Compare commits

...

1 Commits

Author SHA1 Message Date
coah
48df6cf05b decomp(eZiText): match ZiDAWGGetGraphInfo (rx14 astra max) (worker 7f33b2f1) 2026-10-05 10:31:00 +00:00
2 changed files with 458 additions and 4 deletions

View File

@@ -163,9 +163,9 @@ ziU32 ZiDAWGGetGraph(ziPtr context) {
}
ziU32 ZiDAWGGetGraphInfo(zi8DawgCtx* context, ziU8* entry, ziU16* keys) {
ziU32 graph;
ziU8 depth;
ziU32 end;
ziU32 graph;
ziS32 result;
graph = ZiDAWGGetGraph(context);
@@ -199,10 +199,10 @@ checkEntry:
}
result = graph + (ziU32)entry[4] * 0x10000 + (((ziU16)entry[5] << 8) + entry[6]);
((zi8DawgCtx*)context)->endNode =
context->endNode =
graph + (ziU32)entry[7] * 0x10000 + (((ziU16)entry[8] << 8) + entry[9]);
if (((zi8DawgCtx*)context)->endNode == graph) {
((zi8DawgCtx*)context)->endNode = 0;
if (context->endNode == graph) {
context->endNode = 0;
}
if ((((ziU16)entry[0] << 8) + entry[1]) == 0) {

View File

@@ -0,0 +1,454 @@
# rx14 allocator investigation
Branch agent/w1005/rx14, baseline 1a7f21da. Owned units: zmtkey, zi8uwd, zi8match, zidawg1, zoemdata, zkokeyp.
Read lever 22, mwdbg README, f-round/common rules, prior effort notes and current source. Existing Korean data-layout gains must be preserved. User authorizes Matching flips only after all functions in a unit are exact.
Captures and experiment artifacts live in build/rx14. The private mwdbg launcher changes only ROOT resolution and the output-directory boundary so all capture files stay in this worktree. Compiler arguments and debugger driver remain unchanged; the shared debugger lock is retained.
Acceptance: pool identical; per-function objdiff 100 and ctxdiff diffs 0; whole-unit code/data exact before linking; final clean gate and retail DOL hash; no source regressions or forbidden constructs. Every open function receives allocator diagnosis and at least three distinct compiled source trials before a tie can be reported.
Baseline libs/RVLMiddleware/eZiText/src/clib/zi8match: Zi8GetPyFinal 99.245285%. src 0xd4 base 0xd4 insns 53/53; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zi8uwd: Zi8_81480224 99.52586%. src 0x1d0 base 0x1d0 insns 116/116; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zidawg1: ZiDAWGgetCHARattribute 99.0%. src 0xf0 base 0xf0 insns 60/60; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zidawg1: ZiDAWGGetGraphInfo 99.55224%. src 0x218 base 0x218 insns 134/134; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zkokeyp: Zi8_8148302C 99.49152%. src 0xec base 0xec insns 59/59; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zkokeyp: Zi8_81483264 98.902435%. src 0xa4 base 0xa4 insns 41/41; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zkokeyp: Zi8_81483308 99.13793%. src 0xe8 base 0xe8 insns 58/58; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zkokeyp: Zi8_814833F0 99.04256%. src 0xbc base 0xbc insns 47/47; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zkokeyp: Zi8_814834AC 99.65015%. src 0x55c base 0x55c insns 343/343; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zkokeyp: Zi8GetKOcandidates 98.87145%. src 0xa74 base 0xa74 insns 669/669; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zmtkey: Zi8getKeyLayout 99.53297%. src 0x2d8 base 0x2d8 insns 182/182; pool identical.
Baseline libs/RVLMiddleware/eZiText/src/clib/zoemdata: Zi8MatchOEMdata 98.645836%. src 0x300 base 0x300 insns 192/192; pool identical.
Trial Zi8getKeyLayout / keylayout-chained-zero: 99.53297%; insns 182/182, diffs 16; regressions []; source `build/rx14/keylayout-chained-zero/zmtkey.c`.
Trial Zi8getKeyLayout / keylayout-scoped-total: 99.28571%; insns 182/182, diffs 4; regressions ['Zi8getKeyLayout']; source `build/rx14/keylayout-scoped-total/zmtkey.c`.
Trial Zi8getKeyLayout / keylayout-reuse-char-count: 95.6978%; insns 182/182, diffs 42; regressions ['Zi8getKeyLayout']; source `build/rx14/keylayout-reuse-char-count/zmtkey.c`.
Trial Zi8_81480224 / uwd-priority-payload-views: 99.18104%; insns 116/116, diffs 17; regressions ['Zi8_81480224']; source `build/rx14/uwd-priority-payload-views/zi8uwd.c`.
Debugger queue recovery: original captures remained queued for over six minutes behind a capture holding the shared lock for over 35 minutes, with 20 waiters. Stopped only this worker's two waiting captures. Private emulator copy changes only the 16-bit listening-port immediate at ELF VA 0x3cd408, 9001 to 19149; private driver uses the same port and local lock. Shared tools/processes remain unchanged. All compiler bytes, flags, source inputs, and trace hooks remain unchanged. Each successful capture must equal the independent wibo object before allocator conclusions are used. Patch hashes recorded in build/rx14/emulator-port-patch.json.
Trial Zi8GetPyFinal / pyfinal-output-scopes: 99.245285%; insns 53/53, diffs 8; regressions []; source `build/rx14/pyfinal-output-scopes/zi8match.c`.
Trial Zi8GetPyFinal / pyfinal-final-byte-local: 70.962265%; insns 63/53, diffs 52; regressions ['Zi8GetPyFinal']; source `build/rx14/pyfinal-final-byte-local/zi8match.c`.
Trial Zi8GetPyFinal / pyfinal-output-row-local: 62.28302%; insns 58/53, diffs 52; regressions ['Zi8GetPyFinal']; source `build/rx14/pyfinal-output-row-local/zi8match.c`.
Search parser audit: the shared srcsearch parser starts at the first symbol occurrence, which is a forward declaration for Zi8GetPyFinal and Zi8getKeyLayout. Its resulting body belongs to a different function. Discarded those initial searches; private srcsearch-local.py anchors a complete function definition at line start and keeps the original 24-slot lock and all mutation/scoring checks. No shared tool changed. Initial private manual parser also encountered a call to Zi8GetPyFinal; its assertion prevented any trial from compiling until corrected.
Trial Zi8_81480224 / uwd-local-word-ziPtr: 98.01724%; insns 117/116, diffs 110; regressions ['Zi8_81480224']; source `build/rx14/uwd-local-word-ziPtr/zi8uwd.c`.
Trial Zi8_81480224 / uwd-local-word-ziUserWordptr: 98.01724%; insns 117/116, diffs 110; regressions ['Zi8_81480224']; source `build/rx14/uwd-local-word-ziUserWordptr/zi8uwd.c`.
Trial ZiDAWGGetGraphInfo / graphinfo-direct-context-priority: 100.0%; insns 134/134, diffs 0; regressions []; source `build/rx14/graphinfo-direct-context-priority/zidawg1.c`.
Trial ZiDAWGGetGraphInfo / graphinfo-direct-context-end-first: 99.70149%; insns 134/134, diffs 7; regressions []; source `build/rx14/graphinfo-direct-context-end-first/zidawg1.c`.
Trial ZiDAWGGetGraphInfo / graphinfo-scoped-result: 100.0%; insns 134/134, diffs 0; regressions []; source `build/rx14/graphinfo-scoped-result/zidawg1.c`.
Trial ZiDAWGgetCHARattribute / charattr-conditional-decode: 95.583336%; insns 58/60, diffs 50; regressions ['ZiDAWGgetCHARattribute']; source `build/rx14/charattr-conditional-decode/zidawg1.c`.
Trial ZiDAWGgetCHARattribute / charattr-combined-attribute: 98.666664%; insns 60/60, diffs 15; regressions ['ZiDAWGgetCHARattribute']; source `build/rx14/charattr-combined-attribute/zidawg1.c`.
Trial ZiDAWGgetCHARattribute / charattr-condition-at-declaration: 99.0%; insns 60/60, diffs 11; regressions []; source `build/rx14/charattr-condition-at-declaration/zidawg1.c`.
Trial ZiDAWGgetCHARattribute / charattr-key-test-as-unsigned: 98.0%; insns 60/60, diffs 13; regressions ['ZiDAWGgetCHARattribute']; source `build/rx14/charattr-key-test-as-unsigned/zidawg1.c`.
Trial ZiDAWGgetCHARattribute / charattr-key-index-unsigned: 99.0%; insns 60/60, diffs 12; regressions []; source `build/rx14/charattr-key-index-unsigned/zidawg1.c`.
Trial ZiDAWGgetCHARattribute / charattr-key-dereference-fields: 99.0%; insns 60/60, diffs 12; regressions []; source `build/rx14/charattr-key-dereference-fields/zidawg1.c`.
Trial Zi8_81483264 / Zi8_81483264-postincrement-access: 92.560974%; insns 42/41, diffs 22; regressions ['Zi8_81483264']; source `build/rx14/Zi8_81483264-postincrement-access/zkokeyp.c`.
Trial Zi8_81483264 / Zi8_81483264-unsigned-index: 98.902435%; insns 41/41, diffs 8; regressions []; source `build/rx14/Zi8_81483264-unsigned-index/zkokeyp.c`.
Trial Zi8_81483264 / Zi8_81483264-unsigned-bound: 98.902435%; insns 41/41, diffs 8; regressions []; source `build/rx14/Zi8_81483264-unsigned-bound/zkokeyp.c`.
Trial Zi8_81483308 / Zi8_81483308-postincrement-access: 94.74138%; insns 59/58, diffs 34; regressions ['Zi8_81483308']; source `build/rx14/Zi8_81483308-postincrement-access/zkokeyp.c`.
Trial Zi8_81483308 / Zi8_81483308-unsigned-index: 99.13793%; insns 58/58, diffs 9; regressions []; source `build/rx14/Zi8_81483308-unsigned-index/zkokeyp.c`.
Trial Zi8_81483308 / Zi8_81483308-unsigned-bound: 99.13793%; insns 58/58, diffs 9; regressions []; source `build/rx14/Zi8_81483308-unsigned-bound/zkokeyp.c`.
Trial Zi8_814833F0 / Zi8_814833F0-postincrement-access: 93.510635%; insns 48/47, diffs 22; regressions ['Zi8_814833F0']; source `build/rx14/Zi8_814833F0-postincrement-access/zkokeyp.c`.
Trial Zi8_814833F0 / Zi8_814833F0-unsigned-index: 99.04256%; insns 47/47, diffs 8; regressions []; source `build/rx14/Zi8_814833F0-unsigned-index/zkokeyp.c`.
Trial Zi8_814833F0 / Zi8_814833F0-unsigned-bound: 99.04256%; insns 47/47, diffs 8; regressions []; source `build/rx14/Zi8_814833F0-unsigned-bound/zkokeyp.c`.
## ZiDAWGGetGraphInfo exact allocator reconstruction
Target map: entry r31, keys r30, context r29, depth r28, end r27, graph r26, result r25. Baseline early GPR assignment weights were entry 54, keys 12, graph/depth/end 5 each, result/context 4 each. Parameters are considered first, then locals in reverse declaration order; the scan replaces its candidate on equal weight, so the earlier declared local wins a tie. This made context last at r25 and graph first among the weight-5 locals at r29.
The saved-register choices already appear in backend-00 initial PCode, before the interference graph. Graph coloring only assigns scratch registers here; degree and coalescing do not choose these saved homes. Private preallocation hooks at 0x46aeb0, 0x46b1ba and 0x46b1ce read actual parameter/local weights and assignments. These addresses were established from the pinned compiler disassembly, with RegInfo returned by 0x579790.
Removing three redundant receiver casts raises context weight 4 to 7. Declare depth, end, graph in that order so the weight-5 tie becomes depth r28, end r27, graph r26. The confirmation capture gives exactly this order, with result r25. Both 134-instruction streams are identical, objdiff 100%, other functions unchanged. Captured object is byte-identical to independent wibo compilation of the same snapshot. Evidence: build/rx14/graphinfo-direct-context-priority-capture/prealloc.json and build/rx14/graphinfo-direct-context-priority/metrics.json.
Trial Zi8MatchOEMdata / oem-fused-retry-index: 98.645836%; insns 192/192, diffs 47; regressions []; source `build/rx14/oem-fused-retry-index/zoemdata.c`.
Trial Zi8MatchOEMdata / oem-fused-stored-index: 98.645836%; insns 192/192, diffs 47; regressions []; source `build/rx14/oem-fused-stored-index/zoemdata.c`.
Trial Zi8MatchOEMdata / oem-fused-both-indices: 99.166664%; insns 192/192, diffs 29; regressions []; source `build/rx14/oem-fused-both-indices/zoemdata.c`.
Trial Zi8MatchOEMdata / oem-natural-signed-expressions: 98.645836%; insns 192/192, diffs 47; regressions []; source `build/rx14/oem-natural-signed-expressions/zoemdata.c`.
Trial Zi8_8148302C / keylookup-assignment-test: 45.0%; insns 59/59, diffs 24; regressions ['Zi8_8148302C']; source `build/rx14/keylookup-assignment-test/zkokeyp.c`.
Trial Zi8_8148302C / keylookup-count-in-loop: 97.79661%; insns 60/59, diffs 45; regressions ['Zi8_8148302C']; source `build/rx14/keylookup-count-in-loop/zkokeyp.c`.
Trial Zi8_8148302C / keylookup-scalar-entry: 98.47458%; insns 59/59, diffs 7; regressions ['Zi8_8148302C']; source `build/rx14/keylookup-scalar-entry/zkokeyp.c`.
Trial Zi8MatchOEMdata / oem-fused-initial-index: 99.166664%; insns 192/192, diffs 29; regressions []; source `build/rx14/oem-fused-initial-index/zoemdata.c`.
Trial Zi8MatchOEMdata / oem-fallback-explicit-init: 99.166664%; insns 192/192, diffs 29; regressions []; source `build/rx14/oem-fallback-explicit-init/zoemdata.c`.
Trial ZiDAWGgetCHARattribute / charattr-key-conditional-assignments: 95.083336%; insns 62/60, diffs 52; regressions ['ZiDAWGgetCHARattribute']; source `build/rx14/charattr-key-conditional-assignments/zidawg1.c`.
Trial Zi8GetKOcandidates / kocandidates-direct-table: 97.29447%; insns 661/669, diffs 580; regressions ['Zi8GetKOcandidates']; source `build/rx14/kocandidates-direct-table/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-chained-key-clear: 98.146484%; insns 671/669, diffs 637; regressions ['Zi8GetKOcandidates']; source `build/rx14/kocandidates-chained-key-clear/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-table-and-clear: 97.44395%; insns 662/669, diffs 635; regressions ['Zi8GetKOcandidates']; source `build/rx14/kocandidates-table-and-clear/zkokeyp.c`.
Trial Zi8_814834AC / kofilter-byte-record-address: 99.53353%; insns 343/343, diffs 27; regressions ['Zi8_814834AC']; source `build/rx14/kofilter-byte-record-address/zkokeyp.c`.
Trial Zi8_814834AC / kofilter-member-index: 99.591835%; insns 343/343, diffs 25; regressions ['Zi8_814834AC']; source `build/rx14/kofilter-member-index/zkokeyp.c`.
Trial Zi8_814834AC / kofilter-candidate-scoped: 96.09038%; insns 346/343, diffs 338; regressions ['Zi8_814834AC']; source `build/rx14/kofilter-candidate-scoped/zkokeyp.c`.
Trial Zi8_8148302C / keylookup-count-initializer: 99.49152%; insns 59/59, diffs 6; regressions []; source `build/rx14/keylookup-count-initializer/zkokeyp.c`.
Trial Zi8_8148302C / keylookup-loop-value-initializer: 89.81356%; insns 62/59, diffs 54; regressions ['Zi8_8148302C']; source `build/rx14/keylookup-loop-value-initializer/zkokeyp.c`.
Trial Zi8getKeyLayout / keylayout-count-comma-test: 99.36813%; insns 182/182, diffs 22; regressions ['Zi8getKeyLayout']; source `build/rx14/keylayout-count-comma-test/zmtkey.c`.
Trial Zi8getKeyLayout / keylayout-total-explicit-add: 98.98351%; insns 183/182, diffs 71; regressions ['Zi8getKeyLayout']; source `build/rx14/keylayout-total-explicit-add/zmtkey.c`.
Trial Zi8getKeyLayout / keylayout-total-reversed-add: 98.98351%; insns 183/182, diffs 71; regressions ['Zi8getKeyLayout']; source `build/rx14/keylayout-total-reversed-add/zmtkey.c`.
Trial Zi8_8148302C / keylookup-reversed-assignment-test: 45.0%; insns 59/59, diffs 24; regressions ['Zi8_8148302C']; source `build/rx14/keylookup-reversed-assignment-test/zkokeyp.c`.
Trial Zi8_8148302C / keylookup-unsigned-count-and-value: 45.0%; insns 59/59, diffs 24; regressions ['Zi8_8148302C']; source `build/rx14/keylookup-unsigned-count-and-value/zkokeyp.c`.
Trial Zi8_81483308 / scratch-count-update-condition: 97.413795%; insns 59/58, diffs 23; regressions ['Zi8_81483308']; source `build/rx14/scratch-count-update-condition/zkokeyp.c`.
Trial Zi8_81483308 / scratch-expanded-count-update: 98.27586%; insns 59/58, diffs 16; regressions ['Zi8_81483308']; source `build/rx14/scratch-expanded-count-update/zkokeyp.c`.
Trial Zi8getKeyLayout / zmtkey-typed-direct-work: 99.53297%; insns 182/182, diffs 16; regressions []; source `build/rx14/zmtkey-typed-direct-work/zmtkey.c`.
Trial Zi8MatchOEMdata / zoemdata-typed-direct-work: 98.203125%; insns 192/192, diffs 62; regressions ['Zi8MatchOEMdata']; source `build/rx14/zoemdata-typed-direct-work/zoemdata.c`.
Trial Zi8MatchOEMdata / oem-expanded-capacity: 99.114586%; insns 192/192, diffs 30; regressions []; source `build/rx14/oem-expanded-capacity/zoemdata.c`.
Trial Zi8GetKOcandidates / kocandidates-native-key-address: 98.87145%; insns 669/669, diffs 83; regressions []; source `build/rx14/kocandidates-native-key-address/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-address-clear-byte-pair: 98.87145%; insns 669/669, diffs 81; regressions []; source `build/rx14/kocandidates-address-clear-byte-pair/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-address-decode-width: 98.87145%; insns 669/669, diffs 83; regressions []; source `build/rx14/kocandidates-address-decode-width/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-stride-last: 98.82661%; insns 669/669, diffs 84; regressions ['Zi8GetKOcandidates']; source `build/rx14/kocandidates-stride-last/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-base-middle: 96.06129%; insns 670/669, diffs 616; regressions ['Zi8GetKOcandidates']; source `build/rx14/kocandidates-base-middle/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-group-right: 98.82661%; insns 669/669, diffs 84; regressions ['Zi8GetKOcandidates']; source `build/rx14/kocandidates-group-right/zkokeyp.c`.
Trial Zi8GetKOcandidates / kocandidates-base-first: 98.87145%; insns 669/669, diffs 83; regressions []; source `build/rx14/kocandidates-base-first/zkokeyp.c`.
## Per-function allocator evidence
The baseline and confirmation objects were independently rebuilt with wibo using each capture's exact source and compiler arguments. All 36 completed objects compare byte-for-byte equal. Graph rewrite validation reports no mismatches. Saved homes are assigned by the earlier weighted scan, not by graph simplify/coloring; graph cost zero must not be mistaken for zero source weight.
### Zi8GetPyFinal
Target output loads require row stride r7 and table high-address r6. Baseline scratch r52, the second row stride, simplifies at degree 3 and colors before r46. It receives r6 from legal mask 0x1fc0. r46 is the shared lis temporary for both output loads, has seven interference neighbors, and is limited to 0x1f80 after r52, so it receives r7. The first stride r45 then receives r6 from 0x1f40. The coalescing log is empty. Explicit output scopes preserve all three choices; a local result or row pointer adds instructions.
Baseline assignment order: index weight 9 -> r31, row weight 8 -> r30.
Confirmation assignment order: index weight 9 -> r31, row weight 8 -> r30.
Evidence: `build/rx14/Zi8GetPyFinal-base/`, `build/rx14/Zi8GetPyFinal-prebase/`, `build/rx14/pyfinal-output-scopes-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8_81480224
Target homes are work r31, current r30, added r29, word r28, position r27, candidate r26, length r25, previous r24. Baseline word, candidate and position all weigh 8. Local candidate wins the first tie over position and parameter word, giving r28/r27/r26 in the wrong value order. Target also compares word priority first and adds position before each record base. The priority/payload-view trial changes the expression tree but leaves 17 differences. Rebinding the input record introduces a stack home and one extra instruction; rejected.
Baseline assignment order: __zi8_work_data weight 12 -> r31, current weight 11 -> r30, added weight 10 -> r29, candidate weight 8 -> r28, position weight 8 -> r27, word weight 8 -> r26, length weight 7 -> r25, previous weight 6 -> r24.
Confirmation assignment order: __zi8_work_data weight 12 -> r31, current weight 11 -> r30, added weight 10 -> r29, position weight 8 -> r28, length weight 7 -> r27, candidate weight 7 -> r26, word weight 7 -> r25, previous weight 6 -> r24.
Evidence: `build/rx14/Zi8_81480224-base/`, `build/rx14/Zi8_81480224-prebase/`, `build/rx14/uwd-priority-payload-views-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### ZiDAWGgetCHARattribute
Target context/key homes are r31/r30; baseline key and context both weigh 8 and the local key wins over the parameter. The conditional initializer reduces key to weight 7 and gives context r31, key r30, attribute r29, but introduces a weight-3 conditional temporary at r28. That temporary displaces node/work to r27/r26 and is not coalesced with the preassigned key. All 60 instructions remain, but 11 differ. The baseline graph never decides the saved homes.
Baseline assignment order: key weight 8 -> r31, context weight 8 -> r30, attribute weight 7 -> r29, node weight 3 -> r28, __zi8_work_data weight 2 -> r27.
Confirmation assignment order: context weight 8 -> r31, key weight 7 -> r30, attribute weight 7 -> r29, @105 weight 3 -> r28, node weight 3 -> r27, __zi8_work_data weight 2 -> r26.
Evidence: `build/rx14/ZiDAWGgetCHARattribute-base/`, `build/rx14/ZiDAWGgetCHARattribute-prebase/`, `build/rx14/charattr-condition-at-declaration-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8_8148302C
Target i/table/work/count/value homes are r31/r30/r29/r28/r27. Baseline value, count and work all weigh 3; locals win before the work parameter. Folding value assignment into the comparison reduces value to weight 2 and assigns it target r27, but count still wins r29 and work gets r28. The folded comparison also loads the input key before decoding the value. The 59-instruction trial is rejected.
Baseline assignment order: i weight 10 -> r31, table weight 4 -> r30, value weight 3 -> r29, count weight 3 -> r28, __zi8_work_data weight 3 -> r27.
Confirmation assignment order: i weight 10 -> r31, table weight 4 -> r30, count weight 3 -> r29, __zi8_work_data weight 3 -> r28, value weight 2 -> r27.
Evidence: `build/rx14/Zi8_8148302C-base/`, `build/rx14/Zi8_8148302C-prebase/`, `build/rx14/keylookup-assignment-test-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8_81483264
Target param/i/work homes are r31/r30/r29. Param and i each weigh 6; local i wins r31. An explicit unsigned bound preserves the weight and the same eight differences. Postincrement in the actual scratch access reduces i to weight 5 and gets the saved homes right, but requires an old-index copy, producing 42 rather than 41 instructions.
Baseline assignment order: i weight 6 -> r31, param weight 6 -> r30, __zi8_work_data weight 2 -> r29.
Confirmation assignment order: i weight 6 -> r31, param weight 6 -> r30, __zi8_work_data weight 2 -> r29.
Evidence: `build/rx14/Zi8_81483264-base/`, `build/rx14/Zi8_81483264-prebase/`, `build/rx14/Zi8_81483264-unsigned-bound-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8_81483308
Target param/count/i/work/key homes are r31/r30/r29/r28/r27. Count and i each weigh 6; local i wins before parameter count. Moving the increment into the scratch read reduces i to weight 5, and the confirmation dump gives all five target saved homes. It needs 59 rather than 58 instructions to preserve the old index. Compound/expanded count updates likewise add an instruction.
Baseline assignment order: param weight 10 -> r31, i weight 6 -> r30, count weight 6 -> r29, __zi8_work_data weight 2 -> r28, key weight 2 -> r27.
Confirmation assignment order: param weight 10 -> r31, count weight 6 -> r30, i weight 5 -> r29, __zi8_work_data weight 2 -> r28, key weight 2 -> r27.
Evidence: `build/rx14/Zi8_81483308-base/`, `build/rx14/Zi8_81483308-prebase/`, `build/rx14/Zi8_81483308-postincrement-access-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8_814833F0
Target param/i/work homes are r31/r30/r29. Param and i each weigh 6, so local i takes r31. Postincrement inside the scratch read gives the target ranking by reducing i weight, but adds the old-index copy, leaving 48 rather than 47 instructions. Width views preserve the original eight register differences.
Baseline assignment order: i weight 6 -> r31, param weight 6 -> r30, __zi8_work_data weight 2 -> r29.
Confirmation assignment order: param weight 6 -> r31, i weight 5 -> r30, __zi8_work_data weight 2 -> r29.
Evidence: `build/rx14/Zi8_814833F0-base/`, `build/rx14/Zi8_814833F0-prebase/`, `build/rx14/Zi8_814833F0-postincrement-access-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8_814834AC
Target param/count/remaining/i/j/work/candidate/matched/inserted homes are r31/r30/r29/r28/r27/r26/r25/r24/r23. Work and candidate both weigh 10; local candidate wins r26 and work gets r25. Two additional differences reverse the operands of record-address adds. Byte-record/member-index forms preserve this tie and worsen scratch ordering. Scoped candidate initialization introduces a frame pointer and three extra instructions. Existing 180/180 data remains preserved by rejecting these variants.
Baseline assignment order: param weight 47 -> r31, count weight 32 -> r30, remaining weight 28 -> r29, i weight 25 -> r28, j weight 11 -> r27, candidate weight 10 -> r26, __zi8_work_data weight 10 -> r25, matched weight 9 -> r24, inserted weight 4 -> r23.
Confirmation assignment order: param weight 47 -> r31, count weight 32 -> r30, remaining weight 28 -> r29, i weight 27 -> r28, j weight 11 -> r27, candidate weight 10 -> r26, __zi8_work_data weight 10 -> r25, matched weight 9 -> r24, inserted weight 4 -> r23.
Evidence: `build/rx14/Zi8_814834AC-base/`, `build/rx14/Zi8_814834AC-prebase/`, `build/rx14/kofilter-member-index-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8GetKOcandidates
All saved homes already match: param r31, options r30, wordNode r29, keyTable r28, work r27. The differing graph is in scratch values and expression association. For the first key row, source PCode r57 is index<<3, r58 reloads index, r59 is keyTable+r57, and r60 is r58+r59; target adds stride and reloaded index before keyTable. The two initialization pairs also place li/stb differently. Direct record indexing shortens eight address instructions; chained key initialization adds two. Combining them changes 669 to 662 instructions. Native address storage leaves 669/669 and the same 83 differences.
Baseline assignment order: param weight 110 -> r31, options weight 40 -> r30, wordNode weight 28 -> r29, keyTable weight 14 -> r28, __zi8_work_data weight 14 -> r27.
Confirmation assignment order: param weight 110 -> r31, options weight 40 -> r30, wordNode weight 28 -> r29, keyTable weight 14 -> r28, __zi8_work_data weight 14 -> r27.
Evidence: `build/rx14/Zi8GetKOcandidates-base/`, `build/rx14/Zi8GetKOcandidates-prebase/`, `build/rx14/kocandidates-table-and-clear-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8getKeyLayout
Target language/tableCount homes are r27/r26. Baseline tableCount weighs 8 and language 7, so tableCount wins r27. Chaining the zero initialization leaves the weights and all 16 differences unchanged. Splitting the running offset from the API table count gives language the target home, but the API count is no longer selected for a saved register and is stored on the stack twice; only four instruction differences remain, with an objdiff regression. The variant is rejected.
Baseline assignment order: keyIndex weight 21 -> r31, dataAddress weight 15 -> r30, __zi8_work_data weight 12 -> r29, chars weight 10 -> r28, tableCount weight 8 -> r27, language weight 7 -> r26, keyChars weight 6 -> r25.
Confirmation assignment order: keyIndex weight 21 -> r31, dataAddress weight 15 -> r30, __zi8_work_data weight 12 -> r29, chars weight 10 -> r28, tableCount weight 7 -> r27, language weight 7 -> r26, keyChars weight 6 -> r25.
Evidence: `build/rx14/Zi8getKeyLayout-base/`, `build/rx14/Zi8getKeyLayout-prebase/`, `build/rx14/keylayout-chained-zero-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
### Zi8MatchOEMdata
Target word/position/work/pattern/length/index/complete/capacity/fallback homes are r31/r30/r29/r28/r27/r26/r25/r24/r23. Baseline index weighs 13 and outranks work/pattern at 12 and length at 10. Fusing both real index updates into their consumers lowers index weight to 11, fixing work/pattern but leaving index ahead of length. Capacity and fallback each weigh 4, so local fallback still wins over the parameter. The 192-instruction trial rises to 99.166664% with 29 differences, but yields no exact function and is rejected.
Baseline assignment order: word weight 25 -> r31, position weight 18 -> r30, index weight 13 -> r29, __zi8_work_data weight 12 -> r28, pattern weight 12 -> r27, length weight 10 -> r26, complete weight 6 -> r25, fallback weight 4 -> r24, capacity weight 4 -> r23, language weight 3 -> r22, continuation weight 2 -> r21.
Confirmation assignment order: word weight 25 -> r31, position weight 18 -> r30, __zi8_work_data weight 12 -> r29, pattern weight 12 -> r28, index weight 11 -> r27, length weight 10 -> r26, complete weight 6 -> r25, fallback weight 4 -> r24, capacity weight 4 -> r23, language weight 3 -> r22, continuation weight 2 -> r21.
Evidence: `build/rx14/Zi8MatchOEMdata-base/`, `build/rx14/Zi8MatchOEMdata-prebase/`, `build/rx14/oem-fused-both-indices-capture/`. Candidate source was compiled from the named private snapshot and discarded after measurement.
## Intermediate gate
Quick gate passed all six owned units, DOL SHA1 26116613f624061ba99c8d1a299aaa6efa85670d, zero regressions, zero forbidden patterns, zero readability warnings. Only ZiDAWGGetGraphInfo is retained. No unit is all-exact, so configure.py stays unchanged. Clean final gate remains pending until source searches finish.
Prior fuzzy branches checked: `git diff origin/main...agent/w1004/s1` through `s8` contain no diff for any owned source unit; their relevant state is already represented by the current baseline. Private debugger launchers now also set GDB cwd to this worktree. They read the installed shared debugger/runtime but write only beneath build/rx14.
Trial Zi8GetPyFinal / pyfinal-matched-branch: 99.245285%; insns 53/53, diffs 8; regressions []; source `build/rx14/pyfinal-matched-branch/zi8match.c`.
OEM follow-up searches use the best same-count manual source, `oem-fused-both-indices` (99.166664%, 192 instructions), seeds 349,449,549. Original baseline searches remain separate evidence. OEM seed 149 had already started when checked, so it was left running; no process was stopped. Shared 24-slot locking is unchanged.
The final PyFinal branch-shape trial encloses the output stores in `if (index >= 4)` and removes the forward goto. It leaves 53 instructions and the same eight r6/r7 differences. Extra diagnostic captures at AST use-weight site 0x57d9b1 observed no events after CodeGen entry, so they do not add evidence about the earlier use-count traversal. Their three objects are independently byte-identical to wibo; no source change follows from them. The saved-home evidence remains the direct preallocation lists and selection events.
Trial Zi8_81483264 / Zi8_81483264-wide-compound-step: 98.902435%; insns 41/41, diffs 8; regressions []; source `build/rx14/Zi8_81483264-wide-compound-step/zkokeyp.c`.
Trial Zi8_81483264 / Zi8_81483264-wide-explicit-step: 98.902435%; insns 41/41, diffs 8; regressions []; source `build/rx14/Zi8_81483264-wide-explicit-step/zkokeyp.c`.
Trial Zi8_81483308 / Zi8_81483308-wide-compound-step: 99.13793%; insns 58/58, diffs 9; regressions []; source `build/rx14/Zi8_81483308-wide-compound-step/zkokeyp.c`.
Trial Zi8_81483308 / Zi8_81483308-wide-explicit-step: 99.13793%; insns 58/58, diffs 9; regressions []; source `build/rx14/Zi8_81483308-wide-explicit-step/zkokeyp.c`.
Trial Zi8_814833F0 / Zi8_814833F0-wide-compound-step: 99.04256%; insns 47/47, diffs 8; regressions []; source `build/rx14/Zi8_814833F0-wide-compound-step/zkokeyp.c`.
Trial Zi8_814833F0 / Zi8_814833F0-wide-explicit-step: 99.04256%; insns 47/47, diffs 8; regressions []; source `build/rx14/Zi8_814833F0-wide-explicit-step/zkokeyp.c`.
## Compiled manual coverage audit
Counts below use distinct full-source SHA256 values and only successful compilations measured by objdiff and instruction diff. Failed compiler probes and parser failures are excluded.
- Zi8GetPyFinal: 4 distinct compiled sources.
- Zi8_81480224: 3 distinct compiled sources.
- ZiDAWGgetCHARattribute: 7 distinct compiled sources.
- ZiDAWGGetGraphInfo: 3 distinct compiled sources.
- Zi8_8148302C: 7 distinct compiled sources.
- Zi8_81483264: 5 distinct compiled sources.
- Zi8_81483308: 7 distinct compiled sources.
- Zi8_814833F0: 5 distinct compiled sources.
- Zi8_814834AC: 3 distinct compiled sources.
- Zi8GetKOcandidates: 10 distinct compiled sources.
- Zi8getKeyLayout: 7 distinct compiled sources.
- Zi8MatchOEMdata: 8 distinct compiled sources.
Wide 32-bit scratch-loop cursors with byte views at the bound/access and compound or explicit updates still emit the same register swaps. The byte maximum bounds every iteration to 0..254, so this widening does not change the loops’ behavior; neither spelling yields an exact gain.
Local source commit `ca21fa48` records the GraphInfo exact gain after the six-unit quick gate. The final clean gate and completed-search evidence are still pending.
KO scratch-color detail: the first row address colors v60, v59, v58, v57, v56 at zero-based priority positions 189..193. Their simplify degrees are 6, 5, 6, 6, 7; their available masks are 0x1ff8, 0x1ff9, 0x1ff8, 0x1ff1, 0x1ff8, producing r3, r0, r3, r0, r3. None of v56..v60 is coalesced. Earlier call-result copies v43/v44/v45/v54 coalesce into r3, but do not change the earlier PCode association `(keyTable + stride) + index`. The initialization values v35..v38 all color r0; the store-pair ordering divergence already exists before allocation. Thus those KO differences require expression/store-order reconstruction as well as coloring, while the saved-register homes are already correct.
Trial ZiDAWGGetGraphInfo / graphinfo-ablation-direct-context-only: 99.55224%; insns 134/134, diffs 11; regressions []; source `build/rx14/graphinfo-ablation-direct-context-only/zidawg1.c`.
Trial ZiDAWGGetGraphInfo / graphinfo-ablation-declaration-only: 99.25373%; insns 134/134, diffs 19; regressions ['ZiDAWGGetGraphInfo']; source `build/rx14/graphinfo-ablation-declaration-only/zidawg1.c`.
GraphInfo ablations are diagnostic confirmation, not new search coverage: direct context fields alone yield 11 differences; declaration order alone yields 19. Both changes are needed for zero differences. The direct-field-only source had been tested historically; this repeat checks the specific retained combination against this run’s baseline.
Queue scheduling: 19 own waiting search processes had empty logs and no output directory after prolonged shared-slot contention. Restarted only those jobs with staggered launches; arguments, seeds, 1200-second budgets, and the shared 24-slot lock remain unchanged. Jobs that had started producing output were left running. Manifest: build/rx14/requeued-jobs.json.
Trial Zi8MatchOEMdata / oem-capacity-before-guard: 98.359375%; insns 192/192, diffs 22; regressions ['Zi8MatchOEMdata']; source `build/rx14/oem-capacity-before-guard/zoemdata.c`.
Trial Zi8MatchOEMdata / oem-capacity-compound-condition: 99.114586%; insns 192/192, diffs 30; regressions []; source `build/rx14/oem-capacity-compound-condition/zoemdata.c`.
OEM capacity follow-up: moving the necessary capacity decrement into its own statement produces the target capacity/fallback homes, but moves the decrement before the length load and leaves length/index exchanged. It stays 192 instructions with 22 differences and regresses to 98.359375%, so it is discarded. Inline compound decrement stays at 192 instructions with 30 differences (99.114586%), also discarded. Confirmation order: word weight 25 -> r31, position weight 18 -> r30, __zi8_work_data weight 12 -> r29, pattern weight 12 -> r28, index weight 11 -> r27, length weight 10 -> r26, complete weight 6 -> r25, capacity weight 5 -> r24, fallback weight 4 -> r23, language weight 3 -> r22, continuation weight 2 -> r21. This fortieth capture also matches the independent wibo object byte for byte.
## Completed source searches and final verification
Evidence root: `/mnt/drive2/projects/wii-ipl-workers/data-d2/build/rx14`. All requested remaining functions completed seeds 49, 149, and 249 with `--seconds 1200 --mode anneal`, through the unchanged global 24-slot limiter. OEM additionally completed three seeds from its best same-count manual trial. Counts below are search iterations, including rejected or failed compilations, not distinct successful sources. No search supplied another exact function.
| Function | Seeds and iterations |
| --- | --- |
| Zi8GetPyFinal | 49: 2470, 149: 2446, 249: 2423 |
| Zi8_81480224 | 49: 2489, 149: 2068, 249: 2092 |
| ZiDAWGgetCHARattribute | 49: 2554, 149: 3273, 249: 2206 |
| Zi8_8148302C | 49: 2510, 149: 2106, 249: 2085 |
| Zi8_81483264 | 49: 2337, 149: 2181, 249: 2078 |
| Zi8_81483308 | 49: 2098, 149: 2119, 249: 2088 |
| Zi8_814833F0 | 49: 2439, 149: 2107, 249: 2065 |
| Zi8_814834AC | 49: 2157, 149: 2108, 249: 2012 |
| Zi8GetKOcandidates | 49: 2030, 149: 2078, 249: 2056 |
| Zi8getKeyLayout | 49: 2398, 149: 2182, 249: 2164 |
| Zi8MatchOEMdata | 49: 2517, 149: 2389, 249: 2154 |
| Zi8MatchOEMdata | 349: 2114 |
| Zi8MatchOEMdata | 449: 2117 |
| Zi8MatchOEMdata | 549: 2427 |
The last queued CHARattribute seed was relaunched immediately after one of this worker’s active searches exited. It used the same source-search script, arguments, and global slot lock; no other worker process or shared tool was changed.
40 compiler-debugger captures were independently rebuilt with wibo and produced identical object bytes. Each original open function has at least three distinct measured manual trials. The only retained source change is the GraphInfo cast/declaration correction in local commit `ca21fa48`. No unit reached all-exact, so no Matching flag was changed.
| Unit | Exact functions before -> after | Code bytes before -> after | Data bytes before -> after |
| --- | --- | --- | --- |
| zi8match | 9 -> 9/10 | 8044 -> 8044/8256 | 728 -> 728/728 |
| zi8uwd | 3 -> 3/4 | 2192 -> 2192/2656 | 80 -> 80/80 |
| zidawg1 | 4 -> 5/6 | 888 -> 1424/1664 | 80 -> 80/80 |
| zkokeyp | 1 -> 1/7 | 332 -> 332/5200 | 180 -> 180/180 |
| zmtkey | 3 -> 3/4 | 1488 -> 1488/2216 | 60 -> 60/60 |
| zoemdata | 2 -> 2/3 | 208 -> 208/976 | 60 -> 60/60 |
Final clean gate, including its unit lines:
```text
full build: ok
main.dol sha1: 26116613f624061ba99c8d1a299aaa6efa85670d
[libs/RVLMiddleware/eZiText/src/clib/zi8match] pool: IDENTICAL
[libs/RVLMiddleware/eZiText/src/clib/zi8match] objdiff: code 8044/8256 data 728/728 functions 9/10 fuzzy 99.9806 linked code 0
[libs/RVLMiddleware/eZiText/src/clib/zi8match] instruction-exact functions: 9/10
[libs/RVLMiddleware/eZiText/src/clib/zi8match] section .rodata size 528 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zi8match] section .text size 8256 match 99.98062
[libs/RVLMiddleware/eZiText/src/clib/zi8match] section extab size 80 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zi8match] section extabindex size 120 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zi8match] below 100: Zi8GetPyFinal 99.245285
[libs/RVLMiddleware/eZiText/src/clib/zi8match] baseline: code 8044/8256 data 728 functions 9 fuzzy 99.9806
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] pool: IDENTICAL
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] objdiff: code 2192/2656 data 80/80 functions 3/4 fuzzy 99.9172 linked code 0
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] instruction-exact functions: 3/4
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] section .text size 2656 match 99.91717
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] section extab size 32 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] section extabindex size 48 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] below 100: Zi8_81480224 99.52586
[libs/RVLMiddleware/eZiText/src/clib/zi8uwd] baseline: code 2192/2656 data 80 functions 3 fuzzy 99.9172
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] pool: IDENTICAL
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] objdiff: code 1424/1664 data 80/80 functions 5/6 fuzzy 99.8558 linked code 0
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] instruction-exact functions: 5/6
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] section .text size 1664 match 99.85577
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] section extab size 32 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] section extabindex size 48 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] below 100: ZiDAWGgetCHARattribute 99.0
[libs/RVLMiddleware/eZiText/src/clib/zidawg1] baseline: code 888/1664 data 80 functions 4 fuzzy 99.7115
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] pool: IDENTICAL
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] objdiff: code 332/5200 data 180/180 functions 1/7 fuzzy 99.1962 linked code 0
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] instruction-exact functions: 1/7
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] section .data size 40 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] section .text size 5200 match 99.19615
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] section extab size 56 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] section extabindex size 84 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] below 100: Zi8_8148302C 99.49152
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] below 100: Zi8_81483264 98.902435
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] below 100: Zi8_81483308 99.13793
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] below 100: Zi8_814833F0 99.04256
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] below 100: Zi8_814834AC 99.65015
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] below 100: Zi8GetKOcandidates 98.87145
[libs/RVLMiddleware/eZiText/src/clib/zkokeyp] baseline: code 332/5200 data 180 functions 1 fuzzy 99.1962
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] pool: IDENTICAL
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] objdiff: code 1488/2216 data 60/60 functions 3/4 fuzzy 99.8466 linked code 0
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] instruction-exact functions: 3/4
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] section .text size 2216 match 99.84657
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] section extab size 24 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] section extabindex size 36 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] below 100: Zi8getKeyLayout 99.53297
[libs/RVLMiddleware/eZiText/src/clib/zmtkey] baseline: code 1488/2216 data 60 functions 3 fuzzy 99.8466
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] pool: IDENTICAL
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] objdiff: code 208/976 data 60/60 functions 2/3 fuzzy 98.9344 linked code 0
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] instruction-exact functions: 2/3
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] section .text size 976 match 98.934425
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] section extab size 24 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] section extabindex size 36 match 100.0
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] below 100: Zi8MatchOEMdata 98.645836
[libs/RVLMiddleware/eZiText/src/clib/zoemdata] baseline: code 208/976 data 60 functions 2 fuzzy 98.9344
regressions vs baseline: 0
global matched_code_percent: 93.05256 -> 93.07046
global fuzzy_match_percent: 99.78481 -> 99.78490
global complete_code_percent: 77.53828 -> 77.53828
global matched_data_percent: 99.99410 -> 99.99410
forbidden patterns added (net, per file): 0
readability warnings (net, per file; must be 0 in the final result): 0
GATE PASS
```
`ninja -C . progress build/43U/report.json build/43U/ok` also passed after the clean gate. Final report, objects, and DOL are fresh. The broader 4.3U decompilation remains incomplete; eleven assigned functions remain open.
Post-gate direct verification: `ctxdiff.py libs/RVLMiddleware/eZiText/src/clib/zidawg1 ZiDAWGGetGraphInfo` reports `src 0x218 base 0x218 insns 134/134`, `diffs 0: []`. `pool_diff.py` independently reports identical pools for all six rebuilt source/target object pairs. Final source review contains only four additions/four removals in zidawg1.c, with no comments, assembly, carriers, or unrelated edits.
Owned totals: 22 -> 23 exact functions; 13152 -> 13688 matched code bytes; 1188 matched data bytes preserved.