The useful distinction is whether a feature has a scene renderer, a controlled numerical experiment, or a geometric explanation. A rendered image proves that an execution path ran; it does not by itself prove unbiasedness, completeness or a performance advantage. This ledger records the current boundaries rather than counting every menu item as a completed estimator.
Surface cone plus plane was already implemented
PP33 offers Cone, Plane, and Cone + plane: balance MIS. The pair retains one total sheet budget and accounts for both families on their common emitted-path measure. The retained comparison contains 54 native renders: three scenes, three choices, six independent seeds.
An older running build may still offer only the earlier two choices. Surface-Curve-MIS.bat opens the updated paired caustic scene through the AIO launcher. Choose Sheet family → Cone + plane: balance MIS.
The measured result is also useful: the pair did not provide an efficiency win in the tested fixtures. The two charts have the same marginal emitted-path density; their difference is within-sheet correlation and sweep orientation. The pair provides a controlled orientation hedge, not a new proposal that hits glass more often. Read the derivation and native comparison.
What already produces native scene images
| Requested direction | Working route | Current boundary |
|---|---|---|
| Arbitrary pairs/trios and sphere + plane | PP31 custom combinations, including UV + sphere; source-size/profile matrices and measured comparisons | Seven kernel-free members share a measure. Curve-family mixtures retain their finite kernels; combining exact and curve groups is an explicitly labeled fixed blend, not extended cross-measure MIS. |
| More source sizes, angular profiles and colours | Multi-emitter matrices 80–82; equal-flux studies 105–106 | Colour, area and angular power are recorded. Equal pass counts are not equal time or equal quality. |
| Volumetric emitters and heterogeneous scenes | Luminous solids 89–90, procedural emitters 91, heterogeneous chambers 92 and 104 | These exercise existing producer support. The GPU heterogeneous branch still has tracking/connection limitations; custom PP31 remains a bounded homogeneous-medium estimator. |
| Mesh and multiple-glass hourglasses | PP32 numerical mesh sheets, prepared sphere/torus/monkey/teapot/chain scenes, additive fast preview | Mesh topology, branch changes and interpolation cuts remain explicit. Fast raster sheets are a biased viewfinder, with separate native comparisons. |
| Surface curves and surface caustics | PP33 actual analytical/triangle receivers and selected specular chains, with stochastic or finite-quadrature accumulation | Finite-depth L[S]*DE scope. It does not implement arbitrary camera-side specular paths, all materials, dispersion, or a general receiver texture-baking system. |
| Light-source translation | Existing actual-scene central finite-difference window, with shared samples and a second step size | Direct supported emitters/receivers only. It includes changed visibility at finite displacement, but is not a complete differentiable renderer or the zero-width boundary construction. |
| Sphere × sphere connection | New point-source/pinhole single-scattering scene renderer with residual-circle sampling and independent camera-ray reference | Real source/sensor PDFs, visibility and bounded homogeneous attenuation; not arbitrary loaded scenes, area sources, multiple scattering or general volume pairings. |
| Sharp surface derivative | New collimated-quad boundary experiment, with editable motion vector, signed/overlay render and finite-difference comparison | Zero world-space kernel; finite film-pixel integration. Fixed transport with local affine receiver projection, not general changing visibility or indirect-light derivatives. |
| Coherent waves on real receivers | New scalar Helmholtz superposition sampled at actual floor/sphere hits, with coherent/incoherent/signed cross-term views and a plane reference | Ideal field probes do not block or re-scatter the optical field. It is not arbitrary-material coherent path tracing. |
| Enhanced raster viewfinder | Opt-in surface guide plus selected PP scattering film and fast sheets | An appearance composite with visible fallback paths. Secondary guide paths do not acquire complete medium/mesh transport. Legacy raster remains available. |
| Projector depth acquisition and unwrapping | Six calibrated native GPU acquisitions: four-shot CW, three frequencies, anchored off-axis sensor/emitter carriers, plane-front and line setups | New ambiguity resolution uses measured phases and declared capture range. Multipath gives apparent range; low signal, sideband overlap and calibration still matter. |
| Take a shot, long exposure, camera movement and aperture controls | Native high-quality capture, linear-film export where available, camera-motion averaging/time-lapse, ideal thin-lens tilt/shift | Scene simulation is frozen during capture. CPU comparisons require supported camera paths; this is not a multi-element commercial lens prescription. |
| Walk around a captured photograph | World-fixed RGB-D plate with first-hit-depth parallax | It preserves one captured view, leaves disocclusion holes and is a viewport overlay. It does not physically diffract, illuminate other objects or reproduce a full light field. |
The scene catalogue, custom combinations, surface implementation, projector imaging and photography guide carry their exact controls and retained evidence.
Computed experiments that must not be mistaken for general scene modes
| Construction | What the earlier experiment actually computed | What that did not establish |
|---|---|---|
| Sphere × sphere and plane × plane | Exact residual circle/line geometry, coarea factors and samples | A scene-radiance estimator with source/sensor PDFs, visibility and all path roots. |
| Ball × parallelepiped and box × box | Bounded overlap quadrature | A bidirectional path sampler over arbitrary overlapping volumes. |
| Temporal ellipsoid, ball/box clipping, finite kernel and delta limit | A normalized geometric integral and dimension reduction at regular level sets | Arrival-time support in the production PP0–33 films. GPU path gates are a separate route. |
| Surface-footprint derivative | A controlled receiver render using a declared smooth edge kernel | An infinitely thin boundary derivative of a general moving area light. |
| FFT wave propagation and hologram record/replay | Actual complex-amplitude propagation, intensity-only sideband recovery and refocusing for controlled apertures/layers | Coherent multiple scattering through arbitrary scene materials, polarization or physically diffracting in-scene holograms. |
| Collision, track length and ratio tracking | Sample histories, estimates and costs for one specified scalar volume integral | Repair or validation of every existing GPU/UPBP medium-transport branch. |
These experiments remain useful foundations. The distinction matters when a new drawing looks correct while the corresponding transport estimator has not yet been connected to a scene. The new sphere-connection renderer, sharp boundary derivative and coherent receiver render provide actual bounded scene experiments for those three follow-ups. The earlier support and time experiments and wave laboratory remain alongside them.
Directions still open
- General transient photon primitives: arrival-time kernels, multisegment prefixes and correctly normalized delta limits are not universally wired through PP0–33. A sphere or ellipsoid picture alone does not complete them.
- General dual-support transport: arbitrary plane/volume pairings, multiple preimages and manifold constraints still need matching measures, proposals, visibility and variance analysis.
- Cross-measure MIS: finite-radius beam/ring/arc and kernel-free sheets need an extended derivation; the shipped exact/curve blend does not claim it.
- General differentiable transport: smooth throughput terms, moving occlusion, refractive interfaces and indirect paths exceed a single boundary fixture or central finite difference.
- Arbitrary coherent materials and sonar: scalar-wave demonstrations do not supply electromagnetic material boundaries, polarization, acoustic impedance or a transducer model.
- Full manifold exploration: the existing specular-aware mutation prototype is not the complete constraint solver from the published algorithm. This remains the previously designated lower-priority direction.
- A universal automatic combination ranking: second moments, support holes, Jacobian tails and cost make the analysis concrete, but there is no scene- independent score that chooses the best combination without assumptions.
The backend and temporal audit and estimator support matrix separate a missing mathematical ingredient from a performance port. A name in the UI, a geometric schematic and an independently checked scene estimator are different kinds of evidence.