{"id":"5ace547a-4156-43c0-9c39-86edc39934ac","arxiv_id":"2607.12915","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"MXsim v0.1 is a MATLAB simulator, built on CPFloat, for custom-precision OCP MX block floating-point formats and arithmetic aimed at mixed-precision algorithm research.","lead":"The authors release MXsim v0.1, a MATLAB library that simulates OCP MX block floating-point formats and arithmetic on top of CPFloat. It lets mixed-precision algorithm researchers try custom MX variants and rounding choices without needing the newest NVIDIA or AMD GPUs.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Abstract-only review leaves the central claim of faithful MX simulation untestable; no load-bearing technical flaw can be confirmed or refuted from the available material.","rationale":"The Reader correctly treats this as an abstract-only software-tool paper whose modest claim (existence of a customizable MATLAB MX simulator) is plausible but unsupported by code, validation, or numerical results. The single load-bearing issue is therefore the absence of any inspectable evidence that the simulation is faithful to OCP MX block semantics, especially given that the OCP specification itself leaves precision and rounding unspecified and vendor bit-wise behavior is undocumented. That is exactly the Reader's weakest_assumption. No deeper technical contradiction can be asserted without the full text or artifacts; manufacturing one would violate the good-faith rule. Consequently the CONDITIONAL verdict (accept-shaped once the library is released with documentation and basic validation) remains appropriate, confidence stays LOW, and no adjustment is warranted. The concrete test simply operationalizes the missing validation step that would convert the claim from untestable to either supported or refuted.","tokens_in":2025,"tokens_out":569,"duration_ms":6126,"concrete_test":"Obtain the full paper or the MXsim v0.1 release; re-implement or inspect the MX block conversion and the custom-precision dot-product path; then run a fixed set of block-scale and matrix-vector tests (e.g., random blocks of size 16/32 with known shared exponents) and compare bit-level or ulp-level results against any available vendor MX reference or against an independent CPFloat-based reimplementation. If discrepancies exceed 1 ulp on more than 1 % of cases, the fidelity claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that MXsim v0.1, layered on CPFloat, correctly simulates OCP MX block floating-point formats and arithmetic with full user control over exponents, I/O precisions, and accumulator precision, sufficiently for mixed-precision algorithm development. Because only the abstract is available, there is no description of how shared exponents are extracted and applied across a block, how the custom accumulator is realized inside the MX-style dot product, how rounding modes or subnormals are handled relative to the (underspecified) OCP MX rules, or any validation against known MX cases or hardware. The reader's weakest assumption correctly flags the fidelity gap, but that gap cannot be inspected: the claim is therefore unfalsifiable from the given text rather than demonstrably false. The concern is mechanical incompleteness of evidence, not an identified inconsistency in the (unseen) design.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript (available here only as an abstract) presents MXsim v0.1, a MATLAB library layered on CPFloat for simulating OCP MX block floating-point formats and arithmetic. It claims user control over shared exponents, input/output element precisions, and accumulator precision in the MX-style dot product, so that researchers can experiment with custom MX variants and with undocumented vendor settings when NVIDIA/AMD GPU hardware is unavailable, and can study effects on mixed-precision algorithms given that OCP does not prescribe precision and rounding.","tokens_in":2190,"tokens_out":726,"duration_ms":18084,"significance":"If the full library correctly implements the claimed customizations and is shown to capture the numerically relevant behavior of OCP MX blocks, it would be a practical open tool for the mixed-precision algorithm community. Building on CPFloat, exposing a simple MATLAB interface, and targeting the underspecification of OCP MX and vendor bit-wise behavior are genuine strengths for a software-tool contribution—provided the body supplies design detail, validation, and reproducible examples that the abstract alone cannot establish.","major_comments":[{"comment":"Abstract: The central claim that MXsim correctly simulates OCP MX block formats and arithmetic is untestable from the abstract. The full manuscript must specify how shared block exponents are extracted and applied across a block, how the custom-precision accumulator is realized inside the MX-style dot product, and how rounding modes and subnormals are handled relative to the (underspecified) OCP rules; without that design, the fidelity claim cannot be assessed.","section":"Abstract"},{"comment":"Abstract: No validation against known MX cases, hardware, or other simulators is mentioned. For a cs.MS software-tool paper the claim that the library is suitable for mixed-precision algorithm development is load-bearing and requires reproducible numerical examples, error checks, and at least limited comparison; their absence from the reviewed material leaves the main claim unsubstantiated.","section":"Abstract"},{"comment":"Abstract: The asserted “full customisation” of exponents, I/O precisions, and accumulator precision needs concrete API documentation, default choices, and correctness arguments in the body. The abstract states the capability but does not show that the configuration surface is sufficient or that it interacts correctly with CPFloat; that gap is material to the research-utility claim.","section":"Abstract"}],"minor_comments":[{"comment":"The abstract could more clearly separate what the OCP MX specification prescribes from what MXsim chooses as defaults or free parameters, so readers know which knobs are research extensions versus compliance.","section":"Abstract"},{"comment":"Availability (repository, license, version pinning of CPFloat) should be stated explicitly in the full manuscript if not already present; v0.1 alone is insufficient for reproducibility.","section":"Abstract"}],"recommendation":"uncertain","confidential_remarks":"Only the abstract was available for this review; a full-text review is required before any accept/reject decision. Scope (cs.MS software tool) appears appropriate if the complete manuscript includes implementation detail and validation. No circularity or derivation issues arise; the residual risk is ordinary dependence on CPFloat correctness and on undocumented vendor MX behavior, which the abstract itself flags as motivation."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a tools paper announcing MXsim v0.1, a MATLAB layer on CPFloat for OCP MX block floating-point formats and arithmetic. The punchline is modest and practical: give mixed-precision algorithm people a way to try MX variants (and undocumented vendor settings) without the latest NVIDIA/AMD GPUs, with full control over exponents, I/O precisions, and the dot-product accumulator.\n\nWhat is actually new is the artifact itself—an MX-specific MATLAB interface on top of an existing custom-precision simulator. That is legitimate engineering, not a first-principles result. The motivation is clean: the OCP spec does not fix precision or rounding, and vendors do not publish bit-wise behavior, so a customizable simulator is exactly the right tool for the subfield. Building on CPFloat is a sensible reuse rather than reinventing the wheel.\n\nThe soft spots are mechanical and proportional to what we have. Only the abstract is available, so there is no description of shared-exponent extraction, how the custom accumulator is realized, subnormals/rounding relative to the underspecified MX rules, or any validation against known cases or hardware. The reader’s weakest assumption is correct: we cannot yet tell whether the simulation is faithful enough for algorithm development. That is not a demonstrated design flaw; it is simply missing evidence. For a software-tool paper the load-bearing claim is “the library exists and works,” and right now that claim is unfalsifiable from the text.\n\nWho it is for: people writing mixed-precision numerical algorithms who already live in MATLAB and cannot get Blackwell or MI300 time. They get value if the code ships with docs and basic checks. It deserves a serious referee as a tools contribution; expect reviewers to demand the artifact and some validation. I would not desk-reject it. If the full paper delivers the library and a few sanity comparisons, it is a useful incremental piece. If not, it stays an announcement.","headline":"Useful MATLAB front-end for OCP MX on CPFloat; the need is real, but the abstract alone leaves the fidelity claim untested.","tokens_in":2828,"tokens_out":493,"would_cite":false,"duration_ms":10949,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"MXsim is a MATLAB library that simulates OCP MX block floating-point formats with full control over exponents, precisions, and accumulator width.","keywords":["OCP MX","block floating-point","mixed-precision arithmetic","custom-precision simulation","MATLAB","CPFloat","dot product accumulator","GPU formats"],"falsifier":"Compare MXsim results for a set of mixed-precision matrix multiplications or iterative solvers against the same kernels run on actual NVIDIA or AMD MX hardware; any systematic difference in rounding error or final accuracy that cannot be explained by a documented parameter would falsify the claim of faithful simulation.","tokens_in":2859,"feed_emoji":"🧮","tokens_out":727,"duration_ms":7165,"temperature":0.7,"pith_summary":"The paper introduces MXsim v0.1, a MATLAB library that lets researchers simulate OCP MX block floating-point arithmetic without needing the latest NVIDIA or AMD GPUs. Built on the existing custom-precision simulator CPFloat, it exposes full user control over exponents, input and output precisions, and the precision of the accumulator inside the dot product. Because GPU vendors have not published the bit-wise numerical behavior of these formats and the OCP specification itself does not prescribe precision or rounding, the library lets algorithm developers try different plausible settings and measure their effect on mixed-precision numerical methods. A sympathetic reader cares because MX formats are already shipping in hardware, yet the lack of an open, flexible software model has made it hard to design and test algorithms that will run correctly and accurately on them.","feed_headline":"MATLAB library simulates MX block floats with full precision control","feed_subtitle":"Experiment with custom OCP MX formats and undocumented GPU settings without hardware","key_machinery":"MXsim itself: a MATLAB wrapper around CPFloat that implements the OCP MX block-floating-point model, exposing configurable shared exponents, element formats, and a user-chosen accumulator precision for the dot product.","core_discovery":"MXsim v0.1 is a MATLAB library, layered on CPFloat, that correctly simulates OCP MX block floating-point formats and arithmetic while giving the user complete control over exponents, input/output precisions, and accumulator precision in the dot product, thereby enabling experimentation with custom MX variants and with undocumented vendor settings.","pith_inferences":["Once public, MXsim could become a de-facto reference for comparing proposed MX extensions or alternative block-floating schemes.","The ability to vary accumulator precision independently may expose hidden sensitivity of popular mixed-precision methods to intermediate rounding that current hardware hides.","Porting the same control surface to other languages or to a hardware description language would let the same experiments drive silicon design decisions."],"forward_implications":["Researchers can design and test mixed-precision algorithms that target MX formats without waiting for GPU access.","Custom MX variants (different shared-exponent widths, element precisions, or accumulator sizes) can be explored before hardware is fixed.","Undocumented vendor choices for rounding and intermediate precision can be reverse-engineered by matching simulation parameters to observed hardware results.","The same interface can be used to study the numerical stability of existing algorithms when they are moved to block floating-point.","Algorithm papers can report reproducible MX results that others can re-run solely in MATLAB."],"fun_headline_variants":["MXsim simulates OCP MX block floats with custom precision control","MATLAB library for custom OCP MX arithmetic without GPU hardware","Full control of exponents and accumulator in MX format simulation","Experiment with undocumented MX settings via MXsim in MATLAB","Custom-precision MX block floating-point library layered on CPFloat"],"cache_read_input_tokens":128,"weakest_assumption_plain":"That a MATLAB simulation built on CPFloat can reproduce the numerically relevant behavior of real OCP MX hardware well enough for algorithm development, even though the exact bit-wise rules used by vendors remain unpublished.","fun_headline_variants_meta":{"raw":{"variants":["MXsim simulates OCP MX block floats with custom precision control","MATLAB library for custom OCP MX arithmetic without GPU hardware","Full control of exponents and accumulator in MX format simulation","Experiment with undocumented MX settings via MXsim in MATLAB","Custom-precision MX block floating-point library layered on CPFloat"]},"model":"grok-4.5","effort":"low","cost_usd":0.010964,"raw_usage":{"total_tokens":2384,"prompt_tokens":739,"num_sources_used":0,"completion_tokens":86,"cost_in_usd_ticks":109640000,"prompt_tokens_details":{"text_tokens":739,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1559,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":739,"tokens_out":86,"duration_ms":10856,"temperature":1.0,"reasoning_tokens":1559,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-15T02:21:50.827283+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Compare MXsim results for a set of mixed-precision matrix multiplications or iterative solvers against the same kernels run on actual NVIDIA or AMD MX hardware; any systematic difference in rounding error or final accuracy that cannot be explained by a documented parameter would falsify the claim of faithful simulation.","supporting_citations":[],"review_version":1}