dummyindex
Copyright (c) 2026 Ahmed Mulla

dummyindex is distributed under the MIT License (see LICENSE).

This file exists to carry attribution required by third-party licenses for
work incorporated into dummyindex.

--------------------------------------------------------------------------

headroom (Apache License 2.0)
https://github.com/headroomlabs-ai/headroom

    Headroom
    Copyright 2025 Headroom Contributors

    This product includes software developed by the Headroom Contributors.

`dummyindex/context/domains/memory/miner/` implements a deterministic
failure/loop miner whose technique is taken from headroom's
`headroom/learn/` package — principally `loops.py` (two-level grouping of
tool calls by a canonicalized signature, a minimum-occurrence threshold, and
a measured bytes-per-token waste estimate), `plugins/claude.py` and
`_shared.py` (walking a `projects/` transcript store whose location is
resolved through the `CLAUDE_CONFIG_DIR` environment variable), and `base.py`
(the scanner/plugin split).

The port was written against the described technique rather than by copying
files, and it diverges deliberately where dummyindex's inputs differ — most
visibly, structured JSON tool inputs get schema-aware field-dropping instead
of the string-regex normalization headroom applies to shell commands. Even
so, an audit measuring maximal contiguous token runs found short verbatim
spans surviving in the shell-normalization branch and the store-walk loop,
along with several shared private identifiers. That is close enough that
claiming a wholly independent reimplementation would be inaccurate, so this
notice is carried rather than argued away.

headroom's `headroom/learn/analyzer.py` is deliberately NOT ported: it is an
LLM-backed analyzer, and nothing in dummyindex's miner calls a model, opens a
network connection, or shells out.
