 Drew T
|
5674587bcb
|
feat(phase-21): T2 — fuel prefetch, Ghidra-C cache complete for the ROI pool
- headless DecompileFunctions over 263 uncached ROI-pool fns (MCP stopped, R23):
263 ok / 0 fail / 0 no-func. cache 300 -> 563. ROI-pool fuel now 325/325 cached
(28/28 giants, 310/310 reach-134, 7/7 capped) -> the unattended run never needs MCP.
- FIX DecompileFunctions.java: key output by entry address func_<UPPERHEX>.c (Ghidra's
raw-import default is FUN_<lowerhex>, which every consumer + the existing 300 files
miss). Renamed the 263; the tool fix makes re-prefetch reproducible.
- build_fuel_manifest.py --emit-prefetch: reproducible uncached-ROI-pool addr-list
(for the T7 runbook). 349 reach-1 x1-leverage fns intentionally excluded (P9).
- ghidra/ db churn is the MCP-stop no-op save (R23 restart-noise, not staged).
|
2026-06-21 12:00:47 -06:00 |
|
|
Drew T
|
0dd9bf6bac
|
feat(phase-17): calibration harvest wave — fleet 55.04%->55.51% (+0.47%)
- Ultracode wave: 30 high-reach tractable residuals, parallel guided-hand-matching
agents (m2c + Ghidra-C cache + asm + actor-struct + S1/S2/S3a playbook), self-
validate match_one. 18/30 match_one MATCH (60%); 10/30 whole-binary verified (33%)
after sig_unify recovered 2 -> all 10 propagated x134; 2 demo fns caught up
- 136/136 byte-identical (R22 clean check-all); REAL +1,610; 1421 dedup groups
- KEY FINDING: the match_one->whole-binary gap (60%->33%) is 100% SIG CONFLICTS
(shared callees like func_80131CA8 declared inconsistently across parallel agents),
NOT codegen. The canonical-sig layer (deferred as low-value) is the ESSENTIAL
enabler -> would lift 33% toward 60%. 12/30 are the genuine gcc-quirk tail (DIFF).
- Ghidra pre-pass: tools/ghidra_scripts/DecompileFunctions.java (headless batch
decompile -> .run/ghidra_c cache; no /mcp needed); 300 tractable targets sized
- harness: .run/harvest_wave_s3.js (embedded-targets Workflow; args channel doesn't
transit arrays). cost ~1.89M tokens / 30 fns
|
2026-06-19 20:54:45 -06:00 |
|