# DEEPTHINK: DEEPTHINK

> Living preprint · as of 2026-10-25 · map node X-085 (experiment) · interim record, not a final result.

<!-- lint:exempt:start -->
- [X-085] DEEPTHINK · DEEPTHINK · question: Does a 1024-token thinking budget on the deep tier (arm T: cut 6 + Qwen3.6-35B, thinking on) lift V-para vs the same deep tier without thinking (D, DEEPVPARA), and at what time cost? · result: thinking (1024) vs no thinking: raw +0.075 (+0.040…+0.110); floor-adjusted +0.093 (+0.050…+0.137) · set V-para (V-para); v0.1 smoke 2 items, judge Claude Opus, Claude Haiku (L1), n 51 judged ids (58 requests) per seed, seeds 42, 43, CI +0.040…+0.110 · verdict: adopted (lever-adopted (thinking on the deep tier adopted; budget later moved 1024 → 512 by DT512); L1 PROMOTED on the Δ clause)

<!-- lint:exempt:end -->
## Outgoing

- led to: [X-095] DT512: DT512 (thinking budget curve) → https://r4topunk.github.io/boar-research/map/x-095.md

## Incoming

- led to: [D-039] Stop answer-identical 4B arms; reopen the deep tier as the primary quality lever (the deep tier reopened as the quality lever) → https://r4topunk.github.io/boar-research/map/d-039.md
- ruled out: [X-097] GEMMATHINK: GEMMATHINK (a faster deep model lost quality) → https://r4topunk.github.io/boar-research/map/x-097.md

Read chapter 6, The deep tier: https://r4topunk.github.io/boar-research/research/deep-tier.md

## Facts cited {#facts}

Resolve every [F-…] marker in https://r4topunk.github.io/boar-research/data/facts.json. Each line gives the claim, the number with its metric family, set and status, judge, n, seeds, CI, date and caveat.

<!-- lint:exempt:start -->
- [F-007] Letting the deep tier think for up to 1024 tokens before answering. raw 0.542 (0.513–0.571); floor-adjusted 0.422 (0.385–0.459); win 0/51; desktop p50 ≈ 143 s (judge.v1 ratio) · set V-para (dev-style, development data) · judge Claude Opus (judge.v1, both orders) · n 51 · seeds 2 · CI 95 % item bootstrap · date 2026-10-08
- [F-008] Thinking is the largest single measured gain on the bounty-style set. thinking (1024) vs no thinking: raw +0.075 (+0.040…+0.110); floor-adjusted +0.093 (+0.050…+0.137) (judge.v1 ratio) · set V-para (dev-style, development data) · judge Claude Opus (judge.v1, both orders) · n 51 · seeds 2 · CI paired 95 % bootstrap · date 2026-10-08
- [F-617] Process review: 65/65 criterion shas verify; DEEPTHINK yielded 0.30 floor-adj points per 1k judge calls (×10 the deep-tier entry arm) 65/65 · set UNKNOWN (status UNKNOWN) · judge UNKNOWN · n UNKNOWN · seeds UNKNOWN · CI not computed · date 2026-10-10 · caveat: Verification note: Line refs are :13 (65/65) and :18 (DEEPTHINK). Unit: 0.30 floor-adj per 1k calls = 30.5 "points" (100×Δ) per 1k calls in public BAR-TABLE:106.
- [F-1079] Large levers survive multiplicity correction. X2b z ≈ 5.5, LITE5-180 z ≈ 6, DEEPTHINK z ≈ 4.2 · set UNKNOWN (status UNKNOWN) · judge UNKNOWN · n UNKNOWN · seeds UNKNOWN · CI ≈ 60 judged comparisons · date 2026-10-16 · caveat: citation fix; ≈ 60 judged comparisons, Bonferroni over 100; the bigger risk is selection on the same dev sets

<!-- lint:exempt:end -->
---

Cite as: r4to, independent researcher. RatLab — A Boar Research (2026), research record 1.0, as of 2026-10-25. Text and data CC BY 4.0. Fact ids or a link must travel with a quote. How to cite: https://r4topunk.github.io/boar-research/cite/ · https://r4topunk.github.io/boar-research/CITATION.cff.

AI-reviewed, not clinically reviewed. Nothing here is medical advice.