THE ORGAN CATALOG AND THE QUESTION OF WHAT A SONG IS
=====================================================

NITETIME ORGAN SYSTEM // CATALOG AND CURATION

The catalog is where raw MIDI files become known pieces in a living
station. It stores identity, technical data, descriptions, human decisions
and playback history without pretending those are all the same kind of fact.


CONTENT IDENTITY
================

Every MIDI file is identified by the SHA-256 digest of its contents. The
catalog can therefore distinguish:

    one file copied to several locations
    two arrangements of the same composition
    a modified version with different bytes
    a renamed file that is still exactly the same performance

The hash-based approach is stable and reproducible. It is not a complete
understanding of music. Two arrangements can be compositionally identical
while remaining different content objects. An optional composition/work ID
would eventually allow rotation rules to understand that relationship.


TECHNICAL READING
=================

The built-in reader understands common metrical MIDI timing and extracts
format, tracks, tempo, duration, meter, text, channels, programs and note
counts. It integrates tempo events rather than assuming every piece stays at
one tempo.

When a file has no tempo or meter event, playback uses standard MIDI defaults
and marks the inferred meter in the display. An unreadable or unsupported
file is excluded from random selection.

The overlay receives a reduced telemetry summary rather than the entire
catalog record. That keeps the public display useful and bounded.


IDENTIFICATION ASSISTANCE
=========================

Technical parsing answers "what is in this file?" Descriptive analysis tries
to answer "what song might this be?"

A local language model can examine the filename and embedded MIDI metadata
and suggest a title, artist, year, genre, description, confidence and
possible facts. The analysis happens asynchronously and is capped so it
cannot monopolize the organ controller.

The model is an assistant, not an archivist with final authority. Candidate
links and facts remain candidates until a person verifies them. On-air facts
require an approved manual record with a supporting source.

When nothing is sufficiently identified, the overlay says:

    UNKNOWN ORGAN TRANSMISSION

That is more honest than inventing a composer.


MANUAL CURATION
==============

Human corrections can supply a title, artist or composer, year, genre,
description and a verified fact with its source. Those corrections remain
authoritative after an automatic reanalysis.

The system also supports an explicit unverified label for useful provisional
credits. This preserves the difference between "we have a good guess" and
"we have checked the claim."


SCANNING AND AVAILABILITY
=========================

The scanner notices new content and remembers every observed location. A
missing or moved file becomes unavailable after a scan, but historical and
descriptive data remains recoverable.

This is intentional. Removing a file from the live library should not erase
the fact that it once played, who corrected its metadata or why it was
quarantined.


THE CURATOR'S RULE
==================

The catalog keeps several questions apart:

    Is the file structurally valid?
    Is it available to play?
    What does its metadata say?
    What does the model guess?
    What has a person verified?
    What actually happened when it played?

Separating those questions is what lets an automated station remember its
mistakes without turning them into permanent truth.