1# PRISME -- literature verification log and novelty audit23UQO Working Paper 31 (Simon-Pierre Boucher). Audit date: 2026-10-29.4Companion files: `papers.bib` (77 entries), `comparison_matrix.csv` (67 rows).56Method. Every arXiv reference was resolved through the arXiv Atom API7(`https://export.arxiv.org/api/query?id_list=...`): title, full author list,8first-submission and latest-version dates, `comment` / `journal_ref` / `doi`9fields. Code and licences were read from the GitHub REST API10(`api.github.com/repos/<owner>/<repo>`, SPDX id) or from the LICENSE file.11Mechanism details for the closest neighbours were read from the arXiv HTML12full text (v-numbers given below). Nothing in this audit relies on memory13alone; where a venue is "common knowledge" but was not confirmed by a primary14source, it is flagged **NOT verified** here and in the `note` field of the15bib entry.1617---1819## 1. Per-reference verification log2021### 1.1 The twelve requested core references2223| # | Reference | What was fetched | Confirmed | Caveats |24|---|---|---|---|---|25| 1 | Universal Transformers, arXiv:1807.03819 | arXiv API; GitHub `tensorflow/tensor2tensor` | Dehghani, Gouws, Vinyals, Uszkoreit, Kaiser; v1 2018-07-10, v3 2019-03-05; comment "Published at ICLR2019". Weight-shared recurrence over depth + ACT-style per-position halting. Code in tensor2tensor, Apache-2.0. | -- |26| 2 | PonderNet, arXiv:2107.05407 | arXiv API | Banino, Balaguer, Blundell; v1 2021-07-12, v2 2021-09-02; comment "8th ICML Workshop on Automated Machine Learning (2021)". Probabilistic halting with geometric prior, KL regulariser. | No official DeepMind code found. |27| 3 | Perceiver AR, arXiv:2202.07765 | arXiv API; GitHub `google-research/perceiver-ar` | Hawthorne et al. (15 authors); v1 2022-02-15, v2 2022-06-14; comment "ICML 2022". Few latents (last positions) cross-attend causally to the long input; latents re-created per forward. Apache-2.0. | -- |28| 4 | Recurrent Independent Mechanisms, arXiv:1909.10893 | arXiv API; GitHub `anirudh9119/RIMs` | Goyal, Lamb, Hoffmann, Sodhani, Levine, Bengio, Scholkopf; v1 2019-09-24, v6 2020-11-17. Top-k module activation per step; only active modules update. Repo exists, **no licence file**. | ICLR 2021 venue **NOT verified** from a primary source. |29| 5 | Mixture-of-Depths, arXiv:2404.02258 | arXiv API | Raposo, Ritter, Richards, Lillicrap, Humphreys, Santoro; v1 2024-04-02 (only version). Top-k token routing per layer with fixed capacity; unrouted tokens pass through the residual. | No official code found. |30| 6 | Huginn / Recurrent Depth, arXiv:2502.05171 | arXiv API; GitHub `seal-rg/recurrent-pretraining` | Geiping et al. (9 authors); v1 2025-02-07, v2 2025-02-17; comment gives model `tomg-group-umd/huginn-0125` and code repo. Apache-2.0. Zero-shot per-token exit by KL threshold. | No venue listed on arXiv. |31| 7 | Mixture-of-Recursions, arXiv:2507.10524 | arXiv API; GitHub `raymin0223/mixture_of_recursions` | Bae et al. (11 authors); v1 2025-07-14, v3 2025-10-25; comment gives code. Repo description says "(NeurIPS 2025)". Apache-2.0. | Venue taken from the official repo description, not from arXiv. |32| 8 | Ouro, arXiv:2510.25741 | arXiv API; HTML v5 (v6 HTML not rendered); project page ouro-llm.github.io; HF model cards via search | Zhu et al. (33 authors incl. Bengio, Eshraghian); v1 2025-10-29, **v6 2026-10-11**. Weight-tied stack, up to 4 loops; exit gate lambda_t = sigma(Linear(h_t)); Stage I entropy-regularised expected loss; **Stage II (LM frozen): gate trained by BCE against labels w = sigma(k(max(0, L_{t-1}-L_t) - gamma)), k=50, gamma=0.005, i.e. labels derived from the model's own per-token loss improvement**. Weights ByteDance/Ouro-1.4B, -2.6B (+Thinking) Apache-2.0 with research-only notice. | No official training code; community reimplementation `rkstgr/LoopLM` (no licence). No venue. |33| 9 | RecurTrace, arXiv:2609.03379 ("may not exist") | arXiv API; abs page; HTML v3; GitHub `JensenYX/RecurTrace` | **EXISTS.** Wang, Feng, Shen, Xu, Wang, Wu; v1 2026-09-03, v3 2026-10-19. Loop Memory Attention: each looped layer attends over a window W (best W=3) of its own outputs from earlier loops at the same position; empty at loop 1 (returns zero). Halting head p_t = sigma(w^T z_t + b) on pooled state, **sequence-level**, BCE against oracle labels 1[min_{t'>t} L_{t'} < L_t - delta]; threshold tau chosen on held-out data. Baselines named: ACT, PonderNet, CALM, LoopUS-Conf, TaH-Mismatch. Code repo exists (Apache-2.0, created 2026-10-17). | No venue. delta value not read. |34| 10 | Elastic Latent Recurrence, arXiv:2610.12582 ("may not exist") | arXiv API; HTML v1 link scan; HF `nvidia/Dory1.0-4B` (HTTP 401) | **EXISTS.** Ladhak, Acharya, Akle Serano, Sun, Ginsburg, Puvvada; v1 2026-10-08. Recipe: gated state update with learned step size, mixed loop counts per optimizer update, loop-index/budget conditioning (zero-initialised when added late), last-loop KV cache. Reports Koi-4B. | **No code or checkpoint verified.** Paper HTML links only to the `nvidia/nemotron-post-training-v3` HF collection. Third-party blogs claim a public "Dory1.0-4B" release; the HF page returned 401 to the fetcher, so this is **UNVERIFIED** and must not be cited as the paper's release. |35| 11 | Titans, arXiv:2501.00663 | arXiv API; GitHub `lucidrains/titans-pytorch` | Behrouz, Zhong, Mirrokni; v1 2024-12-31 (only version). Neural long-term memory updated at test time by surprise gradient; persistent memory tokens; MAC/MAG/MAL variants. | No official code; unofficial implementation is MIT. Cite year as 2024 (arXiv date) or 2025 consistently. |36| 12 | LayerSkip, arXiv:2404.16710 | arXiv API; GitHub `facebookresearch/LayerSkip` + LICENSE file | Elhoushi et al. (13 authors); v1 2024-04-25, v4 2024-10-18; comment "ACL 2024", DOI 10.18653/v1/2024.acl-long.681. Code licence is **CC BY-NC 4.0** (GitHub reports NOASSERTION; LICENSE file read). | -- |3738### 1.2 Additional references (all verified via arXiv API unless stated)3940Adaptive computation / looped / latent reasoning: Graves 2016 ACT41(1603.08983); Coconut (2412.06769, COLM 2025, code MIT); Giannou et al. 202342(2301.13196; ICML 2023 venue NOT verified); Saunshi et al. 2025 (2502.17416,43ICLR 2025); Fan et al. (2409.15647, ICLR 2025); Relaxed Recursive Transformers44(2410.20672, ICLR 2025); Pause tokens (2310.02226, ICLR 2024); Latent tokens45(2505.12629); PonderLM (2505.20674, ICLR 2026, code Apache-2.0); Inner Thinking46Transformer (2502.13842); Encode-Think-Decode (2510.07358); Parallel Loop47Transformer (2510.24824); Think-at-Hard (2511.08577; **ICML 2026 verified on48PMLR v306 pp. 32129-32154**; title changed between v1 and the proceedings; code49`thu-nics/TaH` Apache-2.0); TaH2 (2609.35748; HTML v1 read: lookahead depth50supervision details below; abstract states code at `thu-nics/TaH`, no separate51TaH2 repo exists); LoopFormer (2602.11451, ICLR 2026); AdaPonderLM52(2603.01914); Latent lookahead (2603.20219); Frey et al. (2603.08391, ICLR 202653workshop); LoopUS (2605.11011); Latent Recurrent Transformer (2605.26797);54LaTER (2605.07315, code Apache-2.0); Sharma & Vu (2606.24898);55Depth-Attention (2606.05014); DiscoLoop (2607.00341); Popescu et al.56(2607.20519); Logan (2607.14427, code MIT); Dau et al. (2609.19934); Huang et57al. (2609.21605); ReFlux (2609.32457; HTML v1 read; code `gooogleshanghai/reflux`58MIT); Olszowiec et al. (2610.18127).5960Slots / workspaces / memory: Slot Attention (2006.15055, NeurIPS 2020, code61Apache-2.0); Shared Global Workspace (2103.01197, ICLR 2022); Perceiver62(2103.03206, ICML 2021); NTM (1410.5401); Memory Networks (1410.3916); Sparse63Access Memory (1610.09027, NIPS 2016); Memformer (2010.06891); RMT (2207.06881,64NeurIPS 2022; repo has no licence file); Token Turing Machines (2211.09119,65CVPR 2023, scenic Apache-2.0); Memory Layers at Scale (2412.09764); Registers66(2309.16588; ICLR 2024 NOT verified).6768Long context / recurrence: Transformer-XL (1901.02860, ACL 2019, Apache-2.0);69Compressive Transformer (1911.05507; ICLR 2020 NOT verified); Mamba70(2312.00752, Apache-2.0).7172Sparse routing: Shazeer et al. 2017 (1701.06538; ICLR 2017 NOT verified);73Switch Transformers (2101.03961, JMLR).7475Value / off-policy: Ernst, Geurts & Wehenkel 2005 (**verified on JMLR**, vol.766, pp. 503-556); DQN workshop paper (1312.5602); Dudik, Langford & Li 201177(1103.4601, ICML 2011); Bottou et al. (1209.2355; JMLR 2013 NOT verified);78Swaminathan & Joachims CRM (1502.02362; ICML 2015 NOT verified).7980Early exit / test-time compute: CALM (2207.07061, NeurIPS 2022 oral);81speculative decoding (2211.17192, ICML 2023 oral); Snell et al. (2408.03314);82Physics of LMs 2.1 (2407.20311).8384Zero-init identity insertion: ReZero (2003.04887); ControlNet (2302.05543;85ICCV 2023 NOT verified); Flamingo (2204.14198, NeurIPS 2022).8687Building blocks: SwiGLU (2002.05202); RoFormer (2104.09864); GQA (2305.13245,88EMNLP 2023); RMSNorm (1910.07467, NeurIPS 2019); Kaplan (2001.08361);89Chinchilla (2203.15556; NeurIPS 2022 NOT verified).9091### 1.3 Things that could not be verified or were dropped9293* `nvidia/Dory1.0-4B` as the public release of ELR / Koi-4B: HF page not94 readable (401); only third-party blogs. Not in the bib.95* "Solve the Loop: Attractor Models for Language and Reasoning"96 (arXiv 2605.12466) surfaced in a search result list only; abstract not97 fetched, so it is omitted.98* Official code for PonderNet, Mixture-of-Depths, Titans, Memformer, Switch99 Transformers, ACT, pause tokens, Shazeer 2017: none found; the bib says so.100* Venues flagged "NOT verified" above are plausible but were not confirmed by101 a primary source during this audit; either confirm them before submission or102 cite the arXiv version.103* No MCP OpenAlex access (server failed to connect); DOIs are therefore given104 only where the arXiv record carries one (LayerSkip) or the JMLR page was105 read.106107---108109## 2. Claim-by-claim novelty table110111Legend for status: **novel** = no predecessor found after targeted search;112**partially novel** = each ingredient exists, the specific combination or113formulation does not; **not novel** = a predecessor implements the same114mechanism; **unknown** = search inconclusive.115116| Proposed claim | Closest source(s) | Common mechanism | Actual distinction | Evidence required | Status |117|---|---|---|---|---|---|118| **C1.** Selected local writes into a temporary slot workspace (16 slots, reset at every token, initialised S0 = e + P h_t, 1-2 slots rewritten per action) | Shared Global Workspace (Goyal et al. 2022); RIMs (2019); Perceiver / Perceiver AR (2021/2022); Slot Attention (2020); ReFlux (2026); SAM (2016); pause / latent / register tokens | Small bank of latent vectors read by cross-attention from a larger source (Perceiver, Slot Attention); write access restricted to a few selected slots or modules (Global Workspace top-k writers, RIMs top-k mechanisms, SAM sparse writes, ReFlux sparse routes); extra ephemeral computation per prediction whose outputs are discarded (pause tokens, latent tokens, registers) | Global Workspace / RIMs / SAM / TTM keep the workspace **across time steps**; PRISME's bank is **non-persistent** (reset per token) and persistence is replaced by a read-only source bank of past backbone states (like Transformer-XL's cache, not a writable memory). Perceiver / Slot Attention update **all** latents each iteration; PRISME updates 1-2. ReFlux selects routes **per block, shared by all tokens**, and writes into the residual stream, not into a per-token slot set. Pause/latent tokens are dense and unselected. No paper found that combines an ephemeral per-token slot bank with editor actions that rewrite only selected slots. | Ablations: (i) 16-slot sparse edits vs. dense update of all slots at equal FLOPs; (ii) reset-per-token vs. carrying slots across tokens; (iii) slot bank vs. plain looping of h_t (Huginn/MoR-style). Must show the sparse-write variant is on a better Pareto frontier, not merely different. | **partially novel** |119| **C2.** Action-conditional counterfactual value supervision: labels g(s,a) = loss(no edit) - loss(after edit a) collected by executing paired edit / no-edit branches, with recorded propensities; heads for cost, uncertainty, harm; greedy Q = g - lambda c - kappa sigma, Q(STOP)=0 | **TaH2** (You et al., 2026-09): "lookahead depth supervision", online per-token labels = loss gain L_m - L_{m+1} obtained from a no-gradient lookahead branch, cost-sensitive BCE; **Ouro** Stage II (2025-10): gate fit to sigma(k(max(0, L_{t-1}-L_t) - gamma)); **RecurTrace** (2026-09): oracle label 1[min future loss < L_t - delta]; TaH (2025-11): token-mismatch oracle; ACT/PonderNet (compute penalty); DQN / fitted Q (greedy argmax over predicted action values); Dudik 2011, Bottou 2013, Swaminathan & Joachims 2015 (logged propensities) | Supervising a stop/continue decision with the **realised change in next-token loss from actually running the extra computation** is now established practice (Ouro, TaH2, RecurTrace, all 2025-26). Greedy selection by predicted value with STOP at value 0 is standard Q-learning. | All three predecessors supervise a **binary** continue/stop decision (one action). PRISME's label is **conditional on which of 25 actions** is taken (which slot or pair to edit), i.e. a counterfactual value per action, and the paper records **sampling propensities** of the logged actions, which none of the three do (TaH2 explicitly has no importance weighting; Ouro/RecurTrace use deterministic oracles). The cost, uncertainty and harm heads are additions. Whether propensities are actually *used* (IPS / DR correction) must be stated, otherwise logging them is not a contribution. | Show that action-conditional labels beat (a) a binary continue/stop decider in TaH2 style applied to the same slot editor, and (b) learning the controller end-to-end from the LM loss (MoD / ReFlux style). Report off-policy evaluation of the learned controller with the logged propensities (DR estimator) to justify the logging. | **partially novel** (binary version: not novel; action-conditional + propensity-logged: no predecessor found) |120| **C3.** Differential readout U(read(S_final) - read(S0)) added to h_t with U = 0 at init, giving an exact backbone output at zero edits | ReZero (2020), ControlNet zero-conv (2023), Flamingo zero-init tanh gates (2022); ReFlux (no active edge => frozen backbone unchanged; additive Delta payload); Mixture-of-Depths (unrouted tokens bypass exactly); TaH / TaH2 / ETD (skipped tokens keep first-pass output); RecurTrace (LMA returns zero at loop 1); ELR (gated update with learned step size, zero-init conditioning, "one-loop tax") | Zero-initialised additive branch so that the untouched model is reproduced exactly at initialisation; exact skip path when the adaptive component is not invoked; subtracting a baseline read so that "no edit" contributes exactly zero | The *differential* form (read after minus read before) guarantees the identity **for all training time**, not only at initialisation, because the two reads cancel exactly when no slot was edited; ReZero/ControlNet/Flamingo lose exactness once the gate is trained, and ELR obtains one-loop parity only empirically. ReFlux has the same all-time property (zero active edges => unchanged) and so does MoD's residual skip, so the property itself is not new; the differential-read implementation is a modest variant. | State precisely what is exact: output equality with the backbone *co-trained with the editor*, not with an independently trained backbone. Show that the fast path really costs zero extra FLOPs (controller cost included) and that the Pareto curve passes through the backbone point. | **not novel** as a mechanism (minor engineering variant); do not list as a standalone contribution |121| **C4.** The combination C1 + C2 + C3 | ReFlux (sparse selected additive writes with exact bypass, but per-block routing and LM-loss training); TaH2 (counterfactual loss labels + exact first-pass bypass, but whole-state dense iteration, binary action); RecurTrace (oracle stop + attention over own trajectory); MoR (token-level routed recurrence as the Pareto baseline) | Each neighbour shares two of the three ingredients. | No work found that has all three: per-token ephemeral slots with 1-2-slot edits, an action-conditional counterfactual value controller, and an exact zero-budget identity. The architecture-level claim therefore survives as a combination, but only if the empirical Pareto improvement over MoR / Huginn / a TaH2-style binary decider on the same backbone is demonstrated at matched FLOPs including controller and editor cost. | Head-to-head at matched training tokens and inference FLOPs against: dense-slot variant, binary-decider variant, Huginn-style global recurrence, MoR. Report quality-compute curves, not single points. | **partially novel** (novel combination; contribution is empirical) |122| **C5.** Fixed action catalog: 16 single-slot edits + 8 fixed-pair edits + STOP (25 actions, max 8 per token) | Switch / Shazeer top-k expert routing; RIMs top-k mechanisms; MoD top-k tokens; DQN discrete action sets | Discrete finite action set with greedy selection and a capacity cap | A particular design choice (which pairs, why 8, why max 8 actions). Nothing in the literature forbids or anticipates this exact catalog, but a fixed catalog is a hyper-parameter, not a mechanism. | Sensitivity analysis over catalog size / pair set / max actions; justify pairs (e.g. learned vs. fixed pairs). | **not novel** (do not claim as a contribution; present as design) |123124---125126## 3. How much novelty survives, and what the contribution statement should say127128Frank assessment. The three mechanisms the paper proposes to claim are each129preceded by well-verified work, and the field moved fast between 2025-10 and1302026-10. Learned stopping supervised by realised loss reduction from an executed131extra branch is now standard: Ouro's Stage II (2025-10) fits its exit gate to132per-token loss improvements, TaH2 (2026-09) calls exactly this "lookahead depth133supervision" with a no-gradient counterfactual branch for already-stopped134tokens, and RecurTrace (2026-09) trains a halting head on an oracle "does more135depth still reduce loss" label. Exact zero-budget bypass is structurally present136in Mixture-of-Depths, in TaH/TaH2 (skipped tokens keep the first pass) and in137ReFlux (no active route leaves the frozen backbone unchanged), and zero-init138identity insertion is a decade-old trick (ReZero, Flamingo, ControlNet). Sparse139competitive writes into a small slot workspace go back to RIMs and the Shared140Global Workspace, and ReFlux (2026-09) already writes a sparse, router-selected141*difference* signal additively with exact bypass.142143What survives: (i) the workspace being **ephemeral per token** while the only144persistent object is a **read-only** projected source bank -- no neighbour has145this split; (ii) the value controller being **action-conditional over which146slot(s) to edit**, with per-action counterfactual labels and recorded147propensities, instead of a binary continue/stop -- no neighbour does this;148(iii) the **combination**. C3 should be demoted to an implementation property149("by construction the zero-edit path returns the backbone output exactly")150rather than a contribution, and C5 to a design choice with a sensitivity study.151152Suggested contribution statement (three items, no more):1531541. *Architecture.* A per-token, non-persistent 16-slot workspace edited by a155 shared sparse editor that rewrites one or two selected slots per action,156 reading a causal, read-only bank of projected backbone states; the readout157 is differential so the zero-edit path is exactly the backbone.1582. *Supervision.* An action-conditional value controller trained from159 counterfactual edit / no-edit branches executed offline, with logged160 propensities, predicting per-action loss gain, cost, uncertainty and harm;161 actions chosen greedily with STOP at zero. Position this explicitly as the162 multi-action generalisation of the binary loss-gain supervision of Ouro163 (Stage II), TaH2 and RecurTrace, and cite them as such.1643. *Evidence.* Quality-compute Pareto curves at matched training tokens and165 inference FLOPs against dense-slot, binary-decider, global-recurrence166 (Huginn-style) and Mixture-of-Recursions baselines on the same backbone,167 with ablations isolating sparse selection, action-conditional labels and168 the exact bypass.169170If the empirical gains over a TaH2-style binary decider on the same slot171editor are small, the honest framing is "a unifying formulation" rather than172"a new mechanism".173174---175176## 4. Near-neighbours published after 2025-06 (chronological)177178| arXiv | Date (v1) | Work | Why it is a neighbour |179|---|---|---|---|180| 2507.10524 | 2025-07-14 | Mixture-of-Recursions (Bae et al.; NeurIPS 2025) | token-level routed recursion; Pareto-frontier baseline; code Apache-2.0 |181| 2510.07358 | 2025-10-08 | Encode, Think, Decode (Koishekenov et al.) | iterate selected layers; per-token adaptive depth |182| 2510.24824 | 2025-10-28 | Parallel Loop Transformer (Wu et al.) | latency-neutral looping; KV sharing |183| 2510.25741 | 2025-10-29 (v6 2026-10-11) | Ouro (Zhu et al.) | loss-improvement-supervised exit gate (Stage II); weights Apache-2.0 |184| 2511.08577 | 2025-11-11 | Think-at-Hard (Fu et al.; ICML 2026) | selective extra latent iteration at few tokens; oracle-distilled decider; code Apache-2.0 |185| 2602.11451 | 2026-02-11 | LoopFormer (Jeddi et al.; ICLR 2026) | budget-conditioned elastic depth |186| 2603.01914 | 2026-03-02 | AdaPonderLM (Song et al.) | token-wise learned halting in pondering LMs |187| 2603.20219 | 2026-03-03 | Latent lookahead (Noci et al.) | extra latent compute at selected positions |188| 2603.08391 | 2026-03-09 | Adaptive loops + memory banks (Frey et al.; ICLR 2026 workshop) | learned halting + gated memory |189| 2605.07315 | 2026-05-08 | LaTER (Li et al.) | supervised halting of a latent phase; code Apache-2.0 |190| 2605.11011 | 2026-05-10 | LoopUS (Park et al.) | confidence head for early exit in recast looped LLMs |191| 2605.26797 | 2026-05-26 | Latent Recurrent Transformer (Huang et al.) | cross-token latent memory from a source layer, one forward per token |192| 2606.05014 | 2026-06-03 | Depth-Attention (Zeng et al.) | attention over earlier-layer keys at the same position |193| 2606.24898 | 2026-06-12 | Readout blind spot (Sharma & Vu) | readout design in looped models |194| 2607.00341 | 2026-07-01 | DiscoLoop (Fu et al.) | looping discrete + continuous states |195| 2607.07128 | 2026-07-08 | Distributed Sparse Interventions (Ernst et al.) | sparse latent interventions (static) |196| 2607.20519 | 2026-07-08 | Diagnosing halting gates (Popescu et al.) | evidence that jointly trained gates can underperform post-hoc readouts |197| 2607.14427 | 2026-07-15 | Per-token fixed-point convergence (Logan) | training-free exit beats trained router; code MIT |198| 2609.03379 | 2026-09-03 | RecurTrace (Wang et al.) | oracle loss-reduction halting; loop-time memory attention; code Apache-2.0 |199| 2609.19934 | 2026-09-17 | Beyond depth truncation (Dau et al.) | evaluation of depth utilisation |200| 2609.21605 | 2026-09-18 | Trading depth for time (Huang et al.) | recurrent transformers |201| 2609.32457 | 2026-09-26 | ReFlux / Write Back the Delta (Ye et al.) | sparse selected additive writes of a difference signal, exact bypass; code MIT |202| 2609.35748 | 2026-09-28 | TaH2 (You et al.) | lookahead (counterfactual) loss-gain supervision of a per-token decider |203| 2610.12582 | 2026-10-08 | Elastic Latent Recurrence (Ladhak et al.) | one-loop-tax removal; gated zero-init update; no code verified |204| 2610.18127 | 2026-10-13 | Monitoring latent reasoning (Olszowiec et al.) | analysis of looped LLM halting |205206Searches run (WebSearch, 2026-10-29): "latent reasoning looped language model207learned halting oracle supervision"; "Think-at-Hard selective latent208iterations"; "slot workspace sparse writes per-token latent edit"; "counterfactual209/ intervention supervision value controller latent edits next-token loss";210"test-time latent computation adaptive recurrence per token learned stopping";211"ephemeral / scratch latent tokens per prediction"; "value head / Q-value selects212which hidden state to edit STOP action counterfactual loss reduction propensity".213The last two returned no paper combining per-action counterfactual labels with214latent-state editing, which is the basis for the "no predecessor found" entries215in Section 2. Absence of evidence after seven targeted searches is not proof of216absence; a final check on arXiv listings for cs.LG/cs.CL in the week before217submission is recommended.218