# CRUCIBLE P0.3: testing the cost of composition

Research brief · JCEE Labs · 2026-09-14 · Version 1.0

The broad operational-burden claim did not survive the tested comparison. A narrower integration-surface result remains.

## The question after P0.2

P0.2 preserved a six-part semantic kernel and demonstrated conventional parity within its synthetic benchmark. P0.3 asked a different question: did the JCEE composition materially reduce operational burden against a strong conventional composition? The earlier semantic result remains a distinct, preserved milestone.

## The terminal result

The terminal disposition is KILL_BROAD_COMPOSITION_COMPRESSION_CLAIM. Both implementations met the safety criteria in the tested synthetic worlds. Late misconfiguration escapes, manual recovery actions, diagnosis burden, and offline audit burden tied. This comparison did not establish a broad operational-burden advantage for JCEE.

A narrow integration-surface difference remained: three versus five integration touchpoints in both the payment and cold-domain evaluations. That observation supports a narrower future question, not a general superiority claim.

## Reproduction and scope

The recorded GitHub Actions reproduction passed on Python 3.10 and 3.12. It checked the frozen source and preregistration identities, 160 payment fault trials, the frozen integration boundary, the subsequent cold-domain reveal, and 20 cold-domain fault trials. Both environments reproduced the terminal disposition.

This is a second execution environment for the same controlled research. It does not establish independent-team authorship, customer economics, production safety, or general transfer to real integrations.

## What follows

P0.3 is frozen. The broad composition-compression claim is retired for this tested comparison. The P0.2 semantic result remains intact.

A further study would need a separately defined question about the narrow integration-surface difference, using heterogeneous real adapters or independent implementers. No such result is claimed here.

## Continue reading

- [Read the preserved P0.2 semantic-kernel report](https://jceelabs.com/research/crucible-semantic-kernel)
- [Current CRUCIBLE status](https://jceelabs.com/registry#crucible-p03)