Where the evidence reaches, and where it fails

Presented form, empirical tests and the limits of transfer

The current-act account has to earn each connection it makes. Evidence can support an actual visual form without identifying its smallest bearer; a physically well-specified predictor can fail without undermining the testimony on which the form was identified. This chapter brings those cases together, developing the actual native applications while testing the reach of their physical interpretation.

10.1 From an extensive form to a modest native application

The presented-form argument concerns the qualitative organisation of an actual presentation. A sufficiently resolved patch can have differently situated aspects within one apparent extent. A location code, an unresolved point, a remembered shape label and an accurate drawing need not have that organisation. The empirical question is therefore whether the record gives reason to accept the extensive presentation, rather than merely a spatially competent response. The subsequent aspect inference and the concrete-performance/mode principle remain separate steps.

Cortically induced extent

In Beauchamp et al. (2020), phosphene mapping included appearance-directed outline, size and location matching. For the acquired-blind participant 03–281, the single-electrode calibration reports apparent enlargement after brightness saturation. This is positive ordinary evidence for some extensive visual instances: it concerns how the patch appeared, beyond which location or letter was selected. The 120 Hz calibration and the separate 60 Hz current-steering experiment must remain distinct. The latter’s ellipses summarise variability in reported location; they do not measure individual phosphene extent or jointness. Neither protocol supplies a dedicated finite-aspect probe or a trialwise jointness rate. The supported inference concerns the calibrated native sensory regime and its extensive instances, rather than every commanded pulse.

This application removes an intact retinal input and an overt two-object comparison as necessary conditions of the minimum target. It does not remove native sensory processing or identify the stimulating electrode with the whole activity. A physically nominated sensory exercise can include a downstream current constructor. A newly formed extensive presentation at uptake $J$ meets the minimum without requiring an earlier image to survive unchanged.

A surface with historical and current suppliers

van Lier et al. (2009) report that, following adaptation to coloured regions around a grey centre, subsequent achromatic outlines determine differently coloured afterimages extending through the outlined area. Their appearance-directed account and centre-colour matching give modest ordinary warrant for an actual filled surface. The illustration of appearance is explanatory, rather than a photograph of a field. Matching alone would not establish spread. The inspected main report leaves the detailed adjustment-display chronology, gaze history and dedicated extent or basing probes unavailable.

Here a later boundary is a possible supplier of the current form. Its presence defeats an inference that the earlier retained state was already a filled field, but need not defeat the positive interpretation of the later surface. The activity to nominate is the history-sensitive current sensory engagement, including its relevant suppliers and inspection, rather than the old adaptor alone. Acceptance of that sensory interpretation permits the spatial-aspect argument; it does not locate first appearance within either 500 ms outline phase, identify a unique circuit, or supply an independent proof of physical inherence.

An endogenous form during REM sleep

LaBerge et al. (2018) combine physiologically staged REM, prearranged eye-movement signals bracketing a tracking task, and subsequent descriptions of following a perceived dream thumb or nail. The sensory testimony gives modest retrospective warrant for some extensive dream-target instances, although less directly than an explicit enlargement or contour probe. Pursuit geometry alone would not do so. The published sensory descriptions and duration/success criteria select the analysed records; their reported success fraction is not an all-attempt relational-occurrence rate. The supplementary narratives were not recovered for this assessment.

A single actually extensive target can meet the minimum without a second dream object or a simultaneously visible motion path. This revises what the evidence must establish, rather than adding an observation to the study. Its time reference remains retrospective: the physiology nominates a sleep engagement, while testimony attributes the visual form to that engagement. First formation at a current dream-time $J$ is compatible with the inference. Assigning all form only to the waking interview instead challenges the accepted retrospective reference; it does not preserve that reference while merely choosing a later phase. Neither interpretation is settled by eye movements alone.

These examples support separately delimited native applications at ordinary phenomenological strength. They do not silently enlarge the original visual predicate $V$ to every sleeper, implanted system or artificial implementation. For each, the appearance evidence, sensory-activity tether, physical nomination, time reference and constitutive placement must be assessed in their own right. The strongest exclusive-bearer alternative can grant all three actual forms and their genuine acquaintance while reserving qualitative inherence to $\mathbb{B}$. The additional applications therefore extend the reach of the defended current-act reasoning without becoming independent experimental proofs of its bearer principle.

10.2 What the basing repair changes in the original archive

Separating current sensory basing from physical-port constitution avoids requiring the empirical archive to settle a metaphysical identity before its comparison can count as appearance-directed. It does not add a missing measurement. In Sections 7.2–7.4, the initial looked-equal staircase, source and annulus variation, and appearance-directed same/different instructions support the ordinary sensory interpretation of the comparison. Joint presentation and the relation’s operative sensory role require the further ordinary current-scene interpretation. Neither was independently measured by a dedicated probe. Agreement between a key and its assigned meaning establishes instructed response semantics and record custody, rather than phenomenal fidelity of an independently isolated transmission channel.

The archive’s prior exposure, retrospective nomination and consistency screening remain as recorded. Trial 2 retains its documented role in nomination history; it supplies no new observation of jointness or basing. Cached-answer and fresh current-uptake alternatives therefore retain their original force. The positive inference remains defeasible current sensory comparison by uptake, followed by the separately defended placement argument. It has acquired a clearer division of obligations, rather than a stronger empirical calibration.

10.3 A fixed physical/report assay that failed its primary criterion

A retrospective analysis of one deposited no-cue EEG sequence provides a useful test of a more specific physical prediction. The source is OpenNeuro ds006171, version 1.0.0, mirrored as NEMAR on006171 (Melcón et al., 2025). It contains near-threshold visual stimulus annotations, subjective seen/unseen responses and EEG. The analysis retained all 400 physical attempts and the original chronological split of 266 training and 134 test attempts. Its primary present-stimulus cohorts were 239 training and 121 test attempts; 27 and 13 catch attempts remained separately accounted for. The reports label a report endpoint. They are not independently calibrated probabilities of actual appearance.

The fixed neural operator used raw ports A15 and A28, each referenced to raw A1, with their nominal standard-cap locations O1, O2 and Cz. Individual wiring and local cortical-generator identity were not established by these names. For each electronic stimulus marker with zero-based sample index $n$, it used the 1024 samples in $[n-1152,n-128)$ at 1024 Hz: one second of data ending 125 ms before the marker. After mean removal and a symmetric Hann taper, the one-sided periodogram was summed over the fixed 8–14 Hz integer bins. Write its two referenced powers $P_L,P_R$ and put

$$D=\frac{P_R-P_L}{P_R+P_L},\qquad G=\log\!\left(1+\frac{P_R+P_L}{1\,\mu\mathrm{V}^2}\right).$$

Exactly zero bilateral energy uses $D=0$ with a separate indicator; no row in the actual sequence had zero energy. Common additive voltage cancels before power calculation, but channel-specific artifacts, reference physiology and unequal mixing do not. The statistic is available at the buffer’s closing cut; it is not an instantaneous phase estimate or a phenomenal clock.

The candidate addition was source alignment $K=HD$, where $H=-1$ for a left and $H=+1$ for a right present stimulus. The 21-column baseline dictionary contained an intercept, training-standardised current contrast $c$ and global power $g$, $c^2$, $g^2$, $D$, $D^2$, $H$, orientation, linear trial position, $Hc$, $Hg$, the zero-energy indicator, the preceding contrast on each side with availability indicators, and the preceding four present-attempt report means on each side with available counts divided by four. Only histories delivered before the current buffer cut were available. Test histories could advance from preceding test responses; fitted coefficients did not update. This accommodates specified adaptive-history proxies rather than recovering the original complete contrast-assignment law.

Training least squares used the minimum-norm SVD solution with singular values retained above $10^{-12}$ times the largest. With training design matrix $X$, let $\beta_0=X^+Y$, $\gamma=X^+K$, and $v=K-X\gamma$. For nondegenerate $v$, the added coefficient is $\theta=v^\top Y/(v^\top v)$. The two probabilities before common clipping to $[0,1]$ are

$$p_0(x)=x\beta_0,\qquad p_1(x,k)=x\beta_0+\theta(k-x\gamma).$$

This is partial regression in a declared finite predictive family. It is not a nonparametric rejection of every global-state or response-criterion account. Bands, ports, sign, window, split and model were fixed before analog-feature inspection. Prior source and outcome exposure nevertheless make this a retrospective computational holdout, not a blinded prospective test.

On the common 121 present test attempts, the primary mean Brier losses were

$$\overline{(p_0-Y)^2}=0.2854130516503272,\qquad \overline{(p_1-Y)^2}=0.29835706393830547.$$

The added-minus-baseline difference was $+0.012944012287978259$: the neural addition was worse on its declared primary criterion. Secondary seen-positive AUC increased from $0.5398907103825137$ to $0.5637978142076503$; that does not reverse the primary outcome. Three test probabilities of the added model were clipped, and 30 test contrasts lay outside the training range; both diagnostics were retained without post-score exclusions. A preceding contrast-only isotonic forecast had also failed: Brier $0.3077816495$ versus its constant baseline $0.2507254066$ on the same 121 present suffix attempts. These are different fitted comparisons and must not exchange their baselines.

The full recording’s byte identity, all 400 nominated electronic code entries, fixed prefixes, and the subsequent matrix/prediction/score calculations were checked during the analysis. Separate implementations verified selected direct Fourier calculations and all present design rows and scores. These are computational checks, not independent human review. During preparation of version 1.1, the small released derived table was refitted and the recorded scores were recomputed; complete recording extraction was not rerun. Versions 1.2, 1.3 and 1.4 retain those results without another fit or extraction. The accompanying reproducibility archive supplies the derived table, fixed predictions, portable refit/score and optional raw-prefix checking code, raw-input acquisition instructions and expected hashes.

The failed addition rejects the precise improvement claim tested. It establishes neither absence of seeing nor irrelevance of neural alpha, and it does not undermine the ordinary native current-act argument. One exposed sequence, adaptive contrast, drift, model misspecification and possible sensory/attention/criterion alternatives restrict extrapolation. No independent-person count, iid interval, permutation significance or causal effect is inferred. Even an improved score would not, by itself, identify same-field sensory basing, exclude first presentation at $J$, or select physical inherence against $\mathbb{B}$.

10.4 A felt act and its bodily execution

Desmurget et al. (2009) supply a pertinent boundary on inferring execution from experience of acting. In seven awake neurosurgical patients, posterior stimulation could elicit reported urge without observed movement or detected activity in the monitored muscles; stronger stimulation could elicit illusory movement reports. Premotor stimulation could elicit movement without reported awareness of it. Responsive-site selection, different clinical cohorts, limited muscle monitoring and the absence here of the detailed supporting protocol restrict the comparison. It is not an all-trigger randomized double dissociation or a measurement of global phenomenal absence.

A genuine feeling of execution can thus misrepresent whether the nominated bodily movement occurred. This does not make the feeling unreal, remove its physical induction, or show that felt motives never contribute to action. Likewise, movement without reported awareness of that movement does not imply an entirely experience-free subject. What the current-act argument must track is the actual sensory or comparative activity, with its operative connection to any further execution independently warranted. The represented act, its present qualitative character and the peripheral bodily result cannot simply be identified. The study supplies a concrete limit on that transfer, rather than a new relational-minimum instance, mode-preserving intervention or bearer test.

10.5 What the additional evidence warrants

The new empirical integration is selective. Extensive-form testimony supports bounded actual native applications of the aspect argument. A failed fixed predictor and a causal execution dissociation discipline the proposed connections between those appearances and physical records. Neither positive testimony nor negative prediction is assigned authority over every other inferential step.

The resulting position remains positive at the defended grain: actual native sensory performances can be warranted as having current relational modes, and the concrete-performance/mode argument supports their physical attribution. Broader occurrence conditions, fine phase identification, dedicated same-field basing and stronger material constitution retain their separate outstanding obligations.