Paying Down the Price of the Bottleneck: Label-Efficient Foundation Codebooks and a Soft Readout for a Deterministic Edge Decision Token

{"Paying":[0],"Down":[1],"the":[2,5,113,126,130,194,300,305,312,316,321,330,359,373,388,391,397,410,426,434,451,461,468,494,505,532,612,617,640,674,681,697,704,712,747,752,760,781,789,796,808,814,820,837,843,854,864,875,902,906,910,920,929,933,936,943,949,955,958,962,968,973,978,985,992,1004,1008,1011,1016,1019,1042,1053,1068,1071,1096,1118,1167,1176,1187,1198,1209,1220,1246,1258,1295],"Price":[3],"of":[4,66,329,341,565,711,1070,1117,1200,1245,1367],"Bottleneck:":[6],"Label-Efficient":[7],"Foundation":[8],"Codebooks":[9],"and":[10,24,92,136,158,176,187,231,239,263,267,292,320,334,361,365,501,510,624,657,679,733,759,860,871,942,947,975,988,991,1022,1036,1046,1050,1062,1076,1091,1135,1145,1152,1172,1186,1202,1219,1225,1233,1262,1266,1277,1283,1294,1308],"a":[11,15,51,89,95,106,177,183,218,240,260,349,393,400,422,445,519,523,552,556,563,595,607,650,658,686,766,777,883,889,893,1023,1037,1077,1081,1193,1243,1251,1353,1368,1378],"Soft":[12],"Readout":[13],"for":[14,77,1080,1273,1319,1329,1343,1358],"Deterministic":[16,1325],"Edge":[17],"Decision":[18],"Token":[19],"Randolph":[20],"James":[21],"Ferlic,":[22,1065],"M.D.":[23],"Kimberly":[25],"Kate":[26],"Ferlic":[27,1061],"(Fieldstone":[28],"Analytics,":[29,1292],"LLC,":[30],"Austin,":[31],"TX,":[32],"USA)":[33],"Preprint":[34],"·":[35,39,42,577,622,689,755,798,846,917,952,1315,1324,1338,1348,1363,1373],"Zenodo":[36],"DOI:":[37],"10.5281/zenodo.22866039":[38],"CC-BY":[40,1303],"4.0":[41,1100,1304],"Community:":[43],"spiral-domain-encoder-campaign":[44],"Abstract":[45],"A":[46,170,295,626,701,1349],"companion":[47],"characterization":[48,564,1244],"established":[49,1156],"that":[50,244,352,1107],"frozen,":[52],"deterministic,":[53],"class-discriminant":[54,1168],"single-token":[55,1169,1320,1345],"encoder":[56,1171],"(features":[57],"→":[58,63,70,475,480,826],"Fisher-discriminant":[59],"⊕":[60],"PCA":[61],"subspace":[62],"k-means":[64],"codebook":[65,172,220,550,629,652,1170,1248,1317],"≤":[67,131,195,301,524,618,813,894],"256":[68],"cells":[69,308,432,707],"nearest-centroid":[71,614,1174],"decision)":[72],"pays":[73],"two":[74],"specific":[75],"bills":[76],"its":[78,210,598,636,1173,1234],"byte-scale,":[79],"auditable,":[80],"sub-milliwatt":[81],"bottleneck:":[82,1072],"it":[83,93,251,289,338,514,569,672,717,792,1255],"is":[84,103,208,221,399,443,561,590,601,634,684,693,749,765,788,795,842,1033,1120,1240,1271],"label-hungry":[85],"(a":[86,277],"cost-setter,":[87],"not":[88,274,355,421,512,685,699,776,802,862,1281,1299,1376],"few-shot":[90,200,242,660,677],"learner)":[91],"carries":[94],"large-N":[96],"accuracy":[97,599,691,773],"tax":[98,600,692,748],"(its":[99],"hard,":[100],"piecewise-constant":[101],"readout":[102,297,375,428,783,879,1079,1224],"coarser":[104],"than":[105,662],"continuous":[107,154,331,520,713,890],"head).":[108],"This":[109,560,1130],"paper":[110],"asks,":[111],"with":[112,162,358,793,819,922,1106],"same":[114,452,506,855],"pre-registration":[115],"discipline,":[116],"whether":[117],"either":[118],"bill":[119,442],"can":[120,283],"be":[121],"paid":[122,580,591,602],"down":[123,581,592,603,1067],"without":[124],"changing":[125],"deterministic":[127,613,1082],"runtime":[128,585,605],"or":[129,336,459,616,881,1113,1124],"256-entry":[132,196,302,619],"auditable":[133,620],"table":[134,180,197,303,323,633],"—":[135,173,207,298,326,409,425,499,664,695,746,780,870,887,919,954,1166,1196,1250,1265],"at":[137,213,222,226,248,269,345,376,381,414,458,485,515,528,545,604,635,645,653,722,738,830,867,872,898],"what":[138],"honest":[139,490,761,848],"cost,":[140],"across":[141,234,253,419,666],"eight":[142],"tasks":[143],"(ECG,":[144],"industrial":[145],"bearing":[146,483],"vibration,":[147,484],"surgical-robot":[148],"kinematics,":[149],"wearable":[150],"EMG,":[151],"financial":[152],"volatility,":[153],"glucose":[155,829],"monitoring;":[156],"binary":[157],"multiclass;":[159],"group-disjoint":[160],"splits":[161],"bootstrap":[163],"confidence":[164],"intervals).":[165],"Two":[166,578],"levers":[167,537],"work.":[168],"(1)":[169],"foundation":[171,360,596,628,877,987,994,1074],"centroids,":[174],"subspace,":[175],"pre-calibrated":[178],"per-cell":[179,553],"frozen":[181,627,876,960],"on":[182,198,280,339,390,450,477,482,504,531,718,828,853,901],"large":[184],"labeled":[185],"population":[186,211,546,637],"deployed":[188],"to":[189,363,370,416,457,731,751,812],"new":[190,572,1138,1268],"subjects,":[191],"refitting":[192],"only":[193,1160],"N":[199,214,227,453,486,530,646,831,874,900],"labels":[201,454,642],"under":[202,1095,1302],"an":[203],"empirical-Bayes":[204,631,1153],"shrinkage":[205,632,998,1154],"rule":[206],"near":[209],"ceiling":[212,333,491,638,715],"≈":[215,347,724],"0":[216],"where":[217],"from-scratch":[219],"chance":[223],"(+0.33–0.36":[224,643],"AUROC":[225,526,644,794,896],"=":[228,383,487,647,740,832],"8),":[229],"monotone":[230],"never-worse,":[232],"decision-shift-robust":[233],"device/site/era/age":[235],"(gaps":[236],"<":[237],"0.025),":[238],"safe":[241,844],"recalibrator":[243,661],"beats":[245,882],"Platt":[246,663,834,1146],"scaling":[247,835],"small":[249,270,838,873],"N;":[250],"replicates":[252],"all":[254,324,1297],"five":[255,667,923,1017],"sensor":[256,668],"domains.":[257,669],"It":[258],"matches":[259,335,673,880],"nearest-class-mean/prototype":[261],"head":[262,678],"tops":[264,680],"logistic":[265],"regression":[266],"k-nearest-neighbors":[268],"N,":[271],"but":[272,683,888,1375],"does":[273,354],"universally":[275],"win":[276],"linear":[278,332,495,714,865],"model":[279,521,886,891],"linearly-separable":[281],"features":[282,507,856],"beat":[284,513,863],"it),":[285],"so":[286],"we":[287],"position":[288],"as":[290,540,1058,1161,1192,1242,1286,1352],"label-efficient":[291,1073],"auditable.":[293],"(2)":[294],"soft":[296,427,608,782,809,878,1078,1184],"averaging":[299,431],"over":[304,372,703],"m":[306,346,382,705,723,739],"nearest":[307,706],"(the":[309,926,965,1183,1217,1332],"filed":[310,567,1201],"distance-vote;":[311],"hard":[313,753],"cell":[314],"assignment,":[315],"emitted":[317],"8-bit":[318],"token,":[319],"static":[322],"unchanged)":[325],"recovers":[327,709,909],"79–92%":[328,710],"exceeds":[337],"2":[340],"5":[342],"domains,":[343],"knee":[344,721],"8,":[348,725],"weighting":[350,726],"scheme":[351,727],"provably":[353,470,728,822],"matter,":[356],"composing":[357],"generalizing":[362,730],"multiclass":[364,732],"every":[366],"modality":[367],"tested":[368],"(+0.10":[369],"+0.37":[371],"constant":[374,462,815],"deployable":[377,516,868],"budgets,":[378,517],"never":[379,736],"worse":[380],"8).":[384],"We":[385],"then":[386],"discipline":[387],"claim":[389],"metric":[392],"reviewer":[394],"would":[395],"attack:":[396],"recovery":[398,764],"genuine":[401],"discrimination":[402,768],"+":[403,630,769],"decision":[404,469,770,821,1084,1369],"gain":[405,412,424,771,779],"(balanced":[406,772],"accuracy,":[407],"macro-F1":[408],"ranking":[411],"translates":[413],"46%":[415],">":[417],"100%":[418],"tasks),":[420],"calibration":[423,456,778,786,811],"de-calibrates":[429],"because":[430],"pulls":[433],"score":[435],"off":[436],"each":[437],"cell's":[438],"empirical":[439],"frequency.":[440],"That":[441],"payable:":[444],"one-parameter":[446],"temperature":[447,805,841,1144],"recalibration":[448,806,945,1002],"fit":[449],"restores":[455,807],"below":[460],"readout's":[463,810,816],"native":[464,817],"level":[465,818],"while":[466],"leaving":[467],"unchanged":[471,823],"(soft-vote":[472],"ECE":[473],"0.36":[474,825],"0.05":[476,827],"glucose,":[478],"0.37":[479],"0.01":[481],"64).":[488,833],"An":[489,847],"check":[492],"shows":[493],"reference":[496,866],"was":[497],"fair":[498],"gradient-boosted":[500],"neural":[502],"models":[503,852],"are":[508,538,857,1027,1040,1048,1155,1197,1280,1284,1305],"few-shot-limited":[509,859],"do":[511,861],"though":[518],"retains":[522],"~0.12":[525,895],"residual":[527,897],"larger":[529,899],"hardest":[533,903],"tasks.":[534,904],"Four":[535,742],"alternative":[536],"reported":[539],"pre-registered":[541],"negatives":[542],"(teacher":[543],"distillation":[544,996],"scale,":[547],"upstream":[548],"supervised":[549,1188],"refinement,":[551],"piecewise-linear":[554],"tuner,":[555],"multi-token":[557,1177,1221,1326,1333],"\\"burst\\"":[558],"ensemble).":[559],"explicitly":[562],"previously-described,":[566],"methods;":[568],"discloses":[570,1136],"no":[571,1029,1109,1137,1267],"algorithmic":[573,1139],"subject":[574,1140,1199,1269],"matter.":[575],"Highlights":[576],"bills,":[579],"from":[582,639,696],"opposite":[583],"sides,":[584],"untouched.":[586],"The":[587,690,756,763,799,1163,1236,1364],"token's":[588],"label-hunger":[589],"offline":[593],"by":[594,606,1127,1208,1257],"codebook;":[597],"readout.":[609],"Neither":[610],"touches":[611],"computation":[615],"table.":[621],"Label-efficient":[623],"never-worse.":[625],"first":[641],"8":[648],"vs":[649,939],"fresh":[651],"chance),":[654],"monotone,":[655],"shift-robust,":[656],"safer":[659],"replicated":[665],"Bounded":[670],"honestly:":[671],"strongest":[675],"simple":[676],"rest,":[682],"universal":[687],"winner.":[688],"recoverable":[694],"readout,":[698],"construction.":[700],"distance-vote":[702],"(filed)":[708],"(matches/exceeds":[716],"2/5":[719],"domains),":[720],"irrelevant,":[729],"six":[734],"modalities,":[735],"hurting":[737],"8.":[741],"construction-side":[743],"alternatives":[744],"fail":[745],"intrinsic":[750],"collapse.":[754],"right":[757,1116],"metric,":[758],"cost.":[762],"real":[767],"/":[774,980,982,995,997,999,1001,1149,1178,1180,1222,1260,1334],"macro-F1),":[775],"de-calibrates.":[784],"Measuring":[785],"separately":[787],"discipline;":[790],"conflating":[791],"error.":[797],"recommendation,":[800],"demonstrated":[801],"asserted.":[803],"One-parameter":[804],"(ECE":[824],"overfits":[836],"label":[839],"set;":[840],"choice.":[845],"ceiling.":[849],"Strong":[850],"nonlinear":[851],"themselves":[858],"budgets":[869],"freshly-trained":[884],"strong":[885],"keeps":[892],"\\"Recovers":[905],"tax\\"":[907],"means":[908],"linear-ceiling":[911],"gap.":[912],"What":[913],"this":[914,1128],"record":[915],"contains":[916],"Manuscript_Paper46.pdf":[918],"manuscript":[921,1020],"figures":[924],"embedded":[925],"two-lever":[927],"scorecard,":[928],"five-domain":[930],"label-efficiency":[931],"comparison,":[932],"accuracy↔auditability":[934],"curve,":[935],"soft-vote":[937,966],"advantage":[938],"label-budget":[940,969],"surface,":[941,972],"post-hoc":[944],"panel),":[946],"Manuscript_Paper46.docx,":[948],"editable":[950],"source.":[951],"PAPER_46_ZENODO_ARCHIVE.zip":[953],"reproducibility":[956],"archive:":[957],"nine":[959],"pre-registrations,":[961],"experiment":[963],"runners":[964],"m-sweep,":[967],"×":[970],"cell-count":[971],"metric-hygiene":[974],"reviewer-proofing":[976],"rounds,":[977],"upstream-refinement":[979],"local-tuner":[981],"multi-token-ensemble":[983],"negatives,":[984],"multi-domain":[986],"reviewer-defense":[989],"batteries,":[990],"Modal-scale":[993],"shift":[1000],"runners),":[1003],"shared":[1005],"feature-extractor":[1006],"loader,":[1007],"figure":[1009],"builder,":[1010],"seventeen":[1012],"per-experiment":[1013],"result":[1014],"records,":[1015],"figures,":[1018],"source,":[1021],"README.":[1024],"All":[1025,1044],"datasets":[1026],"public;":[1028],"raw":[1030],"benchmark":[1031],"data":[1032],"redistributed":[1034],"(sources":[1035],"`PATH_TO_DATA`":[1038],"convention":[1039],"in":[1041],"README).":[1043],"paths":[1045],"identifiers":[1047],"scrubbed":[1049],"leak-scanned":[1051],"per":[1052],"campaign":[1054],"deposit":[1055],"discipline.":[1056],"Cite":[1057],"R.":[1059],"J.":[1060],"K.":[1063,1064],"\\"Paying":[1066],"price":[1069],"codebooks":[1075],"edge":[1083,1238,1360],"token,\\"":[1085],"Zenodo,":[1086],"2026,":[1087],"doi:":[1088],"10.5281/zenodo.22866039.":[1089],"License":[1090,1102],"patent":[1092,1111,1205],"notice":[1093],"Released":[1094],"Creative":[1097],"Commons":[1098],"Attribution":[1099],"International":[1101],"(CC-BY":[1103],"4.0).":[1104],"Consistent":[1105],"license,":[1108],"patent,":[1110],"application,":[1112],"other":[1114],"intellectual-property":[1115],"authors":[1119],"licensed,":[1121],"waived,":[1122],"granted,":[1123],"otherwise":[1125],"conveyed":[1126],"deposit.":[1129],"work":[1131],"characterizes":[1132],"previously-described":[1133],"methods":[1134,1164],"matter;":[1141],"distance-weighted":[1142],"voting,":[1143],"scaling,":[1147],"prototype":[1148,1259],"nearest-class-mean":[1150,1261],"classification,":[1151],"prior":[1157],"art,":[1158],"used":[1159],"tools.":[1162],"characterized":[1165],"monitor,":[1175],"token-ladder":[1179,1223],"soft-readout":[1181,1335],"mechanisms":[1182],"readout),":[1185],"codebook-refinement":[1189],"mechanism":[1190],"(evaluated":[1191],"negative":[1194],"result)":[1195],"pending":[1203],"U.S.":[1204,1212,1229],"applications":[1206,1227],"held":[1207],"authors,":[1210],"including":[1211],"Provisional":[1213],"Application":[1214,1230],"No.":[1215,1231],"64/095,354":[1216],"encoder)":[1218],"supervised-refinement":[1226],"(priority":[1228],"19/467,303":[1232],"continuations).":[1235],"foundation-codebook-with-shrinkage":[1237],"deployment":[1239],"published":[1241],"already-filed":[1247],"method":[1249],"prior-art":[1252],"review":[1253],"found":[1254],"anticipated":[1256],"training-free-calibration":[1263],"literatures":[1264],"matter":[1270],"claimed":[1272],"it.":[1274],"Per-deployment":[1275],"productization":[1276],"deployment-selection":[1278],"know-how":[1279],"disclosed":[1282],"retained":[1285],"trade":[1287],"secrets.":[1288],"©":[1289],"2026":[1290],"Fieldstone":[1291],"LLC":[1293],"authors;":[1296],"rights":[1298],"expressly":[1300],"granted":[1301],"reserved.":[1306],"Licensing":[1307],"collaboration":[1309],"inquiries:":[1310],"[email protected].":[1311],"Companion":[1312],"deposits":[1313],"(spiral-domain-encoder-campaign)":[1314],"Class-discriminant":[1316],"construction":[1318],"signal":[1321],"compression:":[1322],"doi:10.5281/zenodo.20788187":[1323],"token":[1327,1351,1370],"ladder":[1328],"channel-partition":[1330],"compression":[1331],"family):":[1336],"doi:10.5281/zenodo.22003179":[1337],"Label-free":[1339],"inference-time":[1340],"channel":[1341],"fusion":[1342],"drift-robust":[1344],"decisions:":[1346],"doi:10.5281/zenodo.22046713":[1347],"decision-oriented":[1350],"bounded,":[1354],"threshold-free":[1355],"cache":[1356],"key":[1357],"generative":[1359],"outputs:":[1361],"doi:10.5281/zenodo.22148612":[1362],"predictive":[1365],"reach":[1366],"(forecasting/anticipation/fusion):":[1371],"doi:10.5281/zenodo.22736921":[1372],"Non-invertible":[1374],"anonymous:":[1377],"privacy":[1379],"characterizatio":[1380]}

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-09-20
DOI
https://doi.org/10.5281/zenodo.22866038
Primary Topic
Single-cell and spatial transcriptomics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
preprint

Paying Down the Price of the Bottleneck: Label-Efficient Foundation Codebooks and a Soft Readout for a Deterministic Edge Decision Token

Randolph James Ferlic, Kimberly Kate Ferlic
Zenodo (CERN European Organization for Nuclear Research)
Single-cell and spatial transcriptomics
preprint

Paying Down the Price of the Bottleneck: Label-Efficient Foundation Codebooks and a Soft Readout for a Deterministic Edge Decision Token

Randolph James Ferlic, Kimberly Kate Ferlic
preprint en

Abstract

Paying Down the Price of the Bottleneck: Label-Efficient Foundation Codebooks and a Soft Readout for 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.22866039 · CC-BY 4.0 · Community: spiral-domain-encoder-campaign Abstract A companion characterization established that a frozen, deterministic, class-discriminant single-token encoder (features → Fisher-discriminant ⊕ PCA subspace → k-means codebook of ≤ 256 cells → nearest-centroid decision) pays two specific bills for its byte-scale, auditable, sub-milliwatt bottleneck: it is label-hungry (a cost-setter, not a few-shot learner) and it carries a large-N accuracy tax (its hard, piecewise-constant readout is coarser than a continuous head). This paper asks, with the same pre-registration discipline, whether either bill can be paid down without changing the deterministic runtime or the ≤ 256-entry auditable table — and at what honest cost, across eight tasks (ECG, industrial bearing vibration, surgical-robot kinematics, wearable EMG, financial volatility, continuous glucose monitoring; binary and multiclass; group-disjoint splits with bootstrap confidence intervals). Two levers work. (1) A foundation codebook — centroids, subspace, and a pre-calibrated per-cell table frozen on a large labeled population and deployed to new subjects, refitting only the ≤ 256-entry table on N few-shot labels under an empirical-Bayes shrinkage rule — is near its population ceiling at N ≈ 0 where a from-scratch codebook is at chance (+0.33–0.36 AUROC at N = 8), monotone and never-worse, decision-shift-robust across device/site/era/age (gaps < 0.025), and a safe few-shot recalibrator that beats Platt scaling at small N; it replicates across all five sensor domains. It matches a nearest-class-mean/prototype head and tops logistic regression and k-nearest-neighbors at small N, but does not universally win (a linear model on linearly-separable features can beat it), so we position it as label-efficient and auditable. (2) A soft readout — averaging the ≤ 256-entry table over the m nearest cells (the filed distance-vote; the hard cell assignment, the emitted 8-bit token, and the static table all unchanged) — recovers 79–92% of the continuous linear ceiling and matches or exceeds it on 2 of 5 domains, knee at m ≈ 8, a weighting scheme that provably does not matter, composing with the foundation and generalizing to multiclass and every modality tested (+0.10 to +0.37 over the constant readout at deployable budgets, never worse at m = 8). We then discipline the claim on the metric a reviewer would attack: the recovery is a genuine discrimination + decision gain (balanced accuracy, macro-F1 — the ranking gain translates at 46% to > 100% across tasks), not a calibration gain — the soft readout de-calibrates because averaging cells pulls the score off each cell's empirical frequency. That bill is payable: a one-parameter temperature recalibration fit on the same N labels restores calibration to at or below the constant readout's native level while leaving the decision provably unchanged (soft-vote ECE 0.36 → 0.05 on glucose, 0.37 → 0.01 on bearing vibration, at N = 64). An honest ceiling check shows the linear reference was fair — gradient-boosted and neural models on the same features are few-shot-limited and do not beat it at deployable budgets, though a continuous model retains a ≤ ~0.12 AUROC residual at larger N on the hardest tasks. Four alternative levers are reported as pre-registered negatives (teacher distillation at population scale, upstream supervised codebook refinement, a per-cell piecewise-linear tuner, a multi-token "burst" ensemble). This is explicitly a characterization of previously-described, filed methods; it discloses no new algorithmic subject matter. Highlights · Two bills, paid down from opposite sides, runtime untouched. The token's label-hunger is paid down offline by a foundation codebook; its accuracy tax is paid down at runtime by a soft readout. Neither touches the deterministic nearest-centroid computation or the ≤ 256-entry auditable table. · Label-efficient and never-worse. A frozen foundation codebook + empirical-Bayes shrinkage table is at its population ceiling from the first labels (+0.33–0.36 AUROC at N = 8 vs a fresh codebook at chance), monotone, shift-robust, and a safer few-shot recalibrator than Platt — replicated across five sensor domains. Bounded honestly: it matches the strongest simple few-shot head and tops the rest, but is not a universal winner. · The accuracy tax is recoverable — from the readout, not construction. A distance-vote over the m nearest cells (filed) recovers 79–92% of the continuous linear ceiling (matches/exceeds it on 2/5 domains), knee at m ≈ 8, weighting scheme provably irrelevant, generalizing to multiclass and six modalities, never hurting at m = 8. Four construction-side alternatives fail — the tax is intrinsic to the hard collapse. · The right metric, and the honest cost. The recovery is a real discrimination + decision gain (balanced accuracy / macro-F1), not a calibration gain — the soft readout de-calibrates. Measuring calibration separately is the discipline; conflating it with AUROC is the error. · The recommendation, demonstrated not asserted. One-parameter temperature recalibration restores the soft readout's calibration to ≤ the constant readout's native level with the decision provably unchanged (ECE 0.36 → 0.05 on glucose at N = 64). Platt scaling overfits the small label set; temperature is the safe choice. · An honest ceiling. Strong nonlinear models on the same features are themselves few-shot-limited and do not beat the linear reference at deployable budgets — and at small N the frozen foundation soft readout matches or beats a freshly-trained strong model — but a continuous model keeps a ≤ ~0.12 AUROC residual at larger N on the hardest tasks. "Recovers the tax" means recovers the linear-ceiling gap. What this record contains · Manuscript_Paper46.pdf — the manuscript with five figures embedded (the two-lever scorecard, the five-domain label-efficiency comparison, the accuracy↔auditability curve, the soft-vote advantage vs label-budget surface, and the post-hoc recalibration panel), and Manuscript_Paper46.docx, the editable source. · PAPER_46_ZENODO_ARCHIVE.zip — the reproducibility archive: the nine frozen pre-registrations, the experiment runners (the soft-vote m-sweep, the label-budget × cell-count surface, the metric-hygiene and reviewer-proofing rounds, the upstream-refinement / local-tuner / multi-token-ensemble negatives, the multi-domain foundation and reviewer-defense batteries, and the Modal-scale foundation / distillation / shrinkage / shift / recalibration runners), the shared feature-extractor loader, the figure builder, the seventeen per-experiment result records, the five figures, the manuscript source, 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 and identifiers are scrubbed and leak-scanned per the campaign deposit discipline. Cite as R. J. Ferlic and K. K. Ferlic, "Paying down the price of the bottleneck: label-efficient foundation codebooks and a soft readout for a deterministic edge decision token," Zenodo, 2026, doi: 10.5281/zenodo.22866039. 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 previously-described methods and discloses no new algorithmic subject matter; distance-weighted voting, temperature and Platt scaling, prototype / nearest-class-mean classification, and empirical-Bayes shrinkage are established prior art, used only as tools. The methods characterized — the class-discriminant single-token codebook encoder and its nearest-centroid monitor, the multi-token / token-ladder / soft-readout mechanisms (the soft readout), and the supervised codebook-refinement mechanism (evaluated as a negative result) — 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) and the multi-token / token-ladder readout and supervised-refinement applications (priority U.S. Application No. 19/467,303 and its continuations). The foundation-codebook-with-shrinkage edge deployment is published as a characterization of the already-filed codebook method — a prior-art review found it anticipated by the prototype / nearest-class-mean and training-free-calibration literatures — and no new subject matter is claimed for it. 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 (the multi-token / soft-readout family): 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 · Non-invertible but not anonymous: a privacy characterizatio

Zenodo (CERN European Organization for Nuclear Research)
EP Analytics (United States) (US)
Reduced inequalities
Single-cell and spatial transcriptomics
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.