The Price of the Bottleneck: A Pre-Registered, Multi-Domain Deployment Characterization of a Deterministic Edge Decision Token

{"The":[0,237,244,547,601,661,709,902,993,1035,1044,1324],"Price":[1],"of":[2,10,49,149,168,393,438,476,657,904,945,1026,1250,1258,1295,1314,1356],"the":[3,62,80,85,88,202,219,359,394,409,425,430,448,523,535,551,556,678,698,717,723,747,783,797,808,847,880,899,905,912,946,960,974,987,999,1015,1051,1055,1057,1071,1079,1090,1101,1113,1122,1126,1131,1139,1145,1148,1151,1158,1166,1173,1184,1194,1198,1202,1226,1235,1251,1274,1296,1311,1328,1337,1342,1346,1354,1365,1375],"Bottleneck:":[4],"A":[5,489,727,736,766,836],"Pre-Registered,":[6],"Multi-Domain":[7],"Deployment":[8],"Characterization":[9],"a":[11,47,50,56,64,108,143,153,240,254,294,326,346,368,390,404,453,474,493,509,537,541,568,597,616,644,671,704,732,763,776,815,820,824,828,833,885,889,917,942,955,983,1011,1039,1207,1221,1253,1259],"Deterministic":[12],"Edge":[13],"Decision":[14],"Token":[15],"Randolph":[16],"James":[17],"Ferlic,":[18,1247],"M.D.":[19],"and":[20,73,77,87,217,249,267,274,319,340,397,440,592,627,643,695,714,813,861,888,896,908,936,973,980,1028,1096,1130,1137,1172,1180,1206,1220,1232,1244,1269,1317,1333,1345,1358],"Kimberly":[21],"Kate":[22],"Ferlic":[23,1243],"(Fieldstone":[24],"Analytics,":[25],"LLC,":[26],"Austin,":[27],"TX,":[28],"USA)":[29],"Preprint":[30],"·":[31,35,38,488,546,600,660,726,826,901,992,1043,1110,1142],"Zenodo":[32],"DOI:":[33],"10.5281/zenodo.22838118":[34],"CC-BY":[36],"4.0":[37,1278],"Community:":[39],"spiral-domain-encoder-campaign":[40],"Abstract":[41],"Compact,":[42],"deterministic":[43,1260],"\\"decision":[44],"tokens\\"":[45],"—":[46,68,102,138,180,195,351,375,383,446,496,555,608,664,755,819,846,879,907,911,964,979,997,1050,1112,1144,1176,1327,1352],"window":[48],"sensor":[51],"stream":[52,1013],"quantized":[53],"on-device":[54,864,895],"to":[55,316,362,386,624,631,794,801,1062,1076],"single":[57],"~8-bit":[58],"code":[59],"that":[60,462,520,647,796,1285],"carries":[61],"byte":[63],"downstream":[65],"decision":[66,89,251,565,679,718,865,927,1262],"needs":[67,539],"are":[69,467,553,1211,1224,1230,1353],"attractive":[70],"for":[71,93],"edge":[72,1261],"ambient":[74],"intelligence:":[75],"bandwidth":[76],"energy":[78],"collapse,":[79],"raw":[81,1214],"signal":[82],"never":[83],"leaves":[84],"device,":[86,110,264],"is":[90,96,121,135,139,239,246,253,260,280,306,314,344,366,399,452,472,527,530,532,559,567,574,667,680,700,719,884,915,950,982,1038,1059,1074,1082,1094,1098,1217,1298],"auditable.":[91],"Accuracy":[92],"such":[94],"encoders":[95],"routinely":[97],"reported;":[98],"their":[99],"deployment":[100,147,178,422,450,490,499,533,886,995,1120,1128,1256,1312],"behavior":[101,1313],"will":[103],"it":[104,118,122,125,133,136,259,289,305,313,323,329,352,365,398,480,526,573,1104],"hold":[105],"up":[106],"on":[107,298,333,342,436,590,615,675,692,707,986,1010,1024,1066],"different":[109],"site,":[111,265],"population,":[112],"or":[113,1291,1302],"over":[114,408,959],"time;":[115],"when":[116,716],"should":[117],"be":[119,802],"trusted;":[120],"fair;":[123],"does":[124,132,930],"survive":[126],"corrupted":[127],"signal;":[128],"how":[129],"much":[130],"data":[131,1216],"need;":[134],"reproducible":[137,248,561],"not.":[140],"We":[141,175,357,787],"give":[142],"pre-registered,":[144,1254],"adversarial,":[145],"multi-domain":[146,1255],"characterization":[148,475,1257],"one":[150],"concrete":[151],"instance:":[152],"frozen,":[154],"deterministic,":[155],"class-discriminant":[156,1329],"single-token":[157,1330],"encoder":[158,1200,1332],"(features":[159],"→":[160,165,172,1005,1008,1020,1086],"Fisher-discriminant":[161],"⊕":[162],"PCA":[163],"subspace":[164],"k-means":[166],"codebook":[167,748,1331],"≤":[169,255,569,588],"256":[170],"cells":[171,411,963],"nearest-centroid":[173,1335],"decision).":[174],"test":[176,768],"six":[177,497,1155],"axes":[179,221,500],"drift,":[181],"calibration":[182],"/":[183,186,578,580,582,1348,1377],"out-of-distribution":[184],"(OOD)":[185],"abstention,":[187],"subgroup":[188],"fairness,":[189,503],"corruption":[190,355,650],"robustness,":[191],"sample":[192],"efficiency,":[193,505],"reproducibility":[194,1092,1146],"each":[196,507],"with":[197,284,309,508,605,619,682,721,752,969,1115,1284],"frozen":[198,510,557,1149],"Go/No-Go":[199,511],"reads,":[200],"stress":[201,1164],"findings":[203],"through":[204],"four":[205,514,640,1116,1159],"hardening":[206,466,518,1052,1160],"passes":[207],"(strong":[208],"baselines,":[209],"adversarial":[210,378,465,515],"evasion,":[211],"statistical":[212,517],"confidence":[213,285,338,606],"intervals,":[214,286],"honest":[215,242,449,602,1045],"re-evaluation),":[216],"replicate":[218],"core":[220],"across":[222,262,272,335,576,638,750,811],"five":[223],"application":[224],"domains":[225],"(healthcare":[226],"ECG,":[227,676],"industrial":[228],"bearing":[229],"vibration,":[230],"surgical-robot":[231],"kinematics,":[232],"wearable":[233],"EMG,":[234],"financial":[235,696],"volatility).":[236],"result":[238,1196],"sharp,":[241],"map.":[243],"token":[245,417,558,699,977,1058],"bit-exact":[247,560],"its":[250,363,563,909,1334],"function":[252,566],"256-entry":[256,570],"auditable":[257,571],"table;":[258],"stable":[261,575],"acquisition":[263],"era,":[266],"patient":[268,683],"demographics,":[269,583],"discriminatively":[270,584],"fair":[271,585],"sex/age/device,":[273],"in-distribution":[275,593],"calibrated.":[276],"Its":[277],"operating":[278],"envelope":[279],"equally":[281],"specific":[282],"and,":[283],"statistically":[287,701],"characterized:":[288],"drifts":[290],"significantly":[291],"worse":[292,761],"than":[293,670,762],"modern":[295,672,705,764,784,1064],"model":[296,706],"only":[297,651,674,715,771,898],"unseen":[299],"individuals":[300],"in":[301,1225,1341],"identity-entangled":[302],"tasks":[303,697],"(ECG);":[304],"more":[307,609],"label-hungry":[308,610],"no":[310,482,620,753,970,1213,1287,1319],"few-shot/transfer":[311],"shortcut;":[312,622],"brittle":[315,623],"heavy":[317,625],"noise":[318,626],"missing":[320,628],"input":[321],"channels;":[322,629],"cannot":[324],"recognize":[325],"novel":[327,444,632,1032],"class":[328],"was":[330],"not":[331,345,367,457,492,543,823,916,931,1060,1088],"trained":[332],"(universal":[334],"every":[336],"free":[337,641],"signal);":[339],"\\"gate":[341],"novelty\\"":[343],"deployable":[347],"general":[348],"OOD":[349,653],"gate":[350,646,1004,1073],"filters":[353],"gross":[354,649,1077],"only.":[356],"trace":[358],"cross-individual":[360,381,665,913,937],"drift":[361,413,666,686,856,914,966,1124],"mechanism:":[364],"removable":[369,918],"identity":[370,919],"representation":[371,920],"(three":[372],"principled":[373],"front-ends":[374],"identity-axis":[376],"projection,":[377],"patient-invariant":[379],"encoding,":[380],"augmentation":[382,938],"all":[384,639],"fail":[385],"fix":[387],"it)":[388],"but":[389,434,868,929,941,1022],"distributional":[391,943],"limit":[392],"discrete":[395,947],"codebook,":[396],"largely":[400,951],"recoverable":[401,952],"post-hoc":[402,953],"by":[403,954,1305,1364],"soft,":[405],"distance-weighted":[406,956],"readout":[407,958,1379],"nearest":[410,962],"(inter-patient":[412],"falls":[414,967],"~72%,":[415,968],"label-free,":[416],"unchanged).":[418],"Finally,":[419],"an":[420,458,790,803],"integrated":[421,994,1127],"simulation":[423],"shows":[424,746],"recommended":[426,1001],"confidence-gated":[427],"pipeline":[428,810,1002],"cuts":[429,853,1014],"confident-wrong":[431,1016],"rate":[432,1017],"~4–6×":[433,1018],"abstains":[435,1023],"~60%":[437,1025],"inputs":[439,1027],"still":[441,1029],"mis-decides":[442],"genuinely":[443,1031],"classes":[445,633,1033],"so":[447,892],"posture":[451,1037],"high-abstention,":[454,1040],"human-in-the-loop":[455,1041],"assistant,":[456],"autonomous":[459],"decider.":[460],"Results":[461],"changed":[463],"under":[464,1273],"reported":[468],"as":[469,1100,1240],"corrections.":[470],"This":[471,1308],"explicitly":[473],"previously-described,":[477],"filed":[478,848,1357],"methods;":[479],"discloses":[481,1318],"new":[483,1320],"algorithmic":[484,1321],"subject":[485,1322,1355],"matter.":[486,1323],"Highlights":[487],"map,":[491],"benchmark":[494,1215],"score":[495],"pre-registered":[498],"(drift,":[501],"calibration/OOD/abstention,":[502],"corruption,":[504,1078],"reproducibility),":[506],"read,":[512],"plus":[513],"+":[516,731,1135,1162,1169],"tiers":[519,1053,1161],"deliberately":[521,1048],"attack":[522],"paper":[524],"before":[525,540],"written.":[528],"What":[529,1106],"measured":[531],"behavior,":[534],"thing":[536],"partner":[538],"pilot,":[542],"accuracy":[544],"alone.":[545],"strengths,":[548],"believable":[549],"because":[550],"limits":[552],"named":[554],"(float32),":[562],"whole":[564],"table,":[572],"device":[577],"site":[579],"era":[581],"(AUROC":[586],"gap":[587,1081,1156],"0.05":[589],"sex/age/device),":[591],"calibrated":[594],"(ECE":[595],"≈":[596],"logistic":[598],"regression).":[599],"limits,":[603],"sized":[604,1083],"intervals":[607],"(~512":[611],"vs":[612,782],"~32":[613],"labels":[614,972],"hard":[617],"task)":[618],"transfer":[621,1068,1186],"blind":[630],"(novelty":[634,1003],"AUROC":[635,757],"0.57,":[636],"universal":[637],"signals);":[642],"novelty":[645,1072],"catches":[648],"(90%":[652],"recall":[654],"costs":[655],"~70%":[656],"clean":[658],"inputs).":[659],"identity-entanglement":[662],"through-line":[663],"significant":[668],"(worse":[669],"model)":[673],"where":[677],"entangled":[681,720],"morphology":[684],"(token−RF":[685],"CI":[687],"[0.016,":[688],"0.325],":[689],"excluding":[690],"0);":[691],"robotics,":[693],"wearable,":[694],"indistinguishable":[702],"from":[703,843,1192],"drift.":[708],"biggest":[710],"limitation":[711],"bites":[712],"precisely":[713],"who":[722],"individual":[724],"is.":[725,1105],"modest":[728],"cross-database":[729,1133,1185],"tax":[730],"preprocessing-mismatch":[733,778],"sensitivity":[734,779],"(corrected).":[735],"wide":[737,1132],"four-site":[738],"matrix":[739,799,1134],"(PTB-XL/DE,":[740],"CPSC/CN,":[741],"Georgia/US,":[742],"Chapman/CN,":[743],"uniform":[744],"preprocessing)":[745],"transfers":[749],"sites":[751,812],"collapse":[754],"cross-site":[756],"median":[758],"0.735,":[759],"~0.05":[760],"model.":[765],"controlled":[767,798],"(same":[769],"records,":[770,1197],"sampling-rate":[772],"handling":[773],"varied)":[774],"isolates":[775],"token-specific":[777],"(0.044":[780],"drop":[781],"model's":[785],"~0).":[786],"retract,":[788],"verbatim,":[789],"earlier":[791],"two-site":[792],"“collapse":[793],"0.539”":[795],"showed":[800],"unrepresentative":[804],"outlier.":[805],"Rule:":[806],"harmonize/pin":[807],"preprocessing":[809],"budget":[814],"small":[816,837],"per-site":[817],"recalibration":[818],"manageable":[821],"tax,":[822],"barrier.":[825],"Co-channels:":[827],"second":[829],"post-bottleneck":[830,1123,1181],"mitigation,":[831],"at":[832,859,925],"privacy":[834,877,890],"cost.":[835],"co-channel":[838,871],"ensemble":[839],"(K":[840],"parallel":[841],"tokens":[842,872],"feature":[844],"partitions":[845],"channel-partition":[849,882],"ladder,":[850],"averaged":[851],"post-bottleneck)":[852],"MIT-BIH":[854,1097],"inter-patient":[855,965],"~half":[857],"(+0.116→+0.057":[858],"K=2)":[860],"recovers":[862],"int8":[863,1091],"agreement":[866],"0.65→0.98;":[867],"emitting":[869],"those":[870],"sharply":[873],"multiplies":[874],"re-identifiability":[875],"(companion":[876],"paper)":[878],"same":[881],"knob":[883,985],"lever":[887],"risk,":[891],"compute":[893],"co-channels":[894],"egress":[897],"decision.":[900],"price":[903,1249],"bottleneck":[906],"refund":[910],"(identity-axis":[921],"projection":[922],"cleanly":[923],"de-identifies":[924],"zero":[926],"cost":[928],"reduce":[932],"drift;":[933],"domain-adversarial":[934],"invariance":[935],"also":[939],"fail)":[940],"property":[944],"codebook.":[948],"It":[949],"soft":[957,1182],"m":[961,981],"target":[971],"emitted":[975],"8-bit":[976],"unchanged":[978],"tunable":[984],"purity":[988],"↔":[989,1189],"generalization":[990],"tradeoff.":[991],"number":[996],"running":[998],"full":[1000],"margin":[1006],"abstention":[1007],"decide)":[1009],"mixed":[1012],"(0.32":[1019],"0.07),":[1021],"lets":[1030],"through.":[1034],"safe":[1036],"assistant.":[1042],"bound,":[1046],"held":[1047,1363],"conservative":[1049],"set":[1054],"envelope:":[1056],"claimed":[1061],"beat":[1063],"models":[1065],"cross-device":[1067],"(not":[1069],"significant),":[1070],"scoped":[1075],"fairness":[1080],"out-of-sample":[1084],"(0.23":[1085],"0.19,":[1087],"0.03),":[1089],"claim":[1093],"qualified,":[1095],"treated":[1099],"single-feature":[1102],"task":[1103],"this":[1107,1306],"record":[1108],"contains":[1109],"Manuscript_Paper44.pdf":[1111],"manuscript":[1114],"figures":[1117],"embedded":[1118],"(the":[1119,1154,1373],"scorecard,":[1121,1205],"fix,":[1125],"tradeoff,":[1129],"preprocessing-isolation),":[1136],"Manuscript_Paper44.docx,":[1138],"editable":[1140],"source.":[1141],"PAPER_44_ZENODO_ARCHIVE.zip":[1143],"archive:":[1147],"pre-registrations,":[1150],"experiment":[1152],"runners":[1153],"studies,":[1157],"deploy":[1163],"test,":[1165],"five-domain":[1167],"replication":[1168],"bootstrap":[1170],"CIs,":[1171],"drift-mechanism":[1174,1343],"probes":[1175],"identity-suppression,":[1177],"domain-adversarial,":[1178],"scramble,":[1179],"readout),":[1183],"study":[1187],"(PTB-XL":[1188],"Chapman,":[1190],"streamed":[1191],"PhysioNet),":[1193],"per-experiment":[1195],"shared":[1199],"loader,":[1201],"consolidated":[1203],"summary":[1204],"README.":[1208],"All":[1209,1228],"datasets":[1210],"public;":[1212],"redistributed":[1218],"(sources":[1219],"`PATH_TO_DATA`":[1222],"convention":[1223],"README).":[1227],"paths/identifiers":[1229],"scrubbed":[1231],"leak-scanned":[1233],"per":[1234],"campaign":[1236],"deposit":[1237],"discipline.":[1238],"Cite":[1239],"R.":[1241],"J.":[1242],"K.":[1245,1246],"\\"The":[1248],"bottleneck:":[1252],"token,\\"":[1263],"Zenodo,":[1264],"2026,":[1265],"doi:":[1266],"10.5281/zenodo.22838118.":[1267],"License":[1268,1280],"patent":[1270,1289,1361],"notice":[1271],"Released":[1272],"Creative":[1275],"Commons":[1276],"Attribution":[1277],"International":[1279],"(CC-BY":[1281],"4.0).":[1282],"Consistent":[1283],"license,":[1286],"patent,":[1288],"application,":[1290],"other":[1292],"intellectual-property":[1293],"right":[1294],"authors":[1297],"licensed,":[1299],"waived,":[1300],"granted,":[1301],"otherwise":[1303],"conveyed":[1304],"deposit.":[1307],"work":[1309],"characterizes":[1310],"previously-described":[1315],"methods":[1316,1325],"characterized":[1326],"monitor,":[1336],"identity-axis-projection":[1338],"front-end":[1339],"evaluated":[1340],"study,":[1344],"multi-token":[1347,1376],"soft-readout":[1349],"mechanisms":[1350],"referenced":[1351],"pending":[1359],"U.S.":[1360,1368],"applications":[1362],"authors,":[1366],"including":[1367],"Provisional":[1369],"Application":[1370],"No.":[1371],"64/095,354":[1372],"encoder),":[1374],"token-ladder":[1378],"ap":[1380]}

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-09-18
DOI
https://doi.org/10.5281/zenodo.22838117
Primary Topic
Adversarial Robustness in Machine Learning
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
preprint

The Price of the Bottleneck: A Pre-Registered, Multi-Domain Deployment Characterization of a Deterministic Edge Decision Token

Randolph James Ferlic, Kimberly Kate Ferlic
Zenodo (CERN European Organization for Nuclear Research)
Adversarial Robustness in Machine Learning
preprint

The Price of the Bottleneck: A Pre-Registered, Multi-Domain Deployment Characterization of a Deterministic Edge Decision Token

Randolph James Ferlic, Kimberly Kate Ferlic
preprint en

Abstract

The Price of the Bottleneck: A Pre-Registered, Multi-Domain Deployment Characterization of a Deterministic Edge Decision Token Randolph James Ferlic, M.D. and Kimberly Kate Ferlic (Fieldstone Analytics, LLC, Austin, TX, USA) Preprint · Zenodo DOI: 10.5281/zenodo.22838118 · CC-BY 4.0 · Community: spiral-domain-encoder-campaign Abstract Compact, deterministic "decision tokens" — a window of a sensor stream quantized on-device to a single ~8-bit code that carries the byte a downstream decision needs — are attractive for edge and ambient intelligence: bandwidth and energy collapse, the raw signal never leaves the device, and the decision is auditable. Accuracy for such encoders is routinely reported; their deployment behavior — will it hold up on a different device, site, population, or over time; when should it be trusted; is it fair; does it survive corrupted signal; how much data does it need; is it reproducible — is not. We give a pre-registered, adversarial, multi-domain deployment characterization of one concrete instance: a frozen, deterministic, class-discriminant single-token encoder (features → Fisher-discriminant ⊕ PCA subspace → k-means codebook of ≤ 256 cells → nearest-centroid decision). We test six deployment axes — drift, calibration / out-of-distribution (OOD) / abstention, subgroup fairness, corruption robustness, sample efficiency, reproducibility — each with frozen Go/No-Go reads, stress the findings through four hardening passes (strong baselines, adversarial evasion, statistical confidence intervals, honest re-evaluation), and replicate the core axes across five application domains (healthcare ECG, industrial bearing vibration, surgical-robot kinematics, wearable EMG, financial volatility). The result is a sharp, honest map. The token is bit-exact reproducible and its decision function is a ≤ 256-entry auditable table; it is stable across acquisition device, site, era, and patient demographics, discriminatively fair across sex/age/device, and in-distribution calibrated. Its operating envelope is equally specific and, with confidence intervals, statistically characterized: it drifts significantly worse than a modern model only on unseen individuals in identity-entangled tasks (ECG); it is more label-hungry with no few-shot/transfer shortcut; it is brittle to heavy noise and missing input channels; it cannot recognize a novel class it was not trained on (universal across every free confidence signal); and "gate on novelty" is not a deployable general OOD gate — it filters gross corruption only. We trace the cross-individual drift to its mechanism: it is not a removable identity representation (three principled front-ends — identity-axis projection, adversarial patient-invariant encoding, cross-individual augmentation — all fail to fix it) but a distributional limit of the discrete codebook, and it is largely recoverable post-hoc by a soft, distance-weighted readout over the nearest cells (inter-patient drift falls ~72%, label-free, token unchanged). Finally, an integrated deployment simulation shows the recommended confidence-gated pipeline cuts the confident-wrong rate ~4–6× but abstains on ~60% of inputs and still mis-decides genuinely novel classes — so the honest deployment posture is a high-abstention, human-in-the-loop assistant, not an autonomous decider. Results that changed under adversarial hardening are reported as corrections. This is explicitly a characterization of previously-described, filed methods; it discloses no new algorithmic subject matter. Highlights · A deployment map, not a benchmark score — six pre-registered deployment axes (drift, calibration/OOD/abstention, fairness, corruption, efficiency, reproducibility), each with a frozen Go/No-Go read, plus four adversarial + statistical hardening tiers that deliberately attack the paper before it is written. What is measured is deployment behavior, the thing a partner needs before a pilot, not accuracy alone. · The strengths, believable because the limits are named — the frozen token is bit-exact reproducible (float32), its whole decision function is a ≤ 256-entry auditable table, it is stable across device / site / era / demographics, discriminatively fair (AUROC gap ≤ 0.05 on sex/age/device), and in-distribution calibrated (ECE ≈ a logistic regression). · The honest limits, sized with confidence intervals — more label-hungry (~512 vs ~32 labels on a hard task) with no transfer shortcut; brittle to heavy noise and missing channels; blind to novel classes (novelty AUROC 0.57, universal across all four free signals); and a novelty gate that catches gross corruption only (90% OOD recall costs ~70% of clean inputs). · The identity-entanglement through-line — cross-individual drift is significant (worse than a modern model) only on ECG, where the decision is entangled with patient morphology (token−RF drift CI [0.016, 0.325], excluding 0); on robotics, wearable, and financial tasks the token is statistically indistinguishable from a modern model on drift. The biggest limitation bites precisely and only when the decision is entangled with who the individual is. · A modest cross-database tax + a preprocessing-mismatch sensitivity (corrected). A wide four-site matrix (PTB-XL/DE, CPSC/CN, Georgia/US, Chapman/CN, uniform preprocessing) shows the codebook transfers across sites with no collapse — cross-site AUROC median 0.735, ~0.05 worse than a modern model. A controlled test (same records, only sampling-rate handling varied) isolates a token-specific preprocessing-mismatch sensitivity (0.044 drop vs the modern model's ~0). We retract, verbatim, an earlier two-site “collapse to 0.539” that the controlled matrix showed to be an unrepresentative outlier. Rule: harmonize/pin the preprocessing pipeline across sites and budget a small per-site recalibration — a manageable tax, not a barrier. · Co-channels: a second post-bottleneck mitigation, at a privacy cost. A small co-channel ensemble (K parallel tokens from feature partitions — the filed channel-partition ladder, averaged post-bottleneck) cuts MIT-BIH inter-patient drift ~half (+0.116→+0.057 at K=2) and recovers int8 on-device decision agreement 0.65→0.98; but emitting those co-channel tokens sharply multiplies re-identifiability (companion privacy paper) — the same channel-partition knob is a deployment lever and a privacy risk, so compute co-channels on-device and egress only the decision. · The price of the bottleneck — and its refund — the cross-individual drift is not a removable identity representation (identity-axis projection cleanly de-identifies at zero decision cost but does not reduce drift; domain-adversarial invariance and cross-individual augmentation also fail) but a distributional property of the discrete codebook. It is largely recoverable post-hoc by a distance-weighted soft readout over the m nearest cells — inter-patient drift falls ~72%, with no target labels and the emitted 8-bit token unchanged — and m is a tunable knob on the purity ↔ generalization tradeoff. · The integrated deployment number — running the full recommended pipeline (novelty gate → margin abstention → decide) on a mixed stream cuts the confident-wrong rate ~4–6× (0.32 → 0.07), but abstains on ~60% of inputs and still lets genuinely novel classes through. The safe posture is a high-abstention, human-in-the-loop assistant. · The honest bound, held deliberately conservative — the hardening tiers set the envelope: the token is not claimed to beat modern models on cross-device transfer (not significant), the novelty gate is scoped to gross corruption, the fairness gap is sized out-of-sample (0.23 → 0.19, not 0.03), the int8 reproducibility claim is qualified, and MIT-BIH is treated as the single-feature task it is. What this record contains · Manuscript_Paper44.pdf — the manuscript with four figures embedded (the deployment scorecard, the post-bottleneck drift fix, the integrated deployment tradeoff, and the wide cross-database matrix + preprocessing-isolation), and Manuscript_Paper44.docx, the editable source. · PAPER_44_ZENODO_ARCHIVE.zip — the reproducibility archive: the frozen pre-registrations, the experiment runners (the six gap studies, the four hardening tiers + deploy stress test, the five-domain replication + bootstrap CIs, and the drift-mechanism probes — identity-suppression, domain-adversarial, scramble, and post-bottleneck soft readout), the cross-database transfer study (PTB-XL ↔ Chapman, streamed from PhysioNet), the per-experiment result records, the shared encoder loader, the consolidated summary scorecard, and a README. All datasets are public; no raw benchmark data is redistributed (sources and a `PATH_TO_DATA` convention are in the README). All paths/identifiers are scrubbed and leak-scanned per the campaign deposit discipline. Cite as R. J. Ferlic and K. K. Ferlic, "The price of the bottleneck: a pre-registered, multi-domain deployment characterization of a deterministic edge decision token," Zenodo, 2026, doi: 10.5281/zenodo.22838118. 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 deployment behavior of previously-described methods and discloses no new algorithmic subject matter. The methods characterized — the class-discriminant single-token codebook encoder and its nearest-centroid monitor, the identity-axis-projection front-end evaluated in the drift-mechanism study, and the multi-token / soft-readout mechanisms referenced — 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), the multi-token / token-ladder readout ap

Zenodo (CERN European Organization for Nuclear Research)
EP Analytics (United States) (US)
Peace, Justice and strong institutions
Adversarial Robustness in Machine Learning
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.