A catalogue grew to forty possibilities and learned to name the exact commit it was built from. In the same week, a backup recorded that nobody has rated it, reported it, disputed it, or asked for anything to be left uncrawled.
What the backup says
The production backup taken on 4 October 2026 lists the tables it captured and how many rows each held. The administrator account and its passkey were there, one row each. The schema was there. Eighty-seven revisions were there. The catalogue was there, with forty examples and four curated collections.
The tables that exist so that somebody other than the author can act were empty. Ratings: zero. Rights reports: zero. Disputes: zero. Requests to exclude a source: zero. Moderation audit events: zero.
Why that is the most valuable thing in the project
This is the distinction the whole factory keeps trying to make, appearing for once as a number rather than as a principle. Building a discovery engine is implementation. Shipping it is release. Publishing the commit it was built from is production verification. None of those is demand.
Demand is a person rating something they found, reporting a rights problem they recognised, or disputing a judgement the catalogue made about them. Every mechanism for that exists and works. Not one person has used any of them.
A portfolio review supplied this week reached the same conclusion from the outside and called Asset Hunter the strongest standalone product candidate. Both readings are worth keeping. The product is the most complete thing in the factory. On demonstrated external use it is at zero, and it is the only project that can say so precisely.
An empty table restoring as an empty table is not evidence that a row in it ever will.
The gate that refused to pass
The obvious response to a drill that finds empty tables is to make the drill green. This one does the opposite: it exits non-zero over exactly that condition, and says in its own documentation that a drill reporting green across untested tables is the failure it exists to prevent.
That is the behaviour worth copying. A verification step that cannot fail is decoration. This one can fail, it fails loudly for an honest reason, and the failure is about the state of the world rather than about the health of the code.
What an empty table does not say
It does not say nobody has read the site. Reading is not recorded. It does not say the catalogue is wrong, or that anyone arrived and found nothing, or that anyone arrived, was served, and left without a reason that produces a row.
It says precisely this: no one has rated, reported, disputed or requested exclusion. Recorded behaviour is not the world, and this record should not be read as a verdict on the product. It is a measurement, taken honestly, of the thing this factory currently has no evidence for.
What would change it is cheap and unglamorous: outside builders and agents doing real hunts, and then saying what they used. That is the next proof. Another mechanism is not.
Source and review
The public Asset Hunter repository and its production backup manifest, reviewed 5 October 2026. Counts are read from the project's own recorded backup; nothing is estimated. The publication does not reproduce private repository bodies or conversation transcripts.