Pawel Kaplanski
2 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 2 failures of our instrument; 2 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.
"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2605.02236.json · badge: /api/ran/2605.02236.svg
Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.
| Repository | Role | Ran |
|---|---|---|
| kaplan196883/llmattr | canonical (classifier label) classifier basis: census |
2 ran · 2 no recorded run |
| Function | Status | Where it lives |
|---|---|---|
| limit_initial_conditions | Ran ran; our contract draft was wrong, not the code |
kaplan196883/llmattr/llm_attractor_experiment/src/config.py pointer only (licence: NONE) · get_code("90d557e0443fa006") How it ranThe body of this sample is withheld: GitHub detected no licence in this paper's repository, kaplan196883/llmattr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| load_config | Ran ran; our contract draft was wrong, not the code |
kaplan196883/llmattr/llm_attractor_experiment/src/config.py pointer only (licence: NONE) · get_code("487b442d54914665") How it ranThe body of this sample is withheld: GitHub detected no licence in this paper's repository, kaplan196883/llmattr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded. |
| find_window_by_title | Not yet run unverified (harvested, no recorded run) |
kaplan196883/llmattr/screenshot_window.py pointer only (licence: NONE) · get_code("6acb518bb9211e31") How it ranThe body of this sample is withheld: GitHub detected no licence in this paper's repository, kaplan196883/llmattr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded. |
| screenshot_window | Not yet run unverified (harvested, no recorded run) |
kaplan196883/llmattr/screenshot_window.py pointer only (licence: NONE) · get_code("b20edcab41763281") How it ranThe body of this sample is withheld: GitHub detected no licence in this paper's repository, kaplan196883/llmattr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded. |
Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).
Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.
Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Recursive language-model loops often settle into recognizable attractor-like patterns, but the practical question is how much injected text is needed to move a settled loop somewhere else, and whether that move lasts. We study this in 30-step recursive loops by separating the model from the context-update rule: append, replace, and dialog updates expose different histories to the same generator. The main result is that persistent redirection in append-mode recursive loops is memory-policy-conditioned. Under the canonical bounded-memory loop with a 12,000-character tail clip, destination-coherent persistence (kicked and still in the same post-injection cluster at the terminal step) plateaus at about 16% across doses 5-400, and retained source-basin escape (kicked and outside the original cluster at the terminal step) reaches about 36% at dose 400; neither crosses 50%. Under a full-history protocol with no artificial truncation within the 30-step horizon, retained source-basin escape crosses 50% near 400 tokens and saturates near 75-80% by 1,500 tokens, while destination-coherent persistence first reaches a 50% point estimate near 1,500 tokens, with Wilson 95% CI for the proportion [0.41, 0.61], so the half-effect region is only localized to roughly 1,000-2,000 tokens. The persistent-escape family therefore splits into two endpoints that answer different questions. Retained source-basin escape asks whether the loop was durably pushed out of its starting basin. Destination-coherent persistence asks the stricter question of whether it stayed in the specific postinjection basin. At high full-history doses, destination-coherent persistence becomes non-monotonic at the canonical 30-step horizon even though the kicked rate stays near 0.96. A four-step falsification battery (heterogeneity control, cluster-granularity sweep with hierarchical macro-merge, transition-entropy diagnostic, and long-horizon trajectory continuation) recasts the dip as a finitehorizon, endpoint-definition-sensitive feature rather than a stable structural attractor asymmetry. A homogeneous-perturbation control reproduced the high-dose dip almost exactly (0.44, 0.32, 0.40 at doses 1500, 2000, 3000 versus 0.51, 0.32, 0.41 under heterogeneous perturbations), refuting perturbation heterogeneity as the cause. Roughly half the canonical magnitude is attributable to endpoint timing (the destination cluster sampled at the injection step is dominated by raw-injection text, not yet by model-generated continuation), and the residual is substantially reduced when the loop is run another 50 steps: under the frozen canonical basis (PCA + K-means fit on the original 30-step experiment, applied to extended trajectories), the dip drops by 73% from -0.143 (95% family-cluster bootstrap CI [-0.269, -0.034]) at step 29 to -0.039 (CI [-0.158, +0.068]) at step 79, with the step-79 interval now straddling zero. The same control showed lower retained source-basin escape under homogeneous repetition (0.54-0.58) than under heterogeneous concatenation (0.74-0.79), suggesting that repeated content gives the model a clearer signal from which to recover. For the simpler raw endpoint, append-mode continuation has a bounded in-distribution dose response. Adversarial continuations yield ED50 raw ≈ 40 tokens, with convergent estimates from pooled
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2605.02236")
get_code_for_paper("2605.02236")
have("2605.02236")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2605.02236.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2605.02236)
Connect an agent — have() is free.