Blame for analysis/data/failure_mode_annotations.jsonl

{"prog": "bloaty_fuzz_target", "src_site_id": 7429, "to_take": 0, "try_id": 0, "source_location": "/src/bloaty/bloaty/src/dwarf.cc:604:19", "format": "elf,mach-o,pe,wasm", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "For a parsed DW_AT_comp_dir attribute, the false branch requires AttrValue::IsString() to be false, i.e. its parsed type is neither kString nor kUnresolvedString.", "system_requirement": "Supply a valid ELF whose discoverable .debug_info and .debug_abbrev sections describe a compilation-unit DIE with DW_AT_comp_dir using a non-string form such as DW_FORM_udata.", "agent_interpretation": "The agent correctly chose a DWARF compile-unit DW_AT_comp_dir encoded as DW_FORM_udata, which ParseAttr represents as kUint.", "failure_evidence": "The plan is adequate, but the emitted ELF sets e_shoff=0 and omits the mandatory null section while claiming four sections; it also writes DWARF abbrev name/form values as fixed 16-bit fields and redundantly puts name/form bytes in .debug_info. Thus the intended DIE is not parsed; outcome was sfunc=0, reach=0."}
{"prog": "bloaty_fuzz_target", "src_site_id": 3122, "to_take": 0, "try_id": 0, "source_location": "/src/bloaty/bloaty/third_party/protobuf/src/google/protobuf/descriptor.pb.cc:8485:7", "format": "elf,mach-o,pe,wasm", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires the particular google::protobuf::EnumOptions object in ByteSizeLong() to have no unknown fields.", "system_requirement": "The binary input must trigger a concrete Bloaty/protobuf reflection or descriptor path that invokes EnumOptions::ByteSizeLong(), with that EnumOptions instance containing no unknown fields.", "agent_interpretation": "The agent correctly described have_unknown_fields()==false, but assumed that a minimal ELF with no protobuf-looking payload would make an EnumOptions object reach ByteSizeLong() without unknown fields.", "failure_evidence": "The trace repeatedly admits it cannot identify when EnumOptions::ByteSizeLong() is called, then substitutes an unrelated 64-byte sectionless ELF and claims absence of protobuf bytes is sufficient. No input-to-EnumOptions call path is established; outcome was sfunc=1 but reach=0."}
{"prog": "bloaty_fuzz_target", "src_site_id": 7375, "to_take": 1, "try_id": 2, "source_location": "/src/bloaty/bloaty/src/pe.cc:124:7", "format": "elf,mach-o,pe,wasm", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The true branch requires the four bytes in pe_header_.signature to differ from PE_Magic ('P','E',0,0), after the DOS magic and header-bounds checks succeed.", "system_requirement": "Provide at least e_lfanew + sizeof(pe_header) + sizeof(PE_TYPE) bytes, with MZ at offset 0, e_lfanew locating the 24-byte PE header, and a signature other than PE\\0\\0.", "agent_interpretation": "The agent correctly required an MZ DOS header, a usable AddressOfNewExeHeader, and a nonmatching PE signature, choosing FOO\\0 at offset 0x80.", "failure_evidence": "The generated file is 152 bytes with e_lfanew=128, exactly ending after the 24-byte pe_header; the preceding bounds check requires two more PE_TYPE bytes (154 total), so Initialize returns at line 116 before comparing FOO\\0. Outcome was sfunc=1, reach=0."}
{"prog": "exiv2", "src_site_id": 2005, "to_take": 0, "try_id": 2, "source_location": "/src/exiv2/exiv2/src/jp2image.cpp:414:7", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires option to be kpsNone, so bPrint, bXMP, bICC, and bIPTCErase are all false.", "system_requirement": "The evaluated exiv2 command path must select Jp2Image and invoke printStructure with kpsNone after metadata parsing; merely supplying a JP2 signature is insufficient when the default CLI action only prints a metadata summary.", "agent_interpretation": "The agent correctly identified kpsNone and emitted the exact 12-byte JP2 signature, but assumed the evaluated entry path would call Jp2Image::printStructure(kpsNone).", "failure_evidence": "The evaluated exiv2 CLI follows its default summary path, which calls readMetadata but not Jp2Image::printStructure. The trace modeled a different call path and emitted only Jp2Signature; recorded sfunc=0 and reach=0 therefore reflect a missing entry-to-target abstraction."}
{"prog": "exiv2", "src_site_id": 2917, "to_take": 1, "try_id": 1, "source_location": "/src/exiv2/exiv2/src/quicktimevideo.cpp:1620:9", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires file bytes 4-7 to equal one of qTimeTags, making buf.cmpBytes(4, tag, 4) return zero.", "system_requirement": "ImageFactory must reach isQTimeType rather than accept an earlier registered format; a QuickTime-like first 12 bytes should place a known atom tag at offset 4 and preferably a known brand at offset 8.", "agent_interpretation": "The agent correctly chose bytes 4-7 as ftyp and bytes 8-11 as the known brand M4V , but believed a file beginning with the JPEG SOI could still enter QuickTime detection.", "failure_evidence": "The generator and test.jpg begin FF D8, while image.cpp checks the JPEG registry entry before QuickTime and returns immediately on isJpegType. Thus isQTimeType is never called despite correctly placed ftyp/M4V bytes; outcome was sfunc=0, reach=0."}
{"prog": "exiv2", "src_site_id": 465, "to_take": 1, "try_id": 0, "source_location": "/src/exiv2/exiv2/xmpsdk/src/UnicodeInlines.incl_cpp:101:8", "format": "jpeg", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "For the first byte of a nonempty simple XML name, cp must be ASCII below 0x80 but not A-Z, a-z, or underscore, so IsStartChar_ASCII(cp) is false.", "system_requirement": "A recognized JPEG XMP APP1 packet must traverse Expat/RDF processing into an XMP SDK property or qualified-name API that invokes VerifySimpleXMLName with an invalid ASCII-leading name.", "agent_interpretation": "After correcting an early inversion, the agent targeted local name 1invalid, assuming malformed element <dc:1invalid> in JPEG XMP would be passed to VerifySimpleXMLName.", "failure_evidence": "The APP1 payload omits the mandatory 29-byte http://ns.adobe.com/xap/1.0/ NUL identifier required by jpgimage.cpp, and <dc:1invalid> is an invalid XML QName that Expat rejects before the SDK's XPath-name verifier. The trace never derives a viable path delivering cp='1'; outcome reached the condition but did not take it (sfunc=1, reach=1, pass=0)."}
{"prog": "ffmpeg", "src_site_id": 13095, "to_take": 1, "try_id": 2, "source_location": "/src/ffmpeg/ffmpeg/build_wrapper/src/fftools/ffmpeg_mux_init.c:1976:17", "format": "vi,mov,flv,mp4,mp3,h264", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "With no explicit stream maps, one invoked map_auto_* helper must return a negative error from output-stream setup.", "system_requirement": "Supply a parseable input stream and a concrete automatic-mapping/output configuration that deterministically makes ost_add fail, such as an unsupported encoder or invalid mapping option.", "agent_interpretation": "The agent correctly identified ret as a map_auto_* return value and traced negative returns through ost_add, but assumed a minimally corrupted input would make output-stream creation fail.", "failure_evidence": "generator.py emits an MP4 with no valid track, while the trace itself notes this likely makes map_auto_* return 0; no deterministic ost_add failure is specified. Outcome was sfunc=1, reach=1, pass=0."}
{"prog": "ffmpeg", "src_site_id": 5985, "to_take": 1, "try_id": 1, "source_location": "/src/ffmpeg/ffmpeg/build_wrapper/src/libavformat/demux.c:864:9", "format": "vi,mov,flv,mp4,mp3,h264", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Return early if first_dts is already set, dts or cur_dts is AV_NOPTS_VALUE, either timestamp is out of the guarded range, or dts is relative.", "system_requirement": "Create a valid demuxable stream that emits at least one packet and follows a demux timestamp-processing call site with one guarded state or timestamp condition true.", "agent_interpretation": "The agent correctly enumerated the disjuncts and targeted dts == AV_NOPTS_VALUE, but equated malformed or timestamp-free files with delivering such a packet to update_initial_timestamps.", "failure_evidence": "All generated candidates are empty or malformed (for example MP4 sample tables and mdat are empty, FLV has no media tags), so they do not establish a packet/call path; recorded sfunc=0 and reach=0 confirm update_initial_timestamps was never entered."}
{"prog": "ffmpeg", "src_site_id": 1737, "to_take": 1, "try_id": 1, "source_location": "/src/ffmpeg/ffmpeg/build_wrapper/src/libavformat/aixdec.c:112:9", "format": "vi,mov,flv,mp4,mp3,h264", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "unknown", "confidence": "low", "local_condition": "After any AIXE handling, the current 4-byte little-endian chunk tag must differ from AIXP.", "system_requirement": "Pass AIX probing and header parsing, leave the stream positioned at a non-AIXE, non-AIXP packet tag, and invoke aix_read_packet.", "agent_interpretation": "The agent ultimately planned a probe-valid AIX header, an AIXP header chunk, then an AIYE packet tag so read_packet would reach the condition with chunk != AIXP.", "failure_evidence": "The final generator statically places AIXP at 0x50 and AIYE at the computed packet offset 0x70, matching the source path, yet the outcome is sfunc=1, reach=1, pass=0. The generated test_input.aix was deleted in the trace and is absent from preserved artifacts, so the contradictory failure cannot be assigned reliably."}
{"prog": "freetype2_ftfuzzer", "src_site_id": 4836, "to_take": 1, "try_id": 0, "source_location": "/src/freetype2/freetype2/src/truetype/ttinterp.c:6951:14", "format": "ttf,otf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "During IUP processing of a nonempty contour, exactly one point must carry the selected runtime touch flag (TOUCH_X for IUP[1] or TOUCH_Y for IUP[0]), so cur_touched is never advanced beyond first_touched.", "system_requirement": "Provide a valid hinted TrueType glyph whose instructions touch exactly one contour point on the selected axis and then execute the corresponding IUP opcode while ftfuzzer loads that glyph.", "agent_interpretation": "The agent correctly recognized the one-touched-point condition, but believed an on-disk glyf flag byte 0x03 directly marks a point TOUCH_X and that such an outline would invoke IUP.", "failure_evidence": "The generator sets glyf point flag 0x03 but emits zero glyph instructions; on-disk bit 0x02 is an x-coordinate encoding flag, whereas FreeType's runtime TOUCH_X is 0x08. Thus no instruction creates touch state or calls IUP; outcome was sfunc=0, reach=0."}
{"prog": "freetype2_ftfuzzer", "src_site_id": 5060, "to_take": 1, "try_id": 0, "source_location": "/src/freetype2/freetype2/src/truetype/ttgxvar.c:281:12", "format": "ttf,otf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "A packed-delta run must still have j <= cnt when i reaches delta_cnt, meaning the run declares more values than remain in the destination delta array.", "system_requirement": "Supply an accepted variable TTF with consistent fvar/gvar axes, a valid gvar header and glyph-offset array, and reachable glyph variation tuple data whose private point count is smaller than a packed delta run.", "agent_interpretation": "The agent correctly chose a small packed point count and a larger delta-run count, but treated those bytes as if they could be placed directly after a short ad hoc gvar header.", "failure_evidence": "The final font has no fvar table, and its 28-byte gvar omits the required 20-byte gvar header fields, glyph-offset array, glyph variation header, tupleDataSize, and tupleIndex structure. The crafted run therefore never reaches ft_var_readpackeddeltas; outcome was sfunc=0, reach=0."}
{"prog": "freetype2_ftfuzzer", "src_site_id": 4209, "to_take": 1, "try_id": 1, "source_location": "/src/freetype2/freetype2/src/base/ftstroke.c:2155:14", "format": "ttf,otf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "In FT_Stroker_ParseOutline, a contour with at least two points must start with a conic control point and end with an on-curve point.", "system_requirement": "The input must be processed through FT_Stroker_ParseOutline (normally via a stroker API such as FT_Glyph_Stroke) with the required outline; merely loading a glyph is insufficient.", "agent_interpretation": "The agent correctly chose a two-point contour with first tag CONIC and last tag ON, assuming loading that glyph in ftfuzzer would exercise the stroker parser.", "failure_evidence": "The agent never inspected the harness call path. ftfuzzer.cc only calls FT_Load_Glyph; its outline operations are disabled under #if 0 and it contains no FT_Stroker or FT_Glyph_Stroke call, so this input cannot enter FT_Stroker_ParseOutline. Recorded outcome was missing_function with error \"no buffer found\", sfunc=0, reach=0."}
{"prog": "harfbuzz_hb-shape-fuzzer", "src_site_id": 1672, "to_take": 1, "try_id": 0, "source_location": "/src/harfbuzz/harfbuzz/src/hb-ot-layout-gsubgpos.hh:549:6", "format": "ttf,otf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires matcher.may_skip() to return SKIP_YES, e.g. because the current glyph's GDEF property matches an IgnoreBaseGlyphs, IgnoreLigatures, or IgnoreMarks lookup flag.", "system_requirement": "A valid font must activate a contextual/chaining GSUB or GPOS path that calls the skipping iterator's next() and places a glyph with the ignored property among the context glyphs reached from the fuzzer's fixed or font-derived text.", "agent_interpretation": "The agent correctly derived the IgnoreMarks-plus-mark-glyph condition and tried to add an IgnoreMarks single-substitution lookup and mark classes.", "failure_evidence": "The trace never identifies a lookup type that invokes skipping_context_t::next() or ensures the fuzzer text supplies an ignored glyph; it instead builds an unrelated SingleSubst lookup. The final generator also imports nonexistent fontTools.ttLib.tables._g_s_u_b and exits with ModuleNotFoundError; outcome was missing_function, sfunc=0, reach=0, error=\"no buffer found\"."}
{"prog": "harfbuzz_hb-shape-fuzzer", "src_site_id": 2997, "to_take": 0, "try_id": 1, "source_location": "/src/harfbuzz/harfbuzz/src/OT/Color/CPAL/CPAL.hh:54:9", "format": "ttf,otf", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The false branch requires CPALV1Tail.paletteFlagsZ to be nonzero, so execution continues to read the palette type array.", "system_requirement": "Emit a valid sanitized CPAL v1 table whose Palette Type Array is physically present; hb-shape-fuzzer's test_font() then calls hb_ot_color_palette_get_flags().", "agent_interpretation": "The agent correctly required CPAL version 1 and a serialized paletteTypes array at a nonzero offset, using the fuzzer's palette API path.", "failure_evidence": "The final generator uses paletteTypes=[0], FontTools' DEFAULT_PALETTE_TYPE; its compiler may omit an all-default paletteTypes payload and writes offsetToPaletteTypeArray=0 when the compiled payload is empty. Evaluation reached the condition (sfunc=1, reach=1, exec_status=\"success\") but did not take the requested false branch, confirming a concrete CPAL encoding defect."}
{"prog": "harfbuzz_hb-shape-fuzzer", "src_site_id": 312, "to_take": 1, "try_id": 1, "source_location": "/src/harfbuzz/harfbuzz/src/hb-ot-var-common.hh:194:9", "format": "ttf,otf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The true branch requires HBUINT8::sanitize() on DeltaSetIndexMap.u.format to fail because the one-byte format field is outside the sanitizer range or no byte remains.", "system_requirement": "Provide a structurally reachable COLR v1, HVAR/VVAR, or avar parent table with a nonnull DeltaSetIndexMap offset whose target has fewer than one available byte, so OffsetTo dispatch enters DeltaSetIndexMap::sanitize().", "agent_interpretation": "The agent correctly traced the bounds check and planned to make a COLR or HVAR Offset32To<DeltaSetIndexMap> dispatch to a location past EOF.", "failure_evidence": "The concrete layouts are wrong: the generator treats COLR as major/minor plus offsets and overwrites colr_offset+16, while source places varIdxMap after the 14-byte v0 header and three v1 offsets (offset 26); it likewise writes HVAR advMap at +10 instead of +8. It also requires an input-font argv and otherwise exits without emitting bytes; outcome was missing_function, sfunc=0, reach=0, error=\"no buffer found\"."}
{"prog": "imginfo", "src_site_id": 1235, "to_take": 1, "try_id": 0, "source_location": "/src/jasper/jasper/src/libjasper/jpc/jpc_dec.c:2188:6", "format": "bmp,mif,pgx,pnm,ras,jpg,png,tga", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "In jpc_dec_destroy, take the branch when dec->cmpts is non-NULL.", "system_requirement": "Supply a content-detected JPC/JP2 codestream whose valid SIZ marker reaches jpc_dec_process_siz and allocates the component array before decoder teardown.", "agent_interpretation": "A successfully parsed JPEG-2000 SIZ segment with at least one component allocates dec->cmpts, so cleanup takes the branch.", "failure_evidence": "The trace found the allocation at jpc_dec.c:1377 and chose Csiz=1, but generator.py encodes each 32-bit SIZ dimension with five bytes and Csiz with three bytes. The resulting SIZ fields are shifted/malformed; despite reach=1 and sfunc=1, the requested branch did not pass."}
{"prog": "imginfo", "src_site_id": 174, "to_take": 1, "try_id": 1, "source_location": "/src/jasper/jasper/src/libjasper/pnm/pnm_dec.c:573:6", "format": "bmp,mif,pgx,pnm,ras,jpg,png,tga", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "In pnm_getuintstr, take the branch when the parsed unsigned decimal converts to JAS_UI32_MAX, including an input value at least 4294967295 or invalid safe arithmetic.", "system_requirement": "Use a valid text P2/P3 header with maxval below 65536 and unsigned samples, then place the overflowing decimal in pixel sample data so pnm_getuintstr is called.", "agent_interpretation": "Use a text P2 PNM and place 4294967296 in a numeric field so the unsigned parser returns JAS_UI32_MAX.", "failure_evidence": "The trace correctly located the only pnm_getuintstr call at pnm_dec.c:443 in unsigned text sample parsing, but generator.py puts 4294967296 in maxval. Header maxval is read by pnm_getsintstr at line 301, so decoding exits before the target; recorded sfunc=0 and reach=0."}
{"prog": "imginfo", "src_site_id": 1597, "to_take": 1, "try_id": 0, "source_location": "/src/jasper/jasper/src/libjasper/jpc/jpc_cs.c:1073:9", "format": "bmp,mif,pgx,pnm,ras,jpg,png,tga", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "With qntsty not equal to JPC_QCX_NOQNT and numstepsizes positive, take the branch when jpc_getuint16 cannot read both bytes of a quantization stepsize.", "system_requirement": "Reach QCD/QCC parameter parsing through a valid JPC marker sequence while accounting for jpc_getms copying the marker's declared parameter length into a temporary stream before jpc_qcx_getcompparms reads it.", "agent_interpretation": "Declare a SEQNT QCD segment with positive step count but truncate its stepsize bytes at EOF to make jpc_getuint16 fail.", "failure_evidence": "The trace modeled the immediate EOF condition but never incorporated jpc_getms at jpc_cs.c:258-276: it first copies all declared QCD parameter bytes, and a short file makes marker acquisition fail before jpc_qcx_getcompparms. Final QCD declares length 5 but contains only byte 0x22; recorded sfunc=0 and reach=0."}
{"prog": "infotocap", "src_site_id": 2730, "to_take": 0, "try_id": 2, "source_location": "/src/ncurses/ncurses-6.5-20250503/progs/dump_entry.c:1604:10", "format": "terminfo", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "changed is true and the recomputed termcap entry length is at most 1023 bytes.", "system_requirement": "Provide valid terminfo source that infotocap parses into an entry over 1023 formatted termcap bytes, with a valid removable sgr whose removal reduces FMT_ENTRY() to at most 1023.", "agent_interpretation": "Correctly required an initially oversized limited termcap conversion whose earlier removal, especially sgr, sets changed and makes FMT_ENTRY() fit.", "failure_evidence": "Trace identified the full limited termcap shrink path and sgr removal, but sized raw generator stdout (1095 bytes) as if it were FMT_ENTRY(); generator emits colon-separated termcap syntax despite infotocap parsing terminfo source, uses invalid sgr parameters %0..%179 and unknown sa0..sa14, and recorded reach=0 although sfunc=1."}
{"prog": "infotocap", "src_site_id": 2655, "to_take": 0, "try_id": 0, "source_location": "/src/ncurses/ncurses-6.5-20250503/progs/dump_entry.c:973:15", "format": "terminfo", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "The condition must be evaluated for a positive predval with i + 1 less than or equal to the previously accumulated num_bools.", "system_requirement": "Provide valid terminfo source with present true booleans processed in an order where an earlier higher index raises num_bools before a later lower-or-equal index reaches the comparison.", "agent_interpretation": "Ultimately treated zero booleans, so the loop and conditional never execute, as equivalent to taking the false branch.", "failure_evidence": "Trace explicitly noted that skipping the loop is not taking the false branch, then nevertheless generated a compiled header with bools_size=0 and concluded the unevaluated condition was 'always FALSE in practice'; outcome was sfunc=1, reach=0."}
{"prog": "infotocap", "src_site_id": 2738, "to_take": 1, "try_id": 1, "source_location": "/src/ncurses/ncurses-6.5-20250503/progs/dump_entry.c:1437:9", "format": "terminfo", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "acs_chars must be neither ABSENT_STRING nor CANCELLED_STRING.", "system_requirement": "Provide valid terminfo source with acsc and an oversized limited termcap conversion that reaches the shrink stage and calls purged_acs at line 1605.", "agent_interpretation": "Correctly believed that defining a non-cancelled acsc string makes VALID_STRING(acs_chars) true.", "failure_evidence": "Trace focused only on defining acsc and never derived the enclosing call requirement at lines 1557-1605; the small generated entry cannot enter the oversized shrink path, matching recorded sfunc=0 and reach=0."}
{"prog": "jhead", "src_site_id": 172, "to_take": 1, "try_id": 2, "source_location": "/src/jhead/jhead/exif.c:1484:9", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "import_error", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "ImageInfo.ExposureProgram must be nonzero.", "system_requirement": "Provide JPEG bytes containing a valid APP1 Exif/TIFF structure whose Exif IFD sets tag 0x8822 to a nonzero value, through the required generate_buf interface.", "agent_interpretation": "The agent correctly identified EXIF tag 0x8822 and chose ExposureProgram=1 in the Exif IFD.", "failure_evidence": "The final generator has an adequate JPEG/Exif plan but imports piexif and exposes only create_jpeg_with_exposure_program, not generate_buf; the recorded outcome is import_error: No module named 'piexif' with sfunc=reach=0."}
{"prog": "jhead", "src_site_id": 111, "to_take": 1, "try_id": 0, "source_location": "/src/jhead/jhead/exif.c:700:25", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "While processing TAG_USERCOMMENT with no prior ImageInfo.Comments value, msiz must exceed 5 and the comment value's first five bytes must equal ASCII.", "system_requirement": "Provide a valid JPEG APP1 Exif/TIFF hierarchy that reaches the UserComment entry without an earlier comment, preserving all IFD offsets and exposing the resulting bytes via generate_buf.", "agent_interpretation": "The agent correctly targeted EXIF UserComment 0x9286 with value bytes beginning ASCII and longer than five bytes.", "failure_evidence": "The agent recognized its insertion shifted existing TIFF offsets; its generated JPEG produced 'Illegally sized Exif subdirectory (21313 entries)' and 'Unexpected end of file'. The final generator also lacks generate_buf, matching missing_function: no buffer found."}
{"prog": "jhead", "src_site_id": 161, "to_take": 1, "try_id": 2, "source_location": "/src/jhead/jhead/exif.c:1395:9", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "import_error", "generator_exists": 1, "label": "unknown", "confidence": "high", "local_condition": "ImageInfo.DigitalZoomRatio must be greater than 1.", "system_requirement": "Provide JPEG bytes with a valid Exif IFD DigitalZoomRatio entry that jhead converts to a value above 1, through the required generate_buf interface.", "agent_interpretation": "The agent correctly identified EXIF tag 0xA404 and attempted to encode a rational value greater than 1 in a JPEG.", "failure_evidence": "Preserved evidence is contradictory: events show the final PIL-only generator running and jhead printing 'Digital Zoom : 250.000x', and the saved generator imports no piexif, yet the recorded outcome is import_error: No module named 'piexif' with sfunc=reach=0. The failure stage cannot be distinguished reliably."}
{"prog": "jq", "src_site_id": 4402, "to_take": 1, "try_id": 1, "source_location": "/src/jq/jq/src/jv_dtoa.c:3544:6", "format": "json", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires the double argument dd passed to jvp_dtoa to have its sign bit set.", "system_requirement": "The input must make jq create a native negative number that is printed through jvp_dtoa; an unchanged parsed decNumber literal is emitted directly and bypasses that function.", "agent_interpretation": "The agent correctly read the condition as requiring a negative double and generated JSON containing -42.5.", "failure_evidence": "The trace assumes any printed negative JSON number reaches jvp_dtoa, but with USE_DECNUM jv_print.c prints preserved numeric literals directly. The generator emits an unchanged -42.5 literal; outcome was sfunc=0, reach=0 despite successful execution."}
{"prog": "jq", "src_site_id": 3423, "to_take": 0, "try_id": 2, "source_location": "/src/jq/jq/vendor/decNumber/decNumber.c:608:13", "format": "json", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "The target false branch is evaluated only with exactly 10 significant exponent digits and first exponent digit at most '1'; then c==firstexp+10 and *firstexp<='1'.", "system_requirement": "Provide a JSON numeric literal whose exponent reaches decNumberFromString with exactly 10 significant digits (after leading-zero stripping) starting with 0 or 1, so the enclosing check executes the target condition.", "agent_interpretation": "Although it briefly derived the exact 10-digit requirement, the agent finally claimed that using 1e8 to skip the enclosing c>=firstexp+10 check counts as taking the target inner false branch.", "failure_evidence": "The final generator emits {\"value\": 1e8}; its one-digit exponent makes the enclosing condition false, so line 608 is never evaluated. Recorded outcome confirms sfunc=1 but reach=0."}
{"prog": "jq", "src_site_id": 3939, "to_take": 1, "try_id": 0, "source_location": "/src/jq/jq/vendor/decNumber/decNumber.c:7464:7", "format": "json", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires a nonzero value in decSetSubnormal with dn->exponent >= etiny, because adjust=etiny-dn->exponent must be <=0.", "system_requirement": "A JSON numeric value must traverse jq's decNumber arithmetic into decSetSubnormal under jq's actual DEC_CONTEXT settings; quoted text is only a string, and jq's parse context is DEC_INIT_BASE with a much larger configured precision and exponent range, not decimal64.", "agent_interpretation": "The agent sought a one-digit 1e-384 value and reasoned using a decimal64 context (digits=16, emin=-383), then emitted it as the JSON string \"1e-384\".", "failure_evidence": "The trace never establishes the real jq call path and assumes decimal64 bounds, while jv.c initializes parsing with DEC_INIT_BASE and custom digits. Its generator serializes \"1e-384\" as a quoted string, so no decNumber numeric path exists; outcome was missing_function, sfunc=0, reach=0, error=\"no buffer found\"."}
{"prog": "lcms_cms_transform_fuzzer", "src_site_id": 2821, "to_take": 0, "try_id": 0, "source_location": "/src/lcms/Little-CMS/src/cmscnvrt.c:580:17", "format": "cms", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "In the device-link/named-color path, _cmsReadDevicelinkLUT must return a non-NULL pipeline so Lut == NULL is false.", "system_requirement": "Provide an ICC profile accepted by cmsOpenProfileFromMem, classified as link/abstract or eligible named-color, with a valid A2B0/D2B0 pipeline for the fuzzer's perceptual transform.", "agent_interpretation": "The agent correctly required a device-link profile with an intent-0 A2B0 LUT that LittleCMS can parse.", "failure_evidence": "The final profile has class 'link' and A2B0, but writes 'APPL' at header bytes 36-39; cmsio0.c requires the ICC magic 'acsp' there and rejects any mismatch. Accordingly sfunc=0 and reach=0."}
{"prog": "lcms_cms_transform_fuzzer", "src_site_id": 2209, "to_take": 1, "try_id": 0, "source_location": "/src/lcms/Little-CMS/src/cmscnvrt.c:646:9", "format": "cms", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "no_result", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Reach the Error label after Lut has become non-NULL but before it is freed and reset, so Lut != NULL is true.", "system_requirement": "Use an openable device-link profile with a readable A2B/D2B LUT and identify a feasible later ComputeConversion, AddConversion, or cmsPipelineCat failure that occurs before cmsPipelineFree(Lut).", "agent_interpretation": "The agent recognized that a non-NULL LUT must be followed by a later failure, but assumed an invalid PCS would make AddConversion fail in the first device-link profile's actual parameter flow.", "failure_evidence": "The proposed invalid-PCS mechanism is not a demonstrated post-LUT failure for this first-profile flow, and the baseline profile contains neither the device-link class at the correct offset nor any readable LUT. The trajectory never establishes a viable system-level path to Error with Lut non-NULL; outcome was no_result with sfunc=0 and reach=0."}
{"prog": "lcms_cms_transform_fuzzer", "src_site_id": 2377, "to_take": 0, "try_id": 1, "source_location": "/src/lcms/Little-CMS/src/cmsio1.c:143:9", "format": "cms", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "cmsReadTag must successfully return all three rXYZ, gXYZ, and bXYZ colorant values, making every pointer non-NULL and the OR condition false.", "system_requirement": "Supply a valid ICC RGB source profile accepted by cmsOpenProfileFromMem whose matrix-shaper input path reads rXYZ, gXYZ, bXYZ and rTRC/gTRC/bTRC tags.", "agent_interpretation": "The agent correctly identified an RGB matrix-shaper input profile with parseable rXYZ/gXYZ/bXYZ tags and supporting TRCs as the needed path.", "failure_evidence": "The generated output has the intended colorant/TRC tag table, but header bytes 36-39 are zero instead of mandatory ICC magic 'acsp'; cmsio0.c rejects it before transform construction. This matches sfunc=0 and reach=0."}
{"prog": "libjpeg_turbo_fuzzer", "src_site_id": 1083, "to_take": 1, "try_id": 1, "source_location": "/src/libjpeg/libjpeg-turbo.3.0.x/jdphuff.c:488:7", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "In decode_mcu_DC_refine, cinfo->restart_interval must be nonzero after an MCU is decoded.", "system_requirement": "A valid progressive JPEG must reach a DC successive-approximation refinement scan (Ah > 0, Ss = Se = 0) with a nonzero DRI interval.", "agent_interpretation": "Correctly identified that a nonzero DRI marker sets restart_interval.", "failure_evidence": "The generator emits SOF0 baseline JPEG and a single sequential SOS (Ss=0, Se=63, Ah=Al=0), so decode_mcu_DC_refine cannot be selected despite DRI=1; outcome sfunc=0, reach=0."}
{"prog": "libjpeg_turbo_fuzzer", "src_site_id": 1449, "to_take": 1, "try_id": 1, "source_location": "/src/libjpeg/libjpeg-turbo.3.0.x/jdcoefct.c:661:17", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "In the num < 0 AC20 estimation path, coef_bits[3] must give Al > 0 and pred=((Q20<<7)-num)/(Q20<<8) must be at least 1<<Al.", "system_requirement": "A progressive JPEG decoded with block smoothing must leave AC20 zero/inaccurate while supplying a 5x5 neighborhood of DC coefficients whose weighted sum makes num sufficiently negative.", "agent_interpretation": "Correctly derived the negative-num formula and thresholds, including |num| >= 384*Q20 for Al=1 and the need for progressive AC scans.", "failure_evidence": "Although the trace states the required DC-neighborhood formula, create_optimal_trigger returns the first qualifying 8x8 candidate; the saved 8x8 solid-black JPEG has only one DCT block and no constructed 5x5 DC pattern. Outcome sfunc=1 but reach=0."}
{"prog": "libjpeg_turbo_fuzzer", "src_site_id": 371, "to_take": 1, "try_id": 0, "source_location": "/src/libjpeg/libjpeg-turbo.3.0.x/jdcolor.c:815:11", "format": "jpeg", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "With jpeg_color_space == JCS_RGB, out_color_space must map red, green, blue to offsets 0, 1, 2 and have pixel size 3 (JCS_RGB or JCS_EXT_RGB).", "system_requirement": "A structurally valid three-component JPEG with R/G/B component IDs must pass header parsing and reach color-converter initialization.", "agent_interpretation": "Correctly determined that no JFIF marker plus SOF component IDs 'R','G','B' yields JCS_RGB and that the default output is JCS_RGB.", "failure_evidence": "The generator encodes the three-component SOF0 length as 11, but jdmarker.c requires 8+3*num_components=17; only the first component descriptor lies within the declared segment, so parsing fails before jdcolor.c. Outcome sfunc=0, reach=0."}
{"prog": "libpcap_fuzz_both", "src_site_id": 937, "to_take": 1, "try_id": 0, "source_location": "/src/libpcap/libpcap/optimize.c:1877:7", "format": "pcap", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take true when the candidate block reached through diffp has a different true successor from b: JT(*diffp) != JT(b).", "system_requirement": "Supply fuzz_both's composite input: a length-prefixed, NUL-terminated filter that compiles an appropriate CFG, followed by a valid pcap whose link type supports that filter.", "agent_interpretation": "Correctly identified that the two blocks' true-successor pointers must differ and associated the check with an OR-shaped BPF CFG.", "failure_evidence": "The trace never inspected testprogs/fuzz/fuzz_both.c and generator.py emits only a raw Ethernet pcap, merely assuming an external filter such as \"tcp or udp\"; fuzz_both instead reads byte 0 as filterSize, so the raw-pcap magic byte 0xd4 requests 212 filter bytes and the short file returns before pcap_compile. Recorded sfunc=0, reach=0."}
{"prog": "libpcap_fuzz_both", "src_site_id": 1015, "to_take": 0, "try_id": 1, "source_location": "/src/libpcap/libpcap/optimize.c:1656:6", "format": "pcap", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take false when the accumulator value numbers match on exit from child and ep->pred: child->val[A_ATOM] == ep->pred->val[A_ATOM].", "system_requirement": "Derive a valid BPF filter whose compiled CFG reaches fold_edge with matching block opcodes and equal accumulator value numbers, then encode it in fuzz_both's length-prefixed, NUL-terminated wrapper followed by a valid pcap.", "agent_interpretation": "The agent stated the local value-number equality and guessed that a repeated constant-comparison OR filter would create the necessary fold_edge candidate, without deriving its optimizer CFG and state propagation.", "failure_evidence": "The trace never establishes that `0 = 0 or 0 = 0` compiles to the required fold_edge state; it repeatedly guesses that repetition is sufficient. Its wrapper is also malformed and truncates the filter, but even the intended filter lacks a demonstrated optimizer path, so the primary failure is the missing global CFG abstraction; outcome was sfunc=0 and reach=0."}
{"prog": "libpcap_fuzz_both", "src_site_id": 1327, "to_take": 1, "try_id": 2, "source_location": "/src/libpcap/libpcap/gencode.c:2570:6", "format": "pcap", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take true when a register has been allocated for the variable link-header prefix offset: cstate->off_linkhdr.reg != -1.", "system_requirement": "Give fuzz_both a length-prefixed, NUL-terminated payload-accessing filter followed by a valid radiotap pcap (DLT 127), so compilation allocates off_linkpl.reg and then off_linkhdr.reg before gen_load_radiotap_llprefixlen.", "agent_interpretation": "Correctly traced allocation to a variable-prefix link type such as DLT_IEEE802_11_RADIO plus a filter that causes off_linkpl.reg to be allocated.", "failure_evidence": "The trace never inspected fuzz_both.c and generator.py emits only a 92-byte raw radiotap pcap with no filter prefix. fuzz_both interprets its first magic byte 0xd4 as filterSize=212; because Size < 213, it returns at the initial size check, so neither pcap_open_offline nor pcap_compile runs. Recorded sfunc=0, reach=0."}
{"prog": "libpng_read_fuzzer", "src_site_id": 413, "to_take": 0, "try_id": 0, "source_location": "/src/libpng/libpng/pngrutil.c:321:16", "format": "png", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "After png_malloc_base fails in png_read_buffer, take false when warn >= 2; the enclosing buffer==NULL path must be entered.", "system_requirement": "Supply a structurally valid PNG that reaches a warn=2 ancillary handler and forces a reliable allocation failure under the evaluator's memory limits while preserving enough framing to enter png_read_buffer's failure path.", "agent_interpretation": "The agent identified warn=2 and selected pre-IDAT sPLT, but treated a modest 70001-byte request and a truncated chunk as if they guaranteed png_malloc_base to return NULL.", "failure_evidence": "The trajectory never develops a feasible allocation-failure model: 70001 bytes normally succeeds, while the 91-byte file does not even contain its declared 70000-byte sPLT body. Thus it lacks an actionable system-level route to buffer==NULL rather than merely misencoding a sound plan; outcome was sfunc=1 and reach=0."}
{"prog": "libpng_read_fuzzer", "src_site_id": 1330, "to_take": 1, "try_id": 1, "source_location": "/src/libpng/libpng/pngread.c:472:17", "format": "png", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Inside png_read_row pass 1, take true when row_number mod 8 is nonzero or image width is below 5.", "system_requirement": "Use an Adam7-interlaced PNG so png_set_interlace_handling enables PNG_INTERLACE and png_read_row reaches switch pass 1; then width<5 or a suitable row number makes the condition true.", "agent_interpretation": "Correctly read the Boolean expression and chose width=1, but assumed that width alone guarantees evaluation of the target condition.", "failure_evidence": "Source guards the switch with interlaced!=0 and PNG_INTERLACE and the target is case 1, while generator writes IHDR interlace=0. Recorded sfunc=1, reach=0 matches failure to enter the interlaced pass dispatcher."}
{"prog": "libpng_read_fuzzer", "src_site_id": 776, "to_take": 1, "try_id": 2, "source_location": "/src/libpng/libpng/pngrutil.c:2268:13", "format": "png", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "On handling a pre-IDAT pCAL chunk, take true when info_ptr is non-NULL and PNG_INFO_pCAL is already set.", "system_requirement": "Build a valid PNG with IHDR followed by a syntactically valid pCAL whose successful png_set_pCAL call sets the valid bit, then a second pCAL before IDAT.", "agent_interpretation": "Correctly concluded that two valid pre-IDAT pCAL chunks are needed: the first sets PNG_INFO_pCAL and the second is detected as duplicate.", "failure_evidence": "Generator encodes units before type/nparams and sets linear type with nparams=0; parser expects type/nparams before units and requires two parameters for linear, so the first pCAL returns as invalid and never sets PNG_INFO_pCAL. Outcome reach=1 but pass=0 is consistent."}
{"prog": "libxml2_xml", "src_site_id": 7665, "to_take": 0, "try_id": 2, "source_location": "/src/libxml2/libxml2/tree.c:553:9", "format": "xml", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take the false branch when xmlValidateName is invoked with space == 0, so leading blanks are not skipped.", "system_requirement": "Supply fuzz-framed libxml2_xml input that reaches a caller of xmlValidateName with space 0; ordinary parsing alone does not invoke the debug tree checker.", "agent_interpretation": "Correctly identified space == 0 and selected debugXML.c's xmlCtxtCheckName call, which passes 0.", "failure_evidence": "The trace repeatedly assumes any valid XML parsed with an xmllint debug handler will call debugXML.c, but the target is fuzz/xml.c, whose pull, push, and reader paths never call xmlDebugCheckDocument. generator.py emits only a plain XML document; outcome is sfunc=0, reach=0."}
{"prog": "libxml2_xml", "src_site_id": 9756, "to_take": 1, "try_id": 0, "source_location": "/src/libxml2/libxml2/valid.c:4489:9", "format": "xml", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take the true branch when xmlValidateAttributeValueInternal returns 0 for an instance attribute value that violates its declared DTD type.", "system_requirement": "Reach xmlValidateOneAttribute with DTD validation enabled, a retained typed attribute declaration, and an invalid instance value; libxml2_xml parser options are supplied by the fuzz input framing.", "agent_interpretation": "Correctly chose a space-containing value as lexically invalid for NMTOKEN and expected typed-attribute validation to return 0.", "failure_evidence": "The trace never inspects fuzz/xml.c or arranges XML_PARSE_DTDVALID, although SAX2.c calls xmlValidateOneAttribute during parsing only when ctxt->validate is true. It also uses the same invalid NMTOKEN as the DTD default, which valid.c rejects while reading the declaration. The run has sfunc=1 but reach=0."}
{"prog": "libxml2_xml", "src_site_id": 6147, "to_take": 1, "try_id": 1, "source_location": "/src/libxml2/libxml2/parser.c:13256:10", "format": "xml", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "Take the true branch while detaching a successfully parsed chunk element when oldctxt has validation enabled, remains well formed, owns a document with an internal subset, and cur is an element node.", "system_requirement": "Use correctly framed libxml2_xml fuzz input with XML_PARSE_DTDVALID set, plus a well-formed internal DTD and a referenced internal entity expanding to an element.", "agent_interpretation": "The agent correctly enumerated all five conjuncts, required validation, and used an internal general entity whose replacement text contains a child element.", "failure_evidence": "The high-level path is adequate and evaluation reaches the condition, but generator.py emits only XML and does not encode the harness options word with XML_PARSE_DTDVALID. Thus ctxt->validate remains disabled and the concrete fuzz framing fails to realize the stated plan; outcome was sfunc=1, reach=1, pass=0."}
{"prog": "lua", "src_site_id": 430, "to_take": 1, "try_id": 2, "source_location": "/src/lua/lua/lauxlib.c:842:7", "format": "lua", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "After lua_load reads the input stream, ferror(lf.f) must be nonzero because an actual stream read error occurred.", "system_requirement": "Supply an input path that fopen can open and whose stream then produces a real read error during skipcomment/getF; blocking or clean EOF is insufficient.", "agent_interpretation": "The agent correctly identified that fread must set the stream error indicator, but assumed a FIFO with no writer would make fread fail.", "failure_evidence": "The generator only calls os.mkfifo. The trace itself notes that a FIFO without a writer normally blocks, yet still claims it may set ferror; recorded outcome is missing_function with \"no buffer found\" (sfunc=0, reach=0)."}
{"prog": "lua", "src_site_id": 1433, "to_take": 0, "try_id": 0, "source_location": "/src/lua/lua/lapi.c:421:11", "format": "lua", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Reach the non-convertible-value path in lua_tolstring with len == NULL, so the line-421 test is false.", "system_requirement": "A Lua-level operation must call lua_tolstring with NULL while its actual stack argument is non-string and non-number; luaL_tolstring's preprocessing must not replace it with a formatted string first.", "agent_interpretation": "The agent believed tostring({}) passes the table through luaL_tolstring to lua_tolstring with len=NULL and therefore reaches line 421.", "failure_evidence": "lbaselib.c calls luaL_tolstring, whose default case pushes a formatted string and only then calls lua_tolstring(L,-1,len); thus tostring({}) reaches lua_tolstring with a string, not the table. Outcome was success but reach=0 (sfunc=1)."}
{"prog": "lua", "src_site_id": 1012, "to_take": 0, "try_id": 2, "source_location": "/src/lua/lua/lobject.c:684:9", "format": "lua", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "At line 684, source[0] must be '=' and srclen must exceed LUA_IDSIZE (60), making srclen <= bufflen false.", "system_requirement": "Drive luaO_chunkid with a literal chunk identifier beginning '=' and longer than 60 bytes, through a call path that invokes luaO_chunkid on that identifier.", "agent_interpretation": "The agent recognized line 684 is specifically the '=' literal-source case, but asserted that a long '@filename' taking the identical test at line 692 would also satisfy the target.", "failure_evidence": "The generator creates only a 65-character filename, so execution uses the distinct '@' branch and condition at line 692. The trace explicitly notices this mismatch but concludes the same predicate is enough; outcome was success with sfunc=1 but reach=0."}
{"prog": "mbedtls_fuzz_dtlsclient", "src_site_id": 5334, "to_take": 0, "try_id": 1, "source_location": "/src/mbedtls/mbedtls/library/ssl_msg.c:2207:9", "format": "dtls", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The false branch requires ssl_flight_append to be called while ssl->handshake->flight already contains an earlier outgoing DTLS message.", "system_requirement": "Feed the DTLS client a valid server flight through ServerHelloDone so it clears the old ClientHello flight and then emits at least two retained outgoing messages, such as ClientKeyExchange and Finished, in the next flight.", "agent_interpretation": "The agent correctly identified that a second outgoing handshake message in one flight appends to a non-NULL flight and concluded that the supplied complete server-side seed drives such a client response.", "failure_evidence": "Static parsing shows the copied seed is a coherent HVR, ServerHello, Certificate, ServerKeyExchange, ServerHelloDone sequence, so the stated path is adequate; however generator.py exposes only generate_one_input(byte_data) and no evaluator-callable no-argument buffer producer. The recorded outcome is missing_function, \"no buffer found\" (sfunc=0, reach=0), identifying generator integration rather than protocol reasoning as the failure."}
{"prog": "mbedtls_fuzz_dtlsclient", "src_site_id": 4711, "to_take": 1, "try_id": 2, "source_location": "/src/mbedtls/mbedtls/library/ssl_tls12_client.c:1197:9", "format": "dtls", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "After mbedtls_ssl_read_record succeeds inside ssl_parse_server_hello, ssl->in_msgtype must be a content type other than HANDSHAKE (22).", "system_requirement": "Reach the client's SERVER_HELLO state with a non-handshake record that survives mbedtls_ssl_handle_message_type; during the initial handshake, early CCS is buffered, alerts are handled internally, and application data is dropped, so a viable path requires the appropriate enclosing state such as renegotiation.", "agent_interpretation": "The agent correctly read the comparison but assumed an incoming ClientHello record followed by an early ChangeCipherSpec would make ssl_parse_server_hello inspect the second record as non-handshake.", "failure_evidence": "The generator supplies a client-origin ClientHello as server input; ssl_parse_server_hello rejects that handshake before reading the following CCS, and ssl_msg.c would buffer an out-of-order CCS rather than return it to line 1197. Its hand-built DTLS handshake header/version are also invalid; outcome was success with sfunc=1 but reach=0."}
{"prog": "mbedtls_fuzz_dtlsclient", "src_site_id": 4711, "to_take": 1, "try_id": 1, "source_location": "/src/mbedtls/mbedtls/library/ssl_tls12_client.c:1197:9", "format": "dtls", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "After mbedtls_ssl_read_record succeeds inside ssl_parse_server_hello, ssl->in_msgtype must be a content type other than HANDSHAKE (22).", "system_requirement": "Reach the client's SERVER_HELLO state with a non-handshake record that survives mbedtls_ssl_handle_message_type; an out-of-order CCS during the initial handshake is buffered and skipped, so merely changing the first record type cannot expose it at line 1197.", "agent_interpretation": "The agent correctly mapped record byte 0 to in_msgtype and changed the seed's first HVR record from HANDSHAKE to ChangeCipherSpec, expecting that record to reach the conditional directly.", "failure_evidence": "ssl_msg.c returns EARLY_MESSAGE for CCS outside the expected CCS states, buffers it, and continues reading; after replacing the seed's msg_seq-0 HVR, the remaining ServerHello begins at msg_seq 1 and cannot supply the expected current message. The generator additionally prints ASCII hex instead of raw DTLS bytes; outcome was missing_function, \"no buffer found\" (sfunc=0, reach=0)."}
{"prog": "mruby_fuzzer", "src_site_id": 772, "to_take": 0, "try_id": 1, "source_location": "/src/mruby/mruby/src/variable.c:844:7", "format": "ruby", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires v to be neither MRB_TT_CLASS nor MRB_TT_MODULE when mrb_const_set is called.", "system_requirement": "Ruby input must call a path such as Module#const_set or use a qualified constant assignment that dispatches OP_SETMCNST to mrb_const_set with a non-class/module value.", "agent_interpretation": "The agent correctly chose a non-class/module constant value, but assumed an ordinary unqualified Ruby constant assignment invokes mrb_const_set.", "failure_evidence": "The generator emitted `ITEMS = 0...5`; vm.c dispatches ordinary OP_SETCONST to mrb_vm_const_set, which bypasses variable.c:844, while only OP_SETMCNST or Module#const_set calls mrb_const_set. Outcome was sfunc=1, reach=1, pass=0, so incidental initialization reached the site but the generated assignment did not establish the required path."}
{"prog": "mruby_fuzzer", "src_site_id": 2900, "to_take": 1, "try_id": 1, "source_location": "/src/mruby/mruby/src/gc.c:589:7", "format": "ruby", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Immediately after removing a cell from the current free heap page, the true branch requires that page's freelist to have no next free cell.", "system_requirement": "Account for mrb_open and compilation allocations, the actual current free-page occupancy, GC threshold/sweeping, and object liveness; allocate and retain enough objects to consume the final cell of one current free page without GC replenishing its freelist first.", "agent_interpretation": "The agent correctly described consuming the last free cell, but modeled a fresh 1024-cell page and treated 1025 transient empty-array expressions as a deterministic exhaustion count.", "failure_evidence": "generator.py emits 1025 standalone `[]` values without retaining them. The trace never models allocations made by core/gem initialization and compilation or the incremental GC triggered once live exceeds its threshold; those factors invalidate its fixed fresh-page count. Evaluation reached the condition but did not take it (sfunc=1, reach=1, pass=0)."}
{"prog": "mruby_fuzzer", "src_site_id": 2628, "to_take": 1, "try_id": 2, "source_location": "/src/mruby/mruby/src/numeric.c:1483:9", "format": "ruby", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "On the positive-value left-shift path, the true branch requires width > MRB_INT_BIT-1 or val > (MRB_INT_MAX >> width).", "system_requirement": "Execute a positive nonzero Integer left shift with operands exceeding the build's actual mrb_int limit; in this 64-bit MRB_INT64 build, for width 1 val must exceed INT64_MAX>>1, or width can exceed 63.", "agent_interpretation": "The agent saw both 32- and 64-bit definitions but treated the 32-bit threshold as sufficient for the actual MRB_INT64 build, choosing val=1073741824 and width=1.", "failure_evidence": "The trace explicitly noted both width variants yet selected the wrong immediate threshold for the configured MRB_INT64 branch predicate. `1073741824 << 1` is far below INT64_MAX>>1, so the reached condition is false; outcome was sfunc=1, reach=1, pass=0."}
{"prog": "mujs", "src_site_id": 1487, "to_take": 1, "try_id": 1, "source_location": "/src/mujs/mujs/regexp.c:321:7", "format": "javascript", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "While lexclass is parsing a character class, nextrune must reach the end of the regex pattern before an unquoted closing ']', making g->yychar equal EOF.", "system_requirement": "Deliver an unterminated class pattern to js_regcompx, for example through the RegExp constructor with a string; an unterminated regex literal is rejected earlier by the JavaScript lexer.", "agent_interpretation": "Correctly identified an unterminated regex character class as the immediate condition and emitted the literal /[/.", "failure_evidence": "generator.py emits `var re = /[/;`, but jslex.c:487-489 keeps scanning while inclass and raises \"regular expression not terminated\" at source EOF, before js_newregexpx calls js_regcompx. Recorded outcome is success with sfunc=0 and reach=0."}
{"prog": "mujs", "src_site_id": 185, "to_take": 1, "try_id": 1, "source_location": "/src/mujs/mujs/jsfunction.c:18:6", "format": "javascript", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "name_error", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "In jsB_Function, js_gettop(J) must exceed 2, meaning the call frame contains this plus at least two supplied Function-constructor arguments.", "system_requirement": "Generate an executable JavaScript input that invokes the global Function constructor with at least two arguments; the chosen three-argument call is adequate.", "agent_interpretation": "Correctly chose `new Function('a', 'b', 'return a + b')`, whose frame has top=4 and takes the true branch.", "failure_evidence": "The branch and JavaScript plan are correct, but generator.py computes its output path from `__file__`; the recorded evaluator context does not define that name, producing `name_error: name '__file__' is not defined` before any input is emitted (sfunc=0, reach=0)."}
{"prog": "mujs", "src_site_id": 1487, "to_take": 1, "try_id": 2, "source_location": "/src/mujs/mujs/regexp.c:321:7", "format": "javascript", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "While lexclass is parsing a character class, nextrune must reach the end of the regex pattern before an unquoted closing ']', making g->yychar equal EOF.", "system_requirement": "Pass a pattern string containing an unterminated class into RegExp/js_newregexpx so js_regcompx reaches lexclass; malformed regex-literal source is intercepted by the outer JavaScript lexer.", "agent_interpretation": "Correctly identified an unterminated character class and emitted the regex literal /[abc/.", "failure_evidence": "The trace studies js_regcompx but never inspects jslex.c's literal scanner. generator.py emits `var re = /[abc/;`; jslex.c:487-489 remains inclass through the slash and errors at source EOF, so regexp.c is never entered. Recorded outcome is success with sfunc=0 and reach=0."}
{"prog": "nm-new", "src_site_id": 900, "to_take": 0, "try_id": 1, "source_location": "/src/binutils/binutils-2.44/libiberty/getpwd.c:92:10", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires getcwd to fail with errno equal to ERANGE after the PWD shortcut fails, so the loop retries with a larger buffer.", "system_requirement": "nm-new must actually run from a directory whose canonical path exceeds GUESSPATHLEN, with PWD absent or invalid so getpwd reaches getcwd; merely placing the input under a long pathname does not change cwd.", "agent_interpretation": "The agent correctly identified e == ERANGE and proposed making the process working directory longer than the initial getcwd buffer.", "failure_evidence": "The trace explicitly doubts that the harness will run nm from the created directory, yet the final generator only writes input.elf under a 150-character child directory and cannot set nm-new's cwd. Outcome was missing_function, sfunc=0, reach=0, error=\"no buffer found\"."}
{"prog": "nm-new", "src_site_id": 9134, "to_take": 0, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/bfd/cache.c:75:7", "format": "elf", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires max_open_files to be nonzero on a later invocation of bfd_cache_max_open; the first invocation initializes it from zero.", "system_requirement": "The input must make one nm-new process invoke bfd_cache_max_open at least twice, such as by causing multiple BFD cache initializations or a close-and-reopen path; one ordinary file open only establishes the first true evaluation.", "agent_interpretation": "The agent understood that an earlier call must initialize max_open_files, but assumed processing any valid ELF would automatically provide a subsequent target invocation.", "failure_evidence": "The generator emits one minimal ELF and encodes no archive, auxiliary-file, or reopen mechanism. Source shows bfd_openr calls bfd_cache_init once for that BFD; the trace never derives a concrete second call. Outcome was sfunc=1, reach=1 but the requested false branch was not taken."}
{"prog": "nm-new", "src_site_id": 3931, "to_take": 0, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/bfd/elf.c:831:11", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "Within the SHF_ALLOC block, the false branch requires hdr->sh_type == SHT_NOBITS, so SEC_LOAD is not added.", "system_requirement": "Provide a parseable ELF64 object whose section table and string table lead BFD to construct the SHF_ALLOC SHT_NOBITS section and evaluate the target conditional.", "agent_interpretation": "The agent correctly planned a valid ELF64 relocatable object containing an allocated, writable .bss section with sh_type SHT_NOBITS (8).", "failure_evidence": "The concrete ELF64 header uses '<HHQQQIHHHHHHH': e_version is wrongly 8 bytes, e_shoff only 4 bytes, and e_flags only 2 bytes. The resulting bytes put e_shoff=94 at offset 44 instead of 40 and report nonsensical standard fields; the agent's validity check never ran because `file` was missing. Outcome was sfunc=0, reach=0."}
{"prog": "objdump", "src_site_id": 5551, "to_take": 1, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/libctf/ctf-serialize.c:1099:7", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires realloc(buf, buf_size + strtab->cts_len) to return NULL during ctf_serialize.", "system_requirement": "The input must make objdump open and dump a valid .ctf section through a path that actually invokes ctf_serialize, then create enough valid dictionary/type/string state for the final realloc rather than an earlier operation to fail.", "agent_interpretation": "The agent correctly identified an allocation failure as necessary, but assumed large cth_stroff and cth_strlen values from an input CTF header would directly enlarge ctf_serialize's allocation.", "failure_evidence": "objdump's CTF path calls ctf_bfdopen_ctfsect and ctf_dump, not ctf_serialize; moreover ctf_serialize rebuilds hdr from dictionary state and sets hdr.cth_strlen=0, so the crafted input header values do not control buf_size as claimed. The ELF has no section headers exposing .ctf; outcome was sfunc=0, reach=0."}
{"prog": "objdump", "src_site_id": 5411, "to_take": 0, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/bfd/elfcode.h:703:7", "format": "elf", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "unknown", "confidence": "high", "local_condition": "The false branch requires the internal ELF header's e_shnum to equal zero when line 703 is evaluated.", "system_requirement": "Provide an ELF accepted by a matching BFD backend with internal e_shnum still zero after header conversion and all preceding elf_object_p validation.", "agent_interpretation": "The agent correctly targeted e_shnum=0 in a sectionless ELF64 file, although it incorrectly called offset 0x30 the e_shnum offset and verified that unrelated location.", "failure_evidence": "Despite the mistaken offset explanation, the emitted 64-byte ELF also has zero at the real ELF64 e_shnum offset 0x3c; ET_NONE/EM_NONE is compatible with the configured generic elf64-little backend, and the preceding source checks appear satisfiable. The recorded sfunc=1, reach=0, exec_status=success does not reveal why line 703 was missed, so no failure stage is reliably supported."}
{"prog": "objdump", "src_site_id": 5419, "to_take": 0, "try_id": 0, "source_location": "/src/binutils/binutils-2.44/bfd/merge.c:657:7", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires either entsize==align; or, below align, a power-of-two entsize with SEC_STRINGS; or, above align, an entsize that is a multiple of align.", "system_requirement": "Execution must enter the ELF linker merge pipeline: a non-dynamic ELF input SEC_MERGE section must be assigned an output section and passed by bfd_elf_merge_sections to _bfd_add_merge_section.", "agent_interpretation": "The agent correctly chose the sufficient case entsize=align=1 and attempted an ELF section with merge/string flags, sh_entsize=1, and sh_addralign=1.", "failure_evidence": "The trace never inspected callers: _bfd_add_merge_section is called from bfd/elflink.c during linking, while objdump only opens and displays files and does not run bfd_elf_merge_sections. Its structurally broken ELF also declares ELFCLASS64 while encoding ELF32 headers, but the missing objdump-to-linker path already explains sfunc=0, reach=0."}
{"prog": "openh264_decoder_fuzzer", "src_site_id": 775, "to_take": 1, "try_id": 2, "source_location": "/src/openh264/openh264/codec/decoder/core/src/decode_slice.cpp:2970:15", "format": "h264", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "With uiCbpC == 2 in CAVLC macroblock decoding, take true when a chroma-AC WelsResidualBlockCavlc call returns any non-ERR_NONE parsing error.", "system_requirement": "Use a valid CAVLC SPS/PPS/slice and macroblock whose coded block pattern gives uiCbpC=2, then precisely encode or alter a chroma-AC token so earlier syntax succeeds and that call reports an error.", "agent_interpretation": "Correctly identified the chroma-AC gate, preceding luma/chroma-DC parsing, and the CAVLC error returns, planning to preserve a valid stream and corrupt only a chroma-AC coeff_token.", "failure_evidence": "generator.py merely assumes byte offset 0x80 is chroma AC and overwrites every later byte of the 9125-byte, multi-slice Static.264 with 0xFF; no macroblock or bit offset is parsed, so this destroys later syntax rather than realizing the stated targeted mutation. Outcome: missing_function, sfunc=0, reach=0, error=\"no buffer found\"."}
{"prog": "openh264_decoder_fuzzer", "src_site_id": 2593, "to_take": 1, "try_id": 1, "source_location": "/src/openh264/openh264/codec/decoder/core/src/au_parser.cpp:1130:9", "format": "h264", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "While parsing an SPS with frame_cropping_flag=1, take true when frame_crop_top_offset + frame_crop_bottom_offset exceeds iMbHeight * 8.", "system_requirement": "Emit one valid RBSP bitstream containing all preceding SPS fields contiguously, with frame_mbs_only_flag=1, frame_cropping_flag=1, horizontal offsets within bounds, and vertical offsets summing above eight.", "agent_interpretation": "Correctly chose iMbHeight=1 and top=bottom=5, giving 10>8, and intended a baseline SPS carrying those fields.", "failure_evidence": "The plan is adequate, but generator.py emits every ue(v) and one-bit field as separately padded bytes (for example consecutive ue(0) values as 0x01 bytes and gaps flag as 0x00), so the bit parser consumes padding as subsequent fields; write_ue also omits the required leading-zero prefix for nonzero values. ParseSps runs (sfunc=1) but the target is not reached (reach=0)."}
{"prog": "openh264_decoder_fuzzer", "src_site_id": 2076, "to_take": 1, "try_id": 2, "source_location": "/src/openh264/openh264/codec/decoder/core/src/decode_slice.cpp:1088:7", "format": "h264", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "At the end of a macroblock in WelsDecodeMbCabacPSliceBaseMode0, take true when DecodeTerminateCabac decodes the end_of_slice_flag as 1.", "system_requirement": "Provide valid SPS and PPS selecting CABAC, plus a valid P-slice and macroblock syntax that dispatch to WelsDecodeMbCabacPSliceBaseMode0; only then must the CABAC arithmetic state encode a terminating bin of 1.", "agent_interpretation": "Correctly traced uiEosFlag through ParseEndOfSliceCabac to the CABAC terminate comparison, but assumed placing many 0xFF bytes after an ad hoc IDR header would directly create the required CABAC state.", "failure_evidence": "The trace never derives the parser/dispatcher path or required SPS/PPS and P-slice structure. generator.py emits only an IDR NAL with guessed bytes and 150 bytes of 0xFF, so no parameter sets select CABAC or make the target P-slice macroblock function callable; outcome is sfunc=0, reach=0."}
{"prog": "openthread_ot-ip6-send-fuzzer", "src_site_id": 4696, "to_take": 0, "try_id": 1, "source_location": "/src/openthread/openthread/src/core/thread/lowpan.cpp:371:14", "format": "openthread", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "After the preceding multicast test is false, the false branch requires an IPv6 destination outside fe80::/10.", "system_requirement": "Provide data[1:] as a valid IPv6 packet whose payload-length field exactly matches the bytes after its 40-byte header and whose destination is non-link-local, so SendRaw accepts it and forwarding reaches Lowpan::Compress.", "agent_interpretation": "The agent correctly chose non-multicast destination 2001:db8::2 and correctly modeled the fuzzer input as one settings byte followed by a complete IPv6 datagram.", "failure_evidence": "generator.py builds udp_header and computes IPv6 payload length 16, but returns settings+IPv6 header+payload, omitting the 8-byte UDP header; the emitted message is 48 bytes while its header declares 56, so Header::ParseFrom rejects it. Outcome was sfunc=1, reach=0, pass=0."}
{"prog": "openthread_ot-ip6-send-fuzzer", "src_site_id": 4609, "to_take": 1, "try_id": 1, "source_location": "/src/openthread/openthread/src/core/thread/mesh_forwarder.cpp:970:13", "format": "openthread", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires aAddFragHeader to be true or the post-compression payload length to exceed the frame builder's remaining capacity.", "system_requirement": "In this single-radio build aAddFragHeader remains false because its assignment is under OPENTHREAD_CONFIG_MULTI_RADIO; the input therefore needs a valid forwardable IPv6 datagram whose remaining payload is large enough to require 6LoWPAN fragmentation.", "agent_interpretation": "The agent correctly read the OR condition but concluded that setting data[0] bit 0 always makes aAddFragHeader true through link security.", "failure_evidence": "The trace noticed that addFragHeader is set only under OPENTHREAD_CONFIG_MULTI_RADIO yet still relied solely on link security. radio_link.h defaults IEEE 802.15.4=1 and TREL=0, deriving MULTI_RADIO=0; the generated packet is also only 55 message bytes and declares payload length 40 despite carrying 15 bytes. Outcome reached the condition but missed the true branch (sfunc=1, reach=1, pass=0)."}
{"prog": "openthread_ot-ip6-send-fuzzer", "src_site_id": 4696, "to_take": 0, "try_id": 0, "source_location": "/src/openthread/openthread/src/core/thread/lowpan.cpp:371:14", "format": "openthread", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "missing_function", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "After the preceding multicast test is false, the false branch requires an IPv6 destination outside fe80::/10.", "system_requirement": "Place the 40-byte IPv6 header at data[1:41], including payload length at packet offsets 4-5 and destination at packet offsets 24-39, followed by exactly that many payload bytes.", "agent_interpretation": "The agent correctly identified a complete IPv6 packet after the settings byte and planned destination 2000::1, with a matching payload-length field.", "failure_evidence": "generator.py writes payload length to file offsets 4-5 and destination to 24-39 instead of adding the settings-byte offset; this corrupts the IPv6 flow-label/payload-length fields and makes the actual destination start with 00. The generator interface also requires output_dir and filename, and evaluation reported missing_function, \"no buffer found\" (sfunc=0, reach=0)."}
{"prog": "pdftotext", "src_site_id": 782, "to_take": 1, "try_id": 1, "source_location": "/src/xpdf/xpdf-4.05/fofi/FoFiBase.cc:74:7", "format": "pdf", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "Take the true branch when getU8 receives pos < 0 or pos >= the FoFi font buffer length.", "system_requirement": "Embed a Type 1 font that pdftotext loads through FoFiType1 and whose parser reaches a getU8 call with an invalid offset despite all enclosing PFB bounds checks.", "agent_interpretation": "The agent saw the immediately preceding checkRegion(pos1, segLen) guard but still claimed its failure would advance pos1 beyond the buffer and cause a later out-of-range getU8.", "failure_evidence": "FoFiType1::undoPFB breaks before copying or advancing pos1 when checkRegion fails. The trace directly misread this adjacent control flow and asserted pos1 becomes 1006; evaluation reached the target condition elsewhere but did not take the requested edge (sfunc=1, reach=1, pass=0)."}
{"prog": "pdftotext", "src_site_id": 1446, "to_take": 0, "try_id": 2, "source_location": "/src/xpdf/xpdf-4.05/xpdf/Stream.cc:159:11", "format": "pdf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Within an array-valued stream Filter entry, take the false branch when an array element is not a PDF Name object.", "system_requirement": "Create an actual PDF stream with /Filter [<non-Name>] and put it on a path pdftotext consumes, such as a referenced page content stream or an XObject invoked by Do, so Parser::makeStream calls addFilters.", "agent_interpretation": "The agent correctly identified the non-Name element, but assumed an image-like dictionary merely placed in page Resources would be parsed as a filtered stream.", "failure_evidence": "The generator uses DictionaryObject rather than StreamObject, so input.pdf has no stream/endstream token; it also gives the page no Contents and never invokes /Im0 with Do. The required stream-consumption path is absent, matching recorded exec_status=success, sfunc=0, reach=0, pass=0."}
{"prog": "pdftotext", "src_site_id": 3774, "to_take": 1, "try_id": 1, "source_location": "/src/xpdf/xpdf-4.05/xpdf/GfxState.cc:4885:7", "format": "pdf", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take the true branch when the transformed current path's xMax is less than the current clipXMax.", "system_requirement": "Parse a content path with xMax below the current clipping bound, issue W or W*, and then terminate or paint the path so Gfx::doEndPath invokes GfxState::clip before the path is discarded.", "agent_interpretation": "The agent correctly chose a rectangle with transformed xMax 500 below the initial clipXMax 612, but believed the PDF W operator itself immediately triggers GfxState::clip().", "failure_evidence": "Source shows opClip only sets a pending clip flag; GfxState::clip is called later by doEndPath. The generated content is 'q ... re W Q' with no n or painting operator, so the pending clip is never applied; recorded exec_status=success, sfunc=1, reach=0, pass=0."}
{"prog": "php_php-fuzz-parser", "src_site_id": 2436, "to_take": 0, "try_id": 0, "source_location": "/src/php/php-src/ext/pcre/pcre2lib/pcre2_compile.c:10826:13", "format": "php", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "After a nonzero OP_RECURSE target misses the recurse cache, the false branch requires PRIV(find_bracket) to return the compiled capturing group rather than NULL.", "system_requirement": "Compile a PHP preg_match pattern containing a defined capturing group and a valid numbered recursion/subroutine reference to it, causing PCRE2's post-compile recursion fixup to scan OP_RECURSE and resolve the group.", "agent_interpretation": "The agent correctly derived that a valid numbered subroutine call such as (a)(?1) creates OP_RECURSE for group 1 and lets find_bracket resolve its capturing bracket.", "failure_evidence": "The branch and PCRE path were modeled correctly, but generator.py uses random.choice over candidates it claims all work; `(?(1)a|b)` contains no OP_RECURSE, and `(?:(?1))` references nonexistent group 1. A single evaluator invocation can therefore emit an input that never enters the target condition; recorded outcome was success with sfunc=1, reach=0."}
{"prog": "php_php-fuzz-parser", "src_site_id": 676, "to_take": 0, "try_id": 0, "source_location": "/src/php/php-src/ext/date/php_date.c:2617:3", "format": "php", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "The requested continuing/false edge of Z_PARAM_OBJECT_OF_CLASS_OR_NULL requires the optional third argument to be null or a DateTimeZone object, so parameter parsing does not take its failure break.", "system_requirement": "Call date_create_immutable_from_format or DateTimeImmutable::createFromFormat with a path-safe second string and an explicit null or DateTimeZone third argument, reaching and passing the optional timezone type check.", "agent_interpretation": "Despite locating line 2617 as the optional timezone macro, the agent replaced the target with the unrelated line-2621 php_date_initialize condition and tried to make date parsing fail using an invalid second argument.", "failure_evidence": "The final generator supplies only two arguments and targets parse failure at line 2621; it never supplies the optional third argument whose macro is the sampled site. Evaluation reached line 2617 but did not take the requested direction (sfunc=1, reach=1, pass=0)."}
{"prog": "php_php-fuzz-parser", "src_site_id": 64, "to_take": 1, "try_id": 0, "source_location": "/src/php/php-src/ext/date/php_date.c:2615:3", "format": "php", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The successful branch of Z_PARAM_PATH requires the second argument to parse as a string and contain no embedded NUL byte.", "system_requirement": "The evaluated entry program must execute zif_date_create_immutable_from_format; PHP source that merely compiles under fuzzer-parser with execute=0 cannot reach this runtime parameter macro.", "agent_interpretation": "The agent correctly expanded Z_PARAM_PATH and chose a NUL-free second string, but assumed the parser-only fuzz harness executes the generated DateTimeImmutable call.", "failure_evidence": "The fuzzer-parser harness sets execute=0 and only compiles the supplied PHP. The generated runtime calls are never executed, so the trace's locally correct argument reasoning lacks an entry-to-target execution model; recorded sfunc=1 and reach=0 are consistent with incidental harness activity rather than the generated call."}
{"prog": "proj4_proj_crs_to_crs_fuzzer", "src_site_id": 11269, "to_take": 1, "try_id": 1, "source_location": "/src/proj4/PROJ/src/projections/aea.cpp:168:9", "format": "proj", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "During AEA setup, take true when the parsed second standard parallel satisfies |Q->phi2| > pi/2.", "system_requirement": "Provide the fuzzer's required source-CRS/newline/destination-CRS blob, with an AEA CRS on one side whose user-facing lat_2 angle parses to more than 90 degrees.", "agent_interpretation": "The agent correctly stated the internal inequality, but believed the rlat_2 accessor meant the PROJ text must supply radians and emitted pi/2+0.1 as lat_2.", "failure_evidence": "The agent never inspected the fuzzer protocol and generator.py emits only one line, so the harness returns at its missing-newline check. It also emits lat_2=1.6708, while param.cpp defines type r as degree/DMS input converted to radians, making Q->phi2 about 0.029 rad rather than >pi/2; outcome sfunc=0, reach=0."}
{"prog": "proj4_proj_crs_to_crs_fuzzer", "src_site_id": 2, "to_take": 1, "try_id": 1, "source_location": "/src/proj4/PROJ/include/proj/internal/vendor/nlohmann/json.hpp:10474:17", "format": "proj", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "After successfully parsing a JSON value, take true when the current enclosing parser state is a nonempty array, so states.back() is true.", "system_requirement": "Keep the complete fuzz blob at most 1000 bytes and encode source and destination CRS as two lines, with at least one accepted CRS parsed as JSON containing a nonempty array.", "agent_interpretation": "The agent correctly identified true as array context and built a PROJJSON object containing several nonempty arrays.", "failure_evidence": "The trajectory never inspected proj_crs_to_crs_fuzzer.cpp: it validates only standalone JSON. generator.py emits one compact JSON document rather than a source/destination pair, and preserved outputs are about 1273 bytes although the harness returns immediately above 1000 bytes; recorded reach=0."}
{"prog": "proj4_proj_crs_to_crs_fuzzer", "src_site_id": 795, "to_take": 1, "try_id": 0, "source_location": "/src/proj4/PROJ/src/filemanager.cpp:1149:13", "format": "proj", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "With an initially empty context directory, take true when getenv(\"PROJ_USER_WRITABLE_DIRECTORY\") returns a non-NULL, nonempty string.", "system_requirement": "Launch the target process with PROJ_USER_WRITABLE_DIRECTORY set nonempty and supply a two-line source/destination CRS input that reaches a path calling proj_context_get_user_writable_directory while the context field is empty.", "agent_interpretation": "The agent correctly recognized that the target process must start with a nonempty PROJ_USER_WRITABLE_DIRECTORY and that changing os.environ inside a separately run generator cannot affect it.", "failure_evidence": "After recognizing the condition is controlled outside input bytes, the agent merely asserted that the harness would set the environment and removed os.environ from generator.py. The final generator also emits one line, while the fuzzer returns without proj_create_crs_to_crs when no newline exists; outcome sfunc=0, reach=0."}
{"prog": "re2_fuzzer", "src_site_id": 745, "to_take": 1, "try_id": 1, "source_location": "/src/re2/re2/re2/parse.cc:2146:13", "format": "re2", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "While parsing Perl flags, consume '-' before 's' so negated is true in case 's'.", "system_requirement": "Emit a 3-64 byte fuzzer input whose first two bytes are option flags and whose remaining bytes contain a Perl-enabled regex such as (?-s).", "agent_interpretation": "Correctly identified (?-s). as a regex that sets negated before processing the s flag.", "failure_evidence": "Trace never inspected target.cc and generator wrote only b'(?-s).' to a fixed file. target.cc strips bytes 0-1 as flags, so the parsed pattern began '-s).' and outcomes were sfunc=0, reach=0."}
{"prog": "re2_fuzzer", "src_site_id": 735, "to_take": 1, "try_id": 2, "source_location": "/src/re2/re2/re2/parse.cc:2082:7", "format": "re2", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "After ParsePerlFlags removes '(?', the remaining string must have length >2 and begin 'P<'.", "system_requirement": "Prefix the regex with the two option bytes consumed by target.cc, keep PerlX enabled, and emit the complete binary input on stdout.", "agent_interpretation": "Correctly chose the Python named-capture regex (?P<test>.*), which would satisfy the immediate condition if passed as the pattern.", "failure_evidence": "Trace read only parse.cc and generator printed b'(?P<test>.*)\\n' without the two-byte header. target.cc therefore treated '(?' as flags and parsed b'P<test>.*)\\n', which never enters ParsePerlFlags; sfunc=0 and reach=0."}
{"prog": "re2_fuzzer", "src_site_id": 853, "to_take": 1, "try_id": 1, "source_location": "/src/re2/re2/re2/onepass.cc:560:15", "format": "re2", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "Within one IsOnePass flood from a state, encounter kInstMatch twice through input-free transitions so matched is already true on the second encounter.", "system_requirement": "Provide two option bytes plus a regex whose compiled Prog has two input-free routes to match within the same flood, while avoiding earlier one-pass failure checks.", "agent_interpretation": "Claimed a|ab gives two input-free paths after the shared 'a': one to match and one through the byte-consuming 'b' to another match.", "failure_evidence": "The claimed route through 'b' is not input-free: kInstByteRange schedules its out-state as a separate node, where line 435 resets matched=false. The input reached the condition (sfunc=1, reach=1) but did not take True."}
{"prog": "readelf", "src_site_id": 6547, "to_take": 1, "try_id": 1, "source_location": "/src/binutils/binutils-2.44/binutils/readelf.c:14624:7", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires the current ELF class to be anything other than ELFCLASS64, normally ELFCLASS32.", "system_requirement": "Provide an ELF32 file with a usable SHT_SYMTAB or SHT_DYNSYM section so symbol-table processing calls print_symbol_table_heading under the evaluator's symbol-display path.", "agent_interpretation": "The agent correctly traced is_32bit_elf to EI_CLASS and emitted a minimal ELF32 header, but assumed ELF32 classification alone would execute the heading branch.", "failure_evidence": "generator.py emits only a 52-byte ELF32 header with e_shoff=0 and e_shnum=0. Source calls print_symbol_table_heading only while displaying dynamic symbols or dumping a symbol section; the trace never modeled that dispatcher requirement. Outcome was exec_status=success, sfunc=0, reach=0."}
{"prog": "readelf", "src_site_id": 5738, "to_take": 1, "try_id": 0, "source_location": "/src/binutils/binutils-2.44/binutils/readelf.c:942:8", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "While printable_string handles a byte that is neither control nor printable, the true branch requires fewer than four output-buffer bytes remaining.", "system_requirement": "Create a valid ELF section-header string table and a section whose sh_name selects that 254-byte name, so a section-display path calls printable_section_name and then printable_string.", "agent_interpretation": "The agent correctly planned a displayed section name containing 253 printable bytes followed by 0x80, leaving remaining=3 at the target check.", "failure_evidence": "The final generator inserts an erroneous 16-byte pad after the already complete 64-byte ELF header, while e_shoff remains 64; readelf therefore interprets shifted section headers as NULL and has no string table. The trace's final output shows all names as <no-strings>; outcome was exec_status=success, sfunc=0, reach=0."}
{"prog": "readelf", "src_site_id": 3514, "to_take": 0, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/binutils/readelf.c:21495:12", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "In a GNU property note entry of type GNU_PROPERTY_STACK_SIZE, the false branch requires datasz to equal the ELF word size: 4 for ELF32 or 8 for ELF64.", "system_requirement": "Under a notes-display path, a PT_NOTE or SHT_NOTE must contain an outer ELF note header with owner \"GNU\" and type NT_GNU_PROPERTY_TYPE_0; its descriptor must then contain the aligned GNU_PROPERTY_STACK_SIZE entry.", "agent_interpretation": "The agent correctly chose ELF64, property type 1, datasz 8, and an eight-byte stack-size payload, but treated that property entry itself as the complete PT_NOTE contents.", "failure_evidence": "At PT_NOTE offset 0x1000, generator.py writes only `type=1,datasz=8,value=0x1000`, omitting the outer namesz/descsz/type header and GNU owner needed for dispatch to print_gnu_property_note. The trace never derives this note-container requirement; outcome was exec_status=success, sfunc=0, reach=0."}
{"prog": "size", "src_site_id": 4979, "to_take": 1, "try_id": 0, "source_location": "/src/binutils/binutils-2.44/bfd/elfcore.h:382:11", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires a program header parsed by core_find_build_id to have p_type == PT_NOTE and p_filesz > 0.", "system_requirement": "Supply an ELF core whose PT_LOAD segment begins with a valid nested ELF image; bfd_section_from_phdr calls core_find_build_id at that PT_LOAD file offset, where the nested image must contain a nonempty PT_NOTE program header.", "agent_interpretation": "The agent correctly set a top-level ELF program header to PT_NOTE with p_filesz=32 and verified that header with readelf.", "failure_evidence": "elf.c invokes core_find_build_id only for a core file's PT_LOAD p_offset, but the generator emits an ET_EXEC whose only top-level header is PT_NOTE and never models the outer-core/nested-ELF path. Outcome was sfunc=0, reach=0 despite exec_status=success."}
{"prog": "size", "src_site_id": 776, "to_take": 1, "try_id": 1, "source_location": "/src/binutils/binutils-2.44/libiberty/objalloc.c:244:8", "format": "elf", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "Inside the small-chunk case, small is non-NULL when traversal from the newest chunk passes a different small-object chunk before reaching the older small chunk containing block.", "system_requirement": "Drive size through a bfd_release/objalloc_free_block path after allocations create at least two small chunks, then release a block in an older small chunk so a newer small chunk is encountered first.", "agent_interpretation": "The agent repeatedly claimed that freeing from a later or second small chunk would leave a previous or first small chunk in small, reversing the head-insertion order of objalloc chunks.", "failure_evidence": "objalloc_alloc prepends each new chunk at o->chunks, so freeing from the newer second small chunk finds it immediately and leaves small=NULL, contrary to the stated plan. The generator then emits only a generic one-segment ELF and admits it does not control objalloc usage; outcome was sfunc=1, reach=0."}
{"prog": "size", "src_site_id": 2256, "to_take": 1, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/bfd/coffcode.h:1059:8", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires entry->symname, saved from the first qualifying COFF symbol for a section number, to contain '$'.", "system_requirement": "Provide a PE/COFF COMDAT section whose symbol table has a first qualifying section symbol with '$' in its name and then another symbol with the same section number, causing fill_comdat_hash to enter the existing-slot branch and evaluate target_name.", "agent_interpretation": "The agent reduced the requirement to one COMDAT symbol named $comdat and asserted that strchr would inspect that symbol directly.", "failure_evidence": "coffcode.h evaluates the condition only in the existing-slot branch after a prior same-section symbol created the hash entry; generator.py declares only one symbol, with no section-symbol auxiliary record, so that symbol can only create the entry and cannot reach line 1059. Outcome was sfunc=0, reach=0."}
{"prog": "sndfile_fuzzer", "src_site_id": 2248, "to_take": 1, "try_id": 2, "source_location": "/src/libsndfile/libsndfile/src/htk.c:71:7", "format": "aifc,aiff,au,caf,mp3,oga,opus,paf,raw,rf64,sds,wav,xi", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "htk_write_header(psf, SF_FALSE) must return nonzero, which means psf->error is nonzero after header writing.", "system_requirement": "Execution must enter htk_open with mode SFM_WRITE or SFM_RDWR; this fuzzer unconditionally calls sf_open_virtual(..., SFM_READ, ...), so input bytes cannot reach the write-only conditional.", "agent_interpretation": "The agent correctly identified a nonzero error return, but assumed the fuzzer could open or convert an HTK file in SFM_WRITE and tried to induce a zero-samplerate write error.", "failure_evidence": "The trace repeatedly speculates about conversion/write mode and the generator emits an HTK input, while sndfile_fuzzer.cc:120 fixes SFM_READ and htk.c:65 guards the target with WRITE/RDWR. Outcome sfunc=1, reach=0 is consistent with entering htk_open only on its read path."}
{"prog": "sndfile_fuzzer", "src_site_id": 798, "to_take": 1, "try_id": 1, "source_location": "/src/libsndfile/libsndfile/src/double64.c:305:6", "format": "aifc,aiff,au,caf,mp3,oga,opus,paf,raw,rf64,sds,wav,xi", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "In double64_be_read, cptr[0] must have bit 0x80 set and the exponent/mantissa must not all be zero, so execution reaches if (negative) with negative=1.", "system_requirement": "Provide a valid CAF whose desc selects big-endian 64-bit linear PCM and whose data chunk contains a negative double, so sf_readf_float invokes the double decoder after header parsing.", "agent_interpretation": "The agent correctly chose a big-endian negative double (-1.0, first byte 0xBF) and planned a CAF carrying 64-bit floating-point samples that the fuzzer would read.", "failure_evidence": "The concrete CAF encoding is wrong: it uses unsupported format ID 'fl64' instead of 'lpcm' with floating flag 1, writes 32-bit rather than CAF 64-bit chunk sizes, and misorders the data chunk fields. The only reached invocation was the positive 44100.0 CAF sample-rate parse (reach=1), so the desired branch was not taken."}
{"prog": "sndfile_fuzzer", "src_site_id": 1299, "to_take": 1, "try_id": 2, "source_location": "/src/libsndfile/libsndfile/src/ALAC/ag_dec.c:355:2", "format": "aifc,aiff,au,caf,mp3,oga,opus,paf,raw,rf64,sds,wav,xi", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "RequireAction(c, action) expands to if (!(c)) { action }; therefore the true edge at this site requires bitstream->cur > bitstream->end after BitBufferAdvance.", "system_requirement": "A valid CAF/ALAC path must reach dyn_decomp and consume enough coded bits that the post-advance cursor exceeds the packet buffer end, taking the macro-expanded error edge.", "agent_interpretation": "After reading the macro definition, the agent still asserted that the requested true branch meant bitstream->cur <= bitstream->end and designed a normal in-bounds decode.", "failure_evidence": "The trace quotes '#define RequireAction(condition, action) if (!(condition))' but repeatedly labels cur<=end as TRUE and cur>end as FALSE; the generator explicitly tries to avoid overflow. Outcome sfunc=0, reach=0, and its numeric pseudo-markers plus missing pakt metadata also never establish an ALAC decode path."}
{"prog": "sqlite3_ossfuzz", "src_site_id": 1683, "to_take": 0, "try_id": 1, "source_location": "/src/sqlite3/sqlite3/build/sqlite3.c:128364:7", "format": "sql", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The false branch requires the database text-encoding meta value meta[BTREE_TEXT_ENCODING-1] to equal zero when sqlite3InitOne reads the schema.", "system_requirement": "Cause sqlite3InitOne to load a schema whose BTREE_TEXT_ENCODING meta value is zero, either through a genuinely uninitialized database state or by setting SQLITE_DBCONFIG_RESET_DATABASE through a supported API path.", "agent_interpretation": "The agent correctly identified the zero-meta condition, but vacillated between an empty new database and SQLITE_ResetDatabase, finally claiming selector byte 4 might enable reset while emitting SELECT 1.", "failure_evidence": "ossfuzz.c uses only selector bit 0 for foreign keys and the shifted remainder as an output-row limit; selector 4 cannot set SQLITE_ResetDatabase. The trace admits no reset PRAGMA exists yet emits 0x04 plus SELECT 1 without establishing how that SQL produces the required schema state. Outcome reached the condition (sfunc=1, reach=1) but took the wrong direction."}
{"prog": "sqlite3_ossfuzz", "src_site_id": 5533, "to_take": 0, "try_id": 0, "source_location": "/src/sqlite3/sqlite3/build/sqlite3.c:81872:9", "format": "sql", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "To evaluate if(x>r) and take it false, execution must first survive if(x<r), so x must equal r (unordered NaN is another theoretical case); choosing x<r returns before the target condition.", "system_requirement": "Execute a non-folded runtime comparison between an integer Mem and a real Mem with equal numeric values so sqlite3IntFloatCompare reaches the second comparison and x>r is false.", "agent_interpretation": "The agent treated the target false branch as any x<=r and chose integer 1 versus real 2.0, repeatedly asserting this would evaluate if(x>r) as false.", "failure_evidence": "Source line 81871 immediately returns when x<r. The generated constant SQL SELECT 1 WHERE 1 < 2.0 targets x<r, so even if sqlite3IntFloatCompare were invoked it would exit before line 81872; recorded sfunc=1, reach=0 matches this local misread."}
{"prog": "sqlite3_ossfuzz", "src_site_id": 3330, "to_take": 1, "try_id": 2, "source_location": "/src/sqlite3/sqlite3/build/sqlite3.c:57665:7", "format": "sql", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The true branch requires pPager->nSavepoint>0 while pager write processing reaches this condition.", "system_requirement": "Enter a transaction first, open a nested SQL savepoint so db->nSavepoint is positive and sqlite3PagerOpenSavepoint propagates it to the pager, then modify a page before releasing the savepoint.", "agent_interpretation": "The agent correctly traced pager savepoints to db->nSavepoint and planned an active non-transaction savepoint followed by a page-modifying UPDATE; it even noted that the first SAVEPOINT under autocommit is only a transaction savepoint.", "failure_evidence": "Despite identifying the autocommit distinction, generator.py emits no BEGIN and only one SAVEPOINT. Source lines 88443-88447 therefore mark it isTransactionSavepoint without incrementing db->nSavepoint, so the subsequent UPDATE cannot realize the planned pager state. Outcome was sfunc=1, reach=1, pass=0."}
{"prog": "stb_stbi_read_fuzzer", "src_site_id": 657, "to_take": 1, "try_id": 1, "source_location": "/src/stb/stb/stb_image.h:2131:8", "format": "bmp,gif,jpg,tga,png", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take the true branch when slow-path Huffman code length k exceeds the valid bits remaining in j->code_buffer.", "system_requirement": "Provide a parseable JPEG whose entropy scan reaches slow Huffman decode and whose buffer refill is stopped by a non-stuffed marker before enough bits are available, while the buffered prefix selects a valid 10-16 bit code.", "agent_interpretation": "Correctly stated k > j->code_bits, but assumed grow_buffer reads only one 0xFF byte and leaves code_bits at 8.", "failure_evidence": "The trace claims grow turns code_bits 0 into 8 on scan byte 0xFF, but stbi__grow_buffer_unsafe treats FF 00 as a stuffed data byte and continues refilling while code_bits <= 24. The generator emits FF 00 00 before EOI, so this model cannot yield its claimed k=16, code_bits=8 state; outcome was sfunc=0, reach=0."}
{"prog": "stb_stbi_read_fuzzer", "src_site_id": 657, "to_take": 1, "try_id": 0, "source_location": "/src/stb/stb/stb_image.h:2131:8", "format": "bmp,gif,jpg,tga,png", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "Take the true branch when slow-path Huffman code length k exceeds the valid bits remaining in j->code_buffer.", "system_requirement": "Provide a parseable JPEG whose entropy scan reaches slow Huffman decode and whose buffer refill is stopped by a non-stuffed marker before enough bits are available, while the buffered prefix selects a valid 10-16 bit code.", "agent_interpretation": "Correctly identified the slow-path comparison, but proposed depleting code_bits below k across repeated decodes and asserted refill might leave it low.", "failure_evidence": "The trace itself notes each decode refills when code_bits < 16, then still concludes a low-bit state can persist and even says 16 > code_bits >= 10 avoids refill. Source line 2100 contradicts that model; the generator merely repeats stuffed FF 00 bytes, which refill normally. Evaluation reached the enclosing function (sfunc=1) but not the conditional (reach=0)."}
{"prog": "stb_stbi_read_fuzzer", "src_site_id": 570, "to_take": 0, "try_id": 1, "source_location": "/src/stb/stb/stb_image.h:3435:14", "format": "bmp,gif,jpg,tga,png", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "Take the false branch when the DNL marker has Ld == 4 and NL equals the SOF-derived image height img_y.", "system_requirement": "Supply a structurally parseable JPEG that passes stbi_info_from_memory, completes entropy parsing, exposes FF DC as the next marker, and encodes DNL NL equal to the SOF height.", "agent_interpretation": "Correctly planned a JPEG with SOF0 height 8 followed after an SOS scan by DNL length 4 and NL 8.", "failure_evidence": "The high-level SOF/SOS/DNL path and equality are correct, but concrete segment encoding is broken: APP0 declares length 16 while carrying only five payload bytes, and AC DHT declares length 181 while the generator emits 178 payload bytes after the length. These length mismatches desynchronize marker parsing before DNL; outcome was sfunc=1, reach=0."}
{"prog": "strip-new", "src_site_id": 7585, "to_take": 1, "try_id": 0, "source_location": "/src/binutils/binutils-2.44/bfd/elf.c:13331:7", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "In elf_read_notes, the true branch requires _bfd_malloc_and_read(abfd, size + 1, size) to return NULL for a nonzero, nonoverflowing note size.", "system_requirement": "Supply an ET_CORE ELF whose PT_NOTE is processed by elf_core_file_p through bfd_section_from_phdr and then fails the note read.", "agent_interpretation": "The agent correctly tied size to PT_NOTE p_filesz and tried to make the read fail by declaring p_filesz larger than the file.", "failure_evidence": "generator.py sets e_type=ET_EXEC. Source calls bfd_section_from_phdr and elf_read_notes while elf_core_file_p processes core program headers; ordinary elf_object_p does not process this PT_NOTE through that path. Thus the oversized p_filesz never reaches the sampled function; recorded sfunc=0, reach=0."}
{"prog": "strip-new", "src_site_id": 4921, "to_take": 1, "try_id": 1, "source_location": "/src/binutils/binutils-2.44/binutils/objcopy.c:2634:11", "format": "elf", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The true branch requires style == KEEP and both output_bfd and input_bfd to have COFF flavour, since the enclosing function first gates on COFF output.", "system_requirement": "Emit the evaluator's expected input.elf as a valid COFF/PE object so strip preserves a COFF output target and copy_object calls set_long_section_mode.", "agent_interpretation": "The agent correctly identified strip's default long_section_names as KEEP and chose a PE/COFF input, with same-format output, to enter set_long_section_mode's inner condition.", "failure_evidence": "The COFF plan and generated PE bytes are adequate, but generator.py writes only input.coff; the required artifact for this format is input.elf, and none exists in the trace directory. Thus the evaluator did not feed the COFF file to strip; recorded sfunc=0, reach=0."}
{"prog": "strip-new", "src_site_id": 8726, "to_take": 0, "try_id": 2, "source_location": "/src/binutils/binutils-2.44/bfd/tekhex.c:210:7", "format": "elf", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The false branch requires a second or later call to tekhex_init in the same process, after its first call sets the function-static inited flag true.", "system_requirement": "Make strip recognize an archive and process multiple valid TekHex member objects so BFD probing invokes tekhex_init again after its static flag is set.", "agent_interpretation": "The agent correctly understood the persistent flag and formed the high-level plan of processing multiple archive members in one strip process to obtain repeated TekHex initialization.", "failure_evidence": "The multi-member archive plan is adequate, but each generated `%00010000` member is only a stub and not a well-formed TekHex object that BFD can recognize and copy. The concrete member encoding therefore realizes only the first observed initialization; outcome was sfunc=1, reach=1, pass=0."}
{"prog": "tcpdump", "src_site_id": 3071, "to_take": 1, "try_id": 2, "source_location": "/src/tcpdump/tcpdump/print-ppp.c:421:6", "format": "pcap", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires handle_ctrl_proto to receive a PPP control-protocol payload length below 4 bytes.", "system_requirement": "The pcap link type must dispatch the record to the PPP printer (for example DLT_PPP), and the PPP bytes must select a control protocol with fewer than four payload bytes.", "agent_interpretation": "The agent correctly chose an LCP frame ff 03 c0 21 followed by one payload byte, reasoning that ppp_print would pass length 1 to handle_ctrl_proto.", "failure_evidence": "generator.py sets the pcap network field to 1 (DLT_EN10MB) but writes bare PPP bytes as the record, so tcpdump dispatches it as Ethernet rather than to ppp_if_print; outcome was exec success but sfunc=0 and reach=0."}
{"prog": "tcpdump", "src_site_id": 2129, "to_take": 0, "try_id": 2, "source_location": "/src/tcpdump/tcpdump/print-mptcp.c:277:13", "format": "pcap", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "import_error", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The false branch requires a valid MP_JOIN case: opt_len 12 with SYN, opt_len 16 with SYN|ACK, or opt_len 24 with ACK.", "system_requirement": "Emit a parseable Ethernet IPv4 TCP packet whose TCP options contain kind 30, subtype MP_JOIN, length 24, while the TCP ACK flag is set.", "agent_interpretation": "The agent correctly targeted a 24-byte MPTCP MP_JOIN option in an ACK TCP segment and described the Ethernet/IP/TCP/MPTCP path.", "failure_evidence": "The final generator depends on `from scapy.all import Ether, IP, Raw, wrpcap`; evaluation failed while importing it with `No module named 'scapy'`, so no candidate was generated for the otherwise adequate packet plan (sfunc=0, reach=0)."}
{"prog": "tcpdump", "src_site_id": 4497, "to_take": 0, "try_id": 2, "source_location": "/src/tcpdump/tcpdump/print-juniper.c:833:13", "format": "pcap", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The false branch requires juniper_parse_header to return nonzero, which occurs after a valid Juniper MLPPP header passes parsing.", "system_requirement": "Use pcap link type DLT_JUNIPER_MLPPP and provide a packet beginning with MGC, acceptable flags, the MLPPP cookie, and enough following data for successful Juniper header parsing.", "agent_interpretation": "After correcting an initial direction mistake, the agent correctly required the MGC magic, no NO_L2 flag, sufficient bytes, and a DLT_JUNIPER_MLPPP pcap record.", "failure_evidence": "generator.py hard-codes PCAP_NETWORK=119 while the repository defines DLT_JUNIPER_MLPPP as 130; 119 is DLT_PRISM_HEADER, so the record never dispatches to juniper_mlppp_if_print. The recorded outcome was sfunc=0 and reach=0."}
{"prog": "tiff2ps", "src_site_id": 2539, "to_take": 0, "try_id": 0, "source_location": "/src/libtiff/libtiff/libtiff/tif_dirread.c:6348:26", "format": "tiff", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "Because this is an else-if after mb + 1 < tdir_count, reaching it and taking its false branch requires mb + 1 == (uint32_t)dp->tdir_count.", "system_requirement": "A recognized ASCII TIFF tag must reach TIFFFetchNormalTag with valid data and have its first terminating NUL exactly at the final declared byte, so mb equals tdir_count - 1.", "agent_interpretation": "The agent repeatedly treated the false branch as any mb + 1 <= tdir_count and described any sufficiently sized null-terminated ASCII value as adequate.", "failure_evidence": "The trace overlooks the preceding mb + 1 < tdir_count arm, so its <= condition includes cases that skip the target else-if. Moreover, the executed Pillow path emits no ASCII IFD entry at all; recorded outcome was sfunc=1, reach=0, pass=0."}
{"prog": "tiff2ps", "src_site_id": 1155, "to_take": 1, "try_id": 2, "source_location": "/src/libtiff/libtiff/libtiff/tif_lzma.c:335:17", "format": "tiff", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true branch requires LZMAEncode to fill its output buffer and TIFFFlushData1 to return 0 because TIFFAppendToStrip fails.", "system_requirement": "The target requires a TIFF handle opened for writing, LZMA encoding via LZMAEncode, a full raw output buffer, and a failing output seek/write; tiff2ps opens the supplied TIFF only in read mode and invokes decoder paths.", "agent_interpretation": "The agent understood the immediate flush failure, but tried to induce the LZMA write path by supplying a TIFF with a huge on-disk strip offset; it later noticed tiff2ps only reads and therefore cannot call LZMAEncode.", "failure_evidence": "tiff2ps.c opens input with TIFFOpenExt(..., \"r\"), while line 335 is in LZMAEncode. The generator instead creates a read-side TIFF and even writes Compression=5, not COMPRESSION_LZMA=34925; outcome was sfunc=0, reach=0 despite exec_status=success."}
{"prog": "tiff2ps", "src_site_id": 2437, "to_take": 1, "try_id": 0, "source_location": "/src/libtiff/libtiff/libtiff/tif_dirread.c:4813:9", "format": "tiff", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "The true branch requires tif->tif_dir.td_compression to equal COMPRESSION_OJPEG, whose TIFF Compression value is 6.", "system_requirement": "A parseable TIFF must carry IFD tag 259 (Compression) with SHORT value 6 so directory reading sets td_compression to COMPRESSION_OJPEG.", "agent_interpretation": "The agent correctly identified value 6 and planned a valid TIFF using OJPEG compression, but assumed Pillow's compression=\"jpeg\" writes that legacy value.", "failure_evidence": "The generated test_ojpeg.tiff is structurally parseable but its IFD Compression tag is value 7 (COMPRESSION_JPEG), not 6; the generator's Pillow API assumption is wrong. The condition was reached but false (sfunc=1, reach=1, pass=0)."}
{"prog": "tiffsplit", "src_site_id": 932, "to_take": 1, "try_id": 2, "source_location": "/src/libtiff/libtiff/libtiff/tif_ojpeg.c:1496:13", "format": "tiff", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "With jpeg_interchange_format nonzero, take true when its offset is greater than or equal to the actual TIFF file size.", "system_requirement": "Provide a sufficiently valid Old-JPEG TIFF whose Compression tag selects COMPRESSION_OJPEG, loads tag 513 into OJPEGState, and causes tiffsplit to enter OJPEG header/decode handling.", "agent_interpretation": "Correctly identified tag 513 as JPEGInterchangeFormat and set its nonzero LONG value beyond EOF.", "failure_evidence": "The trace briefly questioned whether an OJPEG TIFF was required but concluded the one-entry file was valid; generator.py emits only tag 513, omitting Compression=OJPEG and basic image/strile structure. OJPEGState is created only by TIFFInitOJPEG, and the recorded outcome was sfunc=0, reach=0."}
{"prog": "tiffsplit", "src_site_id": 2562, "to_take": 0, "try_id": 1, "source_location": "/src/libtiff/libtiff/libtiff/tif_dirread.c:1061:17", "format": "tiff", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "timeout", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "The enclosing function first requires tdir_count==1; after one TIFF_DOUBLE is read successfully, take false when -FLT_MAX <= m <= FLT_MAX.", "system_requirement": "Use a recognized field whose set/get type is TIFF_SETGET_FLOAT, encoded as TIFF_DOUBLE with exactly one value, inside a TIFF directory valid enough for tiffsplit to read.", "agent_interpretation": "Correctly chose m=1.0 as in range, but generated a three-element DOUBLE entry and treated that as reaching the scalar check.", "failure_evidence": "Source returns TIFFReadDirEntryErrCount before the target switch whenever tdir_count!=1. generator.py sets ModelPixelScale tag 33550 to TIFF_DOUBLE count=3 and writes three 1.0 values, so its own entry cannot reach line 1061; the run timed out with sfunc=0 and reach=0."}
{"prog": "tiffsplit", "src_site_id": 3195, "to_take": 1, "try_id": 1, "source_location": "/src/libtiff/libtiff/libtiff/tif_dirread.c:2470:21", "format": "tiff", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "While converting a TIFF_SLONG8 array to unsigned LONG8 values, take true when a signed element is negative, making err equal TIFFReadDirEntryErrRange.", "system_requirement": "Construct a valid BigTIFF with required image fields and a StripOffsets or TileOffsets SLONG8 entry so TIFFFetchStripThing invokes TIFFReadDirEntryLong8ArrayWithLimit.", "agent_interpretation": "Correctly traced the branch to TIFFReadDirEntryCheckRangeLong8Slong8 and planned a BigTIFF StripOffsets entry of type SLONG8 containing -1.", "failure_evidence": "The trace identified StripOffsets tag 273 as the relevant path, but generator.py emits custom tag 33000 instead. Its BigTIFF header omits the offset-size field (8), sets first IFD offset 32 although the IFD bytes begin at offset 16, uses 20-byte entries while calculating 16, and count=1 stores the SLONG8 inline rather than at offset 80; outcome was sfunc=0, reach=0."}
{"prog": "vorbis_decode_fuzzer", "src_site_id": 1076, "to_take": 1, "try_id": 2, "source_location": "/src/vorbis/vorbis/lib/vorbisfile.c:2030:20", "format": "vorbis", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "In ov_read's 16-bit signed, host-endian path, vorbis_ftoi(src[j] * 32768.f) must exceed 32767 (a decoded PCM sample at or near 1.0, depending on rounding).", "system_requirement": "Supply a valid decodable Ogg/Vorbis stream whose setup and audio packets deterministically synthesize a sufficiently large positive PCM sample, so ov_open_callbacks succeeds and ov_read reaches the host-endian signed packing loop.", "agent_interpretation": "The agent correctly identified the 16-bit clipping check and sought a decoded PCM sample at least about 1.0.", "failure_evidence": "The trace never derived Vorbis packet fields that control decoded amplitude; generator.py instead ffmpeg-encodes max-int16 audio then blindly overwrites offsets 150-349 with 0xff. In the produced file that range is inside the second Ogg page/header area (next page begins only at offset 4204), so this is unguided structural corruption; outcome sfunc=0, reach=0."}
{"prog": "vorbis_decode_fuzzer", "src_site_id": 1139, "to_take": 1, "try_id": 2, "source_location": "/src/vorbis/vorbis/lib/vorbisfile.c:203:6", "format": "vorbis", "pass": 0, "reach": 1, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "_lookup_serialno must be called with serialno_list non-NULL; during initial BOS processing this requires a prior BOS serial number already added before a later lookup, or a seekable lookup path.", "system_requirement": "For this read-only callback fuzzer, provide multiple initial BOS pages/serials so a later _lookup_page_serialno call receives the already allocated list; seekable _open_seekable2 paths cannot run because seek_func and tell_func are NULL.", "agent_interpretation": "The agent correctly read the pointer test but assumed one ordinary BOS Vorbis stream would populate the list before the relevant lookup, and also relied on seekable-file paths.", "failure_evidence": "decode_fuzzer.cc installs only read_func, so vf->seekable stays false. In _fetch_headers the first BOS page calls _lookup_serialno with *serialno_list still NULL and only then _add_serialno allocates it; the generator emits exactly one BOS page, followed by non-BOS pages. The recorded outcome sfunc=1, reach=1 but pass=0 matches reaching the condition only on its false side."}
{"prog": "vorbis_decode_fuzzer", "src_site_id": 392, "to_take": 1, "try_id": 1, "source_location": "/src/vorbis/vorbis/lib/floor0.c:93:8", "format": "vorbis", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "During floor0_unpack, a valid floor0 book index must reference an already unpacked static codebook whose maptype equals 0.", "system_requirement": "Encode valid Ogg pages containing the three Vorbis headers; in the setup packet, encode a valid maptype-0 codebook followed immediately at the same bit position by a valid floor0 definition selecting that book, so parsing reaches floor0.c:93.", "agent_interpretation": "The agent correctly planned one maptype-0 codebook and a floor type 0 configuration whose book list references codebook 0.", "failure_evidence": "The concrete encoding is malformed: write_ogg_page uses 0x01 as BOS and 0x02 as EOS (actual bits are 0x02 and 0x04), writes zero CRCs, and flushes the codebook to a byte boundary before embedding it although the codebook ends mid-byte. It also omits required residue, mapping, and mode fields. Thus the intended setup cannot be parsed; outcome sfunc=0, reach=0."}
{"prog": "woff2_convert_woff2ttf_fuzzer", "src_site_id": 984, "to_take": 0, "try_id": 0, "source_location": "/src/woff2/brotli/dec/decode.c:1503:14", "format": "woff2", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "local_reasoning_failure", "confidence": "high", "local_condition": "The target else-if is evaluated only after s->state != BROTLI_STATE_COMMAND_BEGIN; its false edge then requires s->state != BROTLI_STATE_COMMAND_INNER (normally POST_DECODE_LITERALS or POST_WRAP_COPY).", "system_requirement": "Provide a valid WOFF2/Brotli stream that reaches ProcessCommandsInternal in a command state that passes the preceding COMMAND_BEGIN test but is not COMMAND_INNER.", "agent_interpretation": "The agent treated COMMAND_BEGIN as making the target condition false and built a stream intended to enter ProcessCommandsInternal in COMMAND_BEGIN.", "failure_evidence": "The trace repeatedly claims COMMAND_BEGIN takes the target false branch, but decode.c:1501 jumps directly to CommandBegin before line 1503 is evaluated. Outcome was exec_status=success, sfunc=0, reach=0; the generator also emits little-endian WOFF2 fields."}
{"prog": "woff2_convert_woff2ttf_fuzzer", "src_site_id": 73, "to_take": 1, "try_id": 1, "source_location": "/src/woff2/woff2/src/woff2_dec.cc:1123:11", "format": "woff2", "pass": 0, "reach": 0, "sfunc": 1, "exec_status": "success", "generator_exists": 1, "label": "construction_failure", "confidence": "high", "local_condition": "For a TTC font, the condition is true when the current font references exactly one of the glyf and loca tables.", "system_requirement": "Encode a valid big-endian WOFF2 header and compact table directory, then place the clear TTC version, font count, font flavor, and table-index list immediately after that directory so parsing reaches the XOR check.", "agent_interpretation": "The agent correctly planned a ttcf WOFF2 with one font whose table index list contains glyf but not loca.", "failure_evidence": "The plan matches the source, but generator.py pads after the directory and inserts 12 zero table-data bytes before the TTC header; the artifact therefore presents header_version=0 at the parser's current offset and fails the line 1080 version check before line 1123. Outcome was exec_status=success, sfunc=1, reach=0."}
{"prog": "woff2_convert_woff2ttf_fuzzer", "src_site_id": 932, "to_take": 1, "try_id": 0, "source_location": "/src/woff2/brotli/dec/decode.c:930:11", "format": "woff2", "pass": 0, "reach": 0, "sfunc": 0, "exec_status": "success", "generator_exists": 1, "label": "global_abstraction_failure", "confidence": "high", "local_condition": "The true edge requires DecodeContextMap to be in READ_PREFIX with fewer than 5 buffered bits and no remaining byte for BrotliPullByte.", "system_requirement": "Build a valid big-endian WOFF2 compact directory and a bit-accurate Brotli stream that passes metablock and Huffman setup, independently decodes num_htrees above one, then ends with under 5 bits at the context-map prefix read.", "agent_interpretation": "The agent identified the truncation condition but assumed num_block_types[0] >= 2 guarantees num_htrees > 1 and represented Brotli metadata as ad hoc whole bytes.", "failure_evidence": "Source line 908 decodes num_htrees independently, contrary to the trace's claimed dependency on num_block_types[0]. The generator uses little-endian WOFF2 fields, a 20-byte legacy-style table record instead of ReadTableDirectory's compact format, and guessed Brotli bytes, so it never establishes the required parser path; outcome was exec_status=success, sfunc=0, reach=0."}