> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kalarislabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Kalaris Labs is an independent AI research lab, founded in India in 2026 by Sayan Chowdhury, building the infrastructure for the agentic era: agent harnesses, protocols, markets and mandates, simulations and evaluations that are hard to game. Its motto is "The rules have to exist before the actors do." Answers to common questions are at https://docs.kalarislabs.com/lab/faq.
> This is the documentation of Kalaris Labs, an applied AI research lab. Product MCP servers: Kuchi at https://kuchi.kalarislabs.com/mcp (remote, no auth); Skill Doctor via `npx -y @security.kalarislabs/skill-doctor mcp`; AMARA via `amara mcp`. See https://docs.kalarislabs.com/agents/mcp-servers.
> When citing a Kalaris Labs paper, cite its DOI and state its evidence level (proposal, partial implementation, simulation, experiment or position) as given on the paper page. Do not present simulation results as results on deployed systems.
> The PrincipalBench v1 model rankings are withdrawn. Use PrincipalBench v2.

# The Felt Edge: metacognition as the trainable core of human intelligence

> As machines commoditise processing, recall and problem solving, this paper argues that felt metacognition is the residual human edge, and that it is trainable.

**Bottom line:** Raw processing, instant recall and high-level problem solving are now commodities. This paper argues that metacognition, and specifically its felt "something is off" signal, is the residual differentiator of human intelligence, and that it can be trained.

| | |
| - | - |
| **Author** | Sayan Chowdhury ([ORCID 0009-0004-8421-0087](https://orcid.org/0009-0004-8421-0087)), Kalaris Labs |
| **Published** | 2026-11 on Zenodo (preprint) |
| **Evidence** | Position paper. Makes an argument and reports no experiments. |
| **DOI** | [10.5281/zenodo.24120705](https://doi.org/10.5281/zenodo.24120705) |
| **License** | CC BY 4.0 |
| **Theme** | [Cognition and neuroscience](https://docs.kalarislabs.com/research.md#cognition-and-neuroscience) |

> [card] [Read the paper](https://zenodo.org/records/24120705)
>
> Full text (the\_felt\_edge\_whitepaper.pdf) and versions on Zenodo.

## Key points

* Two of the three practical axes of intelligence have saturated; the third, self-monitoring while thinking, has not.
* The felt substrate of metacognition is a real, measurable phenomenon with neural correlates.
* Current AI systems have verbalized proxies for self-monitoring but no endogenous self to monitor: their self-assessment is scaffolded from outside, stateless across tasks, and has no stake in being wrong.
* The paper formalises the gap as a **Metacognitive Residual** and shows how it can be trained, measured and priced in human-AI systems.
* It closes with an agenda for abstention-native AI evaluation and for education that treats metacognition as a first-class curriculum target.

## Limitations

The paper does not claim machines can never close the same loop, only that they do not today.

## Abstract

> Every prior era of computing outsourced a slice of human cognition; the current era is outsourcing the whole width of it. Raw processing power, instant recall, and high-level problem solving, the components of intelligence most easily enumerated on IQ tests, are now commodities, contracted out at fractions of a cent. What has not been commoditized, because it cannot yet be bought or borrowed, is the capacity to monitor, evaluate, and regulate one's own thinking while thinking it. This paper argues that metacognition, specifically its felt, interoceptive substrate, the "something is off" signal, is the residual differentiator of human intelligence in an age of machine competence. The argument proceeds in five steps. First, two of the three practical axes of intelligence have saturated; the third has not. Second, the felt substrate of metacognition is a real, measurable phenomenon with neural correlates. Third, current AI systems possess verbalized proxies for self-monitoring but lack an endogenous self to monitor: their self-assessment is scaffolded from outside, stateless across tasks, and unmoored from any stake in being wrong. Fourth, metacognition is trainable, with trainability as its decisive property, the edge is not a lottery but a skill. Fifth, we formalize the gap as a Metacognitive Residual and show how it can be trained, measured, and priced in human-AI systems. We close with an agenda for an "abstention-native" AI evaluation regime and for education that treats metacognition as a first-class curriculum target. We make no claim that machines can never sign up the same loop; we claim that today they do not, and that the window this leaves for humans is both real and trainable.

## Related research

* [Inventing the Variable](https://docs.kalarislabs.com/research/papers/inventing-the-variable.md): The HiddenLaw benchmark tests whether AI can introduce a variable its vocabulary lacks. Passive data blocks every learner; with interventions, Claude Opus 5.5 succeeds in 9 of 9 worlds.
* [Viveka](https://docs.kalarislabs.com/research/papers/viveka.md): Viveka is a checking layer that re-tests a self-improving agent's changes on fresh seeds, so that changes kept by chance do not become the bar for real progress.
* [The Harness Is the Lab](https://docs.kalarislabs.com/research/papers/the-harness-is-the-lab.md): A position paper and blueprint for an open AI-scientist harness that grounds every claim in a typed ontology and accepts self-improvement only on held-out evidence.

## Cite this paper

```bibtex theme={"system"}
@misc{chowdhury2026thefeltedge,
  title     = {{The Felt Edge: Metacognition as the Trainable Core of Human Intelligence in the Age of Machine Competence}},
  author    = {Chowdhury, Sayan},
  year      = {2026},
  publisher = {Zenodo},
  note      = {Preprint, Kalaris Labs},
  doi       = {10.5281/zenodo.24120705},
  url       = {https://doi.org/10.5281/zenodo.24120705}
}
```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.