cs.CLOct 19, 2026

Evaluating Context Compaction on the Questions That Come Later

Authors: Junwei Yao, Yongxi Zhou, Zike Yang, Zhijun Wang, Jianan Liu, Yanxian Chen, Wenbo Ye

Organizations: Independent Researcher

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Language-model agents compact their context for the question at hand, but the compacted context must also serve questions that come later. We evaluate compaction on these later questions, using four corpora that annotate the evidence of several questions per text and recorded agent trajectories, and compare every method with random deletion at the same budget. On MultiRC, keeping the sentences that gpt-6-luna (through the OpenAI Decisions API) judges necessary for the current question raises its accuracy on that question by 11.4 points at 25% of the tokens. On the later questions, however, no method that selects sentences for the current question does detectably better than deleting sentences at random. Deleting words instead of sentences does better: LLMLingua-2, which ignores the question, beats random sentence deletion at the same token count by 12.5 points on the later MultiRC questions. It keeps about twice random deletion's share of the evidence's content words, and at equal content-word retention its gain shrinks to 7.3 points (exploratory). On FEVER, label accuracy can overstate what the kept text supports: the gain of sentence selection is no longer detected when answers are scored against the kept evidence, and the 21.0-point gain of LLMLingua-2 is not shown to rest on the kept text. The ranking also depends on the setting: on Evidence Inference sentence selection beats LLMLingua-2 at a nominal 25% budget, and on ττ-bench trajectories, whose later tool calls need exact identifiers, it beats random deletion while LLMLingua-2 shows no detected gain. Scores on the question a compaction was made for do not reliably show how it serves later ones.

Figures & tables

   accuracy (%) 
modelpolicybudgetknownhiddenhidden −- random
open-decidertyped25%7062−- 0.6 [ −- 1.9 , +0.7 ]
open-decideragnostic25%6363+0.3 [ −- 0.9 , +1.4 ]
open-deciderrandom25%6162—
open-decidertyped50%7266−- 0.6 [ −- 1.7 , +0.6 ]
open-decideragnostic50%6767+0.4 [ −- 0.7 , +1.6 ]
Table 1: Typed selection keeps known-question accuracy above random deletion, while hidden-question accuracy stays near random. Readers: open-decider and gpt-6-luna , the typed readers whose full-text accuracy exceeds the all-“no” rate; agnostic: question-agnostic selection. Paragraph-macro candidate-level accuracy (%) on the known and hidden questions after compaction, and policy −- random hidden accuracy (points, paragraph-bootstrap 95% interval). Table 7 gives the sufficiency check for the same rows.
Figure 2: Hidden questions that share evidence with the known question gain from typed selection; on Evidence Inference, disjoint claims gain too, and more of their evidence is kept. 25% budget. Left: typed −- random hidden accuracy with bootstrap 95% intervals, for questions whose evidence overlaps the known question’s and for disjoint ones. Right: share of disjoint questions’ evidence kept by typed selection and by random deletion (for MultiRC and FEVER the exact expectation for a random subset of the same size). Values and overlap shares in Table 8 .
 MultiRC, gpt-6-lunaMultiRC, open-deciderFEVER, gpt-6-luna
textacc.wordscontentacc.wordscontentacc.wordscontent
random sentences68.326.326.162.126.226.061.225.725.6
typed sentences67.924.424.361.026.326.379.059.159.2
random words70.825.025.262.625.025.167.525.225.2
LLMLingua-280.826.348.068.126.348.082.226.440.7
hybrid76.625.737.365.726.137.684.755.359.2
Table 2: Word-level compression leads on hidden MultiRC and FEVER questions and keeps more of their evidence’s content words than random sentence deletion; on MultiRC, about the same share of its words. 25% of the tokens; FEVER accuracy is label accuracy. Accuracy (%, document-macro) on hidden questions with annotated evidence, and share of the hidden questions’ annotated evidence kept (%), counted over all words and over content words, answer-weighted (content words: words outside a standard English stop list). Random words: a uniform sample of words. Hybrid: typed whole sentences up to half the budget, the rest LLMLingua-2 words (Appendix H ).
Appendix figures & tables21 assets

Supplementary material from the paper’s appendix.

Appendix

analysisappendixpredictionsmetpartly metnot metunresolved
Evidence Inference overlap testE21—1—
KEEP–DROP contrastK44———
LoCoMo testM63——3
Follow-ups A–DG , J , M96111
LLMLingua-2 granularity testH211——
Hybrid compactorH312——
Appendix
Table 3: Outcomes of the prespecified predictions. Partly met: met in some of the cells the prediction covers.
  hidden questions, 25%hidden
corpusknown, 25%alloverlappingdisjoint50%
MultiRC+10.4 [ +7.4 , +13.3 ]−- 0.6 [ −- 1.8 , +0.6 ]+3.5 [ +1.5 , +5.7 ]−- 1.7 [ −- 3.2 , −- 0.2 ]+0.5 [ −- 1.0 , +1.9 ]
FEVER+36.0 [ +28.5 , +43.5 ]+18.5 [ +14.3 , +22.5 ]+27.3 [ +22.7 , +32.0 ]+1.3 [ −- 3.4 , +6.1 ]+10.1 [ +6.6 , +13.9 ]
Ev. Inference+17.4 [ +14.5 , +20.3 ]+11.0 [ +8.9 , +13.1 ]+18.8 [ +14.6 , +23.0 ]+8.5 [ +6.1 , +10.8 ]+4.3 [ +2.7 , +5.9 ]
Appendix
Table 5: qwen3.8-flash as reader: typed −- random accuracy (points, document-bootstrap 95% interval) on the known and hidden questions, with the hidden questions split by evidence overlap, at 25% and (last column) 50%.
corpus, readercontrastbball hiddenoverlappingdisjoint
MultiRC, open-decidertyped −- random25−- 0.1 [ −- 1.1 , +0.8 ]+6.0 [ +4.3 , +7.7 ]−- 2.9 [ −- 4.0 , −- 1.9 ]
  50−- 0.6 [ −- 1.6 , +0.3 ]+3.9 [ +2.4 , +5.4 ]−- 3.2 [ −- 4.4 , −- 1.9 ]
 agnostic −- random25+0.8 [ −- 0.1 , +1.7 ]  
  50+0.4 [ −- 0.5 , +1.3 ]  
MultiRC, gpt-6-lunatyped −- random25−- 0.2 [ −- 1.1 , +0.7 ]+4.4 [ +2.6 , +6.1 ]−- 2.2 [ −- 3.4 , −- 1.0 ]
  50−- 0.2 [ −- 1.3 , +0.8 ]+2.8 [ +1.3 , +4.4 ]−- 1.5 [ −- 2.8 , −- 0.3 ]
Appendix
Table 6: Hidden-question contrasts against random deletion averaged over seeded masks (points, document bootstrap 95% interval), at budget bb (%), over all hidden questions and split by evidence overlap with the known question. Compare Table 8 , which uses one mask.
   sufficiency AUROC vs.hidden acc. after 25% re-read
modelpolicybudgetevidence lostanswer flippedlowest sufficiencyrandom
open-decidertyped25%0.810.696766
open-decideragnostic25%0.790.696866
open-deciderrandom25%0.810.686766
open-decidertyped50%0.780.707169
open-decideragnostic50%0.710.667170
Appendix
Table 7: How well the sufficiency check finds damaged hidden questions, and what re-reading them recovers (MultiRC, the rows of Table 1 ). The AUROC (area under the receiver operating characteristic curve) columns rank hidden questions by low sufficiency and score that ranking against two question-level outcomes: at least one annotated evidence sentence deleted (evidence lost), and at least one candidate answer turned from right to wrong (answer flipped). The last two columns give hidden candidate-level accuracy averaged over questions (not exact match) after re-reading the full text for 25% of hidden questions: those with the lowest sufficiency, or a random 25%.
  typed −- randomoverlapdisjoint kept
corpus, readerbball hiddenoverlappingdisjointsharetyped/random
MultiRC, open-decider25−- 0.6 [ −- 1.9 , +0.7 ]+5.3 [ +2.9 , +7.7 ]−- 3.0 [ −- 4.5 , −- 1.4 ]3514/25
 50−- 0.6 [ −- 1.7 , +0.6 ]+4.2 [ +1.9 , +6.3 ]−- 3.4 [ −- 5.0 , −- 1.8 ] —
MultiRC, gpt-6-luna25−- 0.8 [ −- 1.9 , +0.3 ]+4.3 [ +2.2 , +6.4 ]−- 2.9 [ −- 4.3 , −- 1.4 ]3518/25
 50−- 0.4 [ −- 1.7 , +0.9 ]+2.6 [ +0.5 , +4.7 ]−- 1.6 [ −- 3.3 , +0.1 ] —
FEVER, gpt-6-luna25+17.2 [ +13.4 , +21.2 ]+25.8 [ +21.3 , +30.3 ]+0.2 [ −- 4.8 , +5.2 ]5122/24
Appendix
Table 8: Hidden questions whose evidence overlaps the known question’s gain more from typed selection. Typed −- random hidden accuracy (points, paragraph or document bootstrap 95% interval) at budget bb (%), over all hidden questions and split by evidence overlap with the known question. Overlap share: evidence-classified hidden questions sharing an evidence sentence with the known question (%). Disjoint kept: share of disjoint questions’ evidence kept by typed selection and by random deletion (%; for MultiRC and FEVER the exact expectation for a random subset of the same size, for Evidence Inference the kept set of the random-deletion read). Each accuracy cell averages within documents, over the documents that have such questions, so the two cells need not combine to the all-question contrast by the overlap share. The overlap split is post hoc on MultiRC and FEVER; † on Evidence Inference the overlapping and disjoint contrasts are prespecified, and the all-hidden contrast and the kept column exploratory.
Figure 3: For typed sentence selection, the hidden-question gain follows the extra evidence kept. Typed −- random hidden accuracy against typed minus random evidence kept (points), at 25%, on overlapping (circles) and disjoint (squares) hidden questions. Open markers: the paper’s known question (Figure 2 ); filled: a different known question per document. Dashed: the line specified before the second experiment, 0.366×0.366\times extra evidence + 0.343+\,0.343 .
  hidden questionsoverlapping
corpus, readerknownalloverlappingdisjoint−- disjoint
MultiRC, open-decider+6.2 [ +3.0 , +9.7 ]−- 0.7 [ −- 2.0 , +0.6 ]+4.5 [ +2.3 , +6.9 ]−- 3.3 [ −- 4.9 , −- 1.8 ]+7.6 [ +4.9 , +10.4 ]
MultiRC, gpt-6-luna+8.0 [ +5.0 , +11.1 ]−- 0.4 [ −- 1.5 , +0.6 ]+1.3 [ −- 0.7 , +3.5 ]−- 1.2 [ −- 2.7 , +0.4 ]+3.0 [ +0.3 , +5.8 ]
FEVER, gpt-6-luna+39.5 [ +33.0 , +46.5 ]+16.0 [ +11.9 , +20.3 ]+29.6 [ +24.7 , +34.0 ]−- 6.6 [ −- 12.0 , −- 1.6 ]+35.7 [ +29.0 , +42.7 ]
Ev. Inference, gpt-6-luna+22.6 [ +19.6 , +25.8 ]+11.6 [ +9.5 , +13.8 ]+20.6 [ +16.9 , +24.5 ]+8.1 [ +5.6 , +10.5 ]+12.7 [ +7.6 , +18.1 ]
Appendix
Table 9: With a different known question, typed selection still helps the known question, hidden MultiRC questions stay near random deletion, and overlapping hidden questions gain more than disjoint ones. Typed −- random accuracy at 25% (points, document-bootstrap 95% interval). Overlapping −- disjoint is the paired contrast within the documents that have both kinds of hidden question. Bottom: extra evidence kept (typed −- random, points) and the gain predicted by the line specified before this experiment, 0.366×0.366\times extra evidence + 0.343+\,0.343 ; the observed gains are the overlapping and disjoint columns above.
  hidden questionsknown question
corpus, readerbbL2 −- randomtyped −- L2L2 −- randomtyped −- L2
MultiRC, open-decider25+6.5 [ +5.4 , +7.6 ]−- 6.6 [ −- 8.0 , −- 5.2 ]+6.9 [ +4.6 , +9.3 ]+1.7 [ −- 1.3 , +4.8 ]
 50+5.3 [ +4.4 , +6.3 ]−- 6.0 [ −- 7.3 , −- 4.6 ]+3.4 [ +1.3 , +5.5 ]+2.8 [ +0.3 , +5.2 ]
MultiRC, gpt-6-luna25+12.9 [ +11.8 , +14.1 ]−- 13.1 [ −- 14.5 , −- 11.8 ]+12.8 [ +10.3 , +15.1 ]−- 3.0 [ −- 5.9 , −- 0.1 ]
 50+10.2 [ +9.3 , +11.1 ]−- 10.4 [ −- 11.7 , −- 9.2 ]+9.9 [ +8.0 , +12.0 ]−- 2.3 [ −- 4.3 , −- 0.2 ]
FEVER, gpt-6-luna25+20.8 [ +18.8 , +23.0 ]−- 3.7 [ −- 6.7 , −- 0.6 ]+22.0 [ +16.2 , +27.9 ]+12.5 [ +7.0 , +18.5 ]
Appendix
Table 10: LLMLingua-2 (L2), at a rate targeting random deletion’s share, compared with random deletion averaged over the seeded masks and with typed selection, on hidden and known questions (points, paragraph or document bootstrap 95% interval) at budget bb (%; exploratory).
 MultiRC, gpt-6-lunaMultiRC, open-deciderFEVER, gpt-6-luna
budget25%50%25%50%25%50%
policyknownhiddenknownhiddenknownhiddenknownhiddenknownhiddenknownhidden
typed78.767.983.276.169.061.073.666.192.579.093.085.6
agnostic68.567.576.075.563.362.266.466.462.361.676.076.7
TF-IDF76.967.082.874.865.461.768.266.188.077.290.086.7
random (3 masks)67.368.376.276.061.162.266.766.956.861.272.275.2
Appendix
Table 11: At token-matched budgets, typed selection beats random deletion on the known question, and LLMLingua-2 is highest on hidden questions. Accuracy (%) at 25% and 50% of each document’s o200k_base tokens, on the known and hidden questions. MultiRC: candidate-level, paragraph-macro; FEVER: claim accuracy, document-macro. Random deletion averages three masks. Each level uses every answer its policy gave; differences of these levels are within 1 point of the paired contrasts of Table 12 .
 MultiRC, gpt-6-lunaMultiRC, open-deciderFEVER, gpt-6-luna
policypolicy/randomdifferencepolicy/randomdifferencepolicy/randomdifference
hidden questions, 25% of tokens
typed68.1/68.5−- 0.4 [ −- 1.5 , +0.6 ]61.0/62.2−- 1.1 [ −- 2.1 , −- 0.2 ]79.0/61.2+17.8 [ +15.0 , +20.8 ]
agnostic67.6/68.6−- 1.0 [ −- 1.8 , −- 0.1 ]62.2/62.2+0.1 [ −- 1.0 , +1.2 ]61.6/61.2+0.4 [ −- 2.4 , +3.2 ]
TF-IDF67.1/68.5−- 1.3 [ −- 2.3 , −- 0.4 ]61.7/62.2−- 0.5 [ −- 1.5 , +0.5 ]77.2/61.2+16.0 [ +13.0 , +19.0 ]
LLMLingua-280.7/68.6+12.1 [ +11.0 , +13.3 ]68.0/62.2+5.9 [ +4.7 , +7.0 ]82.2/61.2+21.0 [ +18.6 , +23.2 ]
Appendix
Table 12: On hidden MultiRC questions, every policy except LLMLingua-2 is near or below random deletion; on FEVER, every policy except question-agnostic selection is above it. Hidden questions at token-matched budgets, paired: an answer refused in either read is dropped from both. Policy/random: the two accuracies (%) on the remaining answers, document-macro, with random deletion averaged over its three masks within each document. Difference: policy −- random, which equals the difference of the two levels up to rounding (points, document-bootstrap 95% interval).
  words −- sentencesscores −- random
corpus, readerbbby scores (1)at random (2)sentences (3)words (4)
MultiRC, dec25+4.6 [ +3.2 , +5.9 ]+0.4 [ −- 0.4 , +1.3 ]+1.3 [ +0.2 , +2.3 ]+5.4 [ +4.2 , +6.6 ]
 50+4.3 [ +3.0 , +5.5 ]+0.5 [ −- 0.3 , +1.3 ]+0.5 [ −- 0.4 , +1.5 ]+4.4 [ +3.3 , +5.5 ]
MultiRC, luna25+11.7 [ +10.4 , +13.0 ]+1.2 [ +0.3 , +2.1 ]+0.4 [ −- 0.7 , +1.5 ]+10.0 [ +8.9 , +11.2 ]
 50+9.7 [ +8.6 , +10.9 ]+2.4 [ +1.6 , +3.3 ]+0.1 [ −- 0.9 , +1.0 ]+7.3 [ +6.5 , +8.3 ]
FEVER, luna25+19.1 [ +15.9 , +22.1 ]+6.3 [ +4.4 , +8.2 ]+1.9 [ −- 1.3 , +5.4 ]+14.7 [ +12.7 , +16.8 ]
Appendix
Table 13: The LLMLingua-2 scores help when they choose words, and in five of six cells not when they choose sentences. Top: contrasts on hidden questions at budget bb (% of the tokens; points, document-bootstrap 95% interval), paired on the answers both policies gave. Bottom: hidden-question accuracy (%) of sentence scores and random words; Table 11 gives word scores ( LLMLingua-2 ) and random sentences. Random policies average three masks. luna: gpt-6-luna ; dec: open-decider .
 MultiRC, gpt-6-lunaMultiRC, open-deciderFEVER, gpt-6-luna
policyknownhiddenknownhiddenknownhidden
typed78.767.969.061.092.579.0
LLMLingua-280.980.770.068.079.582.2
hybrid, half82.276.570.765.789.084.7
hybrid, quarter79.679.670.866.881.582.1
random sentences67.368.361.162.256.861.2
Appendix
Table 14: Hybrid compaction at 25% of the tokens. Top: known and hidden accuracy (%); bottom: share of the known and hidden questions’ evidence words kept (%). The hybrid keeps typed whole sentences up to half (or a quarter) of the budget and fills the rest with LLMLingua-2 words. Hidden accuracy here covers all hidden questions; Table 2 covers those with annotated evidence, so the two differ by up to 0.1 points.
cellgaprawallcontent+answerbreadthlinear
MultiRC, lunaLLMLingua-2 −- random sent.+12.5 [ +11.3 , +13.6 ]+10.1+7.3 [ +6.1 , +8.4 ]+4.9+2.6+8.0
MultiRC, lunahybrid −- typed+8.6 [ +7.3 , +9.9 ]+7.2+5.1 [ +3.8 , +6.4 ]+3.2−- 0.7+6.1
MultiRC, decLLMLingua-2 −- random sent.+6.0 [ +4.7 , +7.3 ]+5.5+2.0 [ +0.7 , +3.2 ]+0.6−- 2.2+2.3
MultiRC, dechybrid −- typed+4.7 [ +3.5 , +5.9 ]+4.6+2.5 [ +1.1 , +3.7 ]+1.1−- 2.9+2.8
FEVER, lunaLLMLingua-2 −- random sent.+21.0 [ +18.7 , +23.4 ]+2.7+1.3 [ −- 1.1 , +3.5 ]—−- 4.1+14.9
FEVER, lunahybrid −- typed+5.7 [ +3.6 , +7.9 ]+0.4−- 0.5 [ −- 1.8 , +0.8 ]—−- 5.8+5.7
Appendix
Table 15: Hidden-question gaps (points) at 25% before and after conditioning on a retention measure (exploratory; document-bootstrap 95% intervals). Columns after the raw gap: binned all-word (all), content-word, content-word plus answer and breadth retention, and content-word retention in the linear form (linear). Raw: the model without retention terms, on hidden questions with annotated evidence; the paired contrasts of Appendices H and G differ slightly. luna: gpt-6-luna ; dec: open-decider .
  document label recovered 
corpustextsupportedrefutedbalanced“not enough info”
FEVERfull93.883.588.6–
 typed 25%67.257.362.274.6
 random 25%37.037.137.181.1
 typed 50%78.568.373.471.7
 random 50%60.057.058.575.3
Appendix
Table 16: Typed selection recovers the document’s label more often than random deletion. Three-way read of the hidden claims ( gpt-6-luna ; %, document-macro, over all claims each read answered). Document label recovered: per gold label and label-balanced. “Not enough info”: share of claims whose annotated evidence was all deleted that the reader answers so. Bottom: Label recovery and annotation-derived support accuracy on common claim sets. Hidden claims of the three-way read ( gpt-6-luna ) on the common set of each row: claims answered in both texts and, for support, with a target in both. Claims, documents: size of that set. Typed, random: label-balanced document-macro accuracy StS_{t} (%, defined in the text). Typed −- random: Styped−SrandomS_{\text{typed}}-S_{\text{random}} , equal to the difference of the two levels up to rounding (points, document-bootstrap 95% interval).
corpusclaimsnndocsKEEP −- DROPKEEPDROPtypedrandomfull
FEVERall5,132200+50.0 [ +47.4 , +52.4 ]9545765995
 disjoint2,551158+41.1 [ +36.2 , +45.8 ]9553626293
 overlapping2,581196+50.1 [ +46.7 , +53.5 ]9545886195
 true3,658199—9421684194
 false1,474199—9798989997
Ev. Inf.all2,915792+17.8 [ +15.9 , +19.7 ]9072827090
Appendix
Table 17: KEEP beats DROP at 25% ( gpt-6-luna ). Split by gold label, the gap is on true claims; false claims are answered correctly at 97% or more in every arm. KEEP −- DROP hidden-claim accuracy (points, document bootstrap 95% interval; all and disjoint prespecified, overlapping secondary, label split post hoc) and document-macro accuracy (%) of KEEP, DROP and, descriptively, the main typed-selection, random and full-text reads of the same claims, which asked several claims per request.
  measured, per documentmodeled, per later question
corpuskˉ\bar{k}scoringread 25%read full25%50%fullcheck sep.check jointre-read
MultiRC10.22,60515,08115,4311,2591,3201,446245177361
FEVER27.22,79510,46010,810311426671281169168
Evidence Inf.3.722,1813,3216,4791,5352,7084,8891,4881841,222
LoCoMo149.6180,51064,22096,7426,27011,78322,7216,2181465,680
thousand input tokens per document, typed 25%k=1k{=}1k=3k{=}3k=10k{=}10k=kˉk{=}\bar{k}ratio at kˉ\bar{k}break-even kk
Appendix
Table 18: Input tokens for typed selection at 25% ( gpt-6-luna ). Top: per document, measured scoring and batched reads, and modeled one-question costs (read, sufficiency check as a separate call or joint with the read, and a full re-read charged at a quarter of the full read). Bottom: thousand input tokens per document to answer kk later questions, one request each, the ratio to always reading the full text at k=kˉk=\bar{k} , the corpus’s mean number of later questions, and the break-even kk , the number of later questions at which the policy and the full text cost the same (never: at no kk ). Prompt caching is not modeled.
 accuracy (%)evidence kept,
memoryluna 25%luna 50%qwen 25%25% (%)
typed87.991.983.827.2
agnostic90.494.8—48.6
TF-IDF88.791.485.725.7
random88.389.783.225.4
recency83.187.682.425.3
Appendix
Table 19: LoCoMo hidden-question accuracy at the last checkpoint (conversation-macro %, chance 25) by memory, budget and reader (luna: gpt-6-luna ; qwen: qwen3.8-flash on the same memories), and the share of hidden questions’ evidence turns kept at 25% (%, question mean). Full and empty memory do not depend on the budget. Tests H1–H4: under Test results below.
 gpt-6-lunaqwen3.8-flash
memory (last checkpoint)refusedaccuracyanswered acc.refusedaccuracyanswered acc.
empty, original options10.081.490.60.085.185.1
empty28.460.283.90.071.571.5
full0.590.991.4———
typed 25%1.178.078.90.075.775.7
agnostic 25%2.381.082.80.081.781.7
Appendix
Table 20: With harder options, gpt-6-luna refuses more with empty memory, KEEP stays far above DROP, and typed versus random stays unresolved. LoCoMo hidden questions at the last checkpoint, conversation-macro (%). Refused: share of questions refused. Accuracy: over all questions, refusal counts as wrong (chance 25). Answered acc.: over answered questions. First row: empty memory with the original options, on the same questions. KEEP and DROP are the texts of Section 5 . qwen3.8-flash did not read the full memory.
texthidden, lunaknown, lunahidden, qwenid keptGG
full state91.070.596.4——
oracle93.245.597.3100.01.00
anticipatory75.947.280.077.10.64
ECov †74.849.377.373.60.64
typed67.159.274.066.40.46
recency53.928.956.442.90.19
Appendix
Table 21: On τ\tau -bench retail trajectories at 25% of the tokens, typed selection beats random deletion on hidden questions, LLMLingua-2 does not, and anticipating later calls costs the known question. Accuracy (%, state means) with gpt-6-luna (luna) and qwen3.8-flash (qwen); ECov: expected-coverage selection; id kept: share of hidden questions whose gold identifier survives; GG : share of the random–oracle gap closed on hidden questions (luna). † On a 120-state subsample, where typed selection scores 62.1 (hidden) and 59.9 (known). Further arms are on all states; their contrasts with ECov (M4, M5 in Appendix N ) use the ECov subsample.
 stubsgists
θ\thetaacc.re-readtokensacc.re-readtokens
0.372.365.196.667.764.692.9
0.572.672.8104.569.173.2102.0
0.772.379.5111.769.779.8108.4
Appendix
Table 22: Recoverable arm (retail, 25%, gpt-6-luna ): hidden accuracy (%), share of hidden questions re-read (%) and passage tokens per question as % of the full state’s, by threshold θ\theta .
compactionknownhiddenTT usedcut−- random, hidden−- LLMLingua-2 , hidden−- random, known
random sentences67.368.3     
typed78.767.9     
LLMLingua-280.980.7     
qwen3.8-flash
generic75.476.188.423.7+7.6 [ +6.6 , +8.7 ]−- 4.5 [ −- 5.7 , −- 3.2 ]+8.2 [ +6.1 , +10.3 ]
focused88.565.469.50.7−- 2.9 [ −- 4.0 , −- 1.8 ]−- 15.1 [ −- 16.7 , −- 13.6 ]+21.4 [ +19.2 , +23.5 ]
Appendix
Table 23: Abstractive summaries on MultiRC at 25% of the tokens, read by gpt-6-luna . Left: known and hidden accuracy (%), share of TT used and share of summaries cut at TT (%). Right: summary −- comparator on hidden questions, and −- random deletion on the known question (points, 95% interval).
More in this topic

Context Compaction for LLM Agents

See all 37 papers

Explore similar work

CardsList

  1. Context Compaction Theory

    Aug 2, 2026Hayder Tirmazi, Sam Markelon, Allison Bishop +1LLM Context ManagementContext Compression

    AllSpice Inc. · Boston University · Proof Trading +2

  2. Bud ContextCompact 2.0: Context Compaction for Agent Success and Accuracy with Small Language Models and Constrained Context Windows

    Oct 18, 2026Jithin VG, Ditto PSLLM Context ManagementLLM Context Compression

    Bud Ecosystem

  3. EnComp: Lightweight Encoder-Only Context Compression for Retrieval-Augmented Question Answering

    Mar 10, 2026Thao Do, Dinh Phu Tran, An Vo +2Efficient Language Model InferenceKnowledge-Intensive QA

    Korea Advanced Institute of Science and Technology (KAIST), Daejeon 34141, South Korea