ef1cf9f05e
feat(phase-15): T6 pass-1 — callee-sig-aware harvest of the call-heavy shared core; fleet 25.38% -> 30.48%
...
- tools/gen_harvest_targets.py: callee-signature-aware target manifest (the §14b
extern-conflict wall-breaker). Pre-resolves each callee's EXACT signature from
engine_core.h DEFINE bodies + extern declarations + inline defs, so drafts never
emit a conflicting extern. Lifted call-heavy gate yield from the §14b-naive 16%
to 60-67%.
- ov_SC01_077: 142 functions banked (1716 -> 1574 INCLUDE_ASM stubs) via the
Ultracode parallel-draft + harvest_verify byte-gate loop (cookbook §12). The
draft Workflow was throttled by server-side API rate limiting (stopped at
~246/616); the incremental gate -> dedup_propagate loop is deterministic and
throttle-immune, so the wins banked steadily regardless.
- dedup_propagate --auto-from ov_SC01_077: ~132 functions propagated fleet-wide
(134 overlays each), every round per-overlay byte-gated; dedup 644 -> 776 groups,
dedup_integrate --check 776 validated / 0 failed.
- fleet REAL 85,651 -> 103,217; byte-identical 25.38% -> 30.48%.
- R22 clean rebuild: make clean + extract all 136 + check-all = 136/136 byte-
identical, 0 extract failures.
- cookbook §14c: the wall-breaker + the 4 conflict flavors (defined / declared /
DATA-symbol / narrow-return-widening) + the never-git-checkout-mid-harvest
hazard + the rate-limit-resilient incremental loop. Residual = the data-symbol
conflict class (structural canonical-decls fix queued for next pass) + the
un-started medium band (430).
2026-06-17 00:12:23 -06:00
9d5c087054
feat(phase-15): T5b pass-2 leaf-first harvest (+34 shared fns); fleet 25.38%
...
- leaf-first pass: 52 pure-LEAF unmatched-shared funcs (no jal/%hi/%lo -> match_one is an
exact gate); 5-agent Ultracode Workflow -> 36/52 verified (69% yield, vs 16% mixed) ->
34 propagated fleet-wide (644 groups, dedup-check 0 failed); R22 clean 136/136
- fleet REAL 81,093 -> 85,651; byte-identical 24.05% -> 25.38%
- pure leaves now ~exhausted; remaining shared core is call-heavy (extern-conflict wall)
- CURRENT_PHASE.md: RESUME HANDOFF for a fresh session (T6 callee-aware harvest, T7 giants
+ per-overlay unique, then T8 PhaseEnd) — no PhaseEnd this session (Drew)
- ghidra/ churn NOT staged (R23)
2026-06-16 22:18:06 -06:00
2ac97878b2
feat(phase-15): T5b — Ultracode harvest pass 1 (+49 shared fns); fleet 24.05%
...
- 25-agent Workflow drafted + match_one-verified 300 small unmatched-shared functions;
harvest_verify byte-gate in ov_SC01_077 verified 49 (byte-identical d19c9580);
dedup_propagate spread them fleet-wide (610 groups, dedup-check 0 failed)
- fleet REAL 74,527 -> 81,093; byte-identical 22.14% -> 24.05%; R22 clean 136/136
- yield 16% = the call-heavy hard tail: match_one masks jal/HI16/LO16 (false MATCH on
call-heavy fns); real misses are the extern-type-conflict wall (draft callee extern
!= the callee real macro signature). Leaf fns matched; call-heavy = open-ended tail.
- cookbook 14b documents the wall + the path forward (leaf-first, callee-signature-aware
drafting, link-gated per-fn check)
- ghidra/ churn NOT staged (R23)
2026-06-16 21:16:46 -06:00
aefe2eac18
feat(phase-15): T5a — bulk-propagate 553 shared functions fleet-wide (3.82% -> 22.14%)
...
- dedup_propagate --auto-from ov_SC01_077: propagated 553 already-matched, self-contained
shared engine functions across all 134 overlays, each byte-gated. 9 functions skipped
(use 077-local struct types -> not mechanically liftable; honest, P9). One R22 CLEAN
rebuild: 136/136 byte-identical.
- fleet REAL 947 -> 74,527; byte-identical 13,132 -> 76,174 / 344,010 = 22.14%; 561 dedup
groups, dedup-check 561 validated / 0 failed
- tool hardening for fleet scale: registry SHORTHAND (vram + binaries; dedup.us.yaml ~4k
lines vs ~77k verbose); per-overlay apply (one read/write per file); gate-only-changed;
skip already-registered (additive/resumable); compile pre-filter (body must build with
common.h alone); group_members() reader (both forms) in dedup_integrate + progress;
progress _DEDUP_CACHE (fleet 6m -> 7s)
- the first bulk attempt fail-closed-reverted on a local-typed body (byte-gate working) ->
added the compile filter; cookbook 14a documents the 5 fleet-bulk lessons
- ghidra/ churn NOT staged (R23)
2026-06-16 20:35:39 -06:00
c716eb15e1
feat(phase-15): T4 — fleet progress roll-up + fix cross-report double-count
...
- progress.py: refactor to report(binary) + set_binary() + a per-binary .s index
(fleet run 6m38s -> 7s); add --fleet -> docs/progress.fleet.md (deterministic,
source-derived per-binary table + fleet totals). Single-binary output byte-unchanged.
- fleet now: 947 REAL / 959 LINKED / 13132 byte-identical / 344010 matchable = 3.82%
across 136 binaries (main + resident + 134 overlays)
- dup_report --cross: HONESTY FIX (R14) — an onboarded overlay is both a named BINARIES
entry AND a .run/sig.ov_* glob hit; it was counted twice, inflating collapsible bytes
to 58.7M once the fleet was onboarded. Dedupe by alias -> accurate 9366 h_exact groups
/ 28.6M collapsible (matches the Phase-11 figure)
- Makefile: wire progress.py --fleet into make report (main-only block)
- docs: progress.fleet.md (new), duplicates.cross.md regenerated, SETUP inventory
- make report green; dedup-check 8 validated / 0 failed
- ghidra/ churn NOT staged (R23)
2026-06-16 20:00:19 -06:00
Drew T