Conditional κ .003125282 (+5.22e-7 over #5827): remove helper 468's frame-593 rung with rerouted reads

research/p104-helper468-rung-removal/EXPECT.json

@@ -0,0 +1,38 @@
+{
+ "interval": [
+  "78132227471/25000000000000",
+  "62505781977/20000000000000"
+ ],
+ "paid_calls": 66852,
+ "productive_fields": 13444,
+ "rank_mass_surplus": 760,
+ "events": 25998,
+ "inventory": [
+  770,
+  29,
+  31,
+  13
+ ],
+ "paid_delta": {
+  "1": -4,
+  "4": 1,
+  "6": 1,
+  "7": 1
+ },
+ "helper_delta": {
+  "1": -3,
+  "3": 1,
+  "6": 1,
+  "7": 1
+ },
+ "control_sites": {
+  "add21": 21,
+  "add27": 27
+ },
+ "sign_sites": {
+  "add21": 21,
+  "add27": 27
+ },
+ "base_commit": "05cd36e5af06d7a3790e30ce3f449fa89b853768",
+ "status": "PASS_P104_BASE5827_EDIT_KAPPA_312528200"
+}

research/p104-helper468-rung-removal/PRICING-EDITS.json

@@ -0,0 +1,18 @@
+[
+ [
+  "KAPPA, TESTED = Q('0.00312476680'), Q('0.003124766805')",
+  "KAPPA, TESTED = Q('0.003125282'), Q('0.0031252870')"
+ ],
+ [
+  "PREVIOUS = Q('0.00312418633064')",
+  "PREVIOUS = Q('0.00312476680')"
+ ],
+ [
+  "sum(targets.values()) == 13444 and sum(paid.values()) == 66870",
+  "sum(targets.values()) == 13444 and sum(paid.values()) == 66852"
+ ],
+ [
+  "'kappa_decimal':'0.00312476680','previous_pr5811':PREVIOUS",
+  "'kappa_decimal':'0.003125282','previous_pr5827':PREVIOUS"
+ ]
+]

research/p104-helper468-rung-removal/README.md

@@ -0,0 +1,65 @@
+# Conditional kappa 0.003125282: remove helper 468's frame-593 rung with rerouted reads
+
+This package edits the word of [PR #5827](https://github.com/CrocSwap/integer-mult-bounds/pull/5827)
+(aliromman, head `05cd36e5af06d7a3790e30ce3f449fa89b853768`, conditional kappa 0.00312476680) and re-certifies it with PR #5827's own chain. The complete
+conditional kappa is **0.003125282**, tested at 0.0031252870; the exact raw crossing lies in
+(0.00312528909884, 0.00312528909885) = (78132227471/25000000000000, 62505781977/20000000000000),
+which is +5.223e-07 above PR #5827's raw crossing (.00312476681507). The paid bill falls to
+**66852 calls** (PR #5827: 66,870) at an unchanged 13444 productive targets and rank-mass surplus
+760. Public inputs, owned fields, owned lifetimes and peak occupancy (770, 29, 31, 13) are unchanged;
+the word has 25998 events (retained cap 26,000).
+
+## The edit
+
+The edit removes helper 468's rung at frame 593: its itinerary 3 -> 593 -> 712 becomes 3 -> 712. Helper 703's
+chain 593 -> 859 -> 127 now runs 593 -> 712 -> 127. Three one-hop helpers are retargeted: helper 455 goes from
+3 -> 105 to 3 -> 925, helper 557 from 3 -> 105 to 105 -> 925, and helper 765 from 245 -> 127 to 263 -> 742. Their
+helper spans rise (1 to 7, 6 and 3), which moves paid ranks as listed below. Every other change is the induced
+rewrite of read and write windows: about 650 registers carry a different event sequence, with 1 MOVE fewer and
+15 ADDs more (SCALE, COPY, INIT and CLEAR counts are equal).
+
+Boundary charts changed by the edit: initial 557: 3 -> 105; initial 765: 245 -> 263; final 455: 105 -> 925; final 557: 105 -> 925; final 765: 127 -> 742. Paid-profile change vs PR #5827: rank 1: -4, rank 4: +1, rank 6: +1, rank 7: +1.
+Every public input column, every signed native map and the physical caller are re-checked in full (see Verification).
+
+`EDIT.json.gz` is the explicit edit list: boundary changes plus the new event list as copy segments of
+PR #5827's program and literal events. `construct.py` (standard library only) applies it to PR #5827's word, taken from
+the pinned archive, and regenerates `WORD.json.gz` byte for byte:
+
+```sh
+python3.12 -I -B research/p104-helper468-rung-removal/construct.py
+```
+
+## Verification
+
+```sh
+python3.12 -I -B research/p104-helper468-rung-removal/verify.py
+```
+
+`verify.py` authenticates the pinned PR #5827 package (`inputs/pr5827-source.tar`, per-file SHA256 and Git blob) and its seven-deep source
+chain, asserts that the shipped `pricing.py` equals PR #5827's `pricing.py` plus the literal edits in `PRICING-EDITS.json` (claim,
+previous claim, the paid-call count and two labels), regenerates the word, and runs PR #5827's chain unchanged: exact replay and storage
+inventory, all 799 input columns with a separate transpose, every signed native map (2,921), 208 main adapters, 1,124 helper
+normalizers, the exact physical caller, the complete invoice, both interval engines, compiler and 54 strict plus 4 functional outer
+checks, rejection controls, and the committed `RESULT.json`. `--part build` runs the authentication, regeneration and replay only.
+
+## Conditions inherited
+
+The five framework assumptions retained by the P104 line (native all-column implementation, owned erasure, uniform compiler closure,
+precision and fixed-tape costs, claim accounting) and PR #5827's own checker are inherited unchanged. Author checks are not
+independent human acceptance. The finite local word is materialized; the large uniform word and rounded multiplier are not. No merge,
+closure or selected-frontier change is requested.
+
+## Credits
+
+The base word and verifier are Ali Romman's (aliromman, PR #5827, with substantial OpenAI Codex assistance), built on PR #5811
+(Mason Hale), PR #5786 and the earlier lineage back to PR #5565 (Pierre-Loic Doulcet), with all retained source notices in the pinned
+archive. This package contributes only the edit above and its packaging.
+
+Prepared by DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.
+
+## Open next questions
+
+- Does an independent reviewer confirm the edit list, the regenerated word and the full chain? First step: run `construct.py` and
+  `verify.py` as above, then compare the changed boundary charts with PR #5827's.
+- Do other helpers with the same rung structure admit the same removal, and does any combination fit the 25998-of-26,000
+  event cap? First step: list helpers whose itinerary passes through a frame used by no other reader, then compare event counts.