The permuter-autopsy prerequisite (Drew-directed). Captures the WHAT/WHERE of every
non-match so the classifier/autopsy (Task-13) can mine it, not just a scalar closeness:
- masked_diff.structured_diff(): the per-instruction masked residual [(idx,mine,tgt),...]
- match_one --json: emits {status,closeness,nins,residual} (refactored to share the primitive)
- backlog FIELDS += residual, passes_tried
- gate_stage.match_one_closeness returns + logs the residual on every near/fail (verified
end-to-end: a near-miss's structured residual now lands in backlog.jsonl)
- grinder: durably save the winning C to .run/permuter-winners/ BEFORE gate_stage banks
(this session's lesson — 3 permuter wins were lost to a mid-flight revert)
A study substrate (UNIVERSAL/COMPILER-CLASS/BFM-SPECIFIC tagged) for a future Fable
pass to refine into a general roadmap for any-platform/any-compiler decomps. §7 adds
the permuter-failure-autopsy loop (mine failures -> extend the ILS wrapper -> shrink
LLM to genuine humps) per Drew's 2026-07-20 direction. Companion to Tasks 12-14.
tools/burndown.py snapshots the 3 fleet metrics + remaining substantial frontier
each run (docs/burndown.jsonl, git-HEAD-keyed), computes velocity vs the previous
commit-snapshot, and flags the yield floor (mean instr-pp over last 3 snapshots <
threshold => 'consider closing P29'). Seeded with the session-4-close baseline.
Reads progress.fleet.md + family-hseq.md (regen those first for a true reading).
Previously only auto-isolated on NON-CONTIGUOUS carve collisions; the span-fit
wall ('table starts do not fit the span', when --like's merged span doesn't match
a sibling's jtbl layout) fell through to carve-fail. jr_isolate unblocks it too
(byte-proven: func_8017AE2C ov_SC01_000 member BANKED). The distinct table-count-
drift error ('more rodata .align than pad specs') stays excluded (not isolate-fixable).
The --hseq path templated from pick_exemplar's choice (hard-prefers ov_SC01_077),
silently ignoring --source (wave-2 finding). Now when --source names a different
overlay carrying a MATCHED member of a family, template from IT (self-correcting:
only a non-stub source member is used; whole-binary gate stays the arbiter). The
ov077-default path is unchanged (override skipped).
Propagated the func_8015C32C jump-table exemplar to 110 more overlays via
jtbl_family_bank (carve + ld_interleave + remap, per-sibling whole-binary gated).
Run halted by an uncaught exception at member 111 (ov_SC06_018); 110/119 attempted
banked clean. Remaining ~19 members to follow.
First jtbl core of crack-wave3: gcc emits the switch jump table into .rodata, so
each overlay needs jtbl_carve + ld_interleave. Path validated 9/9 BANKED.
- func_8016B234: §58b self-def — engine_core.h func_8016B234 (void)->() no-proto
(byte-neutral: only 0-arg callers fleet-wide); typedef preamble stripped.
- func_8016C998: §59(1) local struct rename Slot/Blk32->Slot_998/Blk32_998
(collide with TU's Slot) + D_801D9CA0 extern moved to block scope (TU's other
decls of it are all block-scope; a file-scope one collided).
R22 clean-fleet pending (engine_core.h touched).
Exemplar cast (§56b) unblocked propagation: the bare func_80161208(p) call
relied on ov_SC01_077's no-proto decl, but member TUs carry a (void) prototype.
Casting the call member-proofs the templated body. All banks whole-binary byte-gated.
Propagated from the ov_SC01_000 exemplar (void func_8013D9B0(int param_1), a 141-ins
handwritten GTE color-interp loop w/ $s0-$s4 register pins). Each sibling TU carries
extern-void-(void) self-decls of func_8013D9B0 that conflict with the (int) def; NSD is
unsafe here (it reads the else-func_8013D9B0(N) call statements as decls and mis-casts,
§57a), so the self-decls were mechanically reconciled to K&R empty-args (byte-neutral,
compatible with the (int) def + every 0-arg/1-arg/fn-ptr-cast call site) across 132 TUs
before the sweep. 135/135 members banked, 0 failed (pins templated clean across the
structurally-identical engine siblings).
Propagated from the ov_SC01_000 exemplar. remap_hseq's gather_externs synthesizes a
non-volatile `extern u16 D_8011511A;` per sibling, which defeats the §18 volatile-strip
CSE (cast reads reload instead of folding to one load) -> byte-DIFF. Every sibling TU
(jr_8013FFD8) already declares D_8011511A volatile at file scope via the banked
func_801418F8, so the fix is to strip the draft's own decl and let the cast-at-use over
the TU's volatile decl produce the single-load match. 135/135 banked, 0 failed.
func_80141A60 (76): §18 cast-at-use *(u16*)&D_8011511A defeats the TU's file-scope
volatile decl of D_8011511A so gcc CSEs the 3 reads into one load (the draft matched
standalone but DIFFd in-TU due to the volatile reload).
func_8013D9B0 (141): reconciled the TU's two extern-void-(void) self-decls to K&R
empty-args (compatible with the int-param def; byte-neutral for the fn-ptr-cast and
0-arg call sites). NSD was unsafe here (it mis-parses the else-func_8013D9B0(N) call
statements as decls and mis-casts to void(*)(void), the §57a hazard). Both re-gate
BYTE-IDENTICAL.
The jr_8013F350 mega-TU declares func_80128ED8(s32,s32*) and DEFINE-macro func_80146C3C,
conflicting with the draft's carried externs. Dropped both externs + fn-ptr-cast the
func_80128ED8 call at use (byte-identical). Banked ov_SC07_006/007/011. Remaining: 1
o2b member (ov_SC07_010) — -O2 reconstruction TU hits an as --fatal-warnings $at wall.
The 10 jr_8013F350-split residuals failed on 'conflicting types for D_8011512E'
(exemplar decl s16 vs the jr_8013F350 TUs' canonical u16). Reconciled the exemplar
decl to u16 (byte-neutral — sole use is store-0); re-swept. func_801418F8 now 0 stubs.
Remaining: 4 func_8016DC20 SC07 members (byte-DIFF in the o2b/jr_8013F350 split).
Both banked fully (0 stubs remaining). func_8013EF88 templates clean because E4 (its
matching-critical local typedef) is already file-scope in every target TU via the
already-banked sibling func_8013EE10. func_8013D8FC's $s0-$s3 register pins template
fine across the structurally-identical engine siblings. 271/271 members, 0 failed.
hseq family sweep of 2 absent-family cores, templated from the clean ov_SC01_000
bank (s16-array + cast-at-use spellings) instead of the ov077 heavy-struct exemplar
that pick_exemplar defaults to. 258/272 members banked; 14 SC07-layout residuals.
-O0 core, single jump table jtbl_801D82FC carved into the o0 object's .rodata via
jtbl_carve (fit contiguously with the existing o0 carve). ov_SC01_077 byte-identical (R22 d19c9580).
These reference a per-overlay tail work-buffer whose base address DIFFERS per
overlay (h_seq relocated data). remap_hseq keyed the byte-OFFSET addresses
(D_801D9C21..) not the base symbol D_801D9C20/D_801D9C60 the exemplar C uses, so
it left them unresolved -> 0/137. Per overlay: base = symbol_map[D_801D9C21]-1;
declare dlabel D_<base> in config/symbols.<ov>.txt (byte-neutral, re-extract emits
the linker def), remap D_801D9C20->D_<base1> / D_801D9C60->D_<base2>. 3 SC07 stragglers
needed the carried ApplyMatrixSV/RotMatrixYXZ externs dropped (TU already declares
them with a different sig -> conflicting types). Each gated whole-overlay byte-identical.
Drew T