---
id: H-003
title: Crystallization
status: EXPERIMENTING
created: "2026-08-22"
experiments: [E-002]
concepts: [C-crystallization]
---

# H-003 — Crystallization

## Claim

A sufficiently stable agentic execution can be converted into deterministic test
code without losing relevant oracle coverage.

## Falsification condition

Any of:
- the generated deterministic test cannot be produced without human authorship of
  the assertions;
- it drops one or more oracles present in the agentic ancestor;
- it passes where the ancestor would have failed, or vice versa, on the same lab
  version;
- it costs no less to execute than the agentic ancestor.

## Measurement

- Crystallization success rate.
- Oracle set of descendant vs ancestor — must be a superset or equal.
- Verdict agreement between ancestor and descendant across lab versions.
- Execution cost ratio (tokens, wall time) descendant : ancestor.

## Note on the cost criterion

The cost criterion is included deliberately. If crystallization preserves
semantics but saves nothing, the thesis is intact but the product rationale is
not — and that distinction should be visible in the record rather than discovered
by a user.

## Result (TD-WP-0002-T09)

| Criterion | Outcome |
|---|---|
| produced without human authorship of assertions | **met** — assertions are imported, not restated |
| drops no oracle from the ancestor | **met** — descendant's set is a superset |
| agrees with the ancestor across lab versions | **met** — baseline, M15, M17, M16, M13 |
| costs less to execute | **met, but see F-0007** |

Also demonstrated: the descendant still catches a seeded defect (M17), it fails
*loudly* rather than searching when its frozen path breaks (M24), and the agentic
ancestor recovers from that same M24 where the descendant cannot — the
reversibility case from `INTENT.md`, shown in both directions.

**Not falsified. Not yet supported on the criterion that matters most.** The
measured 54 % cost reduction comes entirely from removing local discovery work,
because the T07 runtime consumes no tokens. The saving crystallization actually
claims — model tokens, latency, retry variance — is unmeasured and unmeasurable
with this runtime. See `research/findings/F-0007-...`.

## Status log

- 2026-08-22 `PROPOSED`. No evidence.
- 2026-08-23 `EXPERIMENTING`. Fidelity criteria met; the economic criterion is
  met only in a sense too weak to support the thesis.