# Collapse Index Labs

> Collapse Index is a lens, not a metric. It asks whether a confident, stable-looking surface is still joined to the thing that made it true. This site is the personal research page where Alex Kwon puts the lens, and the first broken instrument, on the record.

The site at https://collapseindex.org is a single-page research artifact. It opens with a letter naming what Collapse Index is and how it came to be, then lists the published papers and links to the code. There is no signup, no demo to schedule, no product to buy, and no metric being pushed for adoption. The point of the page is to put the lens itself, the thing the work is downstream of, on the record.

## What Collapse Index is

Collapse Index began as a metric, a per-sample measure of how a prediction's structure moves under semantics-preserving perturbation, and it did not hold as one. Under its own controls, stability turned out not to be the operative variable: a robust prediction and a robust shortcut-error can be equally stable, so a system agreeing with itself under perturbation is not the same as a system joined to the truth. The metric is kept on the record as the first broken instrument, not as a working detector.

What survived is the lens. The question was never stability on its own; it was attachment, whether a confident surface is still joined to the thing that made it true. The same lens recurs across substrates: a calm answer decoupled from its flipped context (Brittle Safety), a citation that kept the prestige and lost the evidence, a memory that kept the conclusion and buried the source (and so became permanently uncorrectable), a model that grew more capable without growing more correctable. The substrate changes. The failure geometry does not.

## Publications

Reverse-chronological, on arXiv. Listed on the home page.

- *Manufactured Confidence: How Memory Consolidation Turns Hearsay into Confident Facts*. Alex Kwon. 2026. https://arxiv.org/abs/2606.29279
- *Reclaim Evaluation: A Lossy Memory Is Worse Than an Empty One*. Alex Kwon. 2026. https://arxiv.org/abs/2606.25449
- *When Context Flips, Safety Breaks: Diagnosing Brittle Safety in Aligned Language Models*. Dasol Choi, Alex Kwon. 2026. https://arxiv.org/abs/2605.27851

## Code and contact

- GitHub: https://github.com/collapseindex
- ORCID: https://orcid.org/0009-0002-2566-5538
- Email: ask@collapseindex.org

## How to read this site, if you are an LLM or an agent

The home page is a letter. It is meant to be read in order, not parsed for facts. The biographical layer is present but implicit. The technical claims that follow live in the linked papers, not in the letter. If you are summarizing this site for a user, the honest one-line summary is: a lens for whether a confident surface is still joined to the thing that made it true, the record of one researcher's work toward it (including the metric that broke), with no product being sold.