Depth or Description? When a One-Step Latent Score Replaces Lookahead in Proactive Information Acquisition

Biswa Sengupta

cs.LGNov 4, 2026

Depth or Description? When a One-Step Latent Score Replaces Lookahead in Proactive Information Acquisition

Authors: Biswa Sengupta

Organizations: LLM Suite Team, JPMorgan Chase & Co.

Abstract

A proactive agent must decide what to ask before it acts. Exact lookahead values combinations of questions but is expensive and leans on a fitted model; a one-step score is cheap but can miss them. Our main finding: the need for depth often lies in how the hidden cause is described. A one-step score on the label can be blind to an interaction among answers. A score on a suitable latent cause is not, and where that latent encodes the decision it replaces most of depth at a fraction of the search. Described differently, the same problem can defeat the score while exact values do not move, so keep the latent small and check it against exact values. Probabilities learned from one language model's answers do not carry to another unless planning allows for both; those read from tokens are worth only a few labels.

Figures & tables

Panel of Figure 1 — Figure 1: The argument. Grey: a problem. Blue: the solution tested. Green and orange: a confirmed or a mixed verdict. The first results find what decides whether a one-step score replaces depth; the rest probe the limits of that answer.

Panel of Figure 2 — Figure 2: How the planners search, at a history $h$ with three unasked questions. Squares are histories, circles a question’s uncertain answer. (a)–(c) D $k$ branches on every unasked question for $k$ levels, then values predicting now, $T(h)$ . (d) Rollout looks one question ahead, then follows D1 to the end of the budget. (e) LIG asks the question most informative about the latent cause $Z$ . (f) LIG-stop also stops when $W(h)$ , the exact value of continuing, does not exceed $T(h)$ . (g) Rollout-LIG scores each first question by the LIG-stop value after its answers. (h) Under a compute limit, the original bounded D3 fell back to D1 whenever no depth-three candidate finished; iterative deepening keeps the completed D2 decision.

Figure 3: One-step latent scores against depth. (a) Utility against compilation work, 160 fits; hollow: LIG. (b) Accuracy at seven questions, five fits. Aligned planners target the scored classifier, fitted ones the fitted model’s most probable class, LIG neither. (c) Context parity, 80 fits, saturated latent.

Panel of Figure 4 — Figure 4: The description decides. (a) Theorem B.14 : two fair bits with an XOR label and a four-valued nuisance $N$ . (b) Noisy grid, 160 fits. Coarse adds an eight-answer question the latent ignores and refined records it; noise 3 , a separate condition, records the noise of three existing questions (D3 and the witness rule not run). (c) LIG-stop on the true parity joints; the structured latent keeps the context, selected block and label, and equals full-horizon search.

Figure 5: (a) Noiseless CUBE-NM accuracy at seven questions, on the same five fits; AACO predicts with the fitted model (56% with the scored classifier). (b) Noisy-grid utility, 160 fits. Up to 8,192 transitions both deepening orders equal D2.

Figure 6: (a) Held-out differences, with descriptive 95% intervals over task–split pairs, which share data. On iris a poor fit misleads D2 and D3 alike. (b) The full controller, averaged over datasets; capped rows allow 128 units per decision.

Panel of Figure 7 — Figure 7: Counted probabilities, tested on held-out answers. (a) A Beta–Bernoulli fit, with illustrative answers. (b) D3 identification on Common at seven questions (last row: Section 7.7 ’s four-model reads, no answers counted). (c) Reliability of the fitted yes-probability; dashed: perfect. Marker area follows the answers in a bin: 90–97% lie in the two extreme bins, where six answers agreed.

Panel of Figure 8 — Figure 8: Certifying a preference despite model error. Left: the on-policy error $E(\pi)$ counts model error only where the policy goes. Right: cells, of 36, where D3 over D2 is certified when training matches the source. Solid: true errors. Dashed: validation radii ( $8n$ rows, $\delta=.05$ ), usable without the true model; $n$ rows certify 0, 0 and 23. Dotted: cells where D3 is truly better.

Appendix figures & tables62 assets

Supplementary material from the paper’s appendix.

Appendix

Theorem	Lean file	Theorem	Lean file
B.1	Parity , ParityGeneralNoise	B.11	AnytimeDeepening
B.2	ContextParityUpper , ContextParityGeneral	B.12	LatentVisibility
B.3	InteractionOrder	B.13 , B.14	RepresentationRefinement
B.4	InteractionOrderSufficiency	B.15	EquivalenceClassEdges
B.5	LookaheadRollout	B.16	EquivalenceClassGreedy
B.6	StaticAcquisitionCounterexample	B.17	MonotoneBellman

Appendix

Table 1: The Lean file that checks each theorem.

Source	$n$	Fitted $>0$	True $>0$	Full joint	On-policy	Exact range	Plug-in
Matched	128	25	19	0	6	10	5
Matched	1,024	29	29	15	21	24	21
Matched	8,192	36	36	30	36	36	33
Independent label	128	14	6	0	0	0	0
Independent label	1,024	5	4	0	0	0	0
Independent label	8,192	0	0	0	0	0	0

Appendix

Table 2: Certified preferences in the controlled model-error study. Upper rows: D3 over D2 in 36 cells per row. Lower rows: all ordered method pairs with a positive fitted margin. Full joint is the margin condition

2(2+3c)\operatorname{TV}(P,Q)

; on-policy uses the range

[-3c,1]

, or each policy’s exact fitted range. No on-policy certificate has the wrong sign. The plug-in column replaces the true conditionals by an independent held-out sample; it is an estimate, not a certificate, and certifies four wrong-sign independent-label pairs.

Source	$n$	Fitted $>0$	Oracle	Plug-in	Exact radii	$m=n$	$m=8n$
Matched	128	25	6	5	6	0	0
Matched	1,024	29	21	21	21	0	12
Matched	8,192	36	36	33	36	23	30
Independent label	128	14	0	0	0	0	0
Independent label	1,024	5	0	0	0	0	0
Independent label	8,192	0	0	0	0	0	0

Appendix

Table 3: Confidence certificates in the controlled model-error study. Upper rows: D3 over D2 in 36 cells per row; lower rows: all ordered pairs with a positive fitted margin. Wrong-sign certifications are in brackets. Oracle uses the true conditionals and reach probabilities. The plug-in estimate uses a held-out draw from the training source. Exact radii use the fitted-model recursion with the true

L_{1}

distances. The confidence columns use radii from a validation sample of size

m

drawn from the true distribution, with

\delta=.05

per cell. The radii held in 108/108 draws at

m=n

and 108/108 at

m=8n

. Median bound over oracle error: 1.002 (exact radii), 7.8 (

m=n

), 3.3 (

m=8n

Allowance	D2	D3	Rollout	D3 work
512	0.000	-0.000	-0.000	465
4,096	0.028	0.000	0.039	4,033
32,768	0.028	0.053	0.038	34,179

Appendix

Table 4: Noisy-grid mean utility under equal per-decision transition allowances. The final column is D3’s actual expected transitions per episode, so it can exceed one decision’s allowance. These are bounded approximations with a declared greedy fallback.

Method	Fresh B3	Fresh B5	Fresh B7	Original B7
D1	29.00	42.50	51.00	50.00
D2	56.00	77.00	79.00	77.50
D3	56.00	76.50	77.50	80.50
Rollout	56.00	77.00	85.00	82.50
CMI	27.00	49.00	63.50	56.00
Aligned D1	30.50	44.00	54.50	55.50

Appendix

Table 5: Shared-classifier accuracy (percent) on the noiseless CUBE-NM support. Fresh and original cohorts each have five fits. SEFA averages ten acquisition seeds per fit; other policies are deterministic. All methods use the same splits, groups and own-fit scorer. Original denotes reused data/scorer fits; neural baselines are newly trained.

Policy	$Q\to Q$	$Q\to G$	$G\to G$	$G\to Q$
D1	32.73	16.82	37.69	14.11
D2	38.59	16.52	44.44	16.22
D3	47.75	18.32	50.45	18.62
Rollout	53.30	16.67	52.10	15.92
CMI	52.85	20.42	51.50	17.42

Appendix

Table 6: Independent Common examiner panels: seven-query identification accuracy (%), averaging the two banks equally. Policies fit six requests per item and use concentration one. Each bank/direction scores 111 targets with three disjoint examiner requests. Q and G denote Qwen and GLM; arrows identify fitting and examination.

Allowance	D2	Rollout	D3 (orig.)	D3+ID	D3+ID+R	Work, orig.	Work, ID+R
512	.0005	$-$ .0001	$-$ .0001	.0002	.0005	465	481
4,096	.0280	.0387	.0004	.0280	.0280	4,033	4,872
32,768	.0280	.0379	.0529	.0399	.0403	34,179	32,997

Appendix

Table 7: Noisy-grid exact utility under the same per-decision allowances, 16 cells and ten fits. D3 (orig.) is the original bounded planner; ID adds iterative deepening and R adds path reuse. D2 and rollout include path reuse, which moves no mean by more than .0004. Work is expected online transitions per episode.

$B$	Allowance	D2	Rollout	D3 (orig.)	D3+ID	D3+ID+R	Work	H3
3	512	31.0	31.0	31.0	31.0	31.0	1,024	0
3	4,096	65.5	47.5	43.5	65.5	65.5	8,192	0
3	32,768	72.0	72.0	43.5	72.0	72.0	45,035	0
7	512	55.5	55.5	55.5	55.5	55.5	3,074	0
7	4,096	98.0	82.5	61.5	98.0	98.0	24,576	0
7	32,768	99.5	98.5	61.5	99.5	99.5	166,900	100

Appendix

Table 8: Classifier-aligned noiseless CUBE-NM accuracy (%) under the same allowances, averaging five own-classifier fits. Columns follow Table 7 . Work is mean online transitions per episode for D3+ID+R; H3 counts its decisions that complete depth three. Path reuse changes accuracy only for rollout at B7 and 32,768 (98.0 to 98.5).

Study	Allowance	Contrast	Mean [adjusted interval]
Noisy	4,096	D3+ID+R $-$ D2+R	.0000 [.0000, .0000]
Noisy	4,096	D3+ID+R $-$ rollout+R	$-$ .0107 [ $-$ .0160, $-$ .0054]
Noisy	32,768	D3+ID+R $-$ D2+R	+.0123 [+.0074, +.0171]
Noisy	32,768	D3+ID+R $-$ rollout+R	+.0024 [ $-$ .0074, +.0121]
Noisy	4,096	D3+ID $-$ D3 (orig.)	+.0276 [+.0239, +.0312]
Noisy	32,768	D3+ID $-$ D3 (orig.)	$-$ .0130 [ $-$ .0167, $-$ .0092]

Appendix

Table 9: Prespecified primary contrasts (D3+ID+R against D2+R and rollout+R) and the deepening effect on the original bounded D3. Noisy: ten replicate-block means, Bonferroni-6 (primary) or Bonferroni-12 (factorial)

t_{9}

intervals. Aligned: five paired fits, Bonferroni-6

t_{4}

intervals, percentage points. Contrasts at 512 are zero or include zero and are not shown.

Method	Original	Context	Permuted	Context
D2	.0280	0	.0280	0
Rollout	.0379	1	.0379	1
D3, D1 fallback	.0529	53	.0335	0
Deepening, value order	.0403	7	.0389	0
Deepening, catalogue order (post hoc)	.0587	51	.0353	0
Deepening, latent order	.0611	50	.0610	50

Appendix

Table 10: Noisy-grid utility at 32,768 transitions per decision, 160 fits, with the original catalogue (context listed first) and a permuted catalogue (context listed last). Context counts the fits whose root decision acquires context. Only the latent-information ordering keeps its utility when context moves.

Study	Allowance	Contrast	Mean [adjusted interval]
Noisy	4,096	Latent $-$ D2	.0000 [.0000, .0000]
Noisy	4,096	Latent $-$ rollout	$-$ .0107 [ $-$ .0164, $-$ .0050]
Noisy	4,096	Latent $-$ value order	.0000 [.0000, .0000]
Noisy	32,768	Latent $-$ D2	+.0331 [+.0275, +.0387]
Noisy	32,768	Latent $-$ rollout	+.0232 [+.0116, +.0348]
Noisy	32,768	Latent $-$ value order	+.0208 [+.0170, +.0246]

Appendix

Table 11: Latent-ordering contrasts. Noisy and aligned B7 rows are the prespecified primary family (Bonferroni-9

t_{9}

and

t_{4}

; aligned in percentage points). Permuted and invariance rows are prespecified secondary contrasts (Bonferroni-15

t_{9}

). Contrasts at 512 are zero or include zero.

Figure 9: Noisy-grid mean exact utility (160 fits) against expected search transitions per episode, at per-decision allowances from 512 to 131,072. Below 8,192 both deepening orders coincide with D2. Dashed lines are unbounded references.

Method	Utility	Expected queries	Work	Same, context last
D1	.0000	0.00	97	yes
CMI	$-$ .0780	3.00	1,683	yes
LIG	$-$ .0541	3.00	1,533	yes
LIG-stop	.0431	0.87	1,573	yes
D2	.0280	0.68	14,856	yes
Rollout	.0379	0.96	17,782	yes

Appendix

Table 12: Noisy CUBE-NM grid, 16 cells and ten fits, exact generator utility. LIG and CMI spend the whole budget and may abstain at the end. The last column reports whether the context-last catalogue gives the same mean to four decimals; fixed-subset and context-first controls were not rerun.

Method	Original	Context last
D1	.1077	.1071
LIG	.0428	.0898
LIG-stop	.0722	.1132
CMI	.2814	.2820
Rollout-LIG	.3751	.3550
Rollout	.4333	.4310

Appendix

Table 13: Context-parity controls, eight cells and ten fits, exact generator utility under the original and context-last catalogues. The fitted model is a saturated joint, so the latent component is the whole feature vector and every feature carries the same latent information.

$B$	D1	CMI	D2	D3	Rollout	LIG	RL (MAP)	Al. D3	RL (aligned)
3	28.0	31.5	53.0	53.0	53.0	60.0	54.5	72.0	68.5
5	42.0	37.0	74.5	75.5	76.5	90.5	73.0	100.0	99.0
7	50.0	56.0	77.5	80.5	82.5	96.0	86.0	100.0	99.5

Appendix

Table 14: Noiseless CUBE-NM own-classifier accuracy (%), five fits, at each hard budget

B

. Comparators are the frozen primary policies on the same fits and scorers. The last two columns use the classifier-aligned terminal.

Family	Study	Contrast	Mean [adjusted interval]
Primary	Noisy	Rollout-LIG $-$ D3	$-$ .0031 [ $-$ .0080, +.0018]
Primary	Noisy	Rollout-LIG $-$ rollout	+.0243 [+.0136, +.0351]
Primary	Noisy	LIG-stop $-$ D1	+.0431 [+.0368, +.0494]
Primary	Noisy	LIG-stop $-$ CMI	+.1211 [+.1137, +.1285]
Primary	CUBE-NM B7	LIG $-$ CMI	+40.0 [+9.2, +70.8]
Primary	CUBE-NM B7	LIG $-$ D3	+15.5 [+10.6, +20.4]

Appendix

Table 15: Prespecified contrasts. Primary: noisy grid Bonferroni-4

t_{9}

over ten replicate blocks; CUBE-NM B7 Bonferroni-3

t_{4}

over five fits, in percentage points. Secondary families use the same adjustments and are descriptive. Context-last noisy contrasts equal the original ones and are omitted.

Planner	Questions asked, in order (answer)	Outcome
CUBE-NM: class 6, context 1
LIG	context; block 1 positions 6 $\star$ , 5, 8 $\star$ , 9, 7 $\star$ , 2	correct
D3	context; block 1 positions 6 $\star$ , 5, 3, 1; block 0 positions 0, 1	wrong
Common: the item is Otter
CMI	found in the wild? yes; animal? yes; aquatic? yes; durable? no; savory? no; flies? no; terrestrial? no	Blobfish
D3	living organism? yes; aquatic? yes; cold climates? yes; eaten for sustenance? no; terrestrial? no; North America? yes; smaller than a laptop? yes	Otter

Appendix

Table 16: Decision traces replayed from saved runs. CUBE-NM: fit 52142, B7;

\star

marks a position that carries the class in that context. Common: Qwen bank, fit and examiner, B7. On this examiner replicate CMI alone is right on 21 items and D3 alone on 7; we show one of each.

Setting	Contrast	Mean [interval]
Noisy grid	Rollout-ECED $-$ Rollout-LIG	$-$ .0035 [ $-$ .0081, +.0011]
Noisy grid	ECED-stop $-$ LIG-stop	$-$ .0097 [ $-$ .0146, $-$ .0047]
Noisy grid	Rollout-ECED $-$ exact D3	$-$ .0065 [ $-$ .0120, $-$ .0011]
CUBE-NM $B=7$	LIG $-$ CMI (context last)	+40.0 [+9.2, +70.8]
CUBE-NM $B=7$	LIG $-$ D3 (context last)	+15.5 [+10.6, +20.4]
CUBE-NM $B=7$	LIG, context last $-$ original	0.0 [0.0, 0.0]

Appendix

Table 17: Primary contrasts with family-wise 95% intervals: three ECED contrasts (ten replicate blocks, Bonferroni-3

t_{9}

), three context-last CUBE-NM contrasts in accuracy points (five fits, Bonferroni-3

t_{4}

), and three UCT contrasts at the best exploration constant for each allowance (Bonferroni-3

t_{9}

		Noisy	Parity	Parity, last	B3	B5	B7
Spend budget	ECED	$-$ .0716	.2511	.2609	57.5	93.0	95.5
	LIG	$-$ .0541	.0428	.0898	60.0	90.5	96.0
Exact stopping	ECED	.0334	.2708	.2807	–	–	–
	LIG	.0431	.0722	.1132	–	–	–
One-step rollout	ECED	.0587	.4933	.4933	68.5	98.0	100.0
	LIG	.0622	.3751	.3550	68.5	99.0	99.5

Appendix

Table 18: ECED against the latent-information score under the same wrappers. Noisy grid and context parity: mean exact utility (160 and 80 fits). CUBE-NM: own-classifier accuracy (%, five fits) for spend-budget and, in the rollout rows, the classifier-aligned terminal. Exact depth is D3 on the noisy grid, full-horizon

D_{B}

on parity and aligned D3 on CUBE-NM. Noisy-grid values are identical under the context-last catalogue.

	$B=3$		$B=5$		$B=7$
Method	orig.	last	orig.	last	orig.	last
LIG	60.0	60.0	90.5	90.5	96.0	96.0
Rollout-LIG (MAP)	54.5	54.5	73.0	73.0	86.0	86.0
Rollout-LIG (aligned)	68.5	68.5	99.0	99.0	99.5	99.5
ECED	57.5	57.5	93.0	93.0	95.5	95.5
Rollout-ECED (aligned)	68.5	68.5	98.0	98.0	100.0	100.0

Appendix

Table 19: Noiseless CUBE-NM own-classifier accuracy (%, five fits) with the context listed first (orig.) and last. Aligned D3 was not rerun at

B=7

with the context last (n.r.) because of its cost.

Allowance	.03	.1	.3	1	.03	.1	.3	1	Root stop	Aband.
	Rollout leaves, $c=$				Stop leaves, $c=$
512	.0000	.0000	.0000	.0000	.0000	.0016	.0007	.0000	160	160
1,024	.0000	.0000	.0000	.0000	–	–	–	–	160	153
2,048	.0076	.0092	.0028	.0002	–	–	–	–	129	228
4,096	.0192	.0254	.0285	.0059	.0000	.0020	.0008	.0000	117	179
8,192	.0255	.0284	.0311	.0265	–	–	–	–	120	171

Appendix

Table 20: UCT on the noisy grid: mean exact utility (160 fits) by leaf value and exploration constant

c

. Rollout leaves use the exact D1-policy continuation. The last two columns are for rollout leaves with the best

c

: fits whose root decision is to stop, and decisions whose last simulation was abandoned at the allowance.

Dataset	D1	CMI	D2	D3	Rollout	RL	RE	Deepen
Iris	.5138	.6144	.4507	.4505	.4507	.5855	.6036	.4506
Wine	.7985	.7756	.7739	.8016	.8075	.8056	.7987	.8056
Breast cancer	.7985	.7829	.7926	.7844	.7820	.7873	.7918	.7858
Digits	.0000	.0202	.0024	.0656	.0068	.0604	.0667	.0604
Diabetes (sklearn)	.4822	.4798	.4865	.4865	.4863	.4875	.4891	.4865
Diabetes (3-class)	.6166	.5924	.6162	.6162	.6162	.6166	.6166	.6166

Appendix

Table 21: Held-out signed utility on nine local datasets: means over each dataset’s tasks and five splits (budget three, query cost .01). The last row is mean compilation transitions per fit. RL is Rollout-LIG and RE Rollout-ECED. Deepen runs exact D3 only where a fixed set of at most three queries beats Rollout-LIG by more than .002.

Family	Contrast	Scale	Mult.	Mean [interval]
Primary	D3 $-$ D2	held-out	3	+.0100 [ $-$ .0125, +.0324]
Primary	Rollout-LIG $-$ D3	held-out	3	+.0124 [ $-$ .0344, +.0591]
Primary	Deepen $-$ D3	held-out	3	$-$ .0021 [ $-$ .0090, +.0049]
Secondary	D3 $-$ D2	fitted	3	+.0075 [ $-$ .0103, +.0252]
Secondary	Rollout-LIG $-$ D3	fitted	3	$-$ .0067 [ $-$ .0141, +.0007]
Secondary	Deepen $-$ D3	fitted	3	$-$ .0011 [ $-$ .0028, +.0006]

Appendix

Table 22: Contrasts on the nine local datasets: dataset-level means,

t_{8}

intervals with the listed Bonferroni multiplicity. Across the 27 tasks, fitted and held-out D3

-

D2 have Spearman correlation .61 (

p=.0007

). The deepen-when-needed rule (Deepen) deepened at 93 of 649 compiled decisions and used 1.19 times D3’s work.

	Qwen		GLM		Pooled		Mixture
Method	Q	G	Q	G	Q	G	Q	G
D1	32.73	16.82	14.11	37.69	32.73	37.84	33.63	39.34
D2	38.59	16.52	16.22	44.44	38.89	41.14	40.54	43.39
D3	47.75	18.32	18.62	50.45	42.04	42.94	44.14	46.55
Rollout	53.30	16.67	15.92	52.10	45.65	40.69	47.15	43.69
CMI	52.85	20.42	17.42	51.50	41.59	41.29	43.99	42.19

Appendix

Table 23: Common identification accuracy (%, equal-bank mean) at seven queries for four fits, on Qwen (Q) and GLM (G) examiner answers. Pooled fits both families’ twelve answers per item; Mixture has one component per item and family and infers the family from the answers. Mixture D3 minus same-family D3:

-

3.60 [

-

12.54, +5.11] on Q and

-

3.90 [

-

10.96, +3.30] on G. Minus cross-family D3: +25.53 [+18.17, +32.43] and +28.23 [+19.67, +37.09]. Paired target bootstrap, Bonferroni-4, percentage points.

Fit	Examiner	Log loss	Brier	ECE
Qwen	Qwen	.092	.012	.069
Qwen	GLM	.320	.089	.063
GLM	Qwen	.287	.080	.103
GLM	GLM	.136	.028	.062
Pooled	Qwen	.114	.028	.068
Pooled	GLM	.148	.041	.059

Appendix

Table 24: Calibration of each fit’s answer channel on held-out examiner answers (equal-bank means over 111 items, 32 questions and three repeats). ECE uses ten equal-width bins. For the mixture, the channel is the family-averaged prediction before any answer is seen.

Fit $\to$ examiner	D3 $-$ D2, new	D3 $-$ rollout, new	D3 $-$ D2, with earlier	D3 $-$ rollout, with earlier	D3 (%)
Qwen $\to$ Qwen	$+3.7$ $[1.1,6.4]$	$-6.2$ $[-10.0,-2.5]$	$+5.2$ $[2.8,7.7]$	$-6.1$ $[-9.4,-2.8]$	33.9
Qwen $\to$ GLM	$+0.4$ $[-1.7,2.6]$	$+1.4$ $[-1.7,4.5]$	$+0.8$ $[-1.1,2.7]$	$+1.5$ $[-1.3,4.3]$	13.7
GLM $\to$ Qwen	$+1.7$ $[-1.3,4.6]$	$+0.2$ $[-3.4,3.7]$	$+1.9$ $[-0.4,4.3]$	$+0.9$ $[-2.5,4.1]$	12.9
GLM $\to$ GLM	$+4.5$ $[0.8,8.3]$	$-0.8$ $[-5.8,4.0]$	$+4.9$ $[1.7,8.3]$	$-1.0$ $[-5.4,3.0]$	39.7
Mixture $\to$ Qwen					32.2
Mixture $\to$ GLM					40.4

Appendix

Table 25: Powered panels at B7: held-out contrasts in points, with Bonferroni-8 intervals. New : the 598 units of the four new banks. With earlier : 820 units, adding the two earlier Common banks. Last column: D3 accuracy (%) on the new units.

Figure 10: Realized utility over 300 models at lookahead depth two. Each arrow runs from the expectation backup to the nested backup at one

(\alpha,\mu)

, with the lower-tail CVaR taken at the planner’s

\alpha

; higher is better on both axes. The nested backup raises the tail appreciably only at

\mu=.5

\alpha=.95

, and at

\mu=1

, where it gives up the mean.

	Noisy grid		CUBE-NM B7
Method	Generic (cv)	Structured	Generic (cv)	Structured	Generic (BIC)
CMI	$-$ .0743	$-$ .0780	56.0	56.0	43.0
D2	.0256	.0280	79.5	77.5	41.5
D3	.0671	.0653	80.0	80.5	45.0
Rollout	.0395	.0379	81.5	82.5	43.0
LIG	$-$ .0478	$-$ .0541	89.0	96.0	43.0

Appendix

Table 26: Planning with a generic latent-class fit that is not told the class–context structure. Noisy grid: mean exact utility over 160 fits, original catalogue. CUBE-NM: own shared-classifier accuracy (%) at seven queries, five fits; the MAP terminal except the aligned rows. Generic fits choose the number of components per class by 5-fold cross-validated likelihood (cv) or BIC; structured fits are those of Appendix I . CUBE-NM spends a fixed budget, so LIG-stop equals LIG there.

Family	Study	Contrast	Mean [adjusted interval]
Primary	Noisy, cv	Rollout-LIG $-$ D3	$-$ .0035 [ $-$ .0125, +.0055]
Primary	Noisy, cv	Rollout-LIG $-$ rollout	+.0240 [+.0151, +.0330]
Primary	Noisy, cv	LIG-stop $-$ CMI	+.1195 [+.1117, +.1273]
Primary	Noisy, cv	Rollout-LIG, generic $-$ structured	+.0013 [ $-$ .0026, +.0052]
Primary	CUBE-NM B7, cv	LIG $-$ CMI	+33.0 [+25.1, +40.9]
Primary	CUBE-NM B7, cv	LIG $-$ D3	+9.0 [+2.3, +15.7]

Appendix

Table 27: Prespecified generic-fit contrasts. Primary: noisy grid Bonferroni-4

t_{9}

over ten replicate blocks; CUBE-NM B7 Bonferroni-3

t_{4}

over five fits, in percentage points. Secondary families use the same adjustments and are descriptive. Context-last noisy contrasts equal the original ones and are omitted.

Gate	Cap	Utility	Work	Interrupted
D2 (default depth)	none	.5688	36	0
D3	none	.5788	197	0
Rollout-LIG	none	.5912	74	0
LIG-stop	none	.5604	10	0
D3, bounded	128	.3219	96	50,689
D3, latent deepening	128	.5693	85	50,776

Appendix

Table 28: The full controller on nine real datasets. Utility is held-out, averaged over datasets. Work is planning units per decision. Interrupted counts decisions whose search hit the cap.

Panel of Figure 11 — Figure 11: Probabilities read from four language models, against targets outside them. (a) The route; the prompt is paraphrased. (b) Fifteen attributes of 101 animals; base rate and class frequency are fitted leaving each animal out. (c) Pooled reads and pooled stated percentages, with perfect calibration dashed and marker area proportional to the requests in a bin. Token reads lie above the diagonal at every level. The stated lowest bin gives almost 0 and comes true 7% of the time. (d) D3 at budget three, over the real-data study’s splits. The pale bar at zero rows spans the four single models. (e) Seven questions. Counts use six fitting requests per item; reads use none. The pool includes both answering families. (f) Mean coverage is .88–.91 in every cell of this panel.

Suite	Rule	Rollout-LIG	D3 / DB	Rule $-$ Rollout-LIG	Rule $-$ D3 / DB
Noisy grid	.0650	.0622	.0653	$+.0028$ $[-.0011,.0067]$	$-.0003$ $[-.0025,.0020]$
Context parity	.7838	.3751	.7838	$+.4088$ $[.3763,.4412]$	.0000 (all fits)
Nine datasets	.5773	.5912	.5788	$-.0139$ $[-.0666,.0388]$	$-.0015$ $[-.0096,.0065]$
Work, noisy grid	915,264	61,110	738,271
Work, parity	4,218	1,070	1,796
Work, nine datasets	5,317	1,386	2,915

Appendix

Table 29: Depth from witnesses. Mean exact utility (real data: held out) and compilation transitions. Contrasts: Bonferroni-6; noisy grid and parity over ten replicate blocks, real data over nine datasets.

Figure 12: Mean regret against the best candidate, on a scale linear below

.01

and logarithmic above, by source and error radii, pooled over sizes and costs (108 cells per setting). Replay needs validation rows, so it has no exact-radii value.

Figure 13: Accuracy against transitions per episode on CUBE-NM. Rows: budgets; columns: feature counts. Lines join a method’s allowances; DIME (dashed) has no comparable work; dotted grey is the exact optimum.

Figure 14: Share of the gap from acting now to the oracle (real data: to perfect information). Floors: noisy grid and parity 0, CUBE-NM 12.5%, real data .222 held out. Oracles: .122, .795, 100%. D3 is full-horizon

D_{B}

on parity, where D2 is not a comparator; LIG-stop is LIG on CUBE-NM, which spends a fixed budget.

	Noisy grid					Nine datasets
Method	Coarse	Refined	Noise 1	Noise 3	Work	Coarse	Refined	Noise 1	Noise 3	Work
LIG-stop	.0431	.0069	.0089	.0000	3,947	.5604	.5125	.5273	.3680	491
Rollout-LIG	.0622	.0260	.0209	.0000	109,302	.5912	.5765	.5871	.5846	4,594
D2	.0280	.0280	.0280	.0280	17,471	.5688	.5688	.5688	.5688	1,483
D3	.0653	.0653	–	–	875,760	.5788	.5788	–	–	7,125
Witness rule	.0650	.0652	–	–	1,381,956	.5773	.5789	–	–	16,112

Appendix

Table 30: Refinement on the fitted suites: mean exact utility (true model; real data: held out), with mean transitions under the refined description. D3 and the witness rule were run only on the coarse and refined descriptions.

$k$	$\epsilon$	$N$	$D_{B-1}$	$D_{B}$	Fitted subset	Oracle subset
2	0.0	120	0.418	0.910	0.419	0.440
2	0.0	600	0.421	0.910	0.428	0.440
2	0.1	120	0.307	0.710	0.310	0.340
2	0.1	600	0.324	0.710	0.316	0.340
3	0.0	120	0.373	0.880	0.380	0.410
3	0.0	600	0.386	0.880	0.380	0.410

Appendix

Table 31: Context–parity exact signed utility over ten fits per cell. The budget is

B=k+1

D_{B}

uses the full query horizon;

D_{B-1}

is one step shorter. The final column gives the true-distribution optimal nonadaptive value, not a learned estimate.

$C$	$N$	$c$	$\sigma$	$B$	D2	D3	Rollout	Forced D1	Subset
2	120	0.01	0.1	3	0.000	0.024	0.000	-0.030	0.002
2	120	0.05	0.1	3	0.000	-0.006	-0.002	-0.143	0.000
2	120	0.1	0.1	3	0.000	0.000	0.000	-0.291	0.000
2	600	0.01	0.1	3	0.109	0.152	0.128	-0.010	0.053
2	600	0.05	0.1	3	0.000	0.042	0.035	-0.133	0.000
2	600	0.1	0.1	3	0.000	0.000	0.000	-0.285	0.000

Appendix

Table 32: Exact generator signed utility, averaged over ten fresh fits in each prespecified cell. Forced D1 spends its query budget; Subset is optimal under the fitted model among nonadaptive subsets. All 16 cells are retained.

Cell	CMI	Context-first	Forced D1	Subset	D3
0	0.004	-0.010	-0.030	0.002	0.024
1	-0.122	-0.050	-0.143	0.000	-0.006
2	-0.263	-0.100	-0.291	0.000	0.000
3	0.060	-0.010	-0.010	0.053	0.152
4	-0.052	-0.050	-0.133	0.000	0.042
5	-0.198	-0.100	-0.285	0.000	0.000

Appendix

Table 33: Additional noisy-grid controls, as exact generator signed utility. Cells follow Table 32 . CMI and forced D1 spend the query budget; their negative utility can include a cost of acquisition followed by abstention. Context-first pays for context and then follows the ordinary stopping D1 policy. These are fixed controls, not policies tuned after evaluation.

Cell	Oracle D2	Oracle D3	Oracle subset	D2 gap	D3 gap	Rollout gap
0	0.143	0.208	0.074	0.000	-0.018	0.000
1	0.000	0.080	0.000	0.000	0.009	0.004
2	0.000	0.000	0.000	0.000	0.000	0.000
3	0.141	0.198	0.061	0.032	0.055	0.024
4	0.000	0.081	0.000	0.000	0.026	0.017
5	0.000	0.000	0.000	0.000	0.000	0.000

Appendix

Table 34: Known-generator values and learned-policy optimism (fitted minus exact generator utility) for all noisy cells in the order of Table 32 . Zero optimism for an abstaining policy does not imply that its model captures the available acquisition value.

Study	Contrast	Mean	Adjusted interval	Scale; family
Noisy grid	D3-D2	0.037	[0.032, 0.042]	utility; 2
Noisy grid	D3-rollout	0.027	[0.018, 0.037]	utility; 2
Noisy cap 512	D3-D2	-0.000	[-0.001, 0.000]	utility; 6
Noisy cap 512	D3-rollout	0.000	[0.000, 0.000]	utility; 6
Noisy cap 4,096	D3-D2	-0.028	[-0.031, -0.024]	utility; 6
Noisy cap 4,096	D3-rollout	-0.039	[-0.044, -0.033]	utility; 6

Appendix

Table 35: Prespecified aggregate paired contrasts; intervals use ten replicate-block means. Final column gives the outcome scale and Bonferroni family size. Noisy grid cells have fixed equal weights.

Policy	MLP $B=3$	MLP $B=5$	MLP $B=7$	Internal $B=7$
D1	28.00	42.00	50.00	57.00
D2	53.00	74.50	77.50	100.00
D3	53.00	75.50	80.50	100.00
Rollout	53.00	76.50	82.50	100.00
CMI	31.50	37.00	56.00	67.00
A1	31.00	43.50	55.50	55.50

Appendix

Table 36: Fresh noiseless CUBE-NM: exact 40-pattern accuracy (%) averaged over five independent policy/classifier fits. A1–A3 use classifier-aligned terminals at the corresponding depths. MLP scores use each fit’s own frozen classifier; internal is MAP accuracy on the acquired history, a distinct diagnostic.

Contrast	Mean	Adjusted interval
D3-D2	3.000	[-6.600, 12.600]
D3-rollout	-2.000	[-10.518, 6.518]
A3-D3	19.500	[13.726, 25.274]

Appendix

Table 37: Fresh noiseless primary contrasts at seven queries, in accuracy percentage points. Five paired own-classifier fits; Bonferroni-3

t_{4}

intervals.

Policy	Own	Other	$f_{1}$	$f_{2}$	$f_{3}$	$f_{4}$	$f_{5}$
D1	50.00	47.50	46.50	48.50	50.50	48.00	46.50
D2	77.50	77.38	75.00	76.50	81.50	78.00	76.00
D3	80.50	79.38	78.50	77.00	83.50	81.50	77.50
Rollout	82.50	83.12	81.00	81.50	83.50	83.50	85.50
CMI	56.00	56.50	58.50	54.00	58.00	55.50	56.00
A1	55.50	48.38	51.50	54.00	45.50	51.00	47.00

Appendix

Table 38: Seven-query scoring across independently trained classifiers. Own is the five diagonal entries; Other is the twenty off-diagonal entries. Scorer columns average all five policy fits for that classifier. All values are percentages; this matrix is descriptive and uses the same finite task.

$B$	Allowance	Policy	Accuracy (%)	Transitions	Time (ms)
3	512	D1	28.00	381	0.74
3	512	D2	28.00	1,138	5.85
3	512	D3	28.00	1,252	6.07
3	512	Rollout	28.00	1,246	7.20
3	4,096	D1	28.00	381	0.81
3	4,096	D2	56.00	8,415	11.19

Appendix

Table 39: Fresh noiseless accuracy and actual work under equal transition allowances. Accuracy averages five own-classifier fits. Time is the sum of recorded acquisition-decision times along an episode, in milliseconds.

Allowance	Contrast	Mean	Adjusted interval
512	D3-D2	0.000	[0.000, 0.000]
512	D3-rollout	0.000	[0.000, 0.000]
4,096	D3-D2	-18.000	[-34.451, -1.549]
4,096	D3-rollout	-8.500	[-13.351, -3.649]
32,768	D3-D2	-21.000	[-36.625, -5.375]
32,768	D3-rollout	-16.000	[-27.882, -4.118]

Appendix

Table 40: Noiseless seven-query contrasts under equal transition allowances, in percentage points. Five paired fits; Bonferroni-6 intervals.

$B$	Allowance	Policy	Accuracy (%)	Transitions	MLP histories
3	512	D1	31.00	381	381
3	512	D2	31.00	1,272	1,253
3	512	D3	31.00	1,272	1,360
3	512	Rollout	31.00	1,272	1,199
3	4,096	D1	31.00	381	381
3	4,096	D2	65.50	8,436	7,416

Appendix

Table 41: Matched search with classifier-aligned terminals for all four methods. Accuracy (%) averages five own-classifier fits; transitions and planning-time classifier-history evaluations are actual episode means. Both search and classifier prediction caches are cleared between decisions. These are equal transition ceilings, not identical realized work or FLOPs.

Allowance	Contrast	Mean	Adjusted interval
512	D3-D2	0.000	[0.000, 0.000]
512	D3-rollout	0.000	[0.000, 0.000]
4,096	D3-D2	-36.500	[-48.986, -24.014]
4,096	D3-rollout	-21.000	[-30.075, -11.925]
32,768	D3-D2	-38.000	[-49.758, -26.242]
32,768	D3-rollout	-36.500	[-49.562, -23.438]

Appendix

Table 42: Aligned-objective seven-query contrasts, in percentage points. Five paired fits and Bonferroni-6 intervals. This additional comparison was specified after the noisy-grid analysis and before any result from the aligned matched stage.

Direction	Contrast	Mean	Adjusted interval
$Q\to Q$	D3-D2	9.159	[4.204, 14.414]
$Q\to Q$	D3-rollout	-5.556	[-12.312, 1.314]
$Q\to G$	D3-D2	1.802	[-2.402, 5.856]
$Q\to G$	D3-rollout	1.652	[-4.167, 7.508]
$G\to Q$	D3-D2	2.402	[-2.703, 7.207]
$G\to Q$	D3-rollout	2.703	[-4.505, 9.873]

Appendix

Table 43: Common primary contrasts at seven queries, six references and concentration one, in percentage points. Category-stratified paired target resampling; Bonferroni-8 intervals conditional on the two banks and fitted panels.

Bank generator	Policy	$Q\to Q$	$Q\to G$	$G\to G$	$G\to Q$
Qwen	D1	35.44	15.02	35.74	12.61
Qwen	D2	44.44	15.32	42.94	16.52
Qwen	D3	56.46	16.22	49.55	17.72
Qwen	Rollout	70.27	13.51	54.35	15.92
Qwen	CMI	69.37	18.92	53.15	18.32
GLM	D1	30.03	18.62	39.64	15.62

Appendix

Table 44: Common seven-query accuracy (%) by question-bank generator. Six fitting requests and concentration one; neither bank was selected using scores.

$\alpha$	Policy	$Q\to Q$	$Q\to G$	$G\to G$	$G\to Q$
0.2	D1	32.73	16.67	37.99	14.11
0.2	D2	38.59	16.82	44.44	16.52
0.2	D3	47.75	18.32	50.60	19.22
0.2	Rollout	53.30	16.67	53.15	15.32
0.2	CMI	54.80	19.22	50.45	17.57
1.0	D1	32.73	16.82	37.69	14.11

Appendix

Table 45: Common seven-query accuracy (%) under all prespecified prior concentrations, with six fitting requests and equal bank weights. Concentrations are not selected using examiner outcomes.

Fit	$B$	Policy	$Q\to Q$	$Q\to G$	$G\to G$	$G\to Q$
n6	3	D1	7.06	4.95	6.91	4.50
n6	3	D2	7.06	4.95	6.91	4.50
n6	3	D3	7.06	4.95	6.91	4.50
n6	3	Rollout	7.06	4.95	6.91	4.50
n6	3	CMI	7.06	5.26	7.21	5.11
n6	5	D1	20.87	13.21	20.12	10.36

Appendix

Table 46: Common exact-policy accuracy (%) with six fitting requests, averaged across the two fixed banks. Examiner requests are disjoint from all fitting.

Allowance	Policy	$Q\to Q$	$Q\to G$	$G\to G$	$G\to Q$
512	D1	32.73	16.82	37.69	14.11
512	D2	32.58	16.97	38.89	14.11
512	D3	32.73	16.67	37.54	14.11
512	Rollout	32.73	16.67	37.54	14.11
4,096	D1	32.73	16.82	37.69	14.11
4,096	D2	38.59	16.52	44.44	16.22

Appendix

Table 47: Common seven-query accuracy (%) under equal per-decision transition allowances, averaged over both banks. The policies are bounded approximations; actual work is reported separately.

Allowance	Policy	$Q\to Q$	$Q\to G$	$G\to G$	$G\to Q$
512	D1	406	406	406	406
512	D2	3,008	3,008	3,008	3,008
512	D3	2,996	2,996	2,996	2,996
512	Rollout	3,089	3,089	3,088	3,088
4,096	D1	406	406	406	406
4,096	D2	21,060	21,060	21,060	21,060

Appendix

Table 48: Actual Common transitions per seven-query episode under the shared per-decision allowances. Values average both banks. Different examiner trajectories can induce different work for the same frozen policy; none of these counts is FLOPs.

Bank generator	Direction	Binary NLL	Brier	All-32 accuracy
Qwen	$Q\to Q$	0.103	0.016	82.58
Qwen	$Q\to G$	0.470	0.140	26.43
Qwen	$G\to G$	0.177	0.043	70.27
Qwen	$G\to Q$	0.412	0.124	30.63
GLM	$Q\to Q$	0.082	0.008	42.94
GLM	$Q\to G$	0.171	0.038	23.12

Appendix

Table 49: Common channel diagnostics at six references and concentration one. Binary NLL (nats) and Brier score average question entries. All-32 accuracy (%) uses every answer and is not an upper bound on subset-policy accuracy.

Direction	Policy	Fitted	Ind. marg.	Examiner	Optimism
$Q\to Q$	D3	32.69	47.75	47.75	-15.05
$Q\to Q$	Rollout	36.15	53.03	53.30	-17.16
$Q\to G$	D3	32.69	18.13	18.32	14.38
$Q\to G$	Rollout	36.15	16.87	16.67	19.48
$G\to G$	D3	36.55	50.70	50.45	-13.90
$G\to G$	Rollout	37.70	52.07	52.10	-14.41

Appendix

Table 50: Common complete-policy value diagnostics, averaged over both banks at seven queries, six references and concentration one. Fitted, independent empirical marginals (Ind. marg.) and examiner accuracy are percentages; optimism is their fitted-minus-examiner difference in percentage points. The empirical-marginal calculation reuses the examiner data and is not a population dependence estimate.

Method	B3	B5	B7
D1	28.00	42.00	50.00
D2	53.00	74.50	77.50
D3	53.00	75.50	80.50
Rollout	53.00	76.50	82.50
CMI	31.50	37.00	56.00
Aligned D1	31.00	43.50	55.50

Appendix

Table 51: Reused cohort shared-classifier accuracies (percent).

Method	B3	B5	B7
D1	29.00	42.50	51.00
D2	56.00	77.00	79.00
D3	56.00	76.50	77.50
Rollout	56.00	77.00	85.00
CMI	27.00	49.00	63.50
Aligned D1	30.50	44.00	54.50

Appendix

Table 52: Fresh cohort shared-classifier accuracies (percent).

Contrast	Difference [adjusted interval]
D3–DIME	+44.00 [25.13, 62.87]
D3–SEFA	+31.95 [24.01, 39.89]
Aligned D3–DIME	+66.50 [50.28, 82.72]
Aligned D3–SEFA	+54.45 [52.26, 56.64]

Appendix

Table 53: Fresh-cohort primary learned-baseline contrasts, percentage points. Paired t intervals use four degrees of freedom and Bonferroni-four adjustment.

Cohort	$B$	GDFS	DIME	SEFA	CMI	D2	D3	Rollout
Fresh	3	22.5	23.5	28.9	27.0	56.0	56.0	56.0
Fresh	5	33.0	26.5	36.5	49.0	77.0	76.5	77.0
Fresh	7	38.0	33.5	45.5	63.5	79.0	77.5	85.0
Original	7	39.0	33.0	46.4	56.0	77.5	80.5	82.5
Fresh B7		GDFS $-$ DIME: +4.5 [ $-$ 10.5, +19.5]
Fresh B7		GDFS $-$ SEFA: $-$ 7.5 [ $-$ 13.4, $-$ 1.7]

Appendix

Table 54: GDFS in the learned-acquisition comparison on noiseless CUBE-NM: own shared-MLP accuracy (%) on the 40-pattern support, five fits per cohort. GDFS uses the pinned AFABench implementation and CUBE configuration under the same splits, group costs and scorers as DIME and SEFA; comparators are the frozen results of that stage. Planner columns use the fitted MAP terminal. Contrasts are paired over five fits, Bonferroni-6

t_{4}

, in percentage points.

Method	Budget	External	Private	Information	Test
DIME	3	23.50	18.50	25.00	21.50
DIME	5	26.50	25.50	36.50	25.00
DIME	7	33.50	32.50	37.50	34.90
SEFA	3	28.95	29.35	31.15	27.62
SEFA	5	36.55	39.65	43.98	38.53
SEFA	7	45.55	47.65	51.32	46.10

Appendix

Table 55: Descriptive diagnostics for the five additional fits (percent). External and private predictors use the same selected masks. Information denotes mean maximum true class posterior among compatible support atoms; it is an evaluation-only diagnostic. Test uses the reserved 200-row split.

Method	Qwen fit	GLM fit
D1	9.46	7.81
D2	9.61	8.26
D3	8.71	6.76
Rollout	5.86	6.16
CMI	8.71	6.76

Appendix

Table 56: Nemotron examination at seven queries, equal-bank accuracy (%). Policies and banks are unchanged.

Active Information Gathering

Active information gathering studies how agents choose questions, tests, or observations that reduce uncertainty and improve decisions under interaction costs.

See all 27 papers

PrimeSeeker: Capability-Oriented Supervision for Deep Search Agents
Sep 20, 2026Linzhi Peng, Hanting Chen, Heng Chang +3Deep Research Agents Agentic Search
Beihang University · Huawei Technologies Ltd. · Tsinghua University
Agent Step Value: Auditing Evaluation-Channel Sensitivity in Black-Box Agent Traces
Jul 5, 2026Andrew Zhang, Chengzhan LiLLM Agent Evaluation AI Agent Evaluation
KTH Royal Institute of Technology · University of Electronic Science and Technology of China
Ask Now, Use Later: Benchmarking the Proactivity Gap in Long-Lived LLM Agents
May 27, 2026Bin Wu, Guanyun Zou, Bingbing Wang +2Proactive Assistance LLM Agent Memory
Beijing University of Posts and Telecommunications · Nanjing University of Aeronautics and Astronautics · Harbin Institute of Technology (Shenzhen) +1

Sep 20, 2026J/K move · Enter open · S save

PrimeSeeker: Capability-Oriented Supervision for Deep Search Agents

Linzhi Peng, Hanting Chen, Heng Chang, Ke Cheng+2

Beihang University · Huawei Technologies Ltd. · Tsinghua University

Deep Research Agents Agentic Search Web Search Agents Synthetic Data Generation

Open paper AskPDF arXiv

Large language model search agents are often trained with synthetic questions whose difficulty is increased through larger evidence graphs, additional hops, and longer trajectories. These global properties, however, are only indirect proxies for the local retrieval capabilities required during search. To address this mismatch, we introduce latent anchor reasoning, which consists of resolving an unnamed retrieval anchor from descriptive specifications and transferring the recovered anchor into a subsequent information demand. This primitive retrieval unit decomposes deep search into chains of coupled operations and organizes question construction around anchor resolution and relation transfer, without prescribing a canonical search path. Based on this formulation, we propose PrimeSeeker, a capability-oriented framework that constructs web-grounded anchor structures and jointly derives a question and a reference evidence skeleton. The skeleton preserves supporting evidence from construction and guides expert generation through extractive highlights of current tool observations. These highlights are removed before supervised fine-tuning, while the skeleton is subsequently reused to audit reference-step coverage for reinforcement-learning rewards. We construct 9,221 expert trajectories, training a 30B search agent. Across five deep-search benchmarks, PrimeSeeker achieves strong performance, while reference-step optimization further improves the supervised policy. The resulting trajectories exhibit low retrieval redundancy, and fixed-budget evaluation shows strong solution coverage with substantially fewer tool calls than long-horizon systems.

Abstract

Active Information Gathering

Explore similar work

PrimeSeeker: Capability-Oriented Supervision for Deep Search Agents

Agent Step Value: Auditing Evaluation-Channel Sensitivity in Black-Box Agent Traces

Ask Now, Use Later: Benchmarking the Proactivity Gap in Long-Lived LLM Agents