Pith. sign in

REVIEW 2 major objections 5 minor 31 references

Learning sparse representations in reinforcement learning

T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read Adding a k-Winners-Take-All sparse-coding layer to a temporal-difference network lets it solve three control tasks where standard backpropagation TD learning fails.

desk verdict A competent but largely retrospective simulation study: the Puddle-world kWTA advantage is real and statistically supported, but the central mechanism claim is untested because no capacity-matched dense baseline appears. read the letter →

arxiv 1909.01575 v1 pith:7P72UKZV submitted 2019-09-04 cs.LG cs.AIcs.NEcs.ROstat.ML

classification cs.LGcs.AIcs.NEcs.ROstat.ML
keywords reinforcementlearningtemporaldifferencesparserepresentationslateralinhibitionk-Winners-Take-AllSARSAcatastrophicinterferencefunctionapproximation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to show that temporal-difference reinforcement learning fails on simple control tasks when the value function is approximated by a standard neural network because the network's hidden representations are dense, and that adding lateral inhibition in the form of a k-Winners-Take-All nonlinearity fixes the failure by producing sparse conjunctive codes. It demonstrates this with SARSA simulations on three tasks previously reported to be problematic for TD with backpropagation: Puddle-world, Mountain-car, and Acrobot. The kWTA network learns better value functions and policies than regular backpropagation networks or linear networks, and only the kWTA network learns the optimal Acrobot policy. If the claim is right, it resolves a tension in the TD account of dopamine-based learning and offers a simple architectural change that makes TD with neural networks more reliable.

What carries the argument

The load-bearing mechanism is the k-Winners-Take-All (kWTA) nonlinearity applied to the hidden layer of a one-hidden-layer backpropagation network. In the paper's Algorithm 2, the net input vector is sorted, the top $k+1$ values are located, and a bias $b = \eta'_k - q(\eta'_k - \eta'_{k+1})$ is subtracted so that only the top $k$ units receive positive adjusted net input; a sigmoid then makes roughly $k=0.1n$ hidden units strongly active. This replaces smooth hidden activations with a sparse conjunctive code of the state, balancing overlap that supports generalization against separation that prevents interference, at an extra cost of $O(n + k \log k)$ per forward pass.

What would settle it

Train a regular backpropagation SARSA network with hidden units reduced so its number of active units matches the kWTA network's k in each task, e.g., 22/260/840 total hidden units instead of 220/2604/8400, holding learning rate, exploration, and episode limits fixed. If the dense network's value-function error and goal attainment match the kWTA network's, the paper's attribution of success to sparse conjunctive coding rather than reduced capacity is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that SARSA temporal-difference learning succeeds on Puddle-world, Mountain-car, and Acrobot when the value-function network uses a hidden layer governed by k-Winners-Take-All lateral inhibition, while otherwise identical backpropagation networks and linear networks fail on at least some of these tasks. In Puddle-world the kWTA network's accumulated rewards deviate from optimal by less than one reward point and reach the goal from 99.9 percent of starting locations, versus 99.0 percent for the regular network and 93.3 percent for the linear network; in Mountain-car and Acrobot only the kWTA network learned a stable optimal policy. The authors interpret this as evidence that sparse conjunctive state representations, not fixed engineered codes, are what make TD with neural function approximation work, and that cortical lateral inhibition may explain why the brain's TD-like dopamine system does not show the same failures.

Load-bearing premise

The paper assumes that the kWTA network's advantage is caused by sparse conjunctive coding rather than by the much smaller number of active hidden units, and it does not include a control network with comparable active-unit count to rule out reduced model capacity as the cause.

Editorial extensions

If this is right

  • If correct, the three reported failures of TD with neural networks are representational failures, not limits of temporal-difference learning itself.
  • If correct, hard-wired coarse codes are unnecessary: backpropagation plus kWTA can learn state representations that give the same benefit as engineered CMAC codes.
  • If correct, the same mechanism should reduce catastrophic interference whenever nearby states demand very different values, not just in the three benchmark tasks.
  • If correct, the biological hypothesis is supported: cortical lateral inhibition is present precisely where and when dopamine-based TD learning must work.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test whether any hard top-$k$ sparsity layer would produce the same benefit; if top-$k$ ReLU matched kWTA, the causal story would be sparsity, not the specific lateral-inhibition mechanism.
  • Because kWTA is non-smooth, the usual gradient-descent interpretation of backpropagation is invalid at the threshold; a rigorous account would require treating the layer as a projection, so convergence cannot be inferred from standard TD theorems.
  • The three solved tasks are all low-dimensional; the strongest untested extension is whether kWTA in fully connected layers of deep networks prevents catastrophic interference in continual-learning benchmarks, which would follow if the paper's mechanism is the active ingredient.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes that adding a k-Winners-Take-All (kWTA) mechanism to the hidden layer of a one-hidden-layer SARSA network induces sparse conjunctive representations of state, and that this avoids catastrophic interference and supports TD learning. The hypothesis is tested empirically on three control tasks: Puddle-world, Mountain-car, and Acrobot. The authors compare three architectures (linear, regular backpropagation, and kWTA) using the same SARSA update, and report that the kWTA network outperforms the regular and linear networks, especially on Acrobot where the baselines fail. The paper is presented as evidence that lateral-inhibition-like sparsity can compensate for known weaknesses of neural-network TD learning, with implications for cognitive neuroscience. The main contribution is empirical; no new theoretical analysis is offered.

Significance. If the central claim is upheld, the paper would provide a useful and biologically motivated architectural modification for TD learning with function approximation, and it would strengthen the link between cortical sparse coding and reward-based learning. The paper has concrete strengths: it reports multi-seed statistics for Puddle-world (20 simulations per condition, t-tests), it provides the simulation code at a public URL, and it gives detailed environment and network hyperparameters that facilitate replication. The Puddle-world advantage of kWTA over regular backpropagation is statistically significant, and the Acrobot result is qualitatively clear. However, the key mechanistic conclusion, that sparse conjunctive coding is the causal factor, is not yet supported because the kWTA network also has about 10-fold fewer active hidden units than the regular network, confounding sparsity with reduced model capacity.

major comments (2)
  1. [§4.1, §5.1, §5.3] The central claim that sparse conjunctive representations produced by lateral inhibition are beneficial is not separated from a simple capacity reduction. In §4.1 the kWTA network keeps only k=10% of the hidden units active (Section 3.2), so for the Puddle-world, Mountain-car, and Acrobot tasks the kWTA network uses 22, 260, and 840 active hidden units respectively, whereas the regular backpropagation network uses 220, 2604, and 8400 active units. The paper never reports a control condition with a regular network whose number of active hidden units matches that of the kWTA network, nor any other capacity-limiting baseline such as a smaller dense hidden layer or a dropout rate. Without such a control, the observed improvements in Figures 8 and 10 cannot be attributed to sparse structuring rather than to fewer active basis functions; a reduced-capacity dense network may show the same interference-relief effect. This is load-bearing because the abstract and §1.2 explicitly frame the benefit as coming from sparse conjunctive coding, not from reduced capacity.
  2. [§5.2 and §5.3] The quantitative evidence for the Mountain-car and Acrobot advantages is incomplete. For Mountain-car, Figure 9 shows learning curves and value-function plots but no statistical tests or multi-run averages are reported; the text in §5.2 says the kWTA value function is 'closest numerically to optimal Q-table results' and 'the most stable,' but no numerical comparisons or error bars are given. For Acrobot, §5.3 states that only the kWTA network could learn the optimal policy, but the figure presents representative runs without quantitative success rates, average steps to goal, or statistics over seeds. Because the paper's central claim spans all three tasks, the lack of statistical backing for two of the three tasks weakens the evidence. At minimum, the authors should report mean and variance of final performance and a test of significance for Mountain-car and Acrobot, or clearly label these as illustrative single-run demonstrations.
minor comments (5)
  1. [Title page] The affiliation line contains a typo: 'Electrical Engineering and Computer Scinence' should be 'Science'.
  2. [§3.2] The sentence 'k is relatively smaller than n. For example k = 0.1× n is considered for the simulations reported in this chapter' refers to 'this chapter', which is thesis language; it should read 'in this paper'.
  3. [§4.1 and §5.1] The paper says results in Figure 7 were 'initially reported at Rafati and Noelle (2015)' and similar text appears for other figures; the relationship to prior conference papers and the new contribution of this manuscript should be stated more explicitly in the introduction or a dedicated section, so that the novelty is clear to the reader.
  4. [§5.1] The significance values in the text and in Figure 8 are slightly inconsistent: the text reports t(38)=2.205; p<0.05 for the goal-rate comparison, while Figure 8 labels the same statistic as p<0.005. The authors should reconcile these values.
  5. [§3.3] The paper acknowledges that backpropagation through the kWTA threshold is not a true gradient descent and that this discrepancy is ignored. This is a reasonable simplification, but a sentence discussing why small learning rates suffice (e.g., the bias is locally constant) would help the reader judge the validity of the approximation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is an empirical simulation hypothesis, not a derived prediction, and self-citations are provenance rather than load-bearing evidence.

full rationale

The paper makes no formal derivation: the central claim is a hypothesis supported by simulations (Sections 4 and 5). The kWTA architecture is taken from external work (O'Reilly and Munakata, 2001; O'Reilly, 2001) and applied to SARSA; no fitted parameter is later renamed as a prediction. The comparisons against regular and linear networks on Puddle-world, Mountain-car, and Acrobot are self-contained simulations with the code available at http://rafati.net/td-sparse/. Self-citations (Rafati and Noelle, 2015, 2017) are provenance statements, e.g., "These results were initially reported at Rafati and Noelle (2015)", rather than load-bearing evidence: the paper itself reports the simulation design and results. Section 3.3's admission that the kWTA bias is ignored when computing backpropagation gradients is a technical caveat, not circularity. The absence of a capacity-matched dense control (e.g., a dense network with only 22 active hidden units in Puddle-world) is a potential confound for the claim that sparse conjunctive coding, rather than reduced active-unit count, drives the improvement; but a confound is not a circular reduction. No equation in the paper defines the target result in terms of the input, and no fitted parameter is relabeled as a prediction. Therefore the paper is not circular.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The simulations depend on many hand-selected hyperparameters (hidden sizes, k, learning rates, exploration schedules, input encoding widths) and on domain assumptions about the kWTA model of cortical inhibition and the validity of approximate gradients. No free parameters are fit to test data, but the configuration choices are selected by the authors and could favor the kWTA network.

free parameters (5)
  • kWTA sparsity fraction k = 10% of hidden units (22, 260, 840)
    Hand-selected from cognitive neuroscience convention; no sensitivity sweep is reported, and the method's sparsity mechanism depends on this value.
  • Hidden layer size = 220 (Puddle-world), 2604 (Mountain-car), 8400 (Acrobot)
    Set manually with no justification relative to task difficulty, and the regular baselines use the same sizes.
  • Learning rate alpha = 0.005, 0.001, 0.0001
    Per-task values chosen by hand; no tuning protocol or sensitivity analysis is provided.
  • Input encoding mesh and Gaussian width = 21, 61, 21 input units per variable; Gaussian sigma = 1/mesh
    The input representation is hand-designed, and the benefit is attributed to hidden sparsity even though the coarse-coded input also contributes structure.
  • Exploration and termination schedule = epsilon 0.1 or 0.05 with decay; max episodes 44,100 or 200,000; max steps 80, 3000, 2000
    Hand-selected stopping and exploration criteria, with no sensitivity analysis.
assumptions (5)
  • standard math Standard MDP and SARSA convergence properties for tabular value functions.
    The comparison to an optimal policy uses a large look-up table trained with SARSA, relying on standard convergence results for tabular TD. Cited in Sections 2.3 and 5.1.
  • domain assumption kWTA dynamics adequately model cortical lateral inhibition.
    The motivation depends on O'Reilly and Munakata (2001) and Noelle (2008); kWTA is used as a stand-in for GABAergic lateral inhibition without testing other sparsity mechanisms.
  • ad hoc to paper Backpropagation through the kWTA threshold is a good approximation of gradient descent.
    Section 3.3 states the mechanism violates assumptions of backprop as gradient descent and assumes a small learning rate keeps deviations small; no proof or measurement is given.
  • domain assumption Poor Regular network performance reflects catastrophic interference, not poor hyperparameter choice.
    Section 5 interprets Regular network failures as the known instability of nonlinear TD, but the Regular baselines were not tuned; an untuned baseline may understate the baseline's ability.
  • domain assumption The three benchmark tasks instantiate the claimed failure mode.
    Puddle-world, Mountain-car, and Acrobot are taken from Boyan and Moore (1995) and Sutton (1996) as tasks where TD with neural networks fails; no diagnostic verifies that failure is due to interference in these runs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning sparse representations in reinforcement learning." pith.science (2026). https://pith.science/paper/7P72UKZV

@misc{pith2026190901575,
  author       = {Pith},
  title        = {Pith review of: Learning sparse representations in reinforcement learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7P72UKZV}},
  note         = {Machine review of arXiv:1909.01575}
}
read the original abstract

Reinforcement learning (RL) algorithms allow artificial agents to improve their selection of actions to increase rewarding experiences in their environments. Temporal Difference (TD) Learning -- a model-free RL method -- is a leading account of the midbrain dopamine system and the basal ganglia in reinforcement learning. These algorithms typically learn a mapping from the agent's current sensed state to a selected action (known as a policy function) via learning a value function (expected future rewards). TD Learning methods have been very successful on a broad range of control tasks, but learning can become intractably slow as the state space of the environment grows. This has motivated methods that learn internal representations of the agent's state, effectively reducing the size of the state space and restructuring state representations in order to support generalization. However, TD Learning coupled with an artificial neural network, as a function approximator, has been shown to fail to learn some fairly simple control tasks, challenging this explanation of reward-based learning. We hypothesize that such failures do not arise in the brain because of the ubiquitous presence of lateral inhibition in the cortex, producing sparse distributed internal representations that support the learning of expected future reward. The sparse conjunctive representations can avoid catastrophic interference while still supporting generalization. We provide support for this conjecture through computational simulations, demonstrating the benefits of learned sparse representations for three problematic classic control tasks: Puddle-world, Mountain-car, and Acrobot.

Figures

Figures reproduced from arXiv: 1909.01575 by the authors.

Figure 1
Figure 1. The agent/environment interaction in reinforcement learning (Sutton and Barto, 2017). 1.3. Outline of the paper The organization of this paper is as follows. In Section 2, we provide background on the reinforcement learning problem and the temporal difference learning methods. In Section 3, we introduce a method for learning sparse representation in reinforcement learning inspired by the lateral inhibition in the co… view at source ↗
Figure 2
Figure 2. The kWTA neural network architecture: a backpropagation network with a single layer equipped with the k-Winner-Take-All mechanism (from Algorithm 2). The kWTA bias is subtracted from the hidden units net input that causes polarized activity which supports the sparse conjunctive representation. Only 10% of the neurons in the hidden layer have high activation. Compare the population of red (winner) neurons to the oran… view at source ↗
Figure 3
Figure 3. The neural network architectures used as the function approximator for state action values [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: The agent in puddle-world task attempts to reach the goal location (fixed in the Northeast corner) in the least [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: The goal is to drive an underpowered car up a steep hill. The agent received -1 reward for each time step until [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: The goal is to swing the tip (“feet”) above the horizontal by the length of the lower “leg” link. The agent [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: The performance of various learned value function approximators may be compared in terms of their success [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Averaged over 20 simulations of each network type, these columns display the mean squared deviation of [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: The performance of various networks trained to perform the Mountain-car task. The top row contains the [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: The performance of various networks trained to solve the Acrobot control task. The top row results correspond [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 25 canonical work pages

  1. [1]

    S., 1975

    Albus, J. S., 1975. A new approach to manipulator control: The cerebellar model articulation controller CMAC . Journal of Dynamic Systems, Meaasurement, and Control 97 (3), 220--227

  2. [2]

    A., Moore, A

    Boyan, J. A., Moore, A. W., 1995. Generalization in reinforcement learning: Safely approximating the value function. In: Advances in Neural Information Processing Systems 7. MIT Press, Cambridge, MA, pp. 369--376

  3. [3]

    Reinforcement learning: The good, the bad and the ugly

    Dayan, P., Niv, Y., 2008. Reinforcement learning: The good, the bad and the ugly. Current Opinion in Neurobiology 18, 185--196

  4. [4]

    M., 1991

    French, R. M., 1991. Using semi-distributed representations to overcome catastrophic forgetting in connectionist networks. In: Proceedings of the 13th Annual Cognitive Science Society Conference. Lawrence Erlbaum, Hillsdale, NJ, pp. 173--178

  5. [5]

    J., 2012

    Kandel, E., Schwartz, J., Jessell, T., Siegelbaum, S., Hudspeth, A. J., 2012. Principles of Neural Science, 5th Edition. McGraw-Hill, New York

  6. [6]

    The utility of sparse representations for control in reinforcement learning

    Liu, V., Kumaraswamy, R., Le, L., White, M., 2018. The utility of sparse representations for control in reinforcement learning. arXiv e-prints (1811.06626)

  7. [7]

    A., Veness, J., Bellemare, M

    Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al., 2015. Human-level control through deep reinforcement learning. Nature 518 (7540), 529--533

  8. [8]

    R., Dayan, P., Sejnowski, T

    Montague, P. R., Dayan, P., Sejnowski, T. J., 1996. A framework for mesencephalic dopamine systems based on predictive Hebbian learning. Journal of Neuroscience 16, 1936--1947

Show all 31 references
  1. [9]

    C., 2008

    Noelle, D. C., 2008. Function follows form: Biologically guided functional decomposition of memory systems. In: Biologically Inspired Cognitive Architectures — Papers from the 2008 AAAI Fall Symposium

  2. [10]

    C., 2001

    O'Reilly, R. C., 2001. Generaliztion in interactive networks: The benefits of inhibitory competition and Hebbian learning. Neural Computation 13, 1199--1242

  3. [11]

    C., McClelland, J

    O'Reilly, R. C., McClelland, J. L., 1994. Hippocampal conjunctive encoding, storage, and recall: Avoiding a trade-off. Hippocampus 4 (6), 661--682

  4. [12]

    C., Munakata, Y., 2001

    O'Reilly, R. C., Munakata, Y., 2001. Computational Explorations in Cognitive Neuroscience. MIT Press, Cambridge, Massachusetts

  5. [13]

    F., 2019

    Rafati, J., Marcia, R. F., 2019. Deep reinforcement learning via l-bfgs optimization. arXiv e-print (arXiv:1811.02693)

  6. [14]

    C., 2015

    Rafati, J., Noelle, D. C., 2015. Lateral inhibition overcomes limits of temporal difference learning. In: 37th Annual Cognitive Science Society Meeting. Pasadena, CA, USA

  7. [15]

    C., 2017

    Rafati, J., Noelle, D. C., 2017. Sparse coding of learned state representations in reinforcement learning. In: Conference on Cognitive Computational Neuroscience. New York City, NY, USA

  8. [16]

    C., 2019 a

    Rafati, J., Noelle, D. C., 2019 a . Learning representations in model-free hierarchical reinforcement learning. arXiv e-print (arXiv:1810.10096)

  9. [17]

    C., 2019 b

    Rafati, J., Noelle, D. C., 2019 b . Unsupervised methods for subgoal discovery during intrinsic motivation in model-free hierarchical reinforcement learning. In: 33rd AAAI Conference on Artificial Intelligence (AAAI-19), 2nd Workshop on Knowledge Extraction From Games. Honolul...

  10. [18]

    C., 2019 c

    Rafati, J., Noelle, D. C., 2019 c . Unsupervised subgoal discovery method for learning hierarchical representations. In: 7th International Conference on Learning Representations, ICLR 2019 Workshop on ``Structure & Priors in Reinforcement Learning'', New Orleans, LA, USA

  11. [19]

    Learning representations in reinforcement learning

    Rafati Heravi, J., 2019. Learning representations in reinforcement learning. Ph.D. thesis, University of California, Merced. ://escholarship.org/uc/item/3dx2f8kq

  12. [20]

    E., Hinton, G

    Rumelhart, D. E., Hinton, G. E., Williams, R. J., 1986. Learning representations by back-propagating errors. Nature 323, 533--536

  13. [21]

    R., 1997

    Schultz, W., Dayan, P., Montague, P. R., 1997. A neural substrate of prediction and reward. Science 275 (5306), 1593--1599

  14. [22]

    Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., Dieleman, S., Grewe, D., Nham, J., Kalchbrenner, N., Sutskever, I., Lillicrap, T., Leach, M., Kavukcuoglu, K., Graepel, T., H...

  15. [23]

    S., 1988

    Sutton, R. S., 1988. Learning to predict by the methods of temporal differences. Machine Learning 3, 9--44

  16. [24]

    S., 1996

    Sutton, R. S., 1996. Generalization in reinforcement learning: Successful examples using sparse coarse coding. In: Advances in Neural Information Processing Systems 8. MIT Press, Cambridge, MA, pp. 1038--1044

  17. [25]

    S., Barto, A

    Sutton, R. S., Barto, A. G., 1998. Reinforcement Learning: An Introduction, 1st Edition. MIT Press, Cambridge, MA, USA

  18. [26]

    S., Barto, A

    Sutton, R. S., Barto, A. G., 2017. Reinforcement Learning: An Introduction, 2nd Edition. MIT Press, Cambridge, MA, USA

  19. [27]

    Temporal difference learning and TD-Gammon

    Tesauro, G., 1995. Temporal difference learning and TD-Gammon . Communications of the ACM 38 (3)

  20. [28]

    S., Osindero, S., Schaul, T., Heess, N., Jaderberg, M., Silver, D., Kavukcuoglu, K., 2017

    Vezhnevets, A. S., Osindero, S., Schaul, T., Heess, N., Jaderberg, M., Silver, D., Kavukcuoglu, K., 2017. Feudal networks for hierarchical reinforcement learning. In: Proceedings of Thirty-fourth International Conference on Machine Learning (ICML-17)

  21. [29]

    A survey of sparse representation: Algorithms and applications

    Zhang, Z., Xu, Y., Yang, J., Li, X., Zhang, D., 2015. A survey of sparse representation: Algorithms and applications. IEEE Access 3, 490--530

  22. [30]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.senten...

  23. [31]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.