{"id":"https://openalex.org/W7164064005","doi":"https://doi.org/10.48550/arxiv.2606.09662","title":"When Built-in Thinking Helps and Hurts: Constraint-Level Error Shifts in Instruction Following","display_name":"When Built-in Thinking Helps and Hurts: Constraint-Level Error Shifts in Instruction Following","publication_year":2026,"publication_date":"2026-06-08","ids":{"openalex":"https://openalex.org/W7164064005","doi":"https://doi.org/10.48550/arxiv.2606.09662"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.09662","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.09662","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.09662","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138207201","display_name":"Sai Adith Senthil Kumar","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Kumar, Sai Adith Senthil","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5138207201"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11902","display_name":"Intelligent Tutoring Systems and Adaptive Learning","score":0.4713999927043915,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11902","display_name":"Intelligent Tutoring Systems and Adaptive Learning","score":0.4713999927043915,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11516","display_name":"Visual and Cognitive Learning Processes","score":0.11349999904632568,"subfield":{"id":"https://openalex.org/subfields/3205","display_name":"Experimental and Cognitive Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10103","display_name":"Reading and Literacy Development","score":0.06629999727010727,"subfield":{"id":"https://openalex.org/subfields/3204","display_name":"Developmental and Educational Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/relevance","display_name":"Relevance (law)","score":0.6373000144958496},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.6281999945640564},{"id":"https://openalex.org/keywords/trace","display_name":"TRACE (psycholinguistics)","score":0.5735999941825867},{"id":"https://openalex.org/keywords/correlation","display_name":"Correlation","score":0.5540000200271606},{"id":"https://openalex.org/keywords/coding","display_name":"Coding (social sciences)","score":0.5358999967575073},{"id":"https://openalex.org/keywords/aggregate","display_name":"Aggregate (composite)","score":0.5220999717712402},{"id":"https://openalex.org/keywords/constraint","display_name":"Constraint (computer-aided design)","score":0.5206999778747559},{"id":"https://openalex.org/keywords/sign","display_name":"Sign (mathematics)","score":0.4936000108718872}],"concepts":[{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.6373000144958496},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.6281999945640564},{"id":"https://openalex.org/C75291252","wikidata":"https://www.wikidata.org/wiki/Q1315756","display_name":"TRACE (psycholinguistics)","level":2,"score":0.5735999941825867},{"id":"https://openalex.org/C117220453","wikidata":"https://www.wikidata.org/wiki/Q5172842","display_name":"Correlation","level":2,"score":0.5540000200271606},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.5358999967575073},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.5220999717712402},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.5206999778747559},{"id":"https://openalex.org/C139676723","wikidata":"https://www.wikidata.org/wiki/Q1193832","display_name":"Sign (mathematics)","level":2,"score":0.4936000108718872},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.47099998593330383},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.46709999442100525},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.40220001339912415},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.3709000051021576},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3203999996185303},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.3147999942302704},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3127000033855438},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.30169999599456787},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3012999892234802},{"id":"https://openalex.org/C2778061373","wikidata":"https://www.wikidata.org/wiki/Q1315146","display_name":"Predictive coding","level":3,"score":0.29179999232292175},{"id":"https://openalex.org/C11671645","wikidata":"https://www.wikidata.org/wiki/Q5054567","display_name":"Causal model","level":2,"score":0.29010000824928284},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.28130000829696655},{"id":"https://openalex.org/C114115404","wikidata":"https://www.wikidata.org/wiki/Q1993172","display_name":"Divergent thinking","level":3,"score":0.2809000015258789},{"id":"https://openalex.org/C55078378","wikidata":"https://www.wikidata.org/wiki/Q1136628","display_name":"Pearson product-moment correlation coefficient","level":2,"score":0.2648000121116638},{"id":"https://openalex.org/C2778058735","wikidata":"https://www.wikidata.org/wiki/Q4692253","display_name":"Aggregate data","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.2599000036716461},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.25200000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.09662","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.09662","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.09662","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.09662","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"display_name":"Quality Education","score":0.7028070688247681,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"reasoning":[1],"models":[2,23,32,115],"(LRMs)":[3],"often":[4,215],"improve":[5,66],"math":[6],"and":[7,53,96,179],"coding":[8],"performance,":[9],"but":[10,133],"their":[11],"effect":[12],"on":[13],"instruction":[14],"following":[15],"is":[16],"unclear.":[17],"We":[18],"study":[19],"IFEval":[20],"with":[21,141,171,191,226],"Qwen3":[22],"(1.7B-32B),":[24],"using":[25],"same-weights":[26],"Thinking":[27,121],"ON/OFF":[28],"controls;":[29],"four":[30,113,204],"Hunyuan":[31,114],"provide":[33],"directional":[34],"cross-family":[35],"support.":[36],"Aggregate":[37],"pass-rate":[38],"changes":[39,60,123],"are":[40,213],"small":[41,185],"(-0.55":[42],"to":[43],"-3.52":[44],"pp),":[45],"yet":[46],"10-20%":[47],"of":[48,63],"prompts":[49,65],"switch":[50],"between":[51,175],"pass":[52],"fail":[54],"across":[55,203],"modes,":[56],"suggesting":[57],"that":[58,209],"thinking":[59,139],"the":[61,91,104,130,227],"pattern":[62,108],"errors--some":[64],"while":[67],"others":[68],"worsen--rather":[69],"than":[70,198,217],"uniformly":[71],"degrading":[72],"performance.":[73],"Under":[74],"a":[75,134,142,151,184],"post-hoc":[76],"Qwen3-derived":[77],"grouping,":[78],"constraint":[79],"types":[80],"separate":[81],"into":[82],"Planning":[83,158,218],"(global":[84],"counting,":[85],"structure,":[86],"coordination),":[87],"which":[88,101],"improves":[89],"at":[90,230],"class":[92],"level":[93],"under":[94],"thinking,":[95],"Precision":[97,131,182,210],"(exact":[98],"local":[99],"form),":[100],"consistently":[102],"worsens;":[103],"class-level":[105],"Planning/Precision":[106],"sign":[107],"holds":[109],"directionally":[110],"for":[111],"all":[112],"despite":[116,166],"Hunyuan's":[117],"opposite":[118],"aggregate":[119],"direction.":[120],"also":[122],"final-answer":[124,180],"length;":[125],"matched-length":[126],"analyses":[127],"substantially":[128],"reduce":[129],"drop,":[132],"residual":[135],"penalty":[136],"remains.":[137],"Analyzing":[138],"traces":[140],"cross-encoder":[143],"relevance":[144,178,197],"metric":[145],"reveals":[146],"three":[147],"patterns:":[148],"Neutral":[149],"shows":[150,159,183,208],"positive":[152],"relevance-compliance":[153],"link":[154],"(r":[155,163,188],"approximately":[156,164,189],"0.15);":[157],"near-zero":[160],"predictive":[161],"correlation":[162,187],"0.02)":[165],"measurable":[167],"trace":[168,177],"engagement,":[169],"consistent":[170],"an":[172],"execution":[173],"gap":[174,229],"CE-measured":[176],"compliance;":[181],"negative":[186],"-0.05),":[190],"failing":[192],"instances":[193,212,220],"having":[194],"higher":[195],"mean":[196,224],"passing":[199],"ones.":[200],"Activation":[201],"patching":[202],"model":[205],"sizes":[206],"(1.7B-14B)":[207],"flip":[211,219],"more":[214],"restored":[216],"(32-58%":[221],"vs.":[222],"14-40%":[223],"layer-restoration),":[225],"largest":[228],"14B":[231],"(about":[232],"30":[233],"pp).":[234]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-10T00:00:00"}