Verification: Mind Uploading Verification Commons

At L4, The Test Design Itself Is The Bottleneck

For identity claims, preregistering what is being treated as continuous matters more than merely reporting a high score. If you want a beginner guide to memory, values, learning, branching, and longitudinal continuity, see Wiki: Identity Evaluation and Continuity Tests.

If You Get Stuck At The L3 Entry Point

Closed-loop work needs more than offline accuracy. It also needs end-to-end latency, jitter, drift handling, safety-stop design, and an explicit statement of which body / environment loops were preserved or replaced. For a beginner guide to the timing side, see Wiki: Closed Loops, Latency, Jitter, and Safety Stops.

Non-goals

This page is not where mind uploading is declared possible or impossible. What Mind-Upload is building is a verification substrate that can eventually justify such a claim, meaning the rules for measurement, evaluation, and falsification.

A School-Test Analogy

The roles of these four components become easier to see with a school-test analogy. The Data Standard is the same answer sheet, the Benchmark Suite is the same grading rubric, Registry & Prereg is the set of rules distributed before the exam, and Leaderboard & Model Cards is the report card that includes not only scores but also how the test was solved and where mistakes occurred.

If You Get Stuck On The Difference Between Standards, Repositories, And Validators

BIDS, OpenNeuro, PhysioNet, the BIDS Validator, and benchmarks are all parts of research infrastructure, but they do different jobs. For a beginner explanation of that division of labor, see Wiki: Standards, Repositories, Validators, and Benchmarks.

If You Get Stuck On Updates, Branching, And Stop Rules

Even with a registry in place, it is easy to get stuck on how much updating to allow, how to log branches, and how to distinguish stop rules from kill switches. For a beginner guide to those issues, see Wiki: Update, Branching, and Stop Rules.

Raw EEG Alone Is Not A Data Standard

Comparable inputs require more than waveform files. They also need event markers, stimulus logs, time synchronization, and records of bad channels / bad segments. If those remain ambiguous, later replication of the same task becomes impossible. For a beginner guide, see Wiki: Event Synchronization and Measurement Logs.

If This Still Feels Too Abstract, Use The Walkthrough

If you want to see how these four elements actually fit together in a small EEG example, step by step, see Wiki: Verification Example Walkthrough. The public page keeps the blueprint concise, while the tutorial material lives on the wiki side.

If You Want Only The Minimum L0 Artifact Set First

This page is the blueprint for the full public-good stack. If you want the shortest path to a single pack containing BIDS, the Validator, QC, splits, baselines, execution steps, and failure cases, see Wiki: Minimum Artifact Pack For L0.

At L1 And Above, "What Was Directly Observed" Is Part Of The Artifact

The weakness exposed in this re-audit was that even when hidden states were listed, the artifact still did not fix which measurement stack directly observed which variables. Accordingly, for results at L1 and above, we attach an Observability Budget in addition to the standard model card, including the claim ceiling and abstention conditions.

If You Get Stuck On Operational Terms

Baselines, benchmarks, preregistration, model cards, and failure cases all matter, but they do not play the same role. If you want that difference organized from the ground up, see Wiki: Baselines, Preregistration, and Model Cards.

How To Transfer A Pattern Correctly

Patterns such as PDB (a single archive), BIDS plus OpenNeuro (standard plus repository), PhysioNet (data plus evaluation), and OSF/PROSPERO (preregistration) create structures in which progress can be measured even across different fields. WBE especially requires success conditions and falsification conditions to be fixed in advance.

Why We Do Not Copy Historical Cases Unchanged

WBE includes strong issues such as identity and causal equivalence, so it is not completed by data sharing alone. Even so, the ordering itself, putting standards, repositories, benchmarks, registration, and audit in place first, can be borrowed quite strongly from successful cases in other fields.

The Frank Current State

For scientific integrity, we state the current implementation status of each deliverable explicitly. Please do not confuse "the design document is complete" with "the implementation is complete."

The generic gap list was too weak after the identifiability audit

Villaverde (2019) and Villaverde et al. (2019) separate observability from identifiability, Prinz et al. (2004) showed that similar circuit activity can arise from disparate parameters, Rasero et al. (2024) showed that similar human activation patterns can still hide different macroscopic network states, Beiran & Litwin-Kumar (2025) showed that connectome-constrained predictors remain degenerate until additional recordings collapse the compatible dynamics, and Liu et al. (2025) showed that practical identifiability depends on data-collection policy itself. Therefore, the central question for this site is no longer only whether a measurement stack is rich, but whether it actually rules out the main alternative explanations.

2026-03-20 addendum: observability and identifiability must be logged separately

The remaining weakness after adding the Observability Budget was that richer measurement could still be overread as if it had already collapsed the solution set. The primary literature does not support that shortcut. Villaverde (2019) reviewed that observability and structural identifiability are different theoretical questions, and Villaverde et al. (2019) showed that unknown states, parameters, and inputs must often be treated jointly rather than as separable audits. In neuroscience, Prinz et al. (2004) showed that similar circuit activity can arise from disparate parameters, Rasero et al. (2024) showed that similar human activation patterns can still hide different macroscopic network states, and Beiran & Litwin-Kumar (2025) showed that connectome-constrained recurrent networks remain degenerate until additional recordings collapse the space of compatible dynamics. Liu et al. (2025) then showed that practical identifiability depends on experiment design and data-collection policy, not only on the estimation algorithm. Therefore, this site now asks inverse and model-based claims to attach an Identifiability Card on top of the Observability Budget.

2026-03-28 addendum: ambiguity class must be named before a richer protocol counts as informative

The remaining weakness after adding the Identifiability Card was that ambiguity could still be described as one generic scalar called more data needed. The primary literature does not support that shortcut. Massonis & Villaverde (2020) showed that structural unidentifiability can be generated by Lie symmetries and may require symmetry-breaking observables or reformulation, Prinz et al. (2004) and Beiran & Litwin-Kumar (2025) showed that distinct parameters can still generate near-equivalent dynamics even when connectivity is fixed, White et al. (2016) showed that extra experiments can tighten nominal uncertainty while mainly exposing omitted-mechanism error, and Langdon & Engel (2025) showed that preserving causal interactions among task variables can recover computations that correlation-only reductions miss. Therefore, this site now asks authors to name the ambiguity class before calling a new condition or modality informative.

2026-03-28 addendum: informative protocol must target identifiability, stress model mismatch, and declare minimum sufficiency

The remaining weakness after adding an experiment-design leverage row was that authors could still write generic words such as multimodal, naturalistic, or closed-loop without stating why the chosen protocol should actually separate the surviving alternatives, which design objective selected it, or whether the new condition merely exposed model mismatch. The primary literature does not support that shortcut. Diop & Fliess (1991) made explicit that observability / identifiability depend on persistent trajectories rather than on outputs in the abstract, and Raue et al. (2010) showed that identifiability / observability analysis can be used iteratively to design new experiments rather than only criticize old ones. Chis et al. (2016) then showed that sloppiness is not identifiability and that experiment design should optimize explicit identifiability criteria rather than proxy notions of being merely less sloppy. White et al. (2016) showed that complementary experiments can make previously omitted mechanisms relevant, so a design can tighten nominal parameter uncertainty while simultaneously creating large model discrepancy. In neuroscience, Beiran & Litwin-Kumar (2025) showed that recordings from a small targeted subset of neurons can remove degeneracy in connectome-constrained networks and even prioritize which neurons should be recorded next, while Langdon & Engel (2025) showed that preserving causal interactions among task variables can recover behaviorally relevant computations that correlation-only reductions miss. Gevertz & Kareva (2024) further showed that identifiability analysis can be used to derive a minimally sufficient measurement schedule, and Liu et al. (2025) showed that active learning can reach practical identifiability with markedly fewer observations. Therefore, this site now requires experiment-design leverage to explain not only what extra protocol element was added, but also which identifiability objective selected it, how omitted-mechanism stress was checked, and what minimum-sufficiency criterion stopped further data collection.

Minimum operating rule

If this card is missing, this site stops at predictive fit, localized source hypothesis, or model-conditioned mechanism. It does not promote the result to unique internal-state recovery, state-complete reconstruction, or mechanism uniquely identified language.

2026-03-20 addendum: multimodal and atlas-prior results need a Fusion Card

The remaining weakness after adding the Observability Budget was that the words simultaneous, multimodal, or atlas-informed could still be overread as if the fusion step itself had already been validated. The primary literature does not support that shortcut. Kothe et al. (2025) showed that synchronization middleware can align streams, but does not by itself certify device-side delay truth or biological equivalence. Wei et al. (2020) showed that EEG-fMRI fusion remains a model-conditioned inference problem. Nguyen et al. (2016) then made the temporal limit explicit: even in spatiotemporally constrained EEG-fMRI source imaging, the temporal mismatch between EEG and fMRI still persists. Ripp et al. (2021) showed the same issue on the PET side by treating simultaneous FDG-PET/fMRI working-memory data as scan-window averages, reconstructing PET baseline from 44-60 min and task uptake from 63-71 min post-injection under an assumed steady state. Vafaii et al. (2024) and Chen et al. (2025) then showed that even simultaneous multimodal acquisition can reveal both convergent and divergent structure across modalities rather than one self-validating ground truth. Bolt et al. (2025) and Özbay et al. (2019) further showed that low-frequency/global fMRI-linked components can carry autonomic physiology, so a shared factor is still not automatic target-variable specificity. A second operational limit is that a bundle gain is not automatically robust: Rohaut et al. (2024) showed that adding modalities can reduce uncertainty and improve prognostic accuracy in acute brain injury, but Amiri et al. (2023) showed that direct same-sample multimodal comparison can shrink to a 48-patient complete-feature subset, and Manasova et al. (2026) showed that missing-modality handling, centre transfer, and inter-modality disagreement remain active bundle-level issues. Therefore, this site now asks multimodal or atlas-prior claims to attach a Fusion Card on top of the Observability Budget.

Same session is not yet the same temporal object

Even when several streams are acquired together, this site still asks what each stream means in time. Nguyen et al. (2016) explicitly noted that temporal mismatch persists in EEG-fMRI source imaging, and Ripp et al. (2021) showed that simultaneous FDG-PET/fMRI task data still rely on PET windows spanning many minutes rather than event-scale timing. Chen et al. (2025) then made the multi-timescale point concrete in simultaneous EEG-PET-MRI across wakefulness and NREM sleep. Therefore, same-session wording is not enough on this site unless the Fusion Card states whether the paper is aligning one instantaneous state sample, one shared transition, or only coordinated dynamics across different temporal kernels.

Minimum operating rule

If this card is missing, this site reads a multimodal or atlas-prior result at the ceiling of the strongest individually supported stack, not as same-subject, cross-stack, externally calibrated state identification. An atlas prior plus one live measurement remains an atlas-conditioned measurement, not automatic state completeness. A synchronized common factor without shared-vs-specific disclosure remains fused evidence, not one solved biological state variable. A same-session bundle without an explicit effective-window / temporal-kernel relation remains synchronized multi-timescale evidence, not one matched temporal object. A gain shown only on a narrow complete-feature subset, after unresolved missing-modality imputation, or without a disclosed transfer window remains bounded bundle-performance evidence, not robust multimodal state evidence.

2026-03-21 addendum: proxy-rich human evidence must be composed explicitly, not rhetorically

After adding the Observability Budget and the Fusion Card, one weak point still remained: several living-human proxy rows could still be listed side by side as if coverage automatically added up. The primary literature does not support that shortcut. Johansen et al. (2024) provide a 33-participant SV2A atlas calibrated to autoradiography, which is a cohort-level synaptic-density proxy. Lucchetti et al. (2025) define a five-metabolite within-subject similarity graph in 51 healthy adolescents with 13-person site replication, not kinetic flux imaging. Ren et al. (2015) provide a resting 31P metabolite / pH route, Ren et al. (2017) provide a 7 T MT exchange-flux route, Guo et al. (2024) provide a whole-brain intracellular NAD map, and Kaiser et al. (2026) provide a task-evoked 31P fMRS NAD⁺ route; those are already four different inferential objects before the deuterium rows are added. Li et al. (2025) then provide 7 T dynamic DMRSI kinetic maps in five healthy participants. Baadsvik et al. (2024) provide myelin-bilayer mapping in two healthy volunteers on specialized hardware. Hirschler et al. (2025) provide a specialized 7 T CSF-mobility route whose whole-brain rest maps were shown in 20 healthy younger individuals, with driver analyses reported in 11 of 24 total healthy participants. Dagum et al. (2026) infer sleep-linked glymphatic clearance through older-adult crossover cohorts, an investigational wearable, and a compartmental model. These routes differ in direct observable, quantity type, time window, spatial unit, model burden, deployment maturity, and even the likely nuisance sources that can make rows move together. Vafaii et al. (2024) showed both common and divergent organization across simultaneous Ca²⁺ and BOLD, Chen et al. (2025) showed tightly coupled global progressions alongside two distinct network patterns in simultaneous EEG-PET-MRI, Bolt et al. (2025) showed that a major global fMRI mode is substantially coupled to autonomic physiology, and Epp et al. (2025) showed that significant task BOLD changes can coexist with opposite oxygen-metabolism changes across many cortical voxels. Therefore, this site now requires a Human Proxy Composition Card whenever multiple living-human proxy rows are used together to raise a claim ceiling.

A second correction follows from robustness rather than taxonomy alone. Finnema et al. (2018) showed that even a comparatively stable SV2A PET route still needed route-specific kinetic modeling and yielded mean absolute test-retest reproducibility of 3-9% for regional V_T. Holiga et al. (2018) showed that common task-fMRI and resting-fMRI measures span poor to excellent test-retest reliability rather than one uniform level. Wirsich et al. (2021) then showed that some simultaneous EEG-fMRI connectome relationships can reproduce across four centres, 1.5T to 7T, and different EEG layouts, which means cross-centre robustness can be demonstrated but should not be assumed. Amiri et al. (2023) further showed in acute DoC that only 63 of 87 patients had both EEG and fMRI, while direct same-sample model comparison used 48 complete-feature patients, so full bundles are not automatically acquisition-complete. Manasova et al. (2026) validated multimodal models across centres with different acquisition parameters, reported that performance improved with more modalities, and found higher inter-modality disagreement in minimally conscious or improving patients. Therefore, on this site, per-row repeatability, cross-centre transfer, and partial-bundle availability are part of composition rather than afterthoughts.

A third correction is now required inside named proxy families as well. Naganawa et al. (2021) constrain the SV2A quantification route, Johansen et al. (2024) constrain a healthy-human atlas / baseline route, Snellman et al. (2024) constrain a disease / risk-contrast route, Shatalina et al. (2024) constrain a task / cognition association route, Smart et al. (2021) constrain an activation-null timescale boundary, and Holmes et al. (2022) constrain an intervention-response null at 24 h. Therefore, this site no longer accepts SV2A PET as one reusable bundle row. The composition card now has to type the family-internal comparison family before any shared bundle role is inferred.

A fourth correction concerns evidence role rather than quantity type alone. Johansen et al. (2024) is a healthy atlas / cohort-prior route, Snellman et al. (2024) is a cross-sectional risk-contrast route, Finnema et al. (2018) is a same-subject baseline / repeatability route, Smart et al. (2021) is a within-subject activation-change boundary, and Holmes et al. (2022) is a 24 h intervention-response boundary. Those do not define one interchangeable bundle role. Therefore, this site now asks the card to state whether each row is being used as a normative atlas / cohort prior, a cross-sectional contrast, a same-subject baseline readout, a within-subject change witness, or a perturbation-response witness, together with the route and time window that actually justify that role.

The same rule now also applies to human myelin MRI. Arshad et al. (2017) constrain an MWF versus calibrated T1w/T2w comparison route, Hagiwara et al. (2018) constrain a relaxometry / MT_sat comparison route, Baadsvik et al. (2024) constrain a bilayer-sensitive mapping route, Galbusera et al. (2025) constrain a qT1 remyelination-sensitive pathology route, and Colaes et al. (2026) constrain a T1w/FLAIR tissue-health-sensitive ratio route whose safe reading is broader than myelin-specific contrast. Therefore, this site no longer accepts myelin MRI as one reusable bundle row either. The composition card has to type whether the row is an MWF / calibrated T1w:T2w comparison, a relaxometry / MT_sat comparison, a bilayer-sensitive mapping route, a qT1 remyelination-sensitive pathology route, or a T1w/FLAIR tissue-health-sensitive ratio before any shared bundle role is inferred.

The same rule now also applies to human clearance-transport evidence. Fultz et al. (2019) constrain a macroscopic CSF-oscillation route, Kim, Huang, & Liu (2025) constrain a parenchyma-CSF water-exchange route, Lim et al. (2025) constrain a respiration-conditioned net-flow route, Yoo et al. (2025) constrain an exercise-conditioned contrast-influx / parasagittal meningeal-lymphatic route, Eide et al. (2023) constrain an intrathecal-tracer / CSF-to-blood clearance-capacity route, Hirschler et al. (2025) constrain a CSF-mobility route, and Dagum et al. (2026) constrain a model-based biomarker-efflux route. Therefore, this site no longer accepts clearance / immune support as one reusable bundle row either. The composition card has to type whether the row is macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net-flow, exercise-conditioned contrast influx, intrathecal tracer / CSF-to-blood clearance, CSF mobility, or model-based biomarker efflux before any shared bundle role is inferred.

The same rule now also applies to human target-defined neuroimmune PET. Biechele et al. (2023) show that TSPO is not a species-invariant human activation-state meter, Wijesinghe et al. (2025) constrain a TSPO disease-context / validation-bounded route in PSP, Horti et al. (2022) and Ogata et al. (2025) constrain CSF1R route-setting PET under explicit arterial-input modeling, and Yan et al. (2025) constrain an enzyme-defined COX-2 route with celecoxib blockade and named quantification choices. Therefore, this site no longer accepts immune PET as one reusable bundle row either. The composition card has to type whether the row is TSPO disease-context / validation-bounded, CSF1R route-setting, or COX-2 enzyme-defined before any shared bundle role is inferred.

A fourth correction concerns availability geometry rather than bundle size alone. Amiri et al. (2023) already showed that direct bundle comparison could shrink from 87 enrolled patients to a 48-patient complete-feature subset. Manasova et al. (2026) then made the geometry explicit in the main French dataset: EEG-LG was available in 290 patients, dMRI in 151, aMRI in 101, FDG-PET in 53, and fMRI-RS in only 44, while disagreement rose in MCS and improved groups. Therefore, on this site, a bundle must disclose not only the complete-case count but also the row-overlap geometry and whether missingness itself tracks site, severity, tolerance, contraindication, or protocol. Otherwise, an apparent multimodal gain can still be driven by a changing patient subset rather than by tighter same-subject state constraint.

2026-03-26 tightening: promotion now runs through three gates

A fifth correction concerns discordance topology rather than mean gain alone. Rohaut et al. (2024) showed that multimodal assessment can reduce uncertainty and improve prognostic accuracy overall, while also warning that multimodal approaches increase the odds of discrepancies across markers that can produce choice paralysis or biased decisions. Manasova et al. (2026) then showed that pairwise disagreements across modalities were higher in MCS patients and in those who later improved. Therefore, on this site, a bundle must disclose not only whether the average score improved, but also where disagreement concentrates and how discordant cases are handled.

For fast reading, this card now compresses to three promotion gates. First, the bundle must pass a robustness gate: row-level repeatability at the actual operating point, cross-centre / cross-protocol transfer where claimed, and disclosure of the real complete-case slice plus row-overlap geometry and missingness mechanism. Second, it must pass a common-driver / quantity-bridge gate: same-session agreement is not enough unless the paper shows that the rows refer to compatible effective time windows / state axes and physiological or perturbation regimes, survive shared-driver audit, and can be read on an explicitly named biological axis. Third, it must pass an increment gate: the paper must show what the bundle adds beyond the strongest single row under a matched reading rule, and also where the remaining disagreements concentrate plus whether discordant cases were abstained, adjudicated, or silently absorbed into one final score. Without all three, this site keeps the result at the strongest single-row or proxy-rich ceiling.

Atlas is not yet a change tracker

On this site, role-typing is not optional shorthand. Johansen et al. (2024) is a healthy atlas, Snellman et al. (2024) is a cross-sectional risk contrast, Finnema et al. (2018) is a route-local repeatability anchor, Smart et al. (2021) is a within-subject activation-null boundary, and Holmes et al. (2022) is a 24 h intervention-response boundary. Those papers matter, but they do not license one interchangeable synaptic-density bundle role.

Same-session correlation is not yet same quantity

Even if several rows were acquired together, the card still asks whether they measure the same thing on the same temporal object and in the same regime. A bundle that mixes SV2A density, metabolic similarity, kinetic glucose-rate imaging, CSF mobility, and clearance-efflux modeling does not yet define one common state axis by default, and Epp et al. (2025) showed that even hemodynamic and oxygen-metabolism rows can move in opposite directions. Therefore, same-session alignment or positive correlation is not enough on this site unless the paper discloses the quantity bridge, the effective time-window relation, and the regime compatibility explicitly.

Minimum operating rule

If this card is missing, this site reads a living-human proxy bundle at the ceiling of the strongest individually supported row, or at most as proxy-rich but ceiling-limited human evidence. It is not promoted to same-subject, cross-stack, externally calibrated state identification.

Promotion rule beyond row diversity

On this site, row diversity alone is not enough. A bundle is promoted only when the paper discloses whether the apparent agreement survives a shared-driver audit, where the key rows still disagree and how those discordant cases were handled, whether the key rows remain repeatable and transferable outside one setup, and what the bundle adds beyond the strongest single row under a matched-condition reading. In short, the bundle must pass robustness, common-driver / quantity-bridge, and increment gates. Otherwise, the result remains proxy-rich but ceiling-limited rather than same-subject state identification.

2026-03-21 addendum: same-subject is not enough when the bridge itself is sequential

After adding the Fusion Card, the Human Proxy Composition Card, and the Destructive-Structure Route Card, one weak point still remained: a result could still be described as same-subject or same-brain even when the claim depended on bridging across live measurement, later fixation, ex vivo follow-up, or cross-day reacquisition as if those stages sampled one latent state. The primary literature does not support that shortcut. Lu et al. (2023) showed that preservation route and fixation time course alter extracellular-space retention and geometry. Bosch et al. (2022) showed that bridging in vivo two-photon physiology to synchrotron microtomography and serial block-face EM requires a multistage landmark-based correlative workflow. MICrONS Consortium et al. (2025) showed that same-brain function plus EM remains a sequential local pipeline, not simultaneous whole-state capture. Attardo et al. (2015) further showed that adult CA1 spine lifetimes are on the order of 1-2 weeks, so relaxed bridge windows cannot be treated as silent continuity. Therefore, this site now requires a State-Continuity Bridge Card whenever a claim depends on treating measurements from different acquisition regimes as one latent-state sample.

2026-03-28 addendum: the bridge must declare what is supposed to survive

The remaining weakness after adding bridge class and hidden-state-family disclosure was that a submission could still leave the carried object implicit. The primary literature does not support that shortcut. Bosch et al. (2022) and MICrONS Consortium et al. (2025) show local landmark and correspondence objects rather than one global state object. Gallego et al. (2020), Roth & Merriam (2023), Noda et al. (2025), Van De Ville et al. (2021), and Di et al. (2021) show that different population-level objects can remain stable even while raw units, amplitudes, or feature families change. Karpowicz et al. (2025), Wilson et al. (2025), and Wairagkar et al. (2025) further show that stable use across time can depend on alignment, recalibration, or a short fixed-decoder window. Therefore, this card now asks not only when the bridge happened, but also what object is claimed to have survived it and how failure would have been detected.

2026-03-25 addendum: bridge type and exposed state families must be named together

The remaining weakness was that the card could still be filled as if bridge risk were one generic scalar. The primary literature does not support that shortcut. Lu et al. (2023) and Idziak et al. (2023) show transformation-dominated live-to-fix risk, Bosch et al. (2022) and MICrONS Consortium et al. (2025) show landmark- and deformation-heavy local same-brain bridges, Musall et al. (2019), Benisty et al. (2024), and Egger et al. (2024) show waking-state drift within hours, and Hengen et al. (2016) plus Xu et al. (2024) show that sleep/wake crossing changes homeostatic and computational regime. Therefore, bridge submissions on this site now have to name both the bridge class and the first exposed hidden-state families, rather than writing only elapsed time.

Minimum operating rule

If this card is missing, this site reads a sequential cross-regime result at the ceiling of the strongest directly supported live or destructive stage plus, at most, an unvalidated bridge hypothesis. It is not promoted to same-state, same-time, or maintenance-consistent evidence.

Need the longer bridge critique?

For the background logic behind preservation change, cross-day regime mismatch, bridge-validation rungs, and card stacking, see Wiki: State-Continuity Bridge.

2026-03-20 addendum: destructive ultrastructure routes need their own card

The remaining weakness after the Observability Budget and Fusion Card was that words such as nanoscale, petascale, or same-brain could still be overread as if the destructive route had already preserved native state, solved scaling, and certified reconstruction quality in one move. The primary literature does not support that shortcut. Lu et al. (2023) showed that conventional aldehyde fixation collapses extracellular space, that fixation itself has a non-negligible time course, and that even high-pressure freezing preserves extracellular space only in samples thinner than about 200 μm. Shapson-Coe et al. (2024) then showed that extraordinary human ultrastructure can now be reconstructed, but still from a rapidly preserved 1.05 mm³ surgical fragment with 1.8 PB raw data and 326 days of imaging. MICrONS Consortium et al. (2025) showed that same-brain function plus EM is a sequential local pipeline, with in vivo two-photon imaging first and fixation / sectioning / ex vivo EM afterward, producing about 2 Pb of raw data over about 6 months. Dorkenwald et al. (2024) further showed that even the adult-fly whole-brain frontier still depended on proofreading, thresholding, and about 33 person-years of manual correction. Therefore, this site now asks destructive ultrastructure claims to attach a Destructive-Structure Route Card instead of letting resolution language silently stand in for native-state completeness.

2026-03-18 addendum: ESI validation is a ladder, not one checkbox

This site no longer accepts the phrase external validation without naming the validation class. Simulation / phantom audits solver behavior under known generative conditions, intracranial stimulation audits localization error to a known site and time (Mikulan et al., 2020; Unnwongse et al., 2023), simultaneous invasive recording audits concordance under the same event (Hao et al., 2025), and postsurgical outcome audits clinical relevance rather than direct source ground truth (Birot et al., 2014). Passing one rung does not auto-pass the others.

2026-03-19 addendum: ESI claims need a solver-disagreement audit

The remaining weakness was to let one inverse solution stand in for the whole solution set. Mahjoory et al. (2017) showed that inverse-method and software-package choice induces considerable variability and explicitly recommended verifying EEG source findings with more than one source-imaging procedure. Mikulan et al. (2020) then showed on intracranial-stimulation ground truth that only a small fraction of tested solutions reached the session-wise optimum, even though the benchmark was tightly controlled. Vorwerk et al. (2024) further showed that skull and skin conductivity uncertainty shifts reconstructed depth and location, especially for quasi-tangential sources. Therefore, when an anatomical source claim depends on one inverse family or one hand-picked parameter set, this site now asks for a solver-disagreement audit that reports compared solver families / packages, parameter window, and headline-location spread. If that is missing, the result stays at the method-sensitive source-hypothesis level rather than stable anatomical evidence.

2026-03-31 addendum: solver disagreement is not one generic number

The next weak point was to treat inverse-family disagreement as if every family were estimating one interchangeable source object. Current primary literature does not support that shortcut. Luria et al. (2024) expose posterior support and alternative configurations for focal-source hypotheses, Tong et al. (2025) expose debiased estimation and inference for sparse spatial-temporal sources, and Feng et al. (2025) expose empirical-Bayesian uncertainty for extended-source reconstruction. Therefore, the Inverse-Solver Agreement Log on this site now requires the source regime / target object and the uncertainty object to be named alongside the solver family. A disagreement between posterior-support maps, debiased intervals, and extent-overlap maps is not reduced to one winner by default.

2026-03-22 addendum: ESI claims need a field-formation audit

The remaining weakness was upstream of the inverse solver itself. A paper can report a cleaner map, a lower benchmark error, or better regularization and still leave unasked whether the targeted source class was expected to generate a usable scalp field in the first place. Ahlfors et al. (2010) showed with realistic tissue boundaries that source orientation matters far more for MEG than for EEG, with median lowest-to-highest sensitivity ratios of 0.06 for MEG and 0.63 for EEG. Ahlfors et al. (2010) and Goldenholz et al. (2009) showed that extended sources, cortical folding, and cancellation can materially reshape or suppress surface signals, while Piastra et al. (2021) showed that omitting the CSF compartment overestimates EEG SNR and changes cortical / subcortical sensitivity comparisons. Therefore, when a submission claims deep-source recovery or anatomical improvement from EEG/MEG, this site now asks for a field-formation audit before the inverse result is read strongly.

2026-03-18 addendum: hemodynamic modalities need a vascular-state / CVR audit

Hemodynamic stacks do not carry only neural uncertainty; they also carry a vascular transfer state, and newer work shows that transfer-audited amplitude and oxygen-metabolism routes are still different inferential objects. Murphy et al. (2011) showed that inter-subject CBF / CBV differences contribute to BOLD reactivity and that breath-hold-derived vascular-reactivity covariates improve group analyses. Williams et al. (2023) showed that task BOLD magnitude is strongly predicted by hypercapnia-based CVR across multiple cortical regions, Wu et al. (2023) showed that baseline CBF partly explains age-related components of multiple-demand-network BOLD responses, Yucel et al. (2015) and An et al. (2025) showed that short-channel regression remains necessary to reduce superficial autonomic confounds in fNIRS, Epp et al. (2025) showed that about 40% of voxels with significant task BOLD changes can show opposite oxygen-metabolism changes, and Jaroszynski et al. (2025) showed that constrained qBOLD reaches OEF / CMRO2 only through an explicit model stack with separate pCASL-based CBF input. Therefore, when a submission uses BOLD or fNIRS, this site now asks authors to disclose whether the claim lives at uncalibrated amplitude, transfer-side calibrated amplitude, or a model-conditioned oxygen-metabolism / quantity-bridge route; otherwise the result stays at a hemodynamic-limited difference rather than a neural or metabolic claim.

2026-03-22 addendum: tractography connectomes need a tractography route card

For diffusion-MRI-derived connectomes, the modality label alone is not enough. Reveley et al. (2015) showed that superficial white matter can block long-range tracking from roughly half of the cortical surface, Schilling et al. (2018) showed that tractography endpoints are biased toward gyral crowns across deterministic and probabilistic algorithms and even very high-resolution data, Sarwar et al. (2023) showed that filtering gains are modest in complex human-like architectures, He et al. (2024) showed that filtering can significantly shift laterality indices for more than 10% of connections, Gajwani et al. (2023) showed across 40 pipelines and 44 group-representative reconstructions that hub location is highly variable and hub connectivity correlates with regional surface area in 69% of assessed pipelines, McMaster et al. (2025) showed that voxel size changes the resulting connectome and recommended resampling to 1 mm isotropic for robust comparisons, Bramati et al. (2026) showed on the same 3 T scanner with uniform processing that common diffusion-sampling schemes can still shift voxel metrics and tractography outputs, Manzano-Patrón et al. (2025) made fibre-orientation uncertainty explicit rather than silent, and Zhu et al. (2025) improved tractography by fusing MRI with microscopy. Therefore, this site now asks any tractography-derived connectome claim to attach a tractography route card naming what was directly measured, how acquisition and harmonization were handled, how cortical endpoints were assigned, which priors, filtering, parcellation, and weighting choices shaped the graph, how uncertainty and protocol sensitivity were quantified, what external calibration exists, and where abstention begins. The longer public reading rule is summarized in Wiki: tractography route card.

2026-03-19 addendum: connectome-constrained predictors need a conditional-model route card

For connectome-constrained neural predictors, the label alone is also not enough. Lappalainen et al. (2024) showed that connectome structure plus task optimization can yield rich fly visual-system activity predictions, but the model still depended on partial motion-pathway reconstruction, simplified neuron/synapse dynamics, and an ensemble of multiple local optima. Shiu et al. (2024) showed that synapse-level connectivity plus neurotransmitter identity can predict specific fly sensorimotor circuits, while explicitly reading the result as a coarse description of named behaviours. Pospisil et al. (2024) used the connectome as a prior for perturbation-based effect estimation, but explicitly recovered a linear approximation to nonlinear dynamics. Finally, Beiran & Litwin-Kumar (2025) showed that even with the same synaptic weights, different biophysical parameters can still generate divergent recurrent dynamics. Therefore, this site now asks any connectome-constrained model claim to attach a conditional-model route card naming what structure was actually used, what remained fitted, which regime was tested, which mechanisms were omitted, how validation was done, and where abstention begins. The longer public reading rule is summarized in Wiki: conditional-model route card.

Minimum operating rules

If this card is missing, this site treats the result as L0/L1 reproducible analysis or limited decode and does not promote it to L2 or above. For example, the default ceiling is macro-state tracking for EEG / HD-EEG + MRI only, an implant-region local population window for high-density extracellular probe only, structural scaffold for volume EM only, local ex vivo scaffold for human ultrastructure without a destructive-structure route card, molecular / spatial prior for whole-brain atlas only, and still only local conditional prediction even for same-brain calcium + EM. For inverse or model-based claims, a submission without an Identifiability Card is not accepted here as unique internal-state recovery and remains at the predictive / localization / model-conditioned ceiling even if its observability class improved. For living-human multi-stack proxy claims, a submission without a Human Proxy Composition Card is not accepted here as same-subject cross-stack state identification and remains at the strongest single-route or proxy-rich but ceiling-limited level. For sequential cross-regime claims, a submission without a State-Continuity Bridge Card is not accepted here as same-state evidence and remains at the strongest directly supported stage plus an unvalidated bridge. For hemodynamic modalities, a group or cross-day BOLD / fNIRS difference without vascular-state / CVR or short-separation / superficial-bias audit is not accepted here as a neural difference. For diffusion-MRI tractography, a connectome claim without a tractography route card is not accepted here as an edge-complete graph and remains at the macro pathway prior / targeted bundle hypothesis ceiling. For connectome-constrained predictors, a model claim without a conditional-model route card remains at the conditional hypothesis engine / task-bounded predictor ceiling and is not accepted here as unique internal-state recovery. If a chronic-probe result lacks a unit-identity audit, single-unit longitudinal claims are not accepted. Detailed stack-specific ceilings and the state variable × timescale matrix are summarized in Wiki: observability and claim ceiling by measurement stack.

How to use this budget on this site

The latent-state error budget is not a second abstract. It is a submission-side stop-rule table. If a paper adds one extra evidence layer, such as transcriptomics, SV2A PET, receptor / transporter atlas priors, occupancy PET, release-sensitive displacement PET, myelin imaging, a glial perturbation, or a clearance proxy, it should say which error term that layer reduces under the same held-out condition and which latent families remain untouched. This is the site-wide rule behind the augmentation / ablation logic in State variable integrity gate.

2026-03-22 addendum: neuromodulatory evidence also needs a route card

On this site, mixed arousal proxy, local transmitter sensor, receptor / transporter atlas, occupancy PET, and release-sensitive displacement PET are not treated as equivalent. Reimer et al. (2016) showed that pupil fluctuations track both adrenergic and cholinergic activity, Neyhart et al. (2024) showed that local cortical ACh depends on axon activity and local clearance, Hansen et al. (2022) and Goulas et al. (2021) showed that receptor maps are structured regional priors, Wong et al. (2013) showed selected D₂-receptor target engagement by an administered drug, and Koepp et al. (1998), Lippert et al. (2019), and Erritzoe et al. (2020) showed challenge- and window-limited dopamine or serotonin release proxies. Therefore, submissions that depend on neuromodulation now have to state not only which rung was used, but also the claim family, transmitter axis / receptor family, direct observable, challenge or administered-drug route, time window / model burden, and abstention boundary before any claim ceiling is raised. The longer public rule is in Wiki: neuromodulatory route card.

2026-03-18 addendum: PET-based evidence is unreadable without the measurement model

For PET-based routes, naming the modality is still too coarse. Naganawa et al. (2021) showed for human SV2A PET that the quantification route depends on metabolite-corrected arterial input, reference-region choice, compartment model, and scan window. Smart et al. (2021) showed that [¹¹C]UCB-J binding measures stay unchanged during brief visual activation despite blood-flow-driven influx changes, so synaptic-density PET should not be read as momentary synaptic efficacy. Hansen et al. (2022) showed that receptor atlas is built from group-average PET maps from more than 1,200 healthy individuals, Wong et al. (2013) quantified occupancy for an administered antipsychotic, and Koepp et al. (1998), Lippert et al. (2019), and Erritzoe et al. (2020) used challenge-linked binding changes as displacement-based release proxies. This site therefore requires tracer, occupancy-versus-displacement design, challenge or administered drug plus dose when applicable, quantification route, model/window, and partial-volume handling where relevant before any PET-based claim ceiling is interpreted.

2026-03-21 addendum: synaptic-density PET is not one audit item

SV2A PET now gets its own route-card logic on this site. Naganawa et al. (2021) fixes tracer and quantification burden, Johansen et al. (2024) is a healthy-human atlas, Shatalina et al. (2024) is a task / cognition association study, Smart et al. (2021) shows that brief activation does not produce a momentary SV2A state readout, and Holmes et al. (2022) shows that rapid ketamine response need not imply measurable SV2A change at 24 h. Therefore, submissions that cite synaptic-density PET must state whether the paper is an atlas, disease / risk contrast, task / cognition association, intervention / target-engagement design, or something narrower, and must disclose tracer, quantification route, anatomy / partial-volume handling, and abstention boundary before the claim ceiling is raised. The longer route is Wiki: SV2A / synaptic-density PET route card.

2026-04-01 addendum: synaptic-density is not presynaptic release-machinery evidence

Another shortcut needed to be blocked here. Molnár et al. (2016), Sakamoto et al. (2018), Dürst et al. (2022), and Emperador-Melero et al. (2024) together show that release-site number, docked-vesicle organization, active-zone nanostructure, and current release competence are not exhausted by synapse count or regional SV2A density. Therefore, this budget now keeps regional synaptic-density proxy and presynaptic release machinery as separate audit objects. The longer public argument is in Wiki: Why wiring diagrams alone are not enough.

2026-03-18 addendum: sleep history is not sleep architecture

Ngo et al. (2013) showed in humans that in-phase slow-oscillation stimulation enhances spindle coupling and memory whereas out-of-phase stimulation does not, Maingret et al. (2016) and Latchoumane et al. (2017) showed in rodents that fine-tuned slow-oscillation / spindle / ripple coordination causally supports consolidation, Schreiner et al. (2021) showed that SO-spindle coupling precision predicts endogenous reactivation strength in humans, Geva-Sagiv et al. (2023) showed that real-time enhancement of hippocampal-prefrontal synchrony improves overnight human memory, Schreiner et al. (2024) linked spindle-locked ripples to human memory reactivation, and Deng et al. (2025) showed that NREM consolidation itself has a specific intracellular time window. This site therefore records sleep duration/history and sleep architecture / replay-coupling separately.

2026-03-20 addendum: sleep replay evidence also needs a route card

The remaining weakness was that sleep replay evidence could still compress a phase-locked healthy-human stimulation study, an endogenous scalp-decoding study, an intracranial closed-loop synchrony intervention, and an item-selective TMR result into one bucket. The primary literature does not support that shortcut. Schreiner et al. (2021) explicitly notes aggregated SO-spindle events, modest decoding levels, and scalp blindness to ripple ground truth, Duan et al. (2025) shows that strengthening and decaying items can coexist within the same human TMR session, and Shin et al. (2025) shows that behavioral benefit can concentrate in the challenging-memory regime rather than appearing uniformly across items. Therefore, this site now asks cross-day or sleep-replay claims to attach a sleep replay route card naming preparation, event definition, timing / control policy, memory target / selection regime, and abstention boundary. The full public rule is summarized in Wiki: sleep replay route card.

2026-03-21 addendum: myelin evidence also needs a route card

The remaining weakness was that myelin evidence could still compress learning-dependent oligodendrogenesis, node / internode / periaxonal timing-state, developmental plasticity brake, remyelination-to-function recovery, and human quantity-defined myelin proxy rows into one bucket. The primary literature does not support that shortcut. On the human side, Arshad et al. (2017) showed that calibrated T₁w/T₂w can be reliable while still having low criterion validity against MWF, Hagiwara et al. (2018) showed stronger agreement between SyMRI and MT_sat than with T₁w/T₂w, Baadsvik et al. (2024) demonstrated bilayer-sensitive mapping only in two healthy volunteers, Chen et al. (2025) showed that orientation dependence is itself a route variable for conventional MT, Galbusera et al. (2025) showed that qT1 but not MWF or MTR separated demyelinated from remyelinated cortical lesions, and Colaes et al. (2026) showed that T₁w/FLAIR remained only weakly associated with MWF and is safer to read as a broader tissue-health marker. On this site, maintenance-state submissions must therefore state which inferential object they are actually using, what structural unit was directly observed, whether functional recovery occurred with partial versus complete remyelination, whether a route should be read as myelin-sensitive or only tissue-health-sensitive, and whether the human evidence is only a proxy row rather than direct timing-state ground truth. The full public rule is summarized in Wiki: myelin / oligodendrocyte route card.

2026-03-20 addendum: do not collapse phospho-signaling into transcript or protein abundance

Giese et al. (1998), Lee et al. (2003), Rodrigues et al. (2004), Tomita et al. (2005), and Vierra et al. (2023) show that phosphosite-specific and nanodomain-specific signaling states can determine plasticity expression and maintenance-relevant routing even when transcript or bulk protein abundance looks similar. Current human evidence such as the ex vivo phosphoproteome atlas from Biswas et al. (2023) is informative for region-specific ceiling setting, but it is still not a comparable in vivo whole-brain human route. Therefore, when a claim depends on phosphosite occupancy, kinase/phosphatase balance, or local second-messenger state, this site asks authors to disclose whether phospho-signaling was directly measured, causally perturbed, externally calibrated, or silently replaced by transcript/protein abundance alone.

2026-03-20 addendum: phospho-signaling evidence also needs a route card

The remaining weakness was that phosphosite-specific plasticity gates, compartmentalized second-messenger routing, region-structured phosphoproteome atlases, and single-site phospho-mutant memory interventions could still be compressed into one bucket called phospho evidence. The primary literature does not support that shortcut. Havekes et al. (2016) is about compartment-targeted PDE4A5 signaling in hippocampal memory, Altas et al. (2024) is region-specific phosphorylation with synapse-type relocalization in mouse and human samples, Rodriguez et al. (2025) is a single-site phospho-mutant causal memory intervention, and Biswas et al. (2023) is a human ex vivo phosphoproteome atlas. On this site, phospho-signaling claims now have to disclose claim family, biological regime, time axis, assay and direct observable, spatial / compartment scope, causal leverage, and abstention boundary. The full operating rule is in Wiki: phospho-signaling route card.

Do not collapse energetic support into glial support

Rangaraju et al. (2014), Underwood et al. (2023), Rangaraju et al. (2019), Divakaruni et al. (2018), Bapat et al. (2024), Hu et al. (2025), and Vishwanath et al. (2026) support a narrower rule: local ATP supply, respiration, mitochondrial positioning, and energetic micro-organization can change repeated-burst reliability and dendritic plasticity even before one asks whether astrocyte or clearance-state matched. Conversely, human ³¹P-MRS metabolite / pH balance routes (Ren et al., 2015), ³¹P MT exchange-flux routes (Ren et al., 2017), ³¹P NAD-content mapping routes (Guo et al., 2024), localized functional ³¹P NAD-dynamics routes (Kaiser et al., 2026), deuterium metabolite-mapping / absolute-quantification routes (Karkouri et al., 2026), and deuterium kinetic-rate imaging (Li et al., 2025) remain macro energetic proxies rather than direct readouts of branch-local mitochondrial state. Ahmadian et al. (2025) further showed that human-brain deuterium signal depends materially on the administered glucose dose, and Bøgh et al. (2024) showed that repeatability depends on a named acquisition and time-point regime. This site therefore records energetic route and glial route separately.

2026-03-21 addendum: bioenergetic evidence also needs a route card

The remaining weakness was that bioenergetic evidence could still compress presynaptic ATP-linked respiration, dendritic positioning / fission support, synaptic ATP-synthase nano-organization, mitochondrial Ca²⁺-efflux tuning, and a single macro energetic row into one bucket. The primary literature does not support that shortcut. On this site, maintenance-state submissions must now state which inferential object they are actually using, which compartment was directly observed, whether the quantity is respiration, ATP-synthase arrangement, Ca²⁺-efflux-driven metabolic tuning, or only a macro ³¹P metabolite / pH balance, ³¹P MT exchange-flux, ³¹P NAD-content mapping, localized functional ³¹P NAD-dynamics, deuterium metabolite-mapping / absolute-quantification, or deuterium kinetic-rate proxy, what dose / time-point / repeatability regime conditioned any deuterium route, what function target was tested, and which local mitochondrial controller remained latent. The full public rule is summarized in Wiki: bioenergetic / mitochondrial route card.

2026-04-01 addendum: glial substrate-routing is not generic astrocyte background

Suzuki et al. (2011) is a lactate-shuttle route, Silva et al. (2022) is a glia-to-neuron ketone-body route under starvation, Pavlowsky et al. (2025) is an intensive-learning glia-to-neuron fatty-acid route, and Greda et al. (2025) is an apoE / sortilin-dependent lipid-delivery and neuronal fuel-choice route when glucose is limited. Therefore, this site does not let `glial support` stand in for astrocyte network state, neuronal mitochondrial arrangement, or a generic human energetic proxy. Submissions that depend on glial fuel support now have to state claim family, supplier / sink, fuel object / carrier, regime trigger, route, and abstention. The longer public rule is in Wiki: glial substrate-routing route card.

2026-03-20 addendum: astrocyte-state is not generic support background

Cahill et al. (2024) showed minute-scale cortical astrocyte-network encoding of local neurotransmitter input, Williamson et al. (2025) showed that learning-associated astrocyte ensembles can regulate hippocampal memory recall, Dewa et al. (2025) showed that astrocytic ensembles can stabilize emotional memory across multiple days, and Bukalo et al. (2026) showed that basolateral-amygdala astrocyte Ca²⁺ signalling supports fear-memory retrieval / extinction representations. Therefore, this site does not accept a submission that depends on recall, multiday stabilization, or fear-state support while writing only a generic glial-support sentence. Authors now have to state whether the relevant astrocyte-state was directly measured, perturbed, only coarsely proxied, or left latent, and they must keep the rodent-to-human observability gap explicit.

2026-03-21 addendum: astrocyte evidence also needs a route card

The remaining weakness was that astrocyte evidence could still compress minute-scale cortical network encoding, learning-associated recall ensembles, multiday stabilization ensembles, fear-state representations, and target-defined human astrocyte-related PET routes into one bucket. The primary literature does not support that shortcut. Villemagne et al. (2022) is a first-in-human MAO-B SMBT-1 target-validation route, Villemagne et al. (2022) is an AD-spectrum disease-context route in the same tracer family, Hiraoka et al. (2025) is a brain-quantification route, Mesfin et al. (2026) is a whole-body biodistribution route, Matsuoka et al. (2026) is a simplified arterial-free SL25.1188 AD quantification route, Tyacke et al. (2018) is a human I₂BS PET route, Livingston et al. (2022) shows BU99008 uptake can vary with region and impairment stage, Best et al. (2026) shows that SL25.1188 MAO-B binding can move with cohort severity and daily cigarette use, and Jaisa-Aad et al. (2024) shows that MAO-B varies across AD/ADRD classes. On this site, maintenance-state submissions must now state which inferential object they are actually using, what the direct astrocyte observable is, whether the result depends on transporter blockade, ensemble reactivation, gene deletion, astrocyte Ca²⁺ intervention, or only tracer binding, what the functional target actually is, what human target / tracer family / route role / quantity type, brain-quantification regime or whole-body tracer-burden regime, disease / cohort regime, and material covariates remain, and which human astrocyte controller is still latent. The full public rule is summarized in Wiki: astrocyte route card.

2026-03-20 addendum: clearance / immune support is not passive cleanup

Louveau et al. (2015) and Ahn et al. (2019) established meningeal-lymphatic drainage routes, Kim et al. (2025) showed that a meningeal-lymphatics-microglia axis can regulate synaptic physiology, Eide & Ringstad (2021) showed that sleep deprivation impairs molecular clearance in humans, Hirschler et al. (2025) measured region-specific CSF-mobility drivers, Lim et al. (2025) reported respiration-conditioned CSF net flow in awake humans while also warning that plane-specific 2D PC-MRI net flow does not by itself represent whole-brain bulk circulation, and Dagum et al. (2026) linked sleep-active physiology to overnight Aβ / tau clearance to plasma in healthy older adults. Therefore, this site does not accept a submission that depends on multiday recovery, protein clearance, or maintenance support while naming only a generic cleanup story. Authors now have to state whether the relevant evidence was meningeal-lymphatic / CSF-interstitial / microglia-related, whether the human route was macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net-flow MRI, CSF mobility, intrathecal tracer retention / CSF-to-blood clearance, or model-based biomarker efflux, whether it remained only a transport-side human proxy, and which local immune or synaptic-maintenance controller remained latent.

2026-03-21 addendum: clearance / immune evidence also needs a route card

The remaining weakness was that clearance evidence could still compress drainage anatomy, microglia-mediated synaptic mechanism, macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net-flow MRI, human CSF-mobility MRI, intrathecal tracer retention / CSF-to-blood clearance, and model-based human biomarker efflux into one bucket. The primary literature does not support that shortcut. On this site, maintenance-state submissions must now state which inferential object they are actually using, whether it is a transport-side human route or an immune-effector route, what the direct observable is, whether the result depends on a physiology / sleep / lymphatic perturbation route, which human route object / quantity type is in play, what human acquisition or model burden remains, and which local immune or synaptic controller is still latent. The full public rule is summarized in Wiki: clearance / immune route card.

2026-04-03 addendum: target-defined human neuroimmune PET is not the same as clearance transport

Another remaining shortcut had to be blocked. Current human-side immune support evidence is not transport-only, but that does not mean that all human immune evidence now forms one reusable row. Biechele et al. (2023) showed that TSPO is not a species-invariant human activation-state meter, Wijesinghe et al. (2025) constrained a TSPO disease-context / validation-bounded route in PSP, Horti et al. (2022) and Ogata et al. (2025) constrained first-in-human CSF1R PET routes, and Yan et al. (2025) constrained an enzyme-defined COX-2 route with celecoxib blockade. Therefore, this site now reads the human lane as at least two different bounded lanes, transport-side clearance routes and target-defined neuroimmune PET routes, and even inside the PET lane it still requires typing as TSPO disease-context / validation-bounded, CSF1R route-setting, or COX-2 enzyme-defined before any claim ceiling is raised.

Do not collapse cargo delivery into proteostasis or ATP

Park et al. (2006) showed that recycling-endosome exocytosis is required for LTP-associated spine growth, Maas et al. (2009) showed that synaptic activation modifies microtubules that support postsynaptic cargo transport, Yin et al. (2011) and Zhao et al. (2020) showed that kinesin-dependent cargo routes change receptor levels, plasticity, and memory, Swarnkar et al. (2021) linked KIF5C-mediated transport to structural plasticity and long-term memory, and Aiken & Holzbaur (2024) showed that axonal microtubule patterning controls presynaptic cargo delivery. This site therefore records cargo-transport state separately from both proteostasis and bioenergetics. Current human in vivo routes do not directly reveal branch- or bouton-specific cargo pausing, motor engagement, or microtubule traffic state, so this route stays latent unless externally calibrated.

2026-03-21 addendum: cargo evidence also needs a route card

The remaining weakness was that cargo evidence could still compress postsynaptic AMPAR / recycling-endosome delivery, learning-phase microtubule-state gating, activity-dependent vesicle confinement, dendritic / synaptic RNA-granule organization, axonal RNA localization, and presynaptic cargo retention into one bucket. The primary literature does not support that shortcut. Correia et al. (2008) is about myosin-Va-dependent transport of AMPARs into spines during LTP, Uchida et al. (2014) is about learning-phase microtubule stability controlling KIF5-mediated GluA2 localization and memory, Wong et al. (2024) is about local confinement of endogenous GluA1 vesicles near stimulated dendritic regions, Nakayama et al. (2017), Liau et al. (2023), and Espadas et al. (2024) are about dendritic / synaptic RNA cargo organization, de Queiroz et al. (2025) is about axonal RNA localization required for long-term memory in a mature in vivo circuit, and Aiken & Holzbaur (2024) is about presynaptic cargo delivery and pausing in a human-neuron preparation. On this site, maintenance-state submissions must now state which inferential object they are actually using, what cargo was followed, whether the result is about long-range delivery, local pausing / confinement, spine entry, dendritic / synaptic RNA organization, axonal RNA localization, or presynaptic retention, what preparation and time window were used, and whether the strongest human evidence is still limited to narrow preparations rather than living-human in vivo measurement. The full public rule is summarized in Wiki: cargo-transport / cytoskeletal trafficking route card.

2026-03-22 addendum: thermal evidence also needs a route card

The remaining weakness was that thermal evidence could still compress local operating-point physiology, field-potential confound, sequence / rhythm perturbation, device-heating artifact, human passive / task-linked macro thermometry, and human perturbation-conditioned thermal routes into one bucket. The primary literature does not support that shortcut. Hardingham & Larkman (1998) is about temperature-dependent synaptic reliability, Moser et al. (1993) is about field-potential amplitude being masked by temperature variation, Long & Fee (2008) is about sequence timing under local cooling, Owen et al. (2019) is about heating introduced by the perturbation device itself, Rzechorzek et al. (2022), Rogala et al. (2024), plus Tan et al. (2025) are human passive / task-linked macro thermometry routes, Tan et al. (2024) is a systemic heat-perturbation human route, and Inoue et al. (2025) is an intraoperative focal-cooling / neurovascular human route. On this site, maintenance-state submissions must now state which inferential object they are actually using, what the direct thermal observable is, whether the result depends on local cooling / warming, a recording confound audit, a device-heating burden, systemic heat exposure, or intraoperative focal cooling, what time window and function target were tested, what human route class remains, and which local thermal controller is still latent. The full public rule is summarized in Wiki: thermal route card.

2026-03-18 addendum: human maintenance evidence must be class-labeled

On this site, Wiki: Homeostatic plasticity and maintenance-state now distinguishes fixed-tissue structural scaffold, regional synaptic-density proxy, regional receptor / transporter atlas prior, ligand-limited occupancy proxy, challenge-limited displacement / release proxy, macro 1H-MRSI biochemical similarity scaffold, high-resolution 1H-MRSI metabolite-distribution proxy, macro 31P metabolite / pH balance proxy, macro 31P MT exchange-flux proxy, macro 31P NAD-content map proxy, localized functional 31P NAD-dynamics proxy, macro deuterium metabolite-mapping / absolute-quantification proxy, macro deuterium kinetic-rate proxy, quantity-defined macro ionic proxy family, macro thermal / perturbation-conditioned thermal proxy family, quantity-defined macro-myelin proxy family, macro BBB water-exchange / tracer-specific transport proxy family, target-defined astrocyte-related proxy, perturbation-conditioned plasticity proxy, and macro clearance-transport proxy family. Human-side evidence is therefore not submitted as one generic “maintenance signal.” Each route must be class-labeled before the claim ceiling is interpreted, and it must also disclose its calibrator role, meaning which hidden-state family it safely constrains rather than merely co-occurs with.

2026-03-20 addendum: causal relevance and human observability must be bridged, not fused

This site now keeps two axes separate when a maintenance-state claim cites both animal causal work and human proxy work. Hadzibegovic et al. (2025) and Terceros et al. (2026) strengthen controller-state and transcriptional-stabilization causality in rodents, while Zrenner et al. (2018), Hirschler et al. (2025), Lim et al. (2025), and Dagum et al. (2026) strengthen perturbation-conditioned excitability or macro clearance-transport observability in humans. These are not the same inferential step. Therefore, when a submission combines them, it must state which state family the causal paper makes relevant, which human evidence class / proxy class was actually observed, which bridge assumptions remain unvalidated, and which local controller remains latent. Without that bridge disclosure, this site does not promote the result to measured human maintenance-state control. The longer composition rule is now anchored in Wiki: Human Proxy Composition.

2026-03-19 addendum: MRSI metabolic connectome is a similarity graph, not a flux map

Lucchetti et al. (2025) defined the human metabolic connectome as pairwise correlations among five metabolites across gray-matter parcels and reported only weak overall alignment with tractography-based structural connectivity. Bhogal et al. (2020), Wright et al. (2022), and Baboli et al. (2024) show why this route still needs a measurement-model audit: MRSI interpretation depends on SNR, partial-volume / lipid handling, and voxel-specific correction or tissue-water modeling. Therefore, on this site, an MRSI-derived metabolic connectome is read as a macro-biochemical similarity scaffold unless the submission names the metabolite set, parceling unit, correction model, spectral QC thresholds, and whether the object is static similarity, high-resolution metabolite-distribution mapping, or dynamic kinetic rate imaging. If the claim is about glucose-flux or rate mapping, the route must be audited separately against rate-imaging methods such as Li et al. (2025).

2026-03-27 addendum: spectroscopy-derived human routes are not one proxy row

The remaining compression problem was to let ``human spectroscopy'' carry the claim. The primary literature does not support that shortcut. Lucchetti et al. (2025) constrained a five-metabolite parcel-similarity scaffold, Guo et al. (2025) constrained high-resolution 1H-MRSI metabolite-distribution maps under explicit reconstruction and artifact-control burden, Ren et al. (2015) constrained resting ATP synthesis, phosphorus metabolites, and pH balance with ³¹P-MRS in 12 healthy participants, Ren et al. (2017) constrained PCr→γ-ATP and Pi→γ-ATP exchange flux with a band-inversion / MT 5-pool model, Guo et al. (2024) constrained whole-brain intracellular NAD content at 7 T, Kaiser et al. (2026) constrained task-evoked NAD⁺ dynamics in a functionally localized human occipital voxel, Karkouri et al. (2026) constrained absolute deuterated metabolite maps with a dedicated absolute-quantification pipeline, and Li et al. (2025) constrained glucose-transport and kinetic-rate maps under a blood-input kinetic model. These are not the same inferential object. Bhogal et al. (2020), Wright et al. (2022), Baboli et al. (2024), and Guo et al. (2025) further show that even metabolite-map formation itself depends on lipid suppression, tissue-fraction correction, water / relaxation modeling, ghosting, aliasing, and low-SNR handling. Therefore, when a submission cites human spectroscopy here, it must state whether the route is 1H-MRSI similarity, high-resolution 1H-MRSI metabolite-distribution mapping, 31P metabolite / pH balance, 31P MT exchange-flux, 31P NAD-content mapping, 31P functional NAD-dynamics, deuterium metabolite-mapping / absolute quantification, or deuterium kinetic-rate imaging, together with cohort size, hardware burden, and model burden. Without that split, this site does not promote the route beyond a coarse human proxy note.

Minimum operating rules

If this budget is missing, this site allows at most same-session fit, cross-day performance with unresolved maintenance route, or support-proxy-aligned evidence. It does not promote the result to maintenance-consistent, reconsolidation-consistent, or remote-memory-relevant. In particular, if phospho-signaling / second-messenger state, sleep / wake history, sleep architecture / replay-coupling state, timing support, thermal-state, ionic / chloride state, bioenergetic support, cargo-transport / cytoskeletal trafficking support, glial substrate-routing, astrocyte-state, and clearance / immune proxy class are all absent, this site stops the reader from rephrasing a temporal hold as long-horizon maintenance evidence.

2026-03-18 addendum: subject / session fingerprint is an independent shortcut family

Chaibub Neto et al. (2019) showed that diagnostic learning can absorb subject characteristics when repeated measures are not participant-disjoint, while Wang et al. (2020) and Di et al. (2021) showed that resting-state EEG alone can support accurate, time-robust person identification. Gibson et al. (2022) further summarized strong subject-driven components in EEG variability. For that reason, this site treats subject / session fingerprint as a shortcut family on the same level as movement or EMG, and fixes independence unit plus metadata-only baselines as separate required fields.

2026-04-04 correction: acquisition-distribution shortcut needs a recording-frame contract

The older rule correctly treated acquisition distribution as independent from subject fingerprint, but it still left one scientific shortcut open: it let readers imagine harmonization as one generic cleanup step. The official EEG-BIDS specification already separates electrodes, channels, coordinate system, and reference scheme. Hu et al. (2018) showed that reference montage and electrode setup alter the measured scalp potential itself, Melnik et al. (2017) showed that EEG differences arise from system as well as subject and session, Xu et al. (2020) showed that cross-dataset EEG decoding is degraded by environmental variability such as amplifier, cap, sampling rate, and filtering, and Dong et al. (2024) validated an explicit REST-based transformation route for channel-location harmonization with high correlation rather than identity by default. For that reason, this site treats site / device / reference system / electrode layout / coordinate route / protocol distribution as a shortcut family on the same level as subject fingerprint, and it also requires the recording-frame contract: original channel map, coordinate route or template, raw reference plus rereference family, omitted/interpolated-channel policy, and the named harmonization branch. Inference from these sources: common-channel intersection, interpolation to a target montage, and REST / another explicit transform are different translation branches and do not justify one generic `harmonized EEG` claim.

2026-03-18 addendum: the Neural Contribution Card is the language-specific version of this general card

The Specificity & Shortcut Card is the general form that covers motor, memory, biomarker, speech, and related decode settings. The Neural Contribution Card is its language-specific specialization for text / speech / generative reconstruction settings, where language priors, candidate sets, prompts, vocoders, and causal deployment guards become dominant. In speech / brain-to-text work, the general shortcut audit is therefore stacked together with the Neural Contribution Card.

Minimum operating rules

If this card is missing, this site treats the result by default as exploratory decode, behavior-linked biomarker, or nuisance-unresolved classification, and does not promote it to a target-specific neural readout, mechanistic biomarker, or deployable controller. In particular, if any of plausible nuisance routes, nuisance-only baselines, fingerprint / acquisition-distribution audit / independence unit, or slice-wise hold-out is missing, this site stops the reader from rephrasing the result as "we learned what this variable means."

How this differs from the existing cards

The Observability Budget fixes what was directly observed. The Specificity & Shortcut Card fixes which route the predictive information came from. The Temporal Validity Card fixes how far the result can be extrapolated across time, the Calibration & Abstention Card fixes what confidence and fallback mean, and the Intervention Card fixes what was actually changed. On this site, decode / biomarker results therefore submit shortcut audit separately instead of mixing "the sensor contains information" with "the information is target-specific."

Open-vocabulary and generative are not one solved category

This site now separates at least four non-invasive language routes: within-subject semantic reconstruction, segment retrieval from a fixed candidate bank, known-onset word decoding, and prompt-conditioned generation. They improve different objects. Tang (2023) does not fix word-level timing. Défossez (2023) does not remove the candidate-segment structure. d'Ascoli (2025) does not remove known onsets or protocol asymmetry. Ye (2025) does not remove prompt / LLM dependence. Therefore, "open-vocabulary" or "generative" is not accepted here as a shortcut for unrestricted neural language readout.

Invasive language BCIs are not one temporal-validity route either

The same caution now applies to invasive communication papers. Willett et al. (2023), Littlejohn et al. (2025), and Wairagkar et al. (2025) support different throughput / expressivity routes under bounded output contracts. Singh et al. (2025) supports a separate cross-subject transfer-initialization route. Karpowicz et al. (2025) and Wilson et al. (2025) support adaptive rescue / recalibration under accumulating neural change. On this site, those papers do not share one durability claim by default.

Minimum operating rules

If this card is missing, this site treats the result by default as task-conditioned language decode, prior-assisted reconstruction, or communication-subsystem evidence, not as unrestricted thought reading, unique internal-state identification, or WBE-relevant state capture. In particular, if timing / segmentation regime, prior scaffold / prompt budget, brain-minus-prior baselines, or subject route / cooperation / countermeasure is missing, this site stops the reader from rephrasing the result as "the brain content itself was read out."

2026-04-04 correction: overlap audit is multi-axis, not a single yes/no field

The older wording on this card was directionally right but still too compressible. Current primary and official sources do not support treating overlap as one box. Brookshire et al. (2024) show that segment-based evaluation can leak subject information through raw-window ancestry, Chaibub Neto et al. (2019) show that repeated-measure diagnostic learning can absorb subject characteristics, Melnik et al. (2017) and Xu et al. (2020) show that device / setup distribution is a separate variance source, Jiang et al. (2024) and El Ouahidi et al. (2025) show that layout heterogeneity remains a live model-design problem, and the official EEG Challenge data, submission, and leaderboard pages together with Xiong et al. (2025) and Liu et al. (2026) show that benchmark object, operations budget, and protocol choices can all move rankings. Therefore, this site now treats overlap as a set of distinct ancestry axes rather than as one binary disclosure.

Official rules are not capability proof, but they are evidence of what comparability already requires

The EEG Challenge rules do not prove that any one model is best. They do something different and still important: they show that current operations already need pretraining-data disclosure, pretrained-model disclosure, and fine-tuning disclosure before a leaderboard is even interpretable. The later leaderboard postmortem sharpened the same point by showing that sample randomization and hidden grouping structure can change what the leaderboard was measuring at all. On this site, that operational lesson is promoted from benchmark-specific documentation to a general audit rule.

Minimum operating rules

If this card is missing, this site treats the result by default as qualified representation-learning / decoding evidence, not as portable transfer evidence, deployable robustness, source-identifiable recovery, or WBE-relevant state capture. In particular, if corpus identity / overlap audit is not split into raw-recording / window, subject / session, site / device / reference / layout, task / benchmark-object, and benchmark-operations ancestry, or if harmonization policy, adaptation regime, benchmark object / supervision unit, benchmark provenance including split randomness / hidden grouping, or the shortcut-resistance bridge to the Specificity & Shortcut Card is missing, this site stops the reader from rephrasing the result as "generalization is solved."

How this differs from the existing cards

The Observability Budget fixes what the sensor directly observed. The Specificity & Shortcut Card fixes which route predictive information came from. The Pretraining Card fixes what part of the reported transfer came from corpus composition, harmonization, adaptation, benchmark object, benchmark design, and compute allocation. The Temporal Validity Card then fixes how far the result can be extrapolated across time. On this site, a foundation-model result needs all of these boundaries kept separate, because a setup-agnostic or benchmark-winning representation is still not automatically shortcut-resistant.

Practical rules here

connectome-complete does not mean emulation-complete. This gate now mirrors the site's current public taxonomy: a connectome can be augmented by cell labels, synaptic snapshots, or one proxy route without closing the nineteen maintenance-state families or the separate shared electrical-state class. The detailed route-card rules for transcription, RNA, phospho-signaling, intrinsic excitability, sleep, myelin, ECM, ionic, thermal, neuromodulatory, bioenergetic, neurovascular / BBB, cargo, glial substrate-routing, astrocyte, clearance, and proteostasis live in Wiki: Homeostatic plasticity and maintenance state; the separate electrical-state rule lives in Wiki: electrical-state route card.

2026-03-27 addendum: vascular transfer audit is not neurovascular support-state audit

The remaining weakness was that a clean hemodynamic nuisance audit could still be misread as if the biological support layer were solved too. The primary literature does not support that shortcut. Bell et al. (2010), Kisler et al. (2020), Pandey et al. (2023), Swissa et al. (2024), and Mai-Morente et al. (2025) separate pericyte / BBB controller biology from measurement-side vascular transfer, while the current human routes already split further: Padrela et al. (2025) is a multi-echo ASL water-exchange route whose gray-matter age effect disappears after CBF / ATT correction, Morgan et al. (2024) showed that DP-ASL and ME-ASL can yield substantially different BBB water-exchange values and inconsistent age dependence, mouse work by Ohene et al. (2019) showed that multi-TE ASL exchange time is sensitive to AQP4 loss at the blood-brain interface, Padrela et al. (2026) showed lower Tex in SCD / MCI and moderate WMH burden while amyloid-group differences did not survive age / sex adjustment, and Chung et al. (2025) is a tracer-specific PET permeability-surface-area route with no human ground-truth validation yet. Therefore, this site now blocks the move from vascular-state / CVR audit passed to neurovascular support matched unless the biological route itself is named, measured, perturbed, or left explicitly latent, and it also blocks the move from human BBB proxy exists to a generic BBB-permeability meter exists unless the quantity type, carrier, and dominant transport interpretation are named explicitly.

2026-03-25 addendum: this gate now follows family-specific primary literature

The correction here is evidence-driven, not editorial. Hengen et al. (2016) distinguishes firing-rate recovery control from other excitability routes, Whitmore et al. (2022), Baxter et al. (2023), Schreiner et al. (2023), Schreiner et al. (2024), Whitmore et al. (2024), and Deng et al. (2025) distinguish sleep architecture / replay-coupling from sleep duration alone and also separate sleep continuity, physiology gating, and memory-age dependence from one generic replay label, Seidl et al. (2015) and Cohen et al. (2020) distinguish myelin / periaxonal timing support from one scalar delay, Reimer et al. (2016) and Hansen et al. (2022) distinguish mixed arousal proxies from transmitter-specific priors, Rzechorzek et al. (2022) keeps thermal-state separate from generic timing or vascular covariates, Rangaraju et al. (2019) distinguishes local bioenergetic / mitochondrial support from generic glial background, and Louveau et al. (2015), Kim et al. (2025), Hirschler et al. (2025), and Dagum et al. (2026) distinguish clearance / immune support from astrocyte-state and from direct local neural readout. Therefore, this page no longer compresses these families into umbrella rows such as intrinsic excitability/homeostasis/maintenance state or neural modification field.

2026-03-18 addendum: cell-type atlas is not current transcriptional state

Santoni et al. (2024) showed that chromatin plasticity predetermines neuronal eligibility for memory-trace formation, Traunmüller et al. (2025) showed temporally defined and region-specific chromatin / gene-expression changes after novel-environment exposure, Coda et al. (2025) showed cell-type- and locus-specific epigenetic editing of memory expression, and Terceros et al. (2026) showed distinct thalamocortical transcriptional gates for memory stabilization. Therefore, when a claim depends on allocation eligibility, late stabilization, or locus-specific plasticity, this site asks authors to disclose whether transcriptional / chromatin state was measured, perturbed, externally calibrated, or left latent.

2026-03-25 addendum: intrinsic-excitability evidence also needs a route card

The remaining weakness was that allocation / engram-excitability papers, homeostatic set-point papers, human local clinical-unit evidence, and living-human perturbation-conditioned plasticity proxies could still be compressed into one bucket called intrinsic-excitability evidence. The primary literature does not support that shortcut. Hadzibegovic et al. (2025) is about early intrinsic-excitability plasticity of neocortical engram neurons, Hengen et al. (2016) is about firing-rate homeostasis and recovery control across sleep / wake, Tallman et al. (2025) is a human hippocampal single-unit allocation-linked route whose encoding firing increase remains an indirect excitability index, and Huber et al. (2013), Kuhn et al. (2016), Khatri et al. (2025), Fehér et al. (2026), plus Zrenner et al. (2018) remain living-human perturbation-conditioned proxy routes rather than direct readouts of AIS geometry, ion-channel distribution, or cell-specific recovery controllers. On this site, intrinsic-excitability claims therefore have to disclose claim family, physiological locus, direct observable, time axis / intervention window, human evidence class / proxy class, and abstention boundary before any claim ceiling is raised. The full operating rule is summarized in Wiki: intrinsic excitability / homeostatic-set-point route card.

2026-03-19 addendum: transcription / chromatin evidence also needs a route card

The remaining weakness was that allocation eligibility, time-resolved response map, persistent stabilization cascade, and locus-specific causal editability could still be compressed into one bucket called transcriptomic evidence, while chromatin accessibility, histone-acetylation / histone-methylation control, DNA-methylation stabilization, higher-order looping, and locus-specific editing could still be compressed into one bucket called epigenetic evidence. The primary literature does not support either shortcut. Santoni et al. (2024), Guan et al. (2009), Gulmez Karaca et al. (2020), Bharadwaj et al. (2014), Coda et al. (2025), and Terceros et al. (2026) constrain different objects again. On this site, such claims now have to disclose claim family, molecular object family, species / region / task, sampling windows, assay and direct observable, animal-level independence / pseudoreplication handling, human observability ceiling, causal perturbation status, and abstention boundary. The full operating rule is in Wiki: transcription / chromatin route card.

2026-03-20 addendum: transcript count is not post-transcriptional RNA-state

Wang et al. (2015) showed that a neuron-specific LSD1 splice isoform regulates memory formation, Dai et al. (2019) showed that presynaptic neurexin alternative splicing changes postsynaptic receptor balance and contextual memory, Shi et al. (2018) and Li et al. (2025) showed that m6A-reader routes can alter hippocampus-dependent learning and memory, and Peterson et al. (2025) showed that ADAR2-mediated GluA2 editing contributes to homeostatic synaptic plasticity. Therefore, when a claim depends on isoform choice, m6A-dependent translation / degradation, or RNA-editing ratio, this site asks authors to disclose whether post-transcriptional RNA-state was directly measured, causally perturbed, externally calibrated, or silently replaced by gene-level abundance alone. Specialized long-read atlas work such as Joglekar et al. (2024) is informative for the ceiling but still not a comparable in vivo whole-brain human route.

2026-03-21 addendum: post-transcriptional RNA evidence now needs a route card

The remaining weakness was that alternative-splice controller papers, m6A-dependent translation papers, m6A-dependent degradation papers, RNA-editing controller papers, and long-read atlas papers could still be compressed into one bucket called post-transcriptional evidence. The primary literature does not support that shortcut. Wang et al. (2015) is a splice-isoform route whose downstream object is chromatin / transcriptional control, Dai et al. (2019) is a splice-dependent transsynaptic receptor-balance route, Shi et al. (2018) and Li et al. (2025) are distinct m6A translation-versus-degradation routes, Peterson et al. (2025) is an ADAR2 / GluA2 editing route for homeostatic scaling, and Joglekar et al. (2024) is an atlas / observability-ceiling route. On this site, post-transcriptional claims therefore have to disclose claim family, RNA control axis, time axis, assay and direct observable, downstream functional object, causal leverage, human observability ceiling, and abstention boundary before any claim ceiling is raised. The full operating rule is summarized in Wiki: post-transcriptional RNA route card.

2026-03-20 addendum: transcript or protein abundance is not phospho-signaling state

Giese et al. (1998), Lee et al. (2003), Rodrigues et al. (2004), Tomita et al. (2005), and Vierra et al. (2023) show that memory-relevant phosphosite occupancy and signaling nanodomains remain another control layer even when transcript or bulk protein abundance looks similar. Therefore, when a claim depends on kinase/phosphatase balance, phosphosite occupancy, or local second-messenger routing, this site asks authors to disclose whether phospho-signaling / second-messenger state was directly measured, causally perturbed, externally calibrated, or silently replaced by abundance-only evidence. Human phosphoproteome atlas work such as Biswas et al. (2023) is informative for the ceiling but still not a comparable in vivo whole-brain human route.

2026-03-21 addendum: local proteostasis evidence now needs a route card

Frey & Morris (1997) and Shires et al. (2012) are about tag / capture eligibility. Govindarajan et al. (2011) is about branch-level integration. Fonseca et al. (2006) and Parker et al. (2025) are about synthesis-degradation or proteasome-capacity balance. Pandey et al. (2021) and Chang et al. (2024) are about autophagy-linked plasticity routes. Lee et al. (2022) and Thomas et al. (2025) are about turnover-resistant persistence or a candidate tag substrate. Therefore, when a claim depends on late stabilization, reconsolidation, or cross-event capture, this site now asks authors not only whether the local proteostasis / synaptic-tag route was measured or perturbed, but also which claim family, which integrative unit, which direct observable, which turnover window, and which human observability ceiling apply. Current human routes on this site do not directly resolve tagged branches, PRP capture, local ribosome / proteasome / autophagy state, or same-subject whole-brain late-stabilization controllers. The longer operating rule is in Wiki: local proteostasis / synaptic-tagging route card.

2026-03 Addendum: delay is not one scalar

Hardware latency audit and biological timing-state audit solve different problems. Seidl et al. (2015), Dutta et al. (2018), Cohen et al. (2020), Micheva et al. (2021), and Dubey et al. (2022) show that node/internode geometry, periaxonal coupling, astrocyte control, and PV-axon myelination can all alter spike-arrival timing and synchrony. van Blooijs et al. (2023) pushes human tract-scale transmission-speed measurement forward, but it still remains a macro proxy. Therefore, when a claim depends on phase, synchrony, or closed-loop timing, this site asks authors to disclose whether biological timing-state was measured, externally calibrated, absorbed into a constant, or left latent.

2026-03-21 addendum: ECM / PNN evidence also needs a route card

The remaining weakness was that ECM / PNN evidence could still compress several different inferential objects into one bucket. The primary literature does not support that shortcut. Pizzorusso et al. (2002) is about plasticity-window reopening, Frischknecht et al. (2009) is about receptor-mobility constraint, Nguyen et al. (2020) is about microglia-driven ECM remodeling, Alexander et al. (2025) is about cell-type-specific memory support, Mehak et al. (2025) is about an age-linked rescue route, and Lehner et al. (2024) plus Banovac et al. (2025) remain human ex vivo histology. On this site, ECM / PNN claims therefore have to disclose which claim family, which matrix object and cell population, which direct observable, which controller or perturbation route, which recovery boundary, which human observability ceiling, and which abstention boundary apply before any claim ceiling is raised. The full operating rule is summarized in Wiki: ECM / PNN route card.

2026-03-21 addendum: ionic / chloride evidence also needs a route card

The remaining weakness was that ionic evidence could still compress several different inferential objects into one bucket. The primary literature does not support that shortcut. Glykys et al. (2014) is about local chloride set point, Heubl et al. (2017) is about activity-dependent KCC2 regulation, Ding et al. (2016) and Forsberg et al. (2022) are about interstitial / CSF ion composition linked to sleep-wake state, Byvaltsev et al. (2023) is about perisynaptic K⁺ clearance by reverse-mode KCC2, Alfonsa et al. (2025) is about sleep-wake-history-dependent E_GABAA shifts that change LTP induction, and Huberfeld et al. (2007) is a human pathology route. The human sodium row is also split: Qian et al. (2012) is TSC mapping, Fleysher et al. (2013) is SQ+TQF-derived ISMF / ISC / ISVF, Rodriguez et al. (2022) is a repeatable normalized sodium density-weighted route, Azilinon et al. (2023) shows that TSC and short-component fraction f can diverge, and Qian et al. (2025) is mono-/bi-T₂ signal separation. On this site, ionic / chloride claims therefore have to disclose which claim family, which direct ionic observable, which spatial regime, which perturbation or controller route, which human quantity type / compartment model, and which abstention boundary apply before any claim ceiling is raised. The full operating rule is summarized in Wiki: ionic / chloride route card.

2026-03-30 addendum: shared extracellular / electrical-state evidence needs an extracellular-geometry split

The remaining weakness was that electrical-state evidence could still compress several different inferential objects into one bucket, especially by hiding extracellular-space geometry inside the electrical label. The primary literature does not support that shortcut. Galarreta & Hestrin (1999) is about gap-junction coupling networks, Anastassiou et al. (2011) is about endogenous-field / ephaptic spike-timing bias, Graydon et al. (2014) is about local extracellular-volume-fraction geometry and neurotransmitter dilution, Kilb et al. (2006) and Lauderdale et al. (2015) are about osmotic extracellular-space contraction / edema-linked excitability, Burman et al. (2023) is about inhibitory driving-force regime in active cortex, Yang et al. (2024) is about activity-dependent electrical-synapse remodeling, Selfe et al. (2024) is a direct inhibitory-driving-force assay with a specialized local optical route, Xie et al. (2013) is about sleep-linked interstitial-space expansion, Voldsbekk et al. (2020) remains a wakefulness-related human diffusion-MRI proxy clue, Örzsik et al. (2023) adds a sleep-conditioned higher-order diffusion / glymphatic clue under a within-subject sleep-deprivation-plus-zolpidem regime, and Feld et al. (2026) remains a human perturbation-conditioned clue with pharmacological caveats rather than a local whole-brain readout. On this site, shared extracellular / electrical-state claims therefore have to disclose which claim family, which direct extracellular / electrical observable, which spatial regime, which perturbation or calibration route, which human evidence class, and which abstention boundary apply before any claim ceiling is raised. The full operating rule is summarized in Wiki: electrical-state route card.

2026-03 Addendum: Use augmentation / ablation instead of enumeration

The weakness found in this re-audit was that by simply listing the state variables as ``missing,'' it was difficult to convey to the reader what could be added to make the claim even stronger. Therefore, on this site, we will compare the connectome-only baseline and the model with additional variables under the same held-out conditions, and request submissions that show which augmentation reduced which error term.

2026-03 Addendum: Add measurement stack to augmentation claim

Even if the same "transcriptomic label is added" or "same-brain function is added", the variables directly observed by the whole-brain spatial atlas, Patch-seq, volume EM, same-brain calcium+EM, and local transmitter sensor are different. Therefore, on this site, we always include Which measurement stack provided the additional information in the augmentation claim, and also write the claim ceiling for each stack separately. For a summary table, see Wiki: observability and claim ceiling by measurement stack.

Weaknesses found in audit

In the previous text, methods with different characteristics, such as ASR, ZapLine-plus, PCI-ST, Effective Information, and EPR, were listed as the same "essential requirements." But first and foremost, COBIDAS-MEEG and EEG-BIDS emphasizetransparent reporting, shareable metadata, and comparable baselines. Therefore, on this page, we reposition foundation of reproducibility as required, methodology selection recommended, andadditional theory-driven indicators as exploration.

Minimum operating rules

No-report / criterion placement is treated as a construct-validity gate, PCI / PCI-ST is treated as a main benchmark candidate only after it passes a perturbational gate, resting-state complexity / criticality remains an auxiliary proxy until it passes a calibration gate, and a multimodal clinical panel remains exploratory until it passes an incremental / deployability gate. In other words, on this site, the same "awareness index" can have a different claim ceiling depending on which gate it has passed.

What still had to be fixed after the 4-gate split

Even after the 4-gate split, one shortcut remained: a submission could still borrow authority from a familiar label such as IIT, PCI, criticality, or multimodal without exposing which gate was actually passed on which denominator. That shortcut is not supported by the primary literature. Therefore, this site now requires the following card whenever a result is promoted as a consciousness-related readout.

How this card changes site behavior

If the card is missing, this site reads the submission at the lowest gate actually supported by the exposed evidence. A theory-associated metric remains a prediction-linked readout, not a theory verdict. A passive classifier remains a proxy, not a perturbation substitute. A multimodal gain remains bundle-performance evidence, not automatic clinical portability.

What is different from ordinary accuracy evaluation

A normal accuracy evaluation asks whether the system gives the same answer to the same question. The causal perturbation suite goes one step further and compares how performance collapses and how it recovers when conditions are intentionally changed. In other words, it is not only a test of answer matching, but also a test of whether internal mechanisms react in comparable ways.

When you want to organize only the entrance first

If you want an everyday-language explanation of the differences between held-out accuracy, intervention, counterfactual reasoning, and perturbation-based verification, read Wiki: Counterfactual and Perturbation Verification first.

2026-03-28 addendum: co-adaptation must be logged before online gains are interpreted

One remaining shortcut at the L3 entrance was that a paper could still say "online performance improved" without separating user-side learning, decoder updates, and application-side shaping. The primary literature does not support that shortcut. Orsborn et al. (2014) showed that decoder adaptation can itself shape neural plasticity during neuroprosthetic control. Perdikis et al. (2018) and Abu-Rmileh et al. (2019) showed that longitudinal BCI gains can depend on mutual learning and that adaptation rate can trade off against subject learning. Wairagkar et al. (2025) and Wilson et al. (2025) then showed modern speech and cursor loops with session-to-session retraining, blockwise decoder updates, and explicit open-loop probes. Therefore, this site now requires a Co-Adaptation Log before same-session online gains are promoted to fixed-decoder or durable-loop language.

Evaluation metrics

In the first judgment, we prioritize pre-registered effect size, robustness under OOD conditions, calibration error, abstention rate, and uncertainty range, as well as end-to-end latency P50 / P95 / P99, jitter, dropout, recalibration burden, and recovery time after perturbation. PCI-ST spatial distribution, bottleneck distance on persistence diagrams, and Fisher Information Metric (FIM) distances between generated models are left as auxiliary analyses; the main pass / fail judgment does not depend on any single method.

Don't sell the word counterfactual cheaply

If the branching conditions, comparison rules, failure conditions, and stimulus artifact windows are not fixed in advance, we do not refer to it as a "counterfactual equivalence" on this site, but rather as an intervention response test or a perturbation generalization test.

Minimum operating rules

If this card is missing, this site treats the result by default as a task-specific local controller, local subsystem loop, or surrogate-body demonstration. It does not promote the result to solved embodiment, subject-complete closed loop, or WBE-relevant body / environment equivalence.

Fast routes and slow milieu are different audits

Respiration, pupil, HR / HRV, tactile contact, self-motion, and dialogue context are not the same inferential object as circadian phase, corticosteroid exposure, or feeding / insulin regime. This site therefore logs fast loop routes and slow internal-milieu routes separately rather than assuming one disclosure covers both.

How this differs from the Intervention Card

The Intervention Card fixes what was changed and how timing/safety were logged. The Body / Environment Boundary Card fixes what system boundary the loop actually used. On this site, a paper needs both before its L3 wording is allowed to rise.

Handling external dependent tasks

Validations that require real subject intervention (TMS/tDCS), IRB review, and equipment procurement are managed as externally dependent tasks. In this repository, "requirement specifications, judgment conditions, and public log format" will be implemented in advance, and the experiment implementation itself will proceed on a separate track.

2026-03-28 addendum: irreversibility claims also need a reverse-transition support audit

Lynn et al. (2021) estimated entropy-production lower bounds from coarse-grained BOLD state transitions, showed sensitivity to the number of coarse-grained states, and displayed finite-data confidence intervals around the resulting fluxes. de la Fuente et al. (2023) used inversion decoding on ECoG and showed dependence on principal-component choice, feature set, and model complexity. Ishihara & Shimazaki (2025) estimated model-based entropy flow from spike trains while explicitly stating pairwise and conditional-independence limits and using trial-shuffled controls to separate coupling-related contributions from firing-rate dynamics and sampling error. Teza & Stella (2020) and Cocconi et al. (2022) showed that coarse graining is part of the thermodynamic question rather than a harmless implementation detail, and Epp et al. (2025) showed that BOLD changes can oppose oxygen-metabolism changes. A second correction is required by the partial-observation and lacking-data literature. Martínez et al. (2019) showed that waiting-time asymmetry can reveal hidden dissipation even when observable current vanishes, Hartich & Godec (2024) showed that this reading can fail when coarse-graining and time reversal do not commute, Martínez et al. (2024) replied by limiting the original claim to coarse-grainings that are local in time and, where needed, second-order semi-Markov constructions, Blom et al. (2024) showed that coarse lumping can hide dissipative cycles and induce memory so that estimates from partial observations become far too small when the observed trajectory is naively treated as Markov, and Baiesi et al. (2024) showed that when reverse transitions are sparse or unobserved, direct entropy-production estimation can fail and lower-bound strategies become preferable. Therefore, the phrase "thermodynamic result" still does not tell us which quantity was computed, whether an energetic interpretation is grounded, whether the observed trajectory is thermodynamically closed enough for the chosen estimator, or whether the reported irreversibility had adequate reverse-transition support rather than only a clean surrogate test. On this site, any such claim now needs an Irreversibility / Thermodynamic Route Card; the longer public rule is summarized in Wiki: irreversibility route card.

2026-04-02 addendum: operational use also needs stability and bridge audits

The next weakness was that estimator meaning and hidden-degree risk were already separated, but operational stability was still too easy to smuggle in as if it came for free once the mathematics looked sound. Current primary literature does not support that shortcut. Poudel et al. (2024) showed that small motion can materially alter visibility-graph structure and that only low-motion subsets reached moderate-to-high test-retest reliability for selected metrics. Metzen et al. (2024) showed that BOLD variability and complexity measures have markedly different reliability profiles, with some functional-connectivity complexity measures remaining unacceptable-to-moderate. Omidvarnia et al. (2021) showed reproducible multiscale-entropy structure for resting-state fMRI, but that does not automatically transfer to every irreversibility family. A second weakness remained on the physiology side: Chen et al. (2025) showed with simultaneous EEG-PET-MRI that temporal coupling across modalities can be strong while spatial organization and state trajectories remain non-identical, and Epp et al. (2025) showed that task BOLD changes can oppose oxygen-metabolism changes. Therefore, this repository now treats stability / nuisance sensitivity, cross-estimator concordance, and physiology-bridge quality as separate reporting burdens before an irreversibility result is discussed as more than an exploratory auxiliary log.

What to do now with this repository

In this repository, we do not make thermodynamic indicators a "required submission," but rather leave them at the stage where route-card reporting, literature monitoring, estimation-error auditing, stability / nuisance checks, and explicit physiology-bridge logging are in place. This will be treated as an exploratory auxiliary analysis until stable within-modality operation and bridge-quality criteria are confirmed.