Decision log

Append-only. Every decision that shaped this codebase, with its reasoning, so it can survive a context compaction and so nobody re-litigates a settled question from first principles six weeks later.

Read this file at the start of any working session. It is the durable memory of the project; the conversation that produced these decisions is not.

This table is generated. Edit an entry under decisions/, then run tools/split-decisions.sh --index orbistoun. A number resolves to exactly one file.

# decision status date
🟢 D001 Rust, not C++ or C# decided 2026-08-19
🟢 D002 Edition 2024, MSRV 1.85, toolchain pinned to 1.97.1 decided 2026-08-19
🟢 D003 MIT OR Apache-2.0; permissive-only dependency policy decided 2026-08-19
🟢 D004 Fourteen crates arranged as a dependency spine decided 2026-08-19
🟢 D005 Interception is linking, not hooking decided 2026-08-19
🟢 D006 The NID hash suffix is runtime data, not a source constant decided 2026-08-19
🟢 D007 Symbol databases store names; NIDs are derived decided 2026-08-19
🟢 D008 Stub policy defaults to Unimplemented, never Ok decided 2026-08-19
🟢 D009 Placeholder error codes avoid the high bit decided 2026-08-19
🟢 D010 Honest failure over plausible output decided 2026-08-19
🟢 D011 The container parser contains zero unsafe decided 2026-08-19
🟢 D012 Vendor p_type range asserted; individual constants are not decided 2026-08-19
🟢 D013 Lints live in a workspace table, with unsafe discipline at deny decided 2026-08-19
🟢 D014 Provenance is enforced in CI and in the pre-push hook decided 2026-08-19
🟢 D015 No vendor trademarks in prose or in our own API decided 2026-08-19
🟢 D016 Validation is separated from effects decided 2026-08-19
🟢 D017 Handles are per-subsystem and never recycled decided 2026-08-19
🟢 D018 Traces are binary, sequence-numbered, and call-site attributed decided 2026-08-19
🟢 D019 Greenfield: no legacy, no compatibility shims decided 2026-08-19
🟢 D020 Synthetic fixtures are a prerequisite, not a nice-to-have decided 2026-08-19
🟢 D021 The ABI spike happens out of order decided 2026-08-19
🟢 D022 The trace sink is wired at phase 4, not earlier decided 2026-08-19
🟢 D023 Release profile favours speed; debug info kept decided 2026-08-19
🟢 D024 Other projects are reference-only, and get credited decided 2026-08-19
🟢 D025 Generate the symbol name list; never ship a database decided 2026-08-19
🟢 D026 Crunch until the queue is dry, then re-plan decided 2026-08-19
🟢 D027 Linux verification via multipass; Windows is the primary host decided 2026-08-19
🟢 D028 No urgency; take the highest-payoff path decided 2026-08-19
🟢 D029 Contracts and swappable backends, abstracted at guest semantics decided 2026-08-19
🟢 D030 Backend seams are enforced by crate boundaries, not discipline decided 2026-08-19
🟢 D031 Where the abstraction principle deliberately stops decided 2026-08-19
🟢 D032 The guest executes in a child process decided 2026-08-19
🟢 D033 Worker mode is self-reinvocation; no binary is privileged decided 2026-08-19
🟢 D034 CLI, GUI, and worker are interaction shims; logic lives in the crates decided 2026-08-19
🟢 D035 The protocol is serialisable data, defined separately from its transport decided 2026-08-19
🟢 D036 GUI: egui, not Tauri decided 2026-08-19
🟢 D037 Two user-facing binaries, plus a portable GUI artifact decided 2026-08-19
🟢 D038 Portable mode specification decided 2026-08-19
🟢 D039 Configuration formats: TOML and JSON; no database decided 2026-08-19
🟢 D040 GUI is library-first; Inspect before Launch decided 2026-08-19
🟢 D041 Firmware / system-menu launch is the lowest-priority stretch goal decided 2026-08-19
🟢 D042 A manifest-driven test corpus under titles/ decided 2026-08-19
🟢 D043 The accuracy suite is a separate repository, obSCEne decided 2026-08-19
🟢 D044 The open toolchain is a lawful source for interface facts, never for implementations decided 2026-08-19
🟢 D045 obSCEne result model, ordering, and coverage generation decided 2026-08-19
🟢 D046 Three observability channels, not one decided 2026-08-19
🟢 D047 Files first; OTLP export is an opt-in feature, not the substrate decided 2026-08-19
🟢 D048 Per-title overrides: one file per title, three layers, merged per key decided 2026-08-19
🟢 D049 Containers are wrapped; the ELF is not at offset zero decided 2026-08-19
🟢 D050 Revised starting order after real material arrived decided 2026-08-19
🟢 D051 Fixtures are generated, never extracted decided 2026-08-19
🟢 D052 The wrapper descriptor table locates segment data, not the ELF headers decided 2026-08-19
🟢 D053 Imports use standard ELF machinery; only the names are vendor-encoded decided 2026-08-19
🟢 D054 A module is reserved as one contiguous span, not per segment decided 2026-08-19
🟢 D054 Distribution formats stop at the door; mounts are ordered decided 2026-08-19
🟢 D055 The Linux path was broken, and only running it showed that decided 2026-08-19
🟢 D055 Title metadata is cheap, not a second container format decided 2026-08-19
🟢 D056 A remote hardware probe is the top-value future capability decided 2026-08-19
🟢 D056 The guest-to-host call boundary works, and needs extern "sysv64" decided 2026-08-19
🟢 D057 Worker mode is implemented, and both shims go through it decided 2026-08-19
🟢 D058 Image placement, and the host allocation granularity decided 2026-08-19
🟢 D059 Relocations apply; "interception is linking" is now machine code decided 2026-08-19
🟢 D060 Protection is computed per page, as a union decided 2026-08-19
🟢 D061 Thread-local storage is variant II, and the block sits below the pointer decided 2026-08-19
🟢 D062 One stub per import, not one shared target decided 2026-08-19
🟢 D063 Guest code runs on its own stack, entered through a switch decided 2026-08-19
🟢 D064 A worker that dies without a verdict gets a postmortem decided 2026-08-19
🟢 D065 Execute is never dropped, and never means "no access" decided 2026-08-19
🟢 D066 A guest that will not stop is a result, not a hang decided 2026-08-19
🟢 D067 A stub index is worthless; a name is a work list decided 2026-08-19
🟢 D068 Names are generated and proved, never obtained decided 2026-08-19
🟢 D069 Nothing is searched without a suffix, and a miss on standard names indicts it decided 2026-08-19
🟢 D070 The hasher read the digest backwards, from the very first commit decided 2026-08-19
🟢 D071 The hash suffix ships with the repository decided 2026-08-19
🟢 D072 First names, and the first concrete implementation target decided 2026-08-19
🟢 D073 Every name records how it was found, and the record is checked decided 2026-08-19
🟢 D074 Symbol databases accumulate, and are generated by a task decided 2026-08-19
🟢 D075 Standard-library names are harvested from FreeBSD, not remembered decided 2026-08-19
🟢 D076 Names are ours; selection is the module's decided 2026-08-19
🟢 D077 The loop is a verb, and its output is durable decided 2026-08-19
🟢 D078 sweep, not crunch decided 2026-08-19
🟢 D079 run <title> does everything a run needs decided 2026-08-19
🟢 D080 The loop has an objective function decided 2026-08-19
🟢 D081 run checks the machine before it tries anything decided 2026-08-19
🟢 D082 Declarations reach the guest, and the first function is implemented decided 2026-08-19
🟢 D083 The guest ignores return codes; the buffer is the channel decided 2026-08-19
🟢 D084 A tech-debt pass, and the vocabulary grows from our own observations decided 2026-08-19
🟡 D084 Instrument the GPU before translating any of it assumed 2026-08-19
🟢 D085 The hash suffix is supplied, and the file now says so decided 2026-08-19
🟡 D085 The instruction encoding table is data, and says it is unverified assumed 2026-08-19
🟢 D086 Rank blockers by shaders blocked, within an effort tier decided 2026-08-19
🟡 D087 Decoding never fails; it reports assumed 2026-08-19
🟡 D088 The shader corpus is content-addressed, and is the regression suite assumed 2026-08-19
🟢 D089 Verify the encoding table against a reference disassembler decided 2026-08-19
🟡 D090 Instruction names come only from what was observed assumed 2026-08-19
🟡 D091 Register extraction is mechanism; the shader-address map is a hypothesis assumed 2026-08-19
🟡 D092 Report rendering lives in the library assumed 2026-08-19
🟢 D093 Operands are decoded, and the numbering lives in data decided 2026-08-19
🟢 D094 Scalar destinations use the shared operand numbering decided 2026-08-19
🟡 D095 The shader census has a CLI surface assumed 2026-08-19
🟢 D096 Operand layout is a property of the opcode, not of the encoding family decided 2026-08-19
🟢 D097 Per-opcode operand fields are solved from probes, not transcribed decided 2026-08-19
🟢 D098 Predicated translation first; structured reconstruction stubbed loudly decided 2026-08-19
🟢 D099 Translated shaders are executed, not merely validated decided 2026-08-19
🟢 D100 Three wavefront models, kept as a differential oracle decided 2026-08-20
🟡 D101 Guest memory is a second binding, never a second half of the first assumed ~2026-08-20..08-21
🟢 D102 The SPIR-V builder holds the section layout, not its callers decided 2026-08-21
🟢 D103 The builder checks that its identifiers resolve decided 2026-08-21
🟢 D104 A blocked instruction names its dependency; an unwritten one does not decided 2026-08-20
🟢 D105 Hand-written assembly is a fixture source, and it caught a wrong row decided 2026-08-21
🟢 D106 The lane model refuses a mask write; Auto reads the shader to choose, and says so decided 2026-08-20
🟡 D107 The operand solver checks that a family agrees with itself assumed ~2026-08-20..08-21
🟢 D108 An operand the encoding does not carry is recorded, not omitted decided 2026-08-21
🟡 D109 A candidate operand field may not overlap the opcode assumed ~2026-08-21
🟢 D110 Control flow is a dispatch loop, not reconstructed structure decided 2026-08-21
🟢 D111 Function bodies are exempt from the define-before-use check decided 2026-08-21
🟢 D112 The GPU subsystem is driven by submissions, not called by the emulator decided 2026-08-21
🟢 D113 Translated shaders are cached by content, never by address decided 2026-08-20
🟡 D114 A shader read from memory ends where the program ends assumed ~2026-08-20..08-21
🟢 D115 The condition code is state both models hold decided 2026-08-21
🟡 D116 SUPPORTED lists what the translator handles, not what one function dispatches assumed ~2026-08-21..08-19
🟢 D117 Library attribution was fabricated, and looked fine decided 2026-08-19
🟢 D118 The real library table, found by prediction rather than by guessing decided 2026-08-19
🟢 D119 A supplied name is not a published one decided 2026-08-19
🟢 D120 Graphics vocabulary, and the submit function decided 2026-08-19
🟡 D121 Signed overflow is computed in bits, not compared assumed ~2026-08-19..08-20
🟢 D122 Knowledge is a file, and it is the loop's output decided 2026-08-20
🟢 D122 Which sub-encoding an opcode uses decides how its first word is read decided 2026-08-21
🟡 D123 An opcode with no shape row is an error, not a skip assumed ~2026-08-21..08-20
🟢 D124 The biggest wall was C++ static initialisation decided 2026-08-20
🟡 D124 The division sequence is refused, and says which numbers are missing assumed ~2026-08-20
🟢 D125 An error code in a pointer register is a wild pointer decided 2026-08-20
🟡 D125 Two routes to a shader address, and the disagreement is the point assumed ~2026-08-20
🟡 D126 A submission knows which queue it came from assumed ~2026-08-20
🟢 D126 The word list is harvested and cited; a rule I invented cost the most important name decided 2026-08-20
🟡 D127 Source modifiers are read from the raw words, and refused where unimplemented assumed ~2026-08-20
🟢 D127 Two invented rules in one afternoon, and the tests caught both decided 2026-08-20
🟢 D128 A field no probe can reach is widened from its family, and the claim is checked decided 2026-08-21
🟢 D128 The heap, and a fault address that identified itself decided 2026-08-20
🟢 D129 Side effects on hidden state are part of an instruction, not an extra decided 2026-08-21
🟢 D129 Two progress signals, and the verdict says which moved decided 2026-08-20
🟡 D130 The GPU-address assumption is a named function, not a silence assumed ~2026-08-20
🟢 D130 The thread pointer installs, and it is checked rather than assumed decided 2026-08-20
🟡 D131 A capture is a pair: what the call asked for, and the bytes it appended assumed ~2026-08-20
🟡 D132 Generated fixtures and dumped shaders do not share an extension assumed ~2026-08-20
🟡 D133 The subgroup level is blocked on hardware, not on effort assumed ~2026-08-20
🟡 D134 A named immediate the reference appends without a comma is still an operand assumed ~2026-08-20
🟡 D135 The local data share is workgroup storage, and the lane model refuses it assumed ~2026-08-20
🟡 D136 The differential oracle is a property, tested by generation assumed ~2026-08-20
🟡 D137 Out-of-range guest memory is masked, because undefined is not comparable assumed ~2026-08-20
🟡 D138 The generator does not branch assumed ~2026-08-20
🟢 D139 The target is RDNA2; the supported list names instructions rather than numbering them decided ~2026-08-20
🟡 D140 A zero-operand entry may be solved from one sample assumed ~2026-08-20
🟡 D141 Sixty-four-lane wavefronts, and the tables do not care assumed ~2026-08-20
🟢 D142 One Vulkan loader, one instance, one device, for the life of the process decided ~2026-08-20
🟢 D143 Two thirds of the division sequence, and a sharper reason for the third decided ~2026-08-20
🟢 D144 The division pre-scale needs no float controls, and could not have had them decided ~2026-08-20
🟢 D145 Wavefront width is a parameter, and a narrow shader is a different instruction stream decided ~2026-08-20
🟢 D146 The subgroup level is the per-lane model with a ballot, not a third model decided ~2026-08-20
🟢 D147 A descriptor this translator cannot address is forced out of bounds decided ~2026-08-20
🟢 D148 The shader work gets the same progress loop as the imports decided ~2026-08-20
🟢 D149 An address that resolves is evidence, not just a precondition decided ~2026-08-20
🟢 D150 Threads: the guest decides how many, the host decides how fast decided 2026-08-20
🟢 D151 A thread handle is an address, because the guest dereferences it decided 2026-08-20
🟢 D152 The entry point was reading a stray host pointer, and it looked like progress decided 2026-08-20
🟢 D153 The process entry image, and a matrix that ruled out three hypotheses at once decided 2026-08-20
🟢 D154 The ranked list is the wrong view at a wall; the ordered tail is the right one decided 2026-08-20
🟢 D155 Two names confirmed by hash, and the vocabulary extended to derive them decided 2026-08-20
🟢 D156 Nothing reachable from a guest call may panic decided 2026-08-20
🟡 D157 The mapping is parked behind a switch, not deleted and not shipped assumed 2026-08-20
🟢 D158 The fault handler always had the registers; it printed two addresses decided 2026-08-20
🟢 D159 Every guest call was misaligned, and nobody was looking decided 2026-08-20
🟢 D160 A second shim found the logic the first one had absorbed decided 2026-08-20
🟢 D161 egui, chosen for the shape of what it has to draw decided 2026-08-20
🟢 D162 A settings pane for a subsystem that does not exist is a dead control decided 2026-08-20
🟢 D163 Homebrew pkgs are encrypted, measured; pkg support deferred decided 2026-08-20
🟢 D164 The library list was doing file I/O every frame decided 2026-08-20
🟢 D165 The files were always there; there was nothing to hand them over decided 2026-08-20
🟢 D166 The main bisection lever was wired to nothing decided 2026-08-20
🟢 D167 The video-output handle, and what a large negative sweep is worth decided 2026-08-20
🟢 D168 The harvested name list is missing the syscall family decided 2026-08-21
🟢 D169 Three names, a video-out handle, and the wall that has not moved decided 2026-08-21
🟢 D170 Cross-project findings taken, and one declined decided 2026-08-21
🟢 D171 An out-pointer that is never written has no signature decided 2026-08-21
🟢 D172 The fault handler walks the frame chain decided 2026-08-21
🟢 D173 Call sites, which principle 9 asked for and nothing recorded decided 2026-08-21
🟢 D174 Physical memory has to alias itself decided 2026-08-21
🟢 D175 The filesystem is exonerated, and four walls are not one wall decided 2026-08-21
🟢 D176 Both container generations parse; the refusal was never tested decided 2026-08-21
🟢 D177 abort must not return, and it was reporting the opposite of the truth decided 2026-08-21
🟢 D178 The mapping was the witness, not the culprit (resolves D157) decided 2026-08-20
🟢 D179 Findings, because the consumer will not be a person decided 2026-08-21
🟢 D180 Behavioural provenance, because abstinence is not enforceable decided 2026-08-21
🟢 D181 A run records what it was subject to, so a verdict can be evidence decided 2026-08-21
🟢 D182 The compatibility record is the other half of the title file decided 2026-08-21
🟢 D183 snprintf_s refuses rather than renders what it can decided 2026-08-21
🟢 D184 Guards for the instrumentation, because the tests were faithful to the mistake decided 2026-08-21
🟢 D185 A poisoned stack, so "nobody wrote this" can be measured decided 2026-08-21
🟢 D186 printf, because the guest was explaining itself into a void decided 2026-08-21
🟢 D187 The stub policy reached only declared imports, and a conclusion rested on it decided 2026-08-21
🟢 D188 The shipped symbol database is loaded unless told otherwise decided 2026-08-21
🟢 D189 The vocabulary was never missing a word; it was missing a shape decided 2026-08-21
🟢 D190 One allocation path, because alignment cannot be a special case decided 2026-08-21
🟢 D191 The harvest walker ignored the rule written to fix it decided 2026-08-22
🟢 D192 check says everything before it fails, and an advisory tool cannot end the report decided 2026-08-22
🟢 D193 The titles carry their own names, and the generator could not spell one decided 2026-08-22
🟢 D194 The run dumps what the guest was pointing at decided 2026-08-22
🟢 D195 A run widens the grammar it searched with decided 2026-08-22
🟢 D196 The unknowns are a queue, not a candour exercise decided 2026-08-24
🟢 D197 The run reports the fault, instead of leaving it in the trace decided 2026-08-24
🟢 D198 Findings for the commonest outcome, and dumps for the ones already implemented decided 2026-08-24
🟢 D199 A guard that cannot see what it is checking is worse than none decided 2026-08-24
🟢 D200 The submission entry point takes an address, because a guest has one decided ~2026-08-24..08-21
🟢 D201 Decision numbers are checked, because more than one session assigns them decided 2026-08-21
🟢 D202 A candidate operand field may not overlap anything the encoding table already reads decided 2026-08-21
🟢 D203 The typed-buffer format table is measured from the assembler, both directions decided 2026-08-21
🟢 D204 Typed and untyped buffer accesses share one body, and unmeasured formats are refused decided 2026-08-21
🟢 D205 Operands spelled as names have their codes measured, not written down decided 2026-08-21
🟢 D206 Vendor documentation is consumed freely; the LLVM restriction is about oracles, not licences decided 2026-08-21
🟢 D207 orbistoun implements obSCEne's protocol; it does not shape it decided 2026-08-21
🟢 D208 The repository's own layout, audited once the tree stopped being small decided 2026-08-24
🟢 D209 The table generators are a crate, behind a seam that replays a recording decided 2026-08-24
🟢 D210 A semaphore handle is an int, and the type says so now decided 2026-08-24
🟢 D211 The call recorder is the dispatch ring; the crate declaring one is deleted decided 2026-08-24
🟢 D212 A language-model service, in a crate that knows nothing about orbistoun decided 2026-08-24
🟢 D213 Harvesting is categorised by what it observed, and the tiers are checked decided 2026-08-24
🟢 D214 Proposals are paired with an oracle; the first one asks for words, not names decided 2026-08-24
🟢 D215 The toolbar captures the window, and says that is what it captures decided 2026-08-24
🟢 D216 A pattern's size is computed once, not per candidate decided 2026-08-24
🟢 D217 The readable window was a page low, and the wall had been unreadable because of it decided 2026-08-24
🟢 D218 Two experiments that eliminated four things and confirmed nothing decided 2026-08-24
🟢 D219 The inference runtime is downloaded, not compiled in; and a proposer samples decided 2026-08-24
🟢 D220 The diagnostics share plumbing, and one of them is dyed banknotes decided 2026-08-24
🟢 D221 Every environment variable is declared in one crate decided 2026-08-24
🟢 D222 A build says which build it is, and the field for it was never populated decided 2026-08-24
🟢 D223 Three cheap diagnostics, and a hypothesis nobody had raised decided 2026-08-24
🟢 D224 Mapping the faulting address moved the wall, and the reading was wrong decided 2026-08-24
🟢 D225 Asking the probe is a live oracle, and the answer travels with its caveat decided 2026-08-24
🟢 D226 The correction to D224: the address was wrong after all decided 2026-08-25
🟢 D227 Principle 3 applies to the tools, and an intervention says so on the line decided 2026-08-25
⚪ D228 The library folder was resolved against the launch directory, which is not a setting unrecorded ~2026-08-25
🟢 D229 A plant could reach only offset zero, and a fill silently erased it decided 2026-08-25
🟢 D230 Nothing could change what an unnamed function answers, so an elimination had never been measured decided 2026-08-25
🟢 D231 A chooser is slower than running everything it would choose between decided 2026-08-25
🟢 D232 A fault that moves and a guest broken earlier are the same address and opposite results decided 2026-08-25
🟢 D233 The subject of a fault is where the guest died, not what it called decided 2026-08-25
🟢 D234 A forced answer reaches an implemented function, and the implementation still runs decided 2026-08-25
🟢 D235 The initialiser tags are parsed, and the answer was that there are none decided 2026-08-25
🟢 D236 orbistoun answers the probe's protocol, and is a stand-in for itself decided 2026-08-25
🟢 D237 Address translation was only ever implemented for signed containers decided 2026-08-25
🟢 D238 The limit that decides a verdict is a call budget; the clock stays as a backstop decided 2026-08-25
🟢 D239 Two counters, one quantity, and the number a person read was wrong by ten decided 2026-08-25
🟢 D240 The numbers in the documentation are generated, and a check fails when they drift decided 2026-08-25
🟢 D241 One lock per module for one process-wide table, and the run that measured nothing decided 2026-08-25
🟢 D242 A name enters the database only if this repository can re-derive it decided 2026-08-25
🟢 D243 The work list removed what a run solved, not what anything could name decided 2026-08-25
🟢 D244 An illegal instruction has an address, and it is not one the guest asked for decided 2026-08-25
🟢 D245 The probe's by-name census was arriving and going nowhere decided 2026-08-25
🟢 D246 An existence fact is graded like every other fact, and only the target may name decided 2026-08-25
🟢 D247 orbistoun read modules by a path the console does not use decided 2026-08-25
🟢 D248 An unchanged fault yields two different offsets, so it read as disagreement decided 2026-08-25
🟢 D249 Nothing the guest calls answers with the base, through any channel decided 2026-08-25
🟢 D250 A guest gets writable storage, and it is never the title's own directory decided 2026-08-25
🟢 D251 The console's filesystem is a knowledge file, and a title's data layers over it decided 2026-08-25
🟢 D252 A failed open must not look like a descriptor decided 2026-08-25
🟢 D253 The model was told the wrong libraries, and earned names from ones it was never shown decided 2026-08-25
🟢 D254 A wrong proposal is free on disk and expensive in the loop decided 2026-08-25
🟢 D255 Thirty-five per cent of what the model proposed was already in the shipped word list decided 2026-08-25
🟢 D256 The probe's worklist, taken from the top decided 2026-08-25
🟢 D257 A name that does not say how it was found cannot be used decided 2026-08-25
🟢 D258 The vocabulary was never the gap; the shapes were decided 2026-08-25
🟢 D259 Two missing words made seven confirmed names unreproducible decided 2026-08-25
🟢 D260 The audit reads the grammar the binary was built with, not the one on disk decided 2026-08-25
🟢 D261 Shapes are the binding constraint, not vocabulary, by three to one decided 2026-08-25
🟢 D262 Repeated learned became affordable when the list shrank, and nobody noticed decided 2026-08-25
🟢 D263 A diagnostic clause is read from the right, so a target may be qualified decided 2026-08-25
🟢 D264 A shape has two costs, and they rank differently decided 2026-08-25
🟢 D265 The model gets its own binary, not a place in the CLI decided 2026-08-25
🟢 D266 A round's cost depends on which position it grows, by a factor of twenty decided 2026-08-25
🟢 D267 A second guess at where data lives put a downloaded runtime in the repository decided 2026-08-25
🟢 D268 Floating-point arguments never reached an implementation decided 2026-08-25
🟢 D269 A mount replaces and a layer stacks, and the order cost a title its textures decided 2026-08-25
🟢 D270 The C library arrived as one batch because it was one absence decided 2026-08-25
🟢 D271 An error code in a boolean reads as true decided 2026-08-25
🟢 D272 Two shapes a guest hands over, and both were got wrong decided 2026-08-25
🟢 D273 A count, an offset and a descriptor are read as data, not tested against a table decided 2026-08-25
🟢 D274 The first call into guest code, rather than out of it decided 2026-08-25
🟢 D275 A clock that does not advance reads as a sleep that returned instantly decided 2026-08-25
🟢 D276 The expensive half of a watchpoint, built because the cheap half ran out decided 2026-08-25
🟢 D277 A data breakpoint fires after the access, and pretending otherwise would be a lie decided 2026-08-25
🟢 D278 A watchpoint that reads its own address traps itself decided 2026-08-25
🟢 D279 A dump is binary, and the size guard was measuring the wrong thing decided 2026-08-25
🟢 D280 The first six steps of check could hide the other ten decided 2026-08-25
🟢 D281 Five crates export a registration nothing calls decided 2026-08-26
🟢 D282 One body for two arities read an argument it was never passed decided 2026-08-26
🟢 D283 The wall needed two things right at once, and the sweep varied one at a time decided 2026-08-26
🟢 D284 A contract can be measured without knowing what the function means decided 2026-08-26
🟢 D285 The shape that spells a verb over two learned nouns decided 2026-08-26
🟢 D286 The sweep gains a second dimension, and it is a condition rather than a sentinel decided 2026-08-26
🟢 D287 Naming a function silently breaks every experiment aimed at its hash decided 2026-08-26
🟢 D288 The sweep kept its own list of diagnostics, and it was already wrong decided 2026-08-26
🟢 D289 The plan gets a runner, and an out-parameter finding follows itself through decided 2026-08-26
🟢 D290 Every floating-point function works and every one is reported as missing decided 2026-08-26
🟢 D291 What the sweep measured becomes a knowledge entry, and what it did not becomes an assumption decided 2026-08-26
🟢 D292 The merge rule moves to the crate that owns the format decided 2026-08-26
🟢 D293 The dispatcher and the naming loop become two crates decided 2026-08-26
🟢 D294 The first function implemented from a contract the loop measured decided 2026-08-26
🟢 D295 A stub policy that can write, so the loop stops needing a person to type Rust decided 2026-08-26
🟢 D296 Three tiers of automatic fix, and why the third is the smallest decided 2026-08-26
🟢 D297 The learned file becomes a record of measurements, so it can be sent to somebody decided 2026-08-26
🟢 D298 Verification runs against a machine that has learned nothing decided 2026-08-26
🟢 D299 Finding which answer the guest dereferenced, rather than reasoning about it decided 2026-08-26
🟢 D300 One concept: give the guest a region, and say how it arrives decided 2026-08-26
🟢 D301 FURTHER saturates, and a comparison needs headroom to be a comparison decided 2026-08-26
🟢 D302 Make the fix loop look like the naming loop: oracle first, generator second decided 2026-08-26
🟢 D303 The corpus is the oracle; the probe is one member of it decided 2026-08-26
🟢 D304 The repair searched for an index it could compute decided 2026-08-26
🟢 D305 A plain name is a NID nobody hashed yet decided 2026-08-26
🟢 D306 The payloads take their world from rdi, and the stack says nothing decided 2026-08-26
🟢 D307 An import that names data was getting a function decided 2026-08-26
🟢 D308 Ask the guest which field it wants, rather than guessing the structure decided 2026-08-27
🟢 D309 The comparison invented a position the display half refuses to invent decided 2026-08-27
🟢 D310 The window and the guest are different processes, so the shell button had nowhere to go decided 2026-08-27
🟢 D311 The shell holds meanings and refuses to hold the vendor's numbers decided 2026-08-27
🟢 D312 A run helped by named overrides was recorded as an honest measurement decided 2026-08-27
🟢 D313 One word for the shell, and the window had already taken it decided 2026-08-27
🟢 D314 An argument beats a setting, and contradictions are refused rather than resolved decided 2026-08-27
🟢 D315 A submission is a bundle of claims, and the receiver counts them itself decided 2026-08-27
🟢 D316 The decision number was allocated by a convention that races decided 2026-08-27
🟢 D317 "Both sides are Vulkan" was an assumption, and it happened to be true decided 2026-08-27
🟢 D318 The overlay is blocked earlier than "cross-process presentation" decided 2026-08-27
🟢 D319 A gate that cannot be scoped cannot be run in a shared tree decided 2026-08-27
🟢 D320 A curated list regrew to sixty-seven times its size, and only the clock noticed decided 2026-08-26
🟢 D321 Three things were built and inert, and one of them could not be built at all decided 2026-08-27
🟢 D322 A generated patch is safe because promotion is the verification step decided 2026-08-27
🟢 D323 Data imports get storage, not a stub decided 2026-08-27
🟢 D324 Nobody picks a test address, at either level decided 2026-08-27
🟢 D325 The third place nobody wrote, and proving the poison fired decided 2026-08-27
🟢 D326 A controller subsystem that stops at the guest, and says where decided 2026-08-27
🟢 D327 Categories along a row, children down a column decided 2026-08-27
🟢 D328 A promotion generated from a measurement, and the one field it invented decided 2026-08-27
🟢 D329 The one table left hand-copied, and the rule that seemed to forbid generating it decided 2026-08-27
🟢 D330 The harvest could undo the curation, and a one-line list read as seventy-six words decided 2026-08-27
🟢 D331 The diagnostic that said something was the one the summary dropped decided 2026-08-27
🟢 D332 A record with no honest baseline, and the parse test that failed for it decided 2026-08-27
🟢 D333 The model that costs nothing to set up, and the two things not copied with it decided 2026-08-27
🟢 D334 Measuring the engines took three tries, and the first two ranked the wrong one decided 2026-08-27
🟢 D335 The benchmark was wrong three times, and the third nearly retired a working engine decided 2026-08-27
🟢 D336 /no_think suppressed the reasoning and not the tags decided 2026-08-27
🟢 D340 The pad library exports both spellings, and both are imported decided 2026-08-27
🟢 D341 Cross-port key conflicts were invisible, and a keyboard could not move a stick decided 2026-08-27
🟢 D342 A shape can be disabled, and the two names that cost decided 2026-08-27
🟢 D343 Enter at main and the payloads start working decided 2026-08-27
🟢 D344 Guest threads are asked to stop, not made to decided 2026-08-27
🟢 D345 Input crosses the process boundary before anything can read it decided 2026-08-27
🟢 D346 A console has users, and one of their names reaches a guest unencoded decided 2026-08-27
🟢 D347 A record named after a scratch directory decided 2026-08-27
🟢 D348 klogsrv prints its own banner decided 2026-08-27
🟢 D349 The POSIX spellings were unserved, and they are the same functions decided 2026-08-27
🟢 D350 sysctl refuses what it does not know, and says what was asked decided 2026-08-27
🟢 D351 The sweep could not judge a guest that never faults decided 2026-08-27
🟢 D352 The harvest took names because the work was naming decided 2026-08-27
🟢 D353 The harvest is a command, and writing it twice found a bug decided 2026-08-27
🟢 D354 A revision the file states is a claim; a revision the checkout states is a fact decided 2026-08-27
🟢 D355 A turn that measured a contract and wrote nothing decided 2026-08-27
🟢 D356 The dispatcher read what crashed and never what we had written down as unknown decided 2026-08-27
🟢 D357 The loop can close a question when the discriminator is arithmetic decided 2026-08-28
🟢 D358 The loop writes down what it worked out decided 2026-08-29
🟢 D359 Entering at main skips the initialisation the program needed decided 2026-08-29
🟢 D360 .bss markers that name themselves, and what they found decided 2026-08-29
🟢 D361 The documentation route is closed, not untried decided 2026-08-29
🟢 D362 concat! defeats an implicit format capture decided 2026-08-29
🟢 D363 A compile-time constant fails as though the source were broken decided 2026-08-29
🟢 D364 A va_list is a cursor, so the v-forms render what the register forms cannot decided 2026-08-29
🟢 D365 The handoff structure answers by being asked decided 2026-08-29
🟢 D366 sceKernelDlsym is how a payload gets its C library, and the answer is a stub we already had decided 2026-08-29
🟢 D367 A symbol is declared where it is imported, not where its code lives decided 2026-08-29
🟢 D368 Handoff field two is a pointer, and what the unknown fields hold is a setting decided 2026-08-29
🟢 D369 A marker nobody decodes is arithmetic somebody does by hand decided 2026-08-29
🟢 D370 A constant one crate hardcodes is checked by the crate that holds the table decided 2026-08-29
🟢 D371 Files and sockets share one descriptor table, because a guest has one decided 2026-08-29
🟢 D372 Where the shared thing is what is under test, a lock is the fix decided 2026-08-29
🟢 D373 Asking must not take, and waiting must not hold decided 2026-08-29
🟢 D374 The checkout is a newer FreeBSD than the target decided 2026-08-29
🟢 D375 orbistoun runs a payload built with the real toolchain decided 2026-08-29
🟢 D376 A runtime's globals are named, and the last wall is a syscall gadget decided 2026-08-29
🟢 D377 A syscall gadget is not a function decided 2026-08-29
🟢 D378 The syscall boundary, built and not yet reached decided 2026-08-29
🟢 D379 A setting consulted nowhere, for the fourth time decided 2026-08-29
🟢 D380 A fault in our own code should say where in our own code decided 2026-08-29
🟢 D381 The dispatch path runs on the guest's stack decided 2026-08-29
🟢 D382 ftpsrv wants to be root, and that is a wall worth having decided 2026-08-29
🟢 D383 Conforming is not the same as compatible decided 2026-08-29
🟡 D384 A gadget is not reached from a call site a compiler wrote assumed 2026-08-30
🟡 D385 What a harvest skips has to be counted, or the section is a lie assumed 2026-08-30
🟡 D386 The seventh argument is on the stack, and nothing was reading it assumed 2026-08-30
🟡 D387 A guest thread's stack is guest memory, and the diagnostics could not see it assumed 2026-08-30
🟡 D388 A set says which and never when, and when was the whole question assumed 2026-08-30
🟡 D389 /dev/klog has something true in it, and the accuracy caveat that comes with it assumed 2026-08-30
🟡 D390 Ask the guest which fields it uses, one poisoned field per run assumed 2026-08-30
🟡 D391 What real hardware said, and which of it orbistoun had wrong assumed 2026-08-30
🟡 D392 A stub for every import makes "does this symbol exist" unanswerable assumed 2026-08-30
🟡 D393 A capital letter is a different symbol assumed 2026-08-30
🟡 D394 The machine is one setting, not five constants assumed 2026-08-30
🟡 D395 Not derivable is not the same as not measurable assumed 2026-08-30
🟡 D396 A log written after the guest stops is a log no guest can read assumed 2026-08-30
🟡 D397 The kernel's own version is a setting with no default assumed 2026-08-30
🟢 D398 The hardware trip happened, and it moved seven placeholders measured 2026-08-30
🟢 D399 The payloads were never handed what they read, and the instrument could not see it measured 2026-08-30
🟢 D400 The payload builds its own syscall entry, and this hands it something with no inside measured 2026-08-30
🟢 D401 A guest that talks to the kernel directly left no trace on the work list measured 2026-08-30
🟢 D402 A stub that returns from exit turns a clean shutdown into a crash measured 2026-08-30
🟢 D403 The call that was blocking every payload, and what it says the machine is measured 2026-08-30
🟢 D404 The wall behind 649: firmware-specific address arithmetic guest-observed 2026-08-30
🟢 D405 The console answered the sysctl probe, and it says 12.40 measured 2026-08-30
🟡 D406 A firmware skeleton crate, for guests that reach past the interface assumed 2026-08-30
🟢 D407 Word zero is getpid, and libkernel is laid out by measured vaddr guest-observed 2026-08-30
🟢 D408 The handoff, measured whole on a console, and made faithful here measured 2026-08-31
🟡 D409 Layout as a testable plan, console profiles, port reporting, and vaddr provenance assumed 2026-08-31
🟢 D410 The console confirmed five more export vaddrs, and validated the whole base+vaddr model measured ~2026-08-31
🟢 D411 Diagnosis of the image+0x2708 wall in klog.elf: kernel_copyout, setsockopt, and high-half kpipe_addr measured 2026-08-31
🟢 D412 Syscall 477 (mmap) implemented and wired to the syscall table published 2026-08-31
🟢 D413 Emulation of the kernel escape R/W pipe and dynamic symbol resolution measured 2026-08-31
🟡 D414 The test corpus, made a verb: a manifest of sources, fetched and recorded on demand assumed 2026-08-31
🟡 D415 The compatibility table as a tracked markdown file, generated from the records assumed 2026-08-31
🟢 D416 Four HLE fixes from the hardware-vs-orbistoun obSCEne diff measured 2026-08-31
🟢 D417 Three more HLE fixes from the same diff: thread join, mutex type, audio init measured 2026-08-31
🟢 D418 The census control failed because obSCEne's canary leaked into our symbol database measured 2026-08-31
🟢 D419 Five more libkernel vaddrs behaviourally confirmed from the second payload run measured 2026-08-31
🟢 D420 sceKernelGetSystemSwVersion answers 13.09, which is not the 12.40 firmware measured 2026-08-31
🟡 D421 The software version is a profile setting, like the firmware, not a kernel constant assumed 2026-08-31
🟢 D422 The console's writable device paths are a per-title sandbox, in the overlay already built decided 2026-08-31
🟢 D423 The sandbox is one entry point in orbistoun-fs, not orchestration in a consumer decided 2026-08-31
🟢 D424 The flip queue is a counter that completes on submit, and it unblocked the whole suite measured 2026-08-31
🟡 D425 sceVideoOutGetResolutionStatus presents 1080p; the skip is a held output, not a headless one assumed 2026-08-31
🟢 D426 Video-out refuses with its own error family, and refuses a second open of a held output measured 2026-09-01
🟢 D427 The GPU subsystem starts at the command builders: a PM4 writer, and the two dispatch calls measured 2026-09-01
🟢 D428 libSceSysmodule, and the real-title walls triaged measured 2026-09-01
🟢 D429 sceKernelReserveVirtualRange, and the boundary the real titles reach measured 2026-09-01
🟢 D430 Reentrant guest execution, and its first user: std::call_once measured 2026-09-01
🟢 D431 The C-runtime threading family, and a metric that fell as it was fixed measured 2026-09-01
🟢 D432 The fault report carries the faulting instruction, and what the shared wall actually is measured 2026-09-01
🟢 D433 Guest thread-local storage: the block is built, and Windows will not keep the base measured 2026-09-01
🟢 D434 The Windows thread-pointer backstop, and PPSA28061 past its TLS wall measured 2026-09-01
🟢 D435 libSceUlt mutexes and condition variables, and the wall that turned out to be online measured 2026-09-01
🟢 D436 The remaining title walls are one class: a guest allocation returning null, and it needs hardware measured 2026-09-01
🟢 D437 The hardware memory data was already captured; the map does not start at zero measured 2026-09-01
🟢 D438 Cross-section diff against hardware: placeholder error codes replaced with the measured ones measured 2026-09-01
🟢 D439 Cross-section hardware diff: the system-wide divergences worth fixing measured 2026-09-01
🟢 D440 The hardware cross-section diff is mined out for clean fixes measured 2026-09-01
🟢 D441 dlsym handle validation and sysmodule id-0, from user direction on the D440 tail measured 2026-09-01
🟢 D442 The flexible-memory "shared allocator" theory was aimed at the wrong titles measured 2026-09-01
🟢 D443 PPSA02664's allocator wall was orbistoun's policy region sitting on the guest's heap measured 2026-09-01
🟢 D444 obSCEne is orbistoun's oracle now, and it measured the flexible-memory bug exactly measured 2026-09-01
🟢 D445 A readable guard above the stack, and obSCEne now runs its whole suite under orbistoun measured 2026-09-01
🟢 D446 sceKernelVirtualQuery now sees the guest's own image and stack measured 2026-09-01
🟢 D447 sysctlbyname answers the knobs it can source, and kern.osrelease stops being refused measured 2026-09-01
🟢 D448 The obSCEne oracle's clean orbistoun bugs are mined out; the rest are by design measured 2026-09-01
🟢 D449 PPSA02664 regressed to the allocator wall; the policy-region/reserve collision is the mechanism (open) measured 2026-09-01
🟢 D450 PPSA02664's "regression" is a thread race between two concurrent walls, not a regression measured 2026-09-01
🟢 D451 PPSA21564 boots once the sceLibcMspace allocator family (and bcmp) answer real values measured 2026-09-01
🟢 D452 scePthreadGetthreadid answers the thread's handle, and PPSA21564 reaches 0% stubs measured 2026-09-01
🟢 D453 POSIX thread-specific-data keys; PPSA21564's TBB scheduler stops aborting measured 2026-09-01
🟢 D454 localtime/asctime and fgets carry PPSA21564 into main(); it prints, parses its args, and walls on a threading assert measured 2026-09-01
🟢 D455 POSIX unnamed semaphores (sem_init family); the Cond.cpp wall is not a traced HLE call measured 2026-09-01
🟢 D456 Fault reports name privileged instructions and call an emulator bug an emulator bug measured 2026-09-01
🟢 D457 Fault reports name the null base register, not just dump sixteen measured 2026-09-01
🟢 D458 Execute breakpoints, so a guest-computed value can be read where it is used measured 2026-09-01
🟢 D459 The call trace records what each call answered, not only what it was asked measured 2026-09-01
🟢 D460 Mapping direct memory commits into an existing reservation, it does not reserve again measured 2026-09-01
🟡 D461 std::random_device is deterministic here, on purpose assumed 2026-09-01
🟢 D462 A reservation the guest could not make is named in the run report measured 2026-09-01
🟢 D463 The guest mapping arena is moved clear of the thunk data blocks measured 2026-09-01
🟢 D464 Spawned guest threads get their own thread-local storage measured 2026-09-02
🟢 D465 sceKernelWaitEventFlag blocks on the event-flag condvar rather than being absent measured 2026-09-02
🟢 D466 Narrow typed-buffer formats translate through a packed path, built up one kind at a time measured 2026-09-02
🟡 D467 A packed half widens to a float through a real 16-bit float type, not a hand-rolled unpack assumed 2026-09-02
🟢 D468 The ctype tables are measured off hardware, because the documented layout is the wrong one measured 2026-09-02
🟢 D469 PPSA02664's wall is not _Getpctype; the trace was naming the wrong function measured 2026-09-02
🟢 D470 A run that wrote no trace reports nothing, rather than reprinting an old one measured 2026-09-02
🟢 D471 The fault reporter checks the page before reading it, because an execute fault puts rip in the hole measured 2026-09-02
🟢 D472 Published interfaces are ported in bulk; only vendor code is found one wall at a time measured 2026-09-02
🟡 D473 Without an unwinder, a C++ throw is a named stop rather than a return assumed 2026-09-02
🟢 D474 "Implemented" and "finished" are different states, and the report counts them separately measured 2026-09-02
🟡 D475 The posix_-prefixed exports delegate to their unprefixed twins, on an assumed footing assumed 2026-09-02
🟡 D476 The errno table carries published values too, in a tier that says so assumed 2026-09-02
🟢 D477 One sysctlbyname, merged from the two that were both live measured 2026-09-02
🟡 D478 A differential result gets its own tier, because it is not a measurement assumed 2026-09-02
🟢 D479 The differential runs against whatever published implementation is reachable, and names it measured 2026-09-02
🟢 D480 The sign-extension divergence is withdrawn: it was the probe's cast, not the console's measured 2026-09-02
🟢 D481 A measured value that is not a property of the interface gets its own list measured 2026-09-02
🟢 D482 A title's own module is found by the name that imports it, not by a path measured 2026-09-03
🟢 D483 Attribution is half of what identifies an import, and libc is the title's measured 2026-09-03
🟢 D484 One stub table across every module, not one table each measured 2026-09-03
🟢 D485 A measured value is not automatically a constant, and the counter proves it measured 2026-09-03
🟢 D486 A measured register holds configuration and status, and only one is reproducible measured 2026-09-03
🟢 D487 A run is not reproducible, so the progress verdict needs a noise floor measured 2026-09-03
🟢 D488 The title's modules were placed on the guest's own heap arena measured 2026-09-03
🟢 D489 Placing a module does nothing until it is bound, protected and located measured 2026-09-03
🟢 D490 A shared stub table needs shared labels, or the trace names the wrong function measured 2026-09-03
🟢 D491 The null is a module global, and it is not a missing constructor measured 2026-09-03
🟢 D492 Retail title modules do not export module_start measured 2026-09-03
🟢 D493 The null is .bss, and binding made the answer invisible measured 2026-09-03
🟢 D494 Bound calls were always observable, and only two of five are called measured 2026-09-03
🟢 D495 The game asks for its modules by name, and LoadStartModule starts nothing measured 2026-09-03
🟢 D496 These modules declare no code to run at load, and DT_FINI is why that is a finding measured 2026-09-03
🟢 D497 Half the encoder probes measure the probe's own initialiser measured 2026-09-03
🟢 D498 strtol converts the zero in "0x", and a stale work item is not a work item measured 2026-09-03
🟢 D499 The residual oscillation is one allocation round, and it moves the verdict again measured 2026-09-03
🟢 D500 Every A/V/input subsystem implements the lifecycle and not the data path measured 2026-09-03
🟢 D501 The citations were already here, in the wrong field measured 2026-09-03
🟢 D502 Ten outstanding measurements were never going to be claimed decided 2026-09-03
🟢 D503 Seven hardware claims from data already in hand, and nine more non-claims measured 2026-09-03
🟢 D504 The current generation's graphics API, declared as names and nothing else decided 2026-09-03
🟢 D505 The declared surface now matches what guests ask for, and the coverage figure got worse decided 2026-09-03
🟢 D506 The encoder measurements record a non-resolution, and thirty names had no provenance measured 2026-09-03
🟢 D507 A second capture settles the third query field, and demotes thirty measurements measured 2026-09-03
🟢 D508 A raw register is claimable once status is masked off decided 2026-09-03
🟢 D509 The console writes eight bytes where documentation said four measured 2026-09-03
🟢 D510 A message that describes a gap outlives the gap decided 2026-09-03
🟢 D511 Two formatter bugs in one run, both in the padding nobody tested measured 2026-09-03
🟢 D512 vsnprintf closes the list, and the break proves which cases were carrying it measured 2026-09-03
🟢 D513 The oscillation was the guest's own allocator, and the deciding input was ordering measured 2026-09-03
🟢 D514 The wall is a module that was loaded, placed, and never started measured 2026-09-03
🟢 D515 Starting the modules took the wall down, and the guest reached its frame loop measured 2026-09-03
🟢 D516 Nothing is ever pending, and the guest left its frame loop guest-observed 2026-09-03
🟢 D517 dlsym never knew the guest's own exports, and the wall did not move anyway measured 2026-09-03
🟢 D518 The null global has exactly one writer (and the guard I blamed is never reached) measured 2026-09-03
🟢 D519 The guest calls into its module before the memory manager exists, and D518 blamed the wrong guard measured 2026-09-03
🟢 D520 Starting every placed module before entry is not the missing ordering measured 2026-09-03
🟢 D521 The callback has no branch that avoids the read, so one of D519's two readings is dead measured 2026-09-03
🟢 D522 A fault now says what its registers point at, and the label is "None" measured 2026-09-03
🟢 D523 The thread affinity setter, and the one beside it that stays unimplemented guest-observed 2026-09-03
🟢 D524 The event queue exists so a handle means something, not so events can be delivered guest-observed 2026-09-03
🟢 D525 The vendor stat is not the POSIX one under another name, and it opened eight functions guest-observed 2026-09-03
🟢 D526 A positioned read took the guest into the GPU, and the wall finally moved guest-observed 2026-09-03
🟢 D527 The wall is sceAgcCreateShader, and a hardcoded cap was hiding it guest-observed 2026-09-03
🟢 D528 Two walls that need a measurement, and the mechanism that asks for one decided 2026-09-03
🟢 D529 The record format already carried what D512 said it lacked decided 2026-09-04
🟢 D530 The wide family is verified, and one case is the reason it was worth doing measured 2026-09-04
🟢 D531 The interleaved sequence is the only thing that tells strtok_r from strtok measured 2026-09-04
🟢 D532 strftime, and the question that found it measured 2026-09-04
🟢 D533 libm splits in two, and only half of it belongs in a differential measured 2026-09-04
🟢 D534 Two rounding rules that pin each other measured 2026-09-04
🟢 D535 strdup covered, and a missing terminator is only caught by luck measured 2026-09-04
🟢 D536 One clock, two names, and one origin guest-observed 2026-09-04
🟢 D537 The reasoning was in the code, and the record said nothing decided 2026-09-04
🟢 D538 Two thirds of the ask list is forty sentences decided 2026-09-04
🟢 D539 One question, written a hundred and forty-nine ways decided 2026-09-04
🟢 D540 The ask list was asking for a function that does not exist decided 2026-09-04
🟢 D541 A measurement was sitting in the list of things nobody knows decided 2026-09-04
🟢 D542 Six records said nothing was known while holding a measurement, and the axis closes decided 2026-09-04
🟢 D543 dlsym succeeds where the console refuses, and the reason written down was not the reason decided 2026-09-04
🟢 D544 Forty-seven measured values that no test looked at decided 2026-09-04
🟢 D545 The relations are testable where the numbers are not decided 2026-09-04
🟢 D546 A permanent resident of the work queue decided 2026-09-04
🟢 D547 The relation needed a mount, not a guest decided 2026-09-04
⚪ D548 The wall is an experiment, and there is no guest to run unrecorded 2026-09-04
🟢 D549 The oracle the roadmap said did not exist decided 2026-09-04
🟢 D550 The oracle draws decided 2026-09-04
🟢 D551 The export was already decoded, and its blocker had half expired decided 2026-09-04
🟢 D552 The fragment path does not need the feature, and the report must not claim it decided 2026-09-04
🟢 D553 The translator draws decided 2026-09-04
🟢 D554 The oracle carries a varying, before anything interpolates decided 2026-09-04
🟢 D555 The wall was real, the record was twelve days stale, and the library was never empty decided 2026-09-04
🟢 D556 The out-parameter is the whole of the wall, and the capture that could answer it does not exist yet measured 2026-09-04
🟡 D557 A measured answer is not a prop, and a region always was one assumed 2026-09-04
🟢 D558 A rung for the first frame, and where frames may not rank measured 2026-09-04
🟢 D559 The Agc argument classes, and how to get past the arity gate without guessing measured 2026-09-04
🟢 D560 Events are delivered, from a published layout, and the wait does not block guest-observed 2026-09-04
🟢 D561 The per-thread scheduling record D523 said would be needed guest-observed 2026-09-04
🟢 D562 The tail, worked in order of what a placeholder does guest-observed 2026-09-04
🟢 D563 The record could not see the work, and three copies of the ordering had drifted measured 2026-09-04
🟢 D564 A placeholder answered as a size is a very quiet four gigabytes guest-observed 2026-09-04
🟢 D565 The first measured command packets, and a census that contradicts its own report measured 2026-09-04
🟢 D566 The biggest item in the work list has no name, and no return value fixes it measured 2026-09-04
🟢 D567 A placeholder that names its own source measured 2026-09-04
🟢 D568 The tail was not the tail measured 2026-09-04
🟢 D569 A diagnostic nobody asked about, and the record it overwrote measured 2026-09-04
🟢 D570 Six arguments, a refuted hypothesis, and a filter that turned out to be load-bearing measured 2026-09-04
🟢 D571 The ring is circular, and the trade it seemed to require was false measured 2026-09-04
🟢 D572 The unnamed futex had a name in the tree, and the search missed it for a word order measured 2026-09-07
🟢 D573 A wait that returns without waiting is a busy loop, so the wait now waits measured 2026-09-07
🟢 D574 Three tools that reported more than they measured, on one afternoon measured 2026-09-07
🟢 D575 The stack the collector scans, and the wall three titles shared measured 2026-09-07
🟢 D576 A breakpoint was called stub padding without anything looking at the address measured 2026-09-07
🟢 D577 A guest may re-protect the regions somebody else placed for it measured 2026-09-07
🟢 D578 The random devices, and a record of what a guest actually opened measured 2026-09-07
🟢 D579 The mapping arena had no name and its bytes were unreadable measured 2026-09-07
🟢 D580 A watch that killed the run, and a message that said mapped when it meant published measured 2026-09-07
🟢 D581 Every mapping the guest was given, in order measured 2026-09-08
🟢 D582 The clock a guest reads has to repeat, and still move measured 2026-09-08
🟢 D583 The loop checks that a run repeats before it believes any comparison measured 2026-09-08
🟢 D584 Every handle the guest is given comes from one region measured 2026-09-08
🟢 D585 A fresh thread attribute set is not entirely zero, and the console said so measured 2026-09-08
🟢 D586 The loop reads back the structure a call was handed measured 2026-09-08
🟢 D587 The asynchronous file path, named and reported measured 2026-09-08
🟢 D588 Write implies read, the dump ran out of room, and D580 was wrong measured 2026-09-08
🟢 D589 Delivering the file the asynchronous path resolved changes nothing measured 2026-09-08
🟢 D590 The guest explains itself, and nothing was reading it measured 2026-09-08
🟢 D591 The title ships the index the asynchronous file path needs measured 2026-09-08
🟢 D592 The replacement library is guest code, and the index is read measured 2026-09-08
🟢 D593 The modules a title ships were unreadable to every diagnostic measured 2026-09-08
🟢 D594 The wall is a library the dump replaced, not three functions measured 2026-09-08
🟢 D595 A read of four hundred and two bytes printed as zero measured 2026-09-08
🟢 D596 Every message carries where it was said from measured 2026-09-08
🟢 D597 Nothing ever wrote the command, and a second summary read as none measured 2026-09-08
🟢 D598 Two titles are two kinds of dump, and one names the code it wanted measured 2026-09-08
🟢 D599 The dispatcher was measuring a different program measured 2026-09-08
🟢 D600 A bogus finding disappears, and the check passes on a run that still varies measured 2026-09-08
🟢 D601 Guest-visible blocks belong in the memory crate decided 2026-09-08
🟢 D602 The drift is a guest branch, not a mapping we refused - NO, it is a refusal measured 2026-09-08
🟢 D603 The head of the call sequence, and a redirection that measured nothing measured 2026-09-08
🟢 D604 A placement that would not step past a conflict measured 2026-09-08
🟢 D605 The largest record kind in every report, read by nothing measured 2026-09-08
🟡 D606 A name is a seed for the next name assumed 2026-09-08
🟢 D607 The sweep kept the weaker of two true records measured 2026-09-08
🟢 D608 A guard reading a shape that no longer exists measured 2026-09-08
🟢 D609 The corpus was five files of twenty-eight measured 2026-09-08
🟢 D610 The count and the mode were both ignored measured 2026-09-08
🟢 D611 A parser that required a prefix the probe does not always write measured 2026-09-08
🟢 D612 The predicate that was never a predicate measured 2026-09-08
🟢 D613 A wait that did not wait, and a record that answered for somebody else measured 2026-09-08
🟢 D614 The timeout that arrived early measured 2026-09-08
🟢 D615 A queue nobody registered, and two nobody waited on measured 2026-09-08
🟢 D616 The record asked the wrong thread measured 2026-09-08
🟢 D617 A second batch, and the pipeline that was built for it measured 2026-09-08
🟢 D618 Evidence going missing made a claim stronger measured 2026-09-08
🟢 D619 The other work list measured 2026-09-08
🟢 D620 A question put to the guest, and the answer was no measured 2026-09-08
🟢 D621 The report paired a fault with another thread's calls measured 2026-09-08
🟢 D622 The probe runs here too measured 2026-09-08
🟢 D623 The dump answered a different question measured 2026-09-08
🟢 D624 The differential counted items and called them findings measured 2026-09-08
🟢 D625 A dump was taken and never shown measured 2026-09-08
🟢 D626 Seven of the differences were the probe declining to call measured 2026-09-08
🟢 D627 Two spellings of one condition gave the encoding measured 2026-09-08
🟢 D628 The guest names its own functions measured 2026-09-08
🟢 D629 A module handle now narrows the answer, for one handle measured 2026-09-08
🟢 D630 The title's own code was mapped, and stubbed anyway measured 2026-09-08
🟢 D631 Advice that cannot succeed measured 2026-09-08
🟢 D632 One function, two answers, decided by how the guest asked measured 2026-09-08
🟢 D633 Three walls, one subsystem measured 2026-09-08
🟢 D634 The ratchet cannot say "worse" measured 2026-09-08
🟢 D635 The branch that printed nothing measured 2026-09-09
🟢 D636 A record older than the repository measured 2026-09-09
🟢 D637 An answer needs somebody who asked measured 2026-09-09
🟢 D638 One defect, against the leg that compares like for like measured 2026-09-09
🟢 D639 Twenty blank pages nobody mentioned measured 2026-09-09
🟢 D640 Every import accounted for, with a reason measured 2026-09-09
🟢 D641 Four numbers, and the wall moved measured 2026-09-09
🟢 D642 Two tables that answer each other measured 2026-09-09
🟢 D643 One name, and a third title reached the same wall measured 2026-09-09
🟢 D644 The binding reached the relocation measured 2026-09-09
🟢 D645 A signal the guest sends itself measured 2026-09-09
🟢 D646 The clock samples instead of sleeping measured 2026-09-09
🟢 D647 The name a faulting register points at measured 2026-09-09
🟢 D648 Half a signal contract is worth implementing measured 2026-09-09
🟢 D649 A name that cannot be confirmed published 2026-09-09
🟢 D650 The signal goes to another thread measured 2026-09-09
🟢 D651 The thread that was not calling measured 2026-09-09
🟢 D652 The pending-signal slot measured 2026-09-09
🟢 D653 A second reader over one corpus measured 2026-09-09
🟢 D654 Four harness errors and no reader defects measured 2026-09-09
🟢 D655 The symbol level agrees measured 2026-09-09
🟢 D656 The context has to be on a real stack measured 2026-09-09
🟢 D657 Both warnings were mirrors measured 2026-09-09
🟢 D658 What the guest said measured 2026-09-09
🟢 D659 Two walls that hardware cannot reach measured 2026-09-09
🟡 D660 A page per title assumed 2026-09-09
🟢 D661 Three roots, not one measured 2026-09-09
🟢 D662 Category is a permission, not a label published 2026-09-10
🟢 D663 The codenames are the vocabulary measured 2026-09-10
🟢 D664 An origin list, with every failure named measured 2026-09-10
🟢 D665 The index reads the entry, not its title measured 2026-09-10
🟢 D666 Orbis is the machine, neo is the refresh decided 2026-09-10
🟢 D667 The vendor spelling of a socket call, and the errno that travels with it decided 2026-09-10
🟢 D668 Twelve knowledge files nothing loaded decided 2026-09-10
🟢 D669 A title does not resolve by name, and orbistoun was calling itself a payload decided 2026-09-10
🟢 D670 The refusal a guest cannot see is a success decided 2026-09-10
🟢 D671 The pad structure was measured, so the shim D345 deferred could be written decided 2026-09-10
🟢 D672 The audio drain was measurable without a device decided 2026-09-10
🟢 D673 An empty queue is a report, not a refusal decided 2026-09-10
🟢 D674 A high-water mark that could move backwards decided 2026-09-10
🟢 D675 A console's own sysctl knobs belong to its profile, and two belong to the platform decided 2026-09-10
🟢 D676 Weak undefined symbols bind to zero when unanswered decided 2026-09-10
meaning
🟢 settled, and the reasoning rests on something checkable
🟡 assumed or proposed - made without input, and in the review queue
🔴 reversed, superseded or blocked
⚪ no status recorded

A date with ~ is not recorded - it is worked out from the dated entries either side, because an entry between two of them was written between their dates. ~ alone is a day both neighbours agree on; ~a..b is a span, and no day inside it is claimed; ~>a and ~<a are entries with a dated neighbour on only one side. A bare - has no dated entry either side to reason from.