{"id":"2ce9d454-f3e1-4df4-8ae4-1cdf45c0e4ae","arxiv_id":"2412.16735","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"A voxel-based combustion simulator for arbitrary wooden structures that matches the general shape of measured temperature and mass-loss curves from a real burning-block experiment, with two-way coupling to deformation.","lead":"FlameForge is a simulator that lets wood, plastic, and stone structures catch fire, burn, and deform, using a voxel grid plus a signed distance field to model the insulating char layer that slows combustion. It matters because it brings general-geometry fire simulation, including a real-world temperature comparison, to computer graphics and possibly fire-safety preview.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed generalization to arbitrary wooden structures rests on Eq. (8)'s nearest-surface SDF proxy for char insulation; the cube validation cannot distinguish this proxy from the path-integrated char effect, so the qualitative match is geometry-specific evidence.","rationale":"The paper is best read as a systems contribution with a hedged validation claim. The central assertion is not exact quantitative agreement but reproduction of the qualitative signature of char insulation: convex sub-surface versus concave core temperature curves. That signature is produced by Eq. (8), so the SDF proxy for char insulation is load-bearing. The paper itself concedes the proxy is a simplification of a path integral (Section 3.2.3), that exact calibration is only partially addressed and the slow rise of inner sensors in the non-charring case is not reproduced (Section 5.3), that oxygen starvation and fracture are unmodeled (Section 6), and that the two-way coupling is constrained by temperature propagation (Appendix A). These are honest limitations, but they do not by themselves falsify the central claim. The residual risk is that the validation geometry—a solid cube with thermocouples at two depths—is exactly the case where nearest-boundary distance is a reasonable proxy, so it cannot certify the arbitrary-geometry scope. The reader's weakest_assumption identifies this same risk, and the concrete test above would settle it. Given the hedged wording and the acknowledged simplifications, the conditional verdict remains appropriate; no stronger or weaker verdict is indicated.","tokens_in":14443,"tokens_out":10838,"duration_ms":103979,"concrete_test":"Repeat the Section 5.3 protocol on an L-shaped or hollow wood sample, placing one thermocouple in an internal concave corner where the SDF distance is small but the path-integrated char insulation is large, and another in a straight section at the same physical depth. Simulate the same geometry twice: once with the paper's SDF proxy in Eq. (8) and once with a path-averaged insulation integral (or a resolved 1D char-layer model). If the two models agree on the solid cube but diverge for the concave-corner sensor by more than the experimental spread, the cube validation is insufficient for arbitrary geometry. If the measured corner temperature matches the path-integrated prediction and not the SDF proxy, the proxy is falsified for the claimed scope.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.2.3 admits that faithful char insulation for volumetric structures 'would require solving an integral over all possible paths from an inner voxel to the materials surface', then replaces that integral with Eq. (8), using a single signed-distance query h to the nearest boundary: c = cmin + (1 - cmin) exp(-h cr). The text also states that virgin wood and char are treated identically. This distance-only coefficient is the mechanism that produces the claimed validation feature in Section 5.3 (convex sub-surface vs. concave core temperature curves) for the 40 mm cube. For a convex cube with thermocouples at 5 mm and 20 mm depth along the same normal, the nearest-boundary distance is a reasonable proxy for char-layer depth only until the boundary recedes and the SDF is updated. For the 'general wooden structures' claimed in the title and abstract—thin walls, concave corners, internal voids, detached char—nearest-boundary distance is not the char path length. A voxel in a concave notch or behind a detached char piece can have a small SDF distance and therefore c near 1, even when the physically relevant path-integrated insulation is large. The cube experiment cannot distinguish the proxy from the true path-integrated effect, so the match to two thermocouple curves is geometry-specific. The house, boat, bridge, and tower scenes have no sub-surface sensor data, so they do not close this gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents FlameForge, a voxel-based combustion simulator for generalized wooden structures. The system couples a low-resolution compressible-flow air solver with a high-resolution material grid, uses a signed distance field to evaluate char insulation, supports charring and non-charring materials, and demonstrates two-way coupling with position-based dynamics. The method is validated qualitatively on house, boat, tower, tree, and bridge scenes, and quantitatively against a new small-scale experiment in which cubic wood and PMMA samples are burned in an oven while subsurface temperatures and mass loss are recorded.","tokens_in":14663,"tokens_out":4637,"duration_ms":39810,"significance":"If the central claims hold, the paper offers a useful unification: a single volumetric framework for charring and non-charring solids with adaptive grids, SDF-based surface queries, and a working bidirectional coupling to a mechanical rod solver. The numerical building blocks (staggered MacCormack advection, AMGCL projection, fast sweeping for the SDF, OpenVDB data structures) are standard and appear to be implemented carefully. The inclusion of new external experimental data is a genuine strength, and the qualitative reproduction of the convex subsurface versus concave core temperature feature for wood is a promising indicator. However, the quantitative evidence is weaker than the abstract and Section 5.3 claim, and the char-insulation model that drives the headline validation is an acknowledged simplification whose generalization to arbitrary geometry is not directly tested.","major_comments":[{"comment":"The central generalization claim is only supported by a geometry-specific validation, because the char-insulation mechanism in Eq. (8) uses nearest-surface distance rather than the path-integrated insulation that the text itself identifies as physically correct. For concave regions, thin walls, and detached char, the proxy and the true path integral diverge, and the cube experiment cannot detect this divergence.","section":"Section 3.2.3, Eq. (8)"},{"comment":"The paper's headline quantitative claim is based on plots without error bars, metrics, or repeated trials; the admitted PMMA discrepancies show that the match is only qualitative. This is a core validation gap for a paper that claims to \"quantitatively assess\" the volumetric combustion process.","section":"Section 5.3, Figure 8"},{"comment":"Without exact parameter values and a sensitivity analysis, the reader cannot tell whether the good qualitative match on the cube is robust or the result of favorable parameter selection. This is especially important because many parameters are inherited from Pirk et al. [PJH*17] and are only given as ranges.","section":"Table 1 and Section 3.2.4"}],"minor_comments":[{"comment":"The caption labels appear to be copy-pasted from the charring panel and make the figure harder to read.","section":"Figure 8 caption"},{"comment":"This is a citation/presentation issue that should be corrected.","section":"References, [HKV07]"},{"comment":"The unit notation is confusing as written and should be reconciled with the relative-mass formulation.","section":"Table 1, units for TMc and TMv"},{"comment":"A reader has to infer that Eq. (9) actually describes mass-loss rates (negative time derivatives of stored mass).","section":"Eq. (9) and Eq. (11)"}],"recommendation":"major_revision","confidential_remarks":"The experimental data is new and external, which is a genuine strength, but the validation set is small (one cube per material) and the claimed generalization rests on the SDF proxy in Eq. (8). The overlap of authors with Pirk et al. [PJH*17] is not by itself a problem, but the transfer of calibrated parameters and the absence of exact values in Table 1 make the quantitative comparison harder to assess. A revision that adds error metrics, exact parameters, and a non-convex char-insulation test would materially strengthen the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: FlameForge is a well-built systems paper that gives the fire-simulation community a single volumetric pipeline for arbitrary wooden geometry with multiple materials and two-way PBD coupling, and it earns its keep with a small but genuine instrumented experiment. The SDF-only char-insulation proxy is a real simplification, and the cube experiment can't distinguish it from the fuller path-integral effect; but the paper's own hedge about reproducing general trends is defensible.\n\nWhat's new: the combination of a voxel grid for combustion with an SDF for surface distance queries, adaptive multiresolution storage, multi-material support (wood, acrylic, stone), and coupling with position-based dynamics. Prior work (Pirk 2017) was tree-specific; Nielsen 2022 did fluids; Di Blasi gave the pyrolysis model. The glue is the unifying framework and the experimental validation. The implementation is sound: staggered grid, MacCormack advection, AMGCL projection, fast sweeping. The qualitative scenes (house, boat, tower, bridge) show the range.\n\nCredit where due: they ran a real combustion experiment with thermocouples at 5 mm and 20 mm depths plus mass loss, and they compare simulation to experiment honestly, including noting where the sensors detach and the slow PMMA rise isn't reproduced. That's more than most CG papers do.\n\nSoft spots: the quantitative section is thinner than it looks. Time is normalized to full burn duration, so the comparisons lose temporal scale; there are no error metrics; Table 1 lists ranges for key parameters (T_m0, T_m1, epsilon_c) but not the actual values used, which leaves tuning ambiguity. No code or data is released. The char insulation via Eq. (8) is a single SDF distance h to the nearest boundary, while the paper admits the correct computation is a path integral. For a convex cube with thermocouples on the same normal, nearest-boundary distance is a reasonable proxy, so the cube match doesn't validate the proxy for thin walls, concave corners, or detached char. The house/boat/bridge scenes are visual only. These are real limitations, but they don't sink the paper because the claims are hedged and the contribution is primarily a practical simulator, not a physics claim.\n\nThe stress-test note is right about the proxy being unvalidated in complex geometry, but it's not a load-bearing failure: the paper says 'general trends,' and the SDF approximation is a deliberate simplification. A serious referee should push for parameter disclosure and error metrics, and maybe a thicker-wall or two-sensor-depth variant, but the paper deserves a slot in the revision process.\n\nWho it's for: graphics researchers working on fire/destruction simulation, and people who need a fast visual approximation of structural burning. Not for combustion scientists looking for quantitative accuracy.\n\nRecommendation: send to peer review, conditional on addressing the parameter disclosure and adding at least one error metric or a sensitivity study.","headline":"A solid systems paper that unifies existing combustion models into one volumetric pipeline; the SDF char-insulation proxy is a real limitation, but the hedged claims and genuine experiment make it worth refereeing.","tokens_in":15396,"tokens_out":3388,"would_cite":true,"duration_ms":29499,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"FlameForge claims a unified voxel-based simulator reproduces multi-phase combustion of arbitrary wooden structures, matching real interior temperature and mass-loss measurements.","keywords":["combustion simulation","charring materials","signed distance field","voxel grids","fire spread","wood combustion","position-based dynamics","multiresolution"],"falsifier":"Measure sub-surface temperature histories in an oven for a thin-walled wooden box whose wall thickness is comparable to the char layer thickness, or for a block with a detached char layer; if the simulated curve does not reproduce the real curve because the signed-distance proxy over- or under-insulates, the central claim fails.","tokens_in":14069,"feed_emoji":"🔥","tokens_out":4205,"duration_ms":34644,"temperature":0.7,"pith_summary":"FlameForge is a volumetric combustion simulator that claims to capture the full multi-phase burning of arbitrary wooden structures: heating, pyrolysis, char formation, and combustion of volatiles and char on a voxel grid, with a signed distance field standing in for the geometry needed to compute the insulating effect of the char layer. The paper argues that this simplified model is enough to reproduce the general trends of real combustion: simulated mass-loss curves match the parabola shape of charring wood and the S shape of non-charring acrylic, and simulated interior temperatures share the convex sub-surface and concave core structure seen in thermocouple measurements of burning cubes. If true, this gives computer graphics and fire-safety studies a way to predict how whole structures such as houses, boats, towers, and bridges deform, collapse, and spread fire across different materials and wind conditions, without resolving the full chemistry of combustion.","feed_headline":"Simulator matches real wood-fire temperatures inside the material","feed_subtitle":"Volumetric charring, mass loss, and fire spread across arbitrary wooden geometry now line up with oven experiments.","key_machinery":"The engine of the method is the combination of a multiresolution voxel grid for the material and a signed distance field that is iteratively updated as mass is consumed. The char insulation coefficient $c = c_{\\min} + (1 - c_{\\min}) \\exp(-h c_r)$, where $h$ is the distance of a voxel from the material boundary, replaces the exact integral over all possible paths to the surface that true char insulation would require, and virgin wood and char are treated identically. Around this sit a one-step global pyrolysis model converting solid fuel into volatiles and char, a temperature-dependent reaction rate with a smooth interpolation function, and a bidirectional air-material temperature coupling that lets flames ignite material and combusting material feed flames.","core_discovery":"On its own terms, the paper's central claim is that the multi-phase combustion of charring materials can be simulated at the scale of whole structures with a unified volumetric model. Wood pyrolysis is treated as an instantaneous one-step decomposition into volatiles and char; the char layer's insulating effect, which slows combustion of the material underneath, is computed from a signed distance field giving each voxel's distance to the material surface; and the air, temperature, smoke, and material fields are coupled through a multiresolution adaptive voxel grid. The quantitative evidence is a set of 40 mm cube experiments in an oven: real and simulated wood samples both show a parabolic mass-loss curve and the same convex-sub-surface/concave-core temperature split attributed to char insulation, while non-charring PMMA shows an S-shaped mass curve and a rapid inner temperature rise. The paper concludes that the simulator successfully reproduces the general trends of the measured curves, providing the basis for accurately capturing combustion phenomena, while noting that exact calibration to a given real environment is only partially addressed.","pith_inferences":["The distance-to-surface proxy for char insulation is the likeliest point of failure: for thin walls, concave corners, or char layers that detach from the surface, a single signed distance cannot encode the integrated shielding along all paths, so the cube match may be coincidental for geometry that differs from a solid block.","A direct test would be to run the same oven experiment on a thin-walled box or a slab with a deliberate void; the simulator's char-insulation term would predict a different interior temperature history than a path-integral model, and the measured curve would decide.","The model contains no oxygen transport, so in enclosed spaces the simulated fire will likely burn longer than a real fire that self-extinguishes from oxygen starvation; adding a fuel-oxygen reaction term would change extinction behavior.","The claimed generality across wood species rests on a single calibration of a handful of parameters; transferring to different moisture contents or species-specific pyrolysis rates would require re-calibration from new measurements."],"forward_implications":["Arbitrary wooden geometries, from thin roof slats to thick walls, can burn in one framework, with fire spreading across connected and disconnected objects through the air.","Non-charring materials such as acrylic glass and non-combustible materials such as stone can be mixed into the same scene, changing how fire spreads and what remains standing.","Two-way coupling with position-based dynamics makes mechanical deformation, including the sagging of a burning bridge, follow from the combustion state.","Wind can be the deciding factor in whether a fire spreads from one object to a neighboring structure, as shown by the tree-to-house scenes.","The cube experiments give a quantitative baseline that the authors say provides the basis for accurately capturing combustion phenomena, even though exact calibration to a specific environment remains open.","If the simulator's distance-to-surface proxy for char insulation fails for thin walls, concave corners, or detached char layers, the cube-experiment match would be coincidental and the generalization to house, boat, and bridge scenes would not be supported."],"supporting_citations":[{"why":"Supplies the char insulation model of Eq. (8) and the combustion parameter values used in Table 1.","marker":"[PJH∗17]"},{"why":"Provides the one-step global pyrolysis model that converts solid fuel into volatiles and char.","marker":"[DB93]"},{"why":"Provides the physics-based combustion framework for the air flow, density-temperature relation, and temperature advection.","marker":"[NBHSB22]"},{"why":"Supplies the heat release rate from char oxidation used for the remaining parameter values.","marker":"[MLH23]"},{"why":"Provides the fast sweeping method used to compute the signed distance field from the voxel volume.","marker":"[Zha05]"},{"why":"Provides the OpenVDB multiresolution voxel-grid implementation used for the adaptive data structures.","marker":"[Mus13]"}],"fun_headline_variants":["Wood-fire simulation matches real internal temperature curves","Charring-wood combustion model reproduces oven temperature data","Volumetric simulator captures char insulation and fire spread","FlameForge: unified wood combustion simulation validated in 3D"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything quantitative rests on the assumption that the char layer's insulation can be summarized by one voxel's distance to the material surface, with virgin wood and char treated identically; if that proxy fails for thin walls or concave geometry, the match with the cube experiment is a coincidence and the larger scenes are unsupported.","fun_headline_variants_meta":{"raw":{"variants":["Wood-fire simulation matches real internal temperature curves","Charring-wood combustion model reproduces oven temperature data","Volumetric simulator captures char insulation and fire spread","FlameForge: unified wood combustion simulation validated in 3D"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000865,"raw_usage":{"total_tokens":3732,"prompt_tokens":910,"completion_tokens":2822,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":526,"completion_tokens_details":{"reasoning_tokens":2756}},"tokens_in":526,"tokens_out":2822,"duration_ms":22172,"temperature":1.0,"reasoning_tokens":2756,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:16:43.417858+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure sub-surface temperature histories in an oven for a thin-walled wooden box whose wall thickness is comparable to the char layer thickness, or for a block with a detached char layer; if the simulated curve does not reproduce the real curve because the signed-distance proxy over- or under-insulates, the central claim fails.","supporting_citations":[],"review_version":1}