Non-Invertible but Not Anonymous: A Privacy Characterization of a Deterministic Edge Decision Token, and a Decision-Bottleneck Egress Rule

{"Non-Invertible":[0],"but":[1,102,525,966],"Not":[2],"Anonymous:":[3],"A":[4,269,349,664,1203],"Privacy":[5],"Characterization":[6],"of":[7,55,112,183,234,276,333,563,616,646,972,1014,1033,1096,1221],"a":[8,14,53,56,63,72,89,107,116,127,132,156,172,181,189,220,243,261,274,303,367,382,388,411,425,429,457,460,464,476,500,507,513,534,557,569,591,654,694,707,732,746,753,901,915,969,973,979,1049,1207,1222],"Deterministic":[9,1184],"Edge":[10],"Decision":[11],"Token,":[12],"and":[13,22,77,87,135,151,180,317,336,366,377,381,398,437,459,499,512,541,630,649,676,693,699,703,718,721,765,806,809,839,850,883,900,914,926,953,961,978,988,1037,1073,1078,1098,1125,1132,1136,1142,1153,1167],"Decision-Bottleneck":[15],"Egress":[16],"Rule":[17],"Randolph":[18],"James":[19],"Ferlic,":[20,964],"M.D.":[21],"Kimberly":[23],"Kate":[24],"Ferlic":[25,960],"(Fieldstone":[26],"Analytics,":[27,1151],"LLC,":[28],"Austin,":[29],"TX,":[30],"USA)":[31],"Preprint":[32],"·":[33,37,40,469,523,581,618,671,726,782,829,853,933,1174,1183,1192,1202,1217],"Zenodo":[34],"DOI:":[35],"10.5281/zenodo.22819210":[36],"CC-BY":[38,1162],"4.0":[39,997,1163],"Community:":[41],"spiral-domain-encoder-campaign":[42],"Abstract":[43],"On-device":[44],"\\"decision":[45],"tokens\\"":[46],"—":[47,67,192,242,308,428,479,528,574,590,623,679,731,789,831,855,935,1067,1092],"compact,":[48],"fixed-width":[49],"codes":[50],"that":[51,62,122,1004,1053],"summarize":[52],"window":[54,125,287,346],"sensor":[57],"stream":[58,182,271,275,299,385,489],"into":[59],"the":[60,80,85,164,169,193,205,214,226,235,239,253,257,293,312,320,326,357,359,392,405,438,443,447,452,519,550,553,576,585,588,599,608,614,625,633,639,658,680,711,724,766,771,775,780,790,794,811,814,818,822,832,835,841,847,856,859,863,870,874,879,884,888,892,897,920,929,936,949,993,1015,1030,1044,1068,1079,1084,1094,1105,1154],"byte":[61],"downstream":[64],"decision":[65,94,258,263,589,593,635,683,976,1223,1244],"needs":[66],"are":[68,99,347,820,905,918,924,1093,1139,1143,1164],"increasingly":[69],"proposed":[70],"as":[71,957,1145,1206],"privacy-favorable":[73],"substrate":[74],"for":[75,373,651,700,1178,1188,1197,1212],"edge":[76,975,1214,1228,1264],"ambient":[78],"intelligence:":[79],"raw":[81,240,334,647,908],"signal":[82,1180],"never":[83],"leaves":[84],"device,":[86],"only":[88,266,319,632],"small":[90,558],"token":[91,130,170,191,215,294,314,393,414,431,444,551,567,600,627,681,712,735,755,772,1186,1205,1224],"or":[92,342,1010,1021],"its":[93,421,742,1074],"is":[95,171,196,200,216,295,415,424,442,455,531,543,568,583,601,736,745,770,911,1017,1048,1054,1059],"shared.":[96],"Such":[97],"claims":[98],"frequently":[100],"asserted":[101],"rarely":[103],"measured.":[104],"We":[105,154],"give":[106],"pre-registered,":[108],"adversarial":[109,481],"privacy":[110,146,466,473,609,970,1031,1080],"characterization":[111,458,971],"one":[113,286],"concrete":[114],"instance:":[115],"frozen,":[117],"deterministic,":[118],"class-discriminant":[119,1069,1246],"single-token":[120,1070,1179,1199],"encoder":[121,1072],"maps":[123],"each":[124],"to":[126,288,331,533,644],"single":[128,190,412,430,477,733,754],"~8-bit":[129],"by":[131,1024,1061,1104],"Fisher-discriminant":[133],"projection":[134],"vector":[136,1260],"quantization.":[137],"On":[138],"real":[139],"electrocardiogram":[140,1248],"data":[141,910,1240],"(MIT-BIH),":[142],"we":[143],"test":[144],"four":[145,374,480],"properties":[147],"with":[148,834,1003],"strong":[149],"attackers":[150],"fairness":[152],"controls.":[153],"find":[155],"sharp,":[157],"honest":[158,408,728,1272],"split.":[159],"(i)":[160],"Exact-signal":[161],"non-invertibility":[162],"holds:":[163],"best":[165],"possible":[166],"reconstruction":[167,530],"from":[168,329,642],"coarse":[173,535],"1-of-128":[174,536],"codebook":[175,537,1071,1176],"centroid":[176,538],"(normalized":[177],"RMSE":[178],"0.36),":[179],"tokens":[184],"reconstructs":[185],"no":[186,201,544,907,1006,1039],"better":[187],"than":[188,355,613],"~8-bit-per-window":[194],"bottleneck":[195],"real.":[197],"(ii)":[198],"There":[199],"membership-inference":[202],"leak":[203,546,659],"through":[204],"encoder's":[206],"native":[207],"novelty":[208,1076],"channel":[209,1195],"(AUROC":[210,547],"0.501).":[211],"(iii)":[212],"But":[213],"not":[217,245,255,280,463,475,526,571,587,669,817,954,967,1056,1140,1158],"anonymizing:":[218],"in":[219,238,252,256,556,584,846,919,1083],"closed-set":[221,487],"linkage":[222,805],"attack":[223,272],"it":[224,259],"re-identifies":[225,264,552,594],"individual":[227,554],"at":[228,265,285,290,297,758,763],"0.976":[229],"(chance":[230],"0.10),":[231],"preserving":[232,561],"98%":[233,562],"identifiability":[236],"present":[237],"features":[241],"pseudonym,":[244,570],"an":[246,434,572,604,750],"anonym.":[247],"Crucially,":[248],"this":[249,826,1025],"identity":[250,283],"lives":[251],"token,":[254,586,977],"supports:":[260],"one-bit":[262,592],"0.146.":[267],"(iv)":[268],"pre-registered":[270,350,472],"shows":[273],"emitted":[277],"decisions":[278],"does":[279],"materially":[281],"accumulate":[282],"(0.133":[284],"0.331":[289],"fifty),":[291],"while":[292,598],"near-perfect":[296,422,743],"every":[298,504],"length.":[300],"This":[301,454,1027],"motivates":[302],"simple,":[304],"measurable":[305],"design":[306,461,666,1051],"rule":[307,449,622,777,951,1047],"decision-bottleneck":[309,620,980,1045],"egress:":[310],"keep":[311,575,624],"identifying":[313,626],"ephemeral":[315,628],"on-device":[316,629],"emit":[318,631],"minimal":[321,634],"decision,":[322],"cutting":[323],"what":[324,637,821],"crosses":[325,638],"device":[327,1250],"boundary":[328,640],"~98%":[330,643],"~5%":[332,645],"identifiability,":[335],"degrading":[337],"steeply":[338],"if":[339],"even":[340],"two":[341],"three":[343],"bits":[344],"per":[345,928],"emitted.":[348],"hardening":[351,515,791,872],"pass":[352,516,792],"sharpens":[353],"rather":[354],"overturns":[356],"thesis:":[358],"dissociation":[360,684],"reproduces":[361,685],"across":[362,396,673,715],"surgical-robot":[363,688],"kinematics,":[364],"sEMG,":[365],"non-personal":[368,695],"industrial":[369,696,877],"source-fingerprinting":[370,697],"case,":[371,698],"holds":[372,650],"alternative":[375],"tokenizers,":[376,675],"survives":[378],"confidence":[379],"intervals":[380],"stronger":[383,812],"sequence-model":[384],"adversary;":[386,813],"on":[387,686,706,1122],"second":[389,514,708],"ECG":[390,709],"database":[391,710],"links":[394,713],"individuals":[395,714],"recordings":[397,716],"leaks":[399,719],"demographic":[400],"attributes":[401],"(sex,":[402],"age)":[403],"beyond":[404,723],"decision.":[406,725],"One":[407],"bound":[409,730],"emerges:":[410],"8-bit":[413,734],"entropy-capped":[416,737],"(≤":[417,738],"256":[418,739],"states),":[419,740,1124],"so":[420,741],"re-identification":[423,744],"small-closed-set":[426,747],"result":[427,478,894],"cannot":[432],"identify":[433],"open":[435,751],"population,":[436],"durable":[439,767],"large-scale":[440,768],"threat":[441,482,677],"stream,":[445,773,869],"which":[446,774],"egress":[448,621,776,950,981,1046],"keeps":[450,778],"off":[451,779],"wire.":[453,781],"explicitly":[456],"rule,":[462,667],"new":[465,941,1040],"mechanism.":[467],"Highlights":[468],"An":[470],"honest,":[471],"characterization,":[474],"models":[483,678],"(reconstruction,":[484],"membership":[485,545,1234],"inference,":[486,495,808],"re-identification,":[488],"re-identification)":[490],"plus":[491],"cross-session":[492,804,1238],"linkage,":[493],"attribute":[494,807,1236],"population-scaling,":[496],"open-set":[497],"verification,":[498],"stronger-adversary":[501],"robustness":[502],"pass;":[503],"hypothesis":[505],"carried":[506],"frozen":[508,861],"read":[509],"before":[510,521],"running,":[511],"deliberately":[517,668,1055],"attacked":[518],"paper":[520,823],"writing.":[522],"Non-invertible":[524],"anonymous":[527],"exact-signal":[529],"bounded":[532],"(nRMSE":[539],"0.36)":[540],"there":[542],"0.501),":[548],"yet":[549],"(0.976":[555],"closed":[559],"set),":[560],"raw-feature":[564],"identifiability.":[565],"The":[566,619,727,1064,1218],"anonym":[573],"confidentiality":[577],"claim,":[578],"drop":[579],"de-identification.":[580],"Identity":[582],"near":[595],"chance":[596],"(0.146)":[597],"near-perfect.":[602],"What":[603,825],"application":[605],"emits":[606],"determines":[607],"outcome":[610],"far":[611],"more":[612],"fact":[615],"tokenization.":[617],"→":[636,761],"drops":[641],"identity,":[648],"continuous":[652],"streams;":[653],"2–3-bit":[655],"emission":[656],"re-opens":[657],"(a":[660],"measured":[661],"granularity":[662],"cliff).":[663],"data-minimization":[665,1050],"claimed.":[670],"Generalization":[672],"modalities,":[674],">":[682],"ECG,":[687],"kinematics":[689],"(surgeon),":[690],"sEMG":[691],"(subject),":[692],"PCA-only,":[701],"random-projection,":[702],"raw-k-means":[704],"tokenizers;":[705],"(0.71)":[717],"sex":[720],"age":[722],"scale":[729],"number;":[748],"against":[749],"population":[752],"collapses":[756],"(0.82":[757],"10":[759],"patients":[760],"0.001":[762],"5,883),":[764],"identifier":[769],"Corrections":[783],"made":[784],"under":[785,992,1161],"scrutiny,":[786],"reported":[787],"verbatim":[788],"forced":[793],"\\"near-perfect":[795],"biometric\\"":[796],"claim":[797],"inward":[798],"(small-closed-set":[799],"/":[800],"strongly-biometric-modality":[801],"specific),":[802],"confirmed":[803],"survived":[810],"corrected":[815],"claims,":[816],"originals,":[819],"reports.":[824],"record":[827],"contains":[828],"Manuscript_Paper43.pdf":[830],"manuscript":[833,1085],"seven":[836,898],"figures":[837],"embedded,":[838],"Manuscript_Paper43.docx,":[840],"editable":[842],"source.":[843],"DOI":[844],"stamped":[845],"front":[848],"matter":[849],"Cite-as":[851],"line.":[852],"PAPER_43_ZENODO_ARCHIVE.zip":[854],"reproducibility":[857],"archive:":[858],"six":[860],"pre-registrations,":[862],"ten":[864],"experiment":[865],"runners":[866],"(reconstruction/membership/identity,":[867],"ephemeral-egress":[868],"sEMG/MIT-BIH/battery":[871],"trio,":[873],"ROSMA":[875],"+":[876],"extension,":[878],"PTB-XL":[880],"scale/cross-session/attribute-inference":[881],"runner,":[882],"alternate-tokenizer":[885],"generality":[886],"sweep),":[887],"consolidated":[889],"figure-rebuild":[890],"script,":[891],"per-experiment":[893],"JSON":[895],"records,":[896],"figures,":[899],"README.":[902],"All":[903,922],"datasets":[904],"public;":[906],"benchmark":[909],"redistributed":[912],"(sources":[913],"DATA_ROOT":[916],"convention":[917],"README).":[921],"paths/identifiers":[923],"scrubbed":[925],"leak-scanned":[927],"campaign":[930],"deposit":[931],"discipline.":[932],"PAPER43_IP_COORDINATION_AUDIT.md":[934],"pre-deposit":[937],"IP":[938],"audit":[939],"(no":[940],"algorithmic":[942,1041],"subject":[943,1042,1095],"matter;":[944,1043],"all":[945,1156],"referenced":[946,1082],"methods":[947,1036,1065],"filed;":[948],"anticipated":[952,1060],"claimed).":[955],"Cite":[956],"R.":[958],"J.":[959],"K.":[962,963],"\\"Non-invertible":[965],"anonymous:":[968],"deterministic":[974],"rule,\\"":[982],"Zenodo,":[983],"2026,":[984],"doi:":[985],"10.5281/zenodo.22819210.":[986],"License":[987,999],"patent":[989,1008,1101],"notice":[990],"Released":[991],"Creative":[994],"Commons":[995],"Attribution":[996],"International":[998],"(CC-BY":[1000],"4.0).":[1001],"Consistent":[1002],"license,":[1005],"patent,":[1007],"application,":[1009],"other":[1011],"intellectual-property":[1012],"right":[1013],"authors":[1016],"licensed,":[1018],"waived,":[1019],"granted,":[1020],"otherwise":[1022],"conveyed":[1023],"deposit.":[1026],"work":[1028],"characterizes":[1029],"behavior":[1032],"previously":[1034],"described":[1035],"discloses":[1038],"principle":[1052],"claimed":[1057],"(it":[1058],"prior":[1062],"art).":[1063],"characterized":[1066],"nearest-centroid":[1075],"monitor,":[1077],"operations":[1081,1121],"(identity-axis":[1086],"projection,":[1087],"differential-privacy":[1088],"composition,":[1089],"keyed":[1090],"access)":[1091],"filed":[1097],"pending":[1099],"U.S.":[1100,1108,1115,1126],"applications":[1102],"held":[1103],"authors,":[1106],"including":[1107],"Provisional":[1109,1127],"Application":[1110,1117],"No.":[1111,1118],"64/095,354":[1112],"(the":[1113],"encoder),":[1114],"Non-Provisional":[1116],"19/691,234":[1119],"(privacy-preserving":[1120],"encoded":[1123],"Applications":[1128],"Nos.":[1129],"64/084,821,":[1130],"64/084,817,":[1131],"64/127,391.":[1133],"Per-deployment":[1134],"productization":[1135],"deployment-selection":[1137],"know-how":[1138],"disclosed":[1141],"retained":[1144],"trade":[1146],"secrets.":[1147],"©":[1148],"2026":[1149],"Fieldstone":[1150],"LLC":[1152],"authors;":[1155],"rights":[1157],"expressly":[1159],"granted":[1160],"reserved.":[1165],"Licensing":[1166],"collaboration":[1168],"inquiries:":[1169],"[email protected].":[1170],"Companion":[1171],"deposits":[1172],"(spiral-domain-encoder-campaign)":[1173],"Class-discriminant":[1175],"construction":[1177],"compression:":[1181,1190],"doi:10.5281/zenodo.20788187":[1182],"multi-token":[1185],"ladder":[1187],"channel-partition":[1189],"doi:10.5281/zenodo.22003179":[1191],"Label-free":[1193],"inference-time":[1194],"fusion":[1196],"drift-robust":[1198],"decisions:":[1200],"doi:10.5281/zenodo.22046713":[1201],"decision-oriented":[1204],"bounded,":[1208],"threshold-free":[1209],"cache":[1210],"key":[1211],"generative":[1213],"outputs:":[1215],"doi:10.5281/zenodo.22148612":[1216],"predictive":[1219],"reach":[1220],"(forecasting/anticipation/fusion):":[1225],"doi:10.5281/zenodo.22736921":[1226],"Keywords":[1227],"AI":[1229],"privacy;":[1230,1253],"re-identification;":[1231],"pseudonymity;":[1232],"non-invertibility;":[1233],"inference;":[1235,1237],"linkage;":[1239],"minimization;":[1241],"information":[1242],"bottleneck;":[1243],"token;":[1245],"codebook;":[1247],"biometrics;":[1249,1255],"fingerprinting;":[1251],"differential":[1252],"cancelable":[1254],"biometric":[1256],"template":[1257],"protection;":[1258],"unlinkability;":[1259],"quantization;":[1261],"wearable":[1262],"sensors;":[1263],"computing;":[1265],"TinyML;":[1266],"ECG;":[1267],"de-identification;":[1268],"GDPR;":[1269],"HIPAA;":[1270],"pre-registration;":[1271],"negatives":[1273]}

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-09-17
DOI
https://doi.org/10.5281/zenodo.22819209
Primary Topic
Wireless Communication Security Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
preprint

Non-Invertible but Not Anonymous: A Privacy Characterization of a Deterministic Edge Decision Token, and a Decision-Bottleneck Egress Rule

Randolph James Ferlic, Kimberly Kate Ferlic
Zenodo (CERN European Organization for Nuclear Research)
Wireless Communication Security Techniques
preprint

Non-Invertible but Not Anonymous: A Privacy Characterization of a Deterministic Edge Decision Token, and a Decision-Bottleneck Egress Rule

Randolph James Ferlic, Kimberly Kate Ferlic
preprint en

Abstract

Non-Invertible but Not Anonymous: A Privacy Characterization of a Deterministic Edge Decision Token, and a Decision-Bottleneck Egress Rule Randolph James Ferlic, M.D. and Kimberly Kate Ferlic (Fieldstone Analytics, LLC, Austin, TX, USA) Preprint · Zenodo DOI: 10.5281/zenodo.22819210 · CC-BY 4.0 · Community: spiral-domain-encoder-campaign Abstract On-device "decision tokens" — compact, fixed-width codes that summarize a window of a sensor stream into the byte that a downstream decision needs — are increasingly proposed as a privacy-favorable substrate for edge and ambient intelligence: the raw signal never leaves the device, and only a small token or its decision is shared. Such claims are frequently asserted but rarely measured. We give a pre-registered, adversarial privacy characterization of one concrete instance: a frozen, deterministic, class-discriminant single-token encoder that maps each window to a single ~8-bit token by a Fisher-discriminant projection and vector quantization. On real electrocardiogram data (MIT-BIH), we test four privacy properties with strong attackers and fairness controls. We find a sharp, honest split. (i) Exact-signal non-invertibility holds: the best possible reconstruction from the token is a coarse 1-of-128 codebook centroid (normalized RMSE 0.36), and a stream of tokens reconstructs no better than a single token — the ~8-bit-per-window bottleneck is real. (ii) There is no membership-inference leak through the encoder's native novelty channel (AUROC 0.501). (iii) But the token is not anonymizing: in a closed-set linkage attack it re-identifies the individual at 0.976 (chance 0.10), preserving 98% of the identifiability present in the raw features — a pseudonym, not an anonym. Crucially, this identity lives in the token, not in the decision it supports: a one-bit decision re-identifies at only 0.146. (iv) A pre-registered stream attack shows a stream of emitted decisions does not materially accumulate identity (0.133 at one window to 0.331 at fifty), while the token is near-perfect at every stream length. This motivates a simple, measurable design rule — decision-bottleneck egress: keep the identifying token ephemeral on-device and emit only the minimal decision, cutting what crosses the device boundary from ~98% to ~5% of raw identifiability, and degrading steeply if even two or three bits per window are emitted. A pre-registered hardening pass sharpens rather than overturns the thesis: the dissociation reproduces across surgical-robot kinematics, sEMG, and a non-personal industrial source-fingerprinting case, holds for four alternative tokenizers, and survives confidence intervals and a stronger sequence-model stream adversary; on a second ECG database the token links individuals across recordings and leaks demographic attributes (sex, age) beyond the decision. One honest bound emerges: a single 8-bit token is entropy-capped (≤ 256 states), so its near-perfect re-identification is a small-closed-set result — a single token cannot identify an open population, and the durable large-scale threat is the token stream, which the egress rule keeps off the wire. This is explicitly a characterization and a design rule, not a new privacy mechanism. Highlights · An honest, pre-registered privacy characterization, not a single result — four adversarial threat models (reconstruction, membership inference, closed-set re-identification, stream re-identification) plus cross-session linkage, attribute inference, population-scaling, open-set verification, and a stronger-adversary robustness pass; every hypothesis carried a frozen read before running, and a second hardening pass deliberately attacked the paper before writing. · Non-invertible but not anonymous — exact-signal reconstruction is bounded to a coarse 1-of-128 codebook centroid (nRMSE 0.36) and there is no membership leak (AUROC 0.501), yet the token re-identifies the individual (0.976 in a small closed set), preserving 98% of raw-feature identifiability. The token is a pseudonym, not an anonym — keep the confidentiality claim, drop de-identification. · Identity is in the token, not the decision — a one-bit decision re-identifies near chance (0.146) while the token is near-perfect. What an application emits determines the privacy outcome far more than the fact of tokenization. · The decision-bottleneck egress rule — keep the identifying token ephemeral on-device and emit only the minimal decision → what crosses the boundary drops from ~98% to ~5% of raw identity, and holds for continuous streams; a 2–3-bit emission re-opens the leak (a measured granularity cliff). A data-minimization design rule, deliberately not claimed. · Generalization across modalities, tokenizers, and threat models — the token > decision dissociation reproduces on ECG, surgical-robot kinematics (surgeon), sEMG (subject), and a non-personal industrial source-fingerprinting case, and for PCA-only, random-projection, and raw-k-means tokenizers; on a second ECG database the token links individuals across recordings (0.71) and leaks sex and age beyond the decision. · The honest scale bound — a single 8-bit token is entropy-capped (≤ 256 states), so its near-perfect re-identification is a small-closed-set number; against an open population a single token collapses (0.82 at 10 patients → 0.001 at 5,883), and the durable large-scale identifier is the token stream, which the egress rule keeps off the wire. · Corrections made under scrutiny, reported verbatim — the hardening pass forced the "near-perfect biometric" claim inward (small-closed-set / strongly-biometric-modality specific), confirmed cross-session linkage and attribute inference, and survived the stronger adversary; the corrected claims, not the originals, are what the paper reports. What this record contains · Manuscript_Paper43.pdf — the manuscript with the seven figures embedded, and Manuscript_Paper43.docx, the editable source. DOI stamped in the front matter and Cite-as line. · PAPER_43_ZENODO_ARCHIVE.zip — the reproducibility archive: the six frozen pre-registrations, the ten experiment runners (reconstruction/membership/identity, ephemeral-egress stream, the sEMG/MIT-BIH/battery hardening trio, the ROSMA + industrial extension, the PTB-XL scale/cross-session/attribute-inference runner, and the alternate-tokenizer generality sweep), the consolidated figure-rebuild script, the per-experiment result JSON records, the seven figures, and a README. All datasets are public; no raw benchmark data is redistributed (sources and a DATA_ROOT convention are in the README). All paths/identifiers are scrubbed and leak-scanned per the campaign deposit discipline. · PAPER43_IP_COORDINATION_AUDIT.md — the pre-deposit IP audit (no new algorithmic subject matter; all referenced methods filed; the egress rule anticipated and not claimed). Cite as R. J. Ferlic and K. K. Ferlic, "Non-invertible but not anonymous: a privacy characterization of a deterministic edge decision token, and a decision-bottleneck egress rule," Zenodo, 2026, doi: 10.5281/zenodo.22819210. License and patent notice Released under the Creative Commons Attribution 4.0 International License (CC-BY 4.0). Consistent with that license, no patent, patent application, or other intellectual-property right of the authors is licensed, waived, granted, or otherwise conveyed by this deposit. This work characterizes the privacy behavior of previously described methods and discloses no new algorithmic subject matter; the decision-bottleneck egress rule is a data-minimization design principle that is deliberately not claimed (it is anticipated by prior art). The methods characterized — the class-discriminant single-token codebook encoder and its nearest-centroid novelty monitor, and the privacy operations referenced in the manuscript (identity-axis projection, differential-privacy composition, keyed access) — are the subject of filed and pending U.S. patent applications held by the authors, including U.S. Provisional Application No. 64/095,354 (the encoder), U.S. Non-Provisional Application No. 19/691,234 (privacy-preserving operations on encoded states), and U.S. Provisional Applications Nos. 64/084,821, 64/084,817, and 64/127,391. Per-deployment productization and deployment-selection know-how are not disclosed and are retained as trade secrets. © 2026 Fieldstone Analytics, LLC and the authors; all rights not expressly granted under CC-BY 4.0 are reserved. Licensing and collaboration inquiries: [email protected]. Companion deposits (spiral-domain-encoder-campaign) · Class-discriminant codebook construction for single-token signal compression: doi:10.5281/zenodo.20788187 · Deterministic multi-token token ladder for channel-partition compression: doi:10.5281/zenodo.22003179 · Label-free inference-time channel fusion for drift-robust single-token decisions: doi:10.5281/zenodo.22046713 · A decision-oriented token as a bounded, threshold-free cache key for generative edge outputs: doi:10.5281/zenodo.22148612 · The predictive reach of a decision token (forecasting/anticipation/fusion): doi:10.5281/zenodo.22736921 Keywords edge AI privacy; re-identification; pseudonymity; non-invertibility; membership inference; attribute inference; cross-session linkage; data minimization; information bottleneck; decision token; class-discriminant codebook; electrocardiogram biometrics; device fingerprinting; differential privacy; cancelable biometrics; biometric template protection; unlinkability; vector quantization; wearable sensors; edge computing; TinyML; ECG; de-identification; GDPR; HIPAA; pre-registration; honest negatives

Zenodo (CERN European Organization for Nuclear Research)
EP Analytics (United States) (US)
Reduced inequalities, Peace, Justice and strong institutions
Wireless Communication Security Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.