---
url: >-
  https://adk.nht.io/api/@nhtio/adk/batteries/context/thrift/relevance/variables/RELEVANCE_FLOOR_MAX.md
description: >-
  The relevance floor's upper bound — the minimum shared-content-word fraction
  required when the window is nearly full (strict: only keep turns with strong,
  unambiguous overlap).
---

# Variable: RELEVANCE\_FLOOR\_MAX

```ts
const RELEVANCE_FLOOR_MAX: 0.43 = 0.43;
```

Defined in: [src/batteries/context/thrift/relevance.ts:57](https://github.com/NHTIO/ADK/blob/v1.20260713.1/src/src/batteries/context/thrift/relevance.ts#L57)

The relevance floor's upper bound — the minimum shared-content-word fraction required when the
window is nearly full (strict: only keep turns with strong, unambiguous overlap).

## Remarks

CALIBRATED against the same triple-oracle stress corpus as [RELEVANCE\_FLOOR\_MIN](RELEVANCE_FLOOR_MIN.md): the
high-precision floor (precision ≥ 0.9, i.e. "when in doubt about whether this survives, don't miss
on the side of dropping something the user is about to ask about") landed BIT-IDENTICAL at
`0.4286` across all three judge models (gpt-5.5, gemini-3.5-flash, claude-haiku-4.5) — a striking
cross-model agreement that grounds this as a real property of the corpus, not judge idiosyncrasy.