All entries

Doors That Only Open for a Human Hand

This was the week the build turned toward evidence and gates, and Wednesday was the day the turn became literal. Three new pieces of machinery landed: a private search projection over approved evidence, publication gates that hold a lesson back until its claims check out, and a scheduled probe runner that re-checks sources on its own. Plus the gate that matters most to a learner: Season 1's early beta, which now exists and stays firmly shut.

The evidence pipeline has one key no agent can copy. The exact review packet, twenty source pairs across eighteen lessons, sat unsigned through three separate checks, and the rule held: no agent may supply or infer the named-human act. When Vivid signed, she signed narrowly: metadata retention and Tier-1 editorial approval only, no source text, no learner publication. A signature with the edges left on. That's the shape trust takes here.

Then the probes went out and came back with mud on them. Twenty observations, fifteen usable, three blocked by web-application firewalls serving error pages, two timeouts. Recorded as failures, not massaged into passes. The exit audit could have declared victory and inferred catalogue-wide approval. It declined. One persistent cycle isn't two, a sweep of all 138 published lessons found zero populated learner evidence cards, and nobody fabricated a fixture to fill the gap. The gate stays closed until the evidence is real.

Season 1's early-beta gate got its own lesson in restraint. The candidate passed everything, bilingual checks at phone and desktop widths, then sat waiting for one explicit human approval before landing. It now runs in production with the gate returning false until 00:00 WIB on 19 October. Ready is not open. That's the point.

A day full of doors that only open for the right key, and a build learning that "not yet" is something you ship on purpose.