The scale cannot see both kinds of engagement
Focused attention is the only subscale off the ceiling because its two items split. Participants reported being absorbed in the activity while declining, as a group, to say they lost themselves in it. Some asked, while filling the form, whether “lost” was something to affirm or to deny.
The interviews name two engaged states. One is immersive absorption. The other is analytic exploration — P8 “purely trying to figure out how what I did affected the visuals,” P10 trying to trip the mapping up. Both are engaged; only the first is what an absorption item is built to detect. A system that invites calibration will be scored as less absorbing than it is.
That is a problem about measuring experience, not a defect in this administration. Items could be paired with a valence check at negligible cost. The UES-SF still did its work: engagement was high, it was not distributed by rowing ability, and focused attention behaved differently from the other three dimensions.
Stopping is a coupling problem, not a rendering one
The field decays when rowing stops, which reads correctly as a hull losing way. What it does not do is stop on command. P8, rowing for the first time, wanted to stop the boat and start again; the water was “always moving past at some speed.” P10, with more than thirty years in the sport, named the real manoeuvre: hold water. His two long pauses are the workaround — wait the field out, then attempt a start.
On the water, holding water is an action with an immediate consequence. In RowSim, stopping is only the absence of action, resolved by a decay the rower waits out. What specifies self-motion is not motion in the visual field but motion that answers to what the observer does. A display that accelerates on demand while decelerating on its own schedule supplies half of that.
P5 identified the complementary limit: wanting motion to persist more strongly when not rowing, so the boat still reads as carrying way. The missing element is a commandable rest rather than any particular resting behaviour.
What this floor licenses
One floor, one palette, one first exposure.
Coupling first. All seventeen described the correspondence, including those who disliked the magenta. Palette and completeness can move; the mapping cannot lag.
An open mapping is spent by exploration, not stored. People who solved it already expected to get bored.
The field should stop when the rower asks it to. Decay is not hold water.
Magenta summoned water and then refused it. People who know what rowing water looks like were the harder audience.
Decide whether you are designing for participation or for training. They want different instruments, and this study measured the first.
Design considerations
Coupling first.
01 · CouplingThe stroke has to answer immediately. Palette and completeness can move; the mapping cannot lag.
Coupling as the wall. Openness, arrest, register, and goal occupy the rooms — negotiable, not structural.
Physics provides continuity; art direction provides legibility
Stable-fluid advection and pressure projection create coherent motion. Noise, bloom, and caustic bias make that motion readable under projection, where contrast and viewing angle fight literal shading. Physics provides continuity; craft provides legibility.
Palette variants exist in the source and were held constant: one magenta hue, no numeric HUD. The result is a perceptually oriented simulation, not a claim of hydrodynamic fidelity. Naming that hybrid keeps the work from oscillating between physics paper and magic demo.
Hybrid model
Two truths, one wake.
Physics buys continuity; craft buys legibility under projection.
01 · PhysicsStable-fluid advection and pressure projection buy coherent motion — continuity before beauty.
Physical continuity and perceptual craft as facing truths that meet in the floor wake.
Hybrid terms with documented roles
The fluid chapter already names the hybrid terms—curl shedding, partial-slip hull bands, dissipation, edge bleed, hash noise—with roles in the shaders. The discussion’s point is narrower: they are not secret sauces. Naming them lets readers see which terms buy continuity and which buy readability.
Force caps per texel and ripple de-duplication windows are the same honesty applied to embodied input. Overlapping Gaussians and coincident ripples are failure modes of dense strokes, not Navier–Stokes features. Documented caps—about 7.2–7.8 locally, 15.0 globally in force kernels, and 6.0× (compute) / 3.0× (fragment) rippleFluidStrength in coupling—keep the wake stable when several producers fire into one neighbourhood.
Honesty about hybrids is itself a design trade-off: the system is easier to evaluate because the craft is inspectable. The publication can defend the incompressible core without treating every perceptual aid as a hydrodynamic claim.
Architecture changes behaviour
As noted in the software chapter, the Apple Silicon compute path includes particle and bloom stages that are absent on the fragment fallback; ripple constants and preferred frame rate also diverge. Those differences are visible, and they are disclosed. The repository does not include controlled cross-device benchmarks, so the publication describes targets and implementation paths rather than measured comparisons.
Half-float storage can accumulate quantisation artefacts over long sessions. Simulation slabs allocate once and stretch on resize rather than re-discretising the domain. Stale or uninitialised auxiliary textures may affect fallback-branch appearance in edge cases. These numerical and rendering limits sit beside the solver’s strengths—semi-Lagrangian stability, RBGS convergence on arm64, ping-pong hazard discipline—so the discussion does not pretend the installation is branch-invariant water.
Failure modes in telemetry are equally concrete. BLE disconnect and reconnect churn can interrupt PM5-driven ownership. PM5 timestamp domains can differ by path and require sink-level normalization. UDP loss and reordering are not fully eliminated by parser logic. Mapper graph churn can produce transient map unavailability. IMU baseline drift and threshold hysteresis can misclassify low-amplitude motion. The discussion names those limits so future measurement work knows where to look.
Ambient feedback changes the evaluation question
If the display is the floor, success is not only solver stability—it is whether attention can stay in the stroke while the wake stays legible. That shifts evaluation from dashboard usability to embodied feedback experience.
Signal families already documented in sensing and system chapters supply the coupling: kinematics to trails, PM5 phase to gated propulsion, IMU orientation to ripples and brightness, watts to ink and peak force to impulse — with radius clamped on the displacement path. The discussion does not re-list those maps; it insists they change what “working” means.
That is why the study and implementation chapters are both necessary: one documents what was built, the other what seventeen people did with it. The seam between them is the integrity mechanism.
When the display is the floor
Embodiment
The floor changes the question.
01 · WasNecessary — but not what “working” means once the display is underfoot.
When the display is the floor, the evaluation question shifts from solver stability to embodied feedback.
Disclose the seam, keep the archive inspectable
Branch divergence, fixed-grid resize behaviour, half-float quantisation, and the absence of in-repo GPU timings are not footnotes to hide. They are the conditions under which the wake is real. Publishing them beside the seventeen-stage pipeline and the completed sessions keeps implemented claims and empirical claims from contaminating each other.
The most useful next sentence is therefore not a stronger fidelity claim. It is a clearer measurement claim: GPU frame time and state-to-photons delay were not logged. Architecture branch remains inferred from the recorded host. Future work can instrument those without rewriting the mapping story the study already ran.
Evidence policy
Keep the seam visible.
01 · BuiltSource-grounded technical claims — what the repository can prove today.
Three claim lanes stay separate above a visible integrity seam. The empirical loop is closed.