Research Models
===============

Ordinary users do not select an algorithm. The public command and Python API use the current solver internally under the generic
name ``coposit``.

A complete source build exposes literature baselines and coposit-created models through the research ``--model`` option. Model
identifiers remain stable so that stored results, scripts, and citations continue to work when source directories are reorganized.

The table below is generated in the browser directly from the authoritative ``models/MODELS.tsv`` file. It therefore follows every
catalogue update automatically; there is no separately maintained documentation table.

Model catalogue
---------------

.. raw:: html

   <div class="model-catalogue" data-model-catalogue data-source="MODELS.tsv">
     <label for="model-catalogue-filter">Filter models</label>
     <input id="model-catalogue-filter" type="search" placeholder="Name, family, backend, mode…" disabled>
     <span data-model-status aria-live="polite">Loading model catalogue…</span>
     <div class="model-catalogue-table" tabindex="0">
       <table>
         <thead></thead>
       </table>
     </div>
     <noscript>
       <p>JavaScript is disabled. <a href="MODELS.tsv">Open the raw model catalogue</a>.</p>
     </noscript>
   </div>

The `raw TSV catalogue <MODELS.tsv>`_ remains available for downloading and machine use. It records the human name, status,
lineage, principal change, Boolean backend, pruning directions, supported modes, replacement status, and algorithm-document path
for every retained source model.

Source layout
-------------

* Faithful literature and source baselines are under ``models/00-baselines/``.
* Every coposit-created or adapted model is directly under ``models/<model-id>/``.
* Every model has an authoritative ``ALGORITHM.md`` beside its implementation.

Status meanings
---------------

``baseline``
   Faithful published or source algorithm.

``active``
   Active research candidate.

``retained``
   Completed experiment still available for controlled comparisons.

``superseded``
   Replaced by a later model but still built for historical comparison.

``archived``
   Preserved in source but excluded from normal builds and research model lists.

The catalogue contains no benchmark timings. Runtime results belong to the results database because they change independently of a
model's identity and lineage.