L5 / Prefix-cache invalidation

01 / QUESTION

Three seeds and six conditions compare unchecked and keyed prefix reuse; numerical equality does not establish permission to share across scopes.

METHOD

An untrained two-layer network, three seeds × six conditions; compare fresh computation, unchecked reuse and keyed reuse.

RESULT

All 18 guarded paths have measured error 0 on this float64 run; unchecked reuse produces nonzero errors after prefix, weight or window changes.

OBSERVATIONS

A changed scope still yields identical numbers, but the key denies reuse. Continuation projects 24 rows on hits and 72 on misses.

CONCLUSION

Prefix identity needs computation settings and application scope; numerical equivalence does not establish permission to reuse.

LIMITATIONS

No language, GPU or latency measurements. Scope is a caller-supplied string, not authentication. Full weight hashing is appropriate only for this teaching model.

REPRODUCE

The pinned repository includes protocol, raw results and independent auditing; the article provides reproduction commands.

1. Mechanisms

   LLM mechanics (4): when a prefix changes, is the old KV cache still valid?

   Three seeds and six conditions compare unchecked and keyed prefix reuse; numerical equality does not establish permission to share across scopes.

   2026-10-039 MINPublished
2. PROJECT / 002

   Java LLM Practice

   Reproducible Java conversation, retry and persistence boundaries, alongside attention, cache-cost and quantization-error experiments.

   Building
3. Mechanisms

   LLM mechanics II: what does KV caching save, and how do we verify it?

   Compare prefix recomputation, token decoding and chunk caching with output errors, projected-row counts and actual array payloads.

   2026-10-038 MINPublished
4. Mechanisms

   LLM mechanics III: why is a four-slot cache not four-token recomputation?

   Separate absolute positions, cache slots and retained representations with a two-layer sliding-attention experiment.

   2026-10-038 MINPublished