entry only |
+
+On2026-11-10 the bounded live PR searches covered CRT in titles, encoded CRT,
+CRT topology, encoded CRT descriptors and three-epoch text. They returned
+the above actual-entry chain and the retained fixed-geometry repair/key
+chain; no encoded balanced-node numerical iterator successor was identified
+in those returned results. The canonical research-root metadata and exact
+relevant README/proof/code sections were also inspected. These searches do
+not certify absence of unpublished work or all arbitrary closed-PR diffs.
+
+No upstream PR or placeholder claim issue is created. The run's local
+allocation/checkpoint manifest is maintained by the orchestrator. Source
+bytes are read from one shared sparse clone through isolated worktrees.

research/encoded-balanced-crt-entry/PR-DESCRIPTION.md

@@ -0,0 +1,39 @@
+## Change and mathematical scope
+
+Replace the compiled node-selection and precision boundary of the actual CRT numerical entry with a **tape-driven balanced-node frontend**. The controller receives the tagged physical axis stream, counts and halves its records on tapes, maintains node and half-size cursors, constructs canonical current-coordinate inputs and the exact precision guide, and invokes the unchanged #5219 Entry and #5053 numerical producer. Runtime receives no width list, ell, start, level or precision integer. **No new kappa.**
+
+The new construction uses the unchanged #5505 bank and all its real acquisition/release checks. Eight protected zero words are physically made from the current guide. One actual numerical result remains live until its lease closes; the encoded cursor advances only after retirement. The physical interface is linear-use: consume the current lease once and relinquish its borrowed aliases. Historical Python-handle identity guards are outside that interface. All source bytes/head return, and all historical backing is counted across the complete tree. There are **78 owned roles plus the supplied source**.
+
+The source-relative theorem covers every admitted dyadic, mixed-width axis source. It pays root count manufacture, each source seek and rewind, all cursor/level transitions, current guard preparation, actual numerical work, output lifetime and cleanup. Additional frontend work is at most `C_front(M+2)^2` for a fixed input-independent constant, plus the sum of the inherited numerical Entry and bank tariffs over the d−1 nodes. Its raw local Port actions satisfy the explicit `2^16(M+2)^2` envelope; no numerical finite-control expansion constant or explicit complete-control majorant is certified. The complete numerical-node traversal is `O(p^4)` for total address width at most p; owned retained storage is `O(p^2)` plus the entire supplied source backing, including older inaccessible tails.
+
+## Sources and verification
+
+Base: `e54523517d0f03e58352e647c403a5af3e525a57`. `SOURCES.json` pins 12 scientific inputs and historically pins catalogue/selected provenance separately. #5353 already supplies the canonical projection for a compiled node; #5219 already supplies guide-driven numerical entry; #5505 already supplies varying-precision bank reuse. The new step manufactures node selection and the guide from actual source tapes and binds them to that exact consumer. No predecessor bytes are modified.
+
+```sh
+python3.12 -B scripts/research_package.py --package research/encoded-balanced-crt-entry
+```
+
+Expected marker: `PASS_ENCODED_BALANCED_CRT_ENTRY nodes=14 owned_roles=78 controls=8`.
+
+Four complete physical fixtures execute 14 actual numerical nodes, including all seven nodes of an eight-axis tree, nonzero node starts, changing half-size and precision, invalid-left disabling and padded targets. A separate exhaustive congruence oracle checks the actual lower/product words and output records. Runtime geometry/decode names are poisoned and constructors are sealed after initial fixed acquisition. Source snapshots, current guard widths, fixed backing identities and all retired historical tails are checked. The eight-axis fixture retains 3,807 cells including 97 older source-tail cells.
+
+Eight controls cover supplied host geometry, live reentry/close, a duplicate close before a later entry, stale history, rejected nondyadic shape with actual cleanup, omitted half doubling and a Boolean/integer receipt substitution. The receipt compares a raw-action envelope, not a complete-control numeric constant. Registered normal/optimized runs and the precise review status are recorded in `VALIDATION.md`; finite execution corroborates the written proof.
+
+## Limits and credit
+
+This is a **numerical-node iterator**, not the complete CRT state or payload permutation. #5496's row/target selection and h/layer ROM headers, #5625's source-update selection, three-epoch reservoir transport, canonical source production and complete binary numerical lowering remain open. The source-operation count is not a binary-TM tariff. A new constructor does not free an old bank; same-bank whole-iterator restart is not supplied by this API. The selected conditional kappa `.00363005817` and all five assumptions remain unchanged.
+
+Prepared for **@CubeJerry (Jerry Chen)** with substantial **OpenAI assistance**. Credit Ali Romman's #5353 projection, JoyboyBrian's #5219/#5053/#5505 entry/arithmetic/owner, the #4195 mixed caller, Cinna Davis's #4100 construction, Hiroki KUMAZAKI, ToikY, Tekalign, Jose Manuel, Ross Richardson, Douglas Colkitt, John-Michael Smith and all retained notices. The existing allocator, numerical CRT identity and general historical-capacity principle are not claimed anew. No human or proof-kernel acceptance is represented.
+
+## Separate review
+
+An independent worker reviewed the corrected source and proof, deriving the
+cursor invariant, canonical-input substitution, exact14P+29 guard lifecycle,
+record scan bounds and complete historical storage. A different CRT-sum
+oracle checks six actual numerical nodes. Separately,57 topology/input nodes
+use actual Entry/Bank with explicitly dummy arithmetic. Normal and optimized
+checks pass all78 role histories,16 record calibrations and three controls.
+The review required the linear-use lease scope and separation of the exact
+raw action envelope from C_front. Its proof, source pins and portable checker
+are under independent-review; no human or kernel acceptance is claimed.

research/encoded-balanced-crt-entry/PR-TITLE.txt

@@ -0,0 +1 @@
+Derive balanced CRT node selection and precision from the physical source before invoking the numerical producer (no new kappa)