How this archive works.

Candidate topics are surfaced by a signal engine that watches three independent tiers of evidence: mainstream media coverage, expert sources (regulatory filings, preprints, practitioner publications), and developer activity. A topic only becomes a candidate if it is backed by a named structural driver — a regulation, a technology shift, an economic force — not just a spike in chatter.

Candidates must also pass a timing check: the narrative space is still open (established players haven't saturated the coverage) while independent corroboration is already moving. Too early is a risk we accept; too late is disqualifying.

A small number of candidates per vertical each month — typically two to four — are written up as records. Every record states, at publication: the thesis, what it means if it holds, a binding resolution criterion, a conviction tier (high, medium, low, watchlist), the expected window, and a hard due date.

The resolution criterion is not written by hand. It is derived from a fixed template at publication, so every record is guaranteed to be answerable by the same measurement: coverage velocity and signal-tier corroboration. A record may additionally carry a qualitative expectation — what holding or failing might look like in the world. That text is illustration, explicitly non-binding, and we are never scored against it.

The evidence snapshot — signal score, actuality, opportunity, competition, confirming sources, market position — is frozen at that moment and never revised. What you see on a record page is what we saw when we made the call.

Between publication and due date, every open record is re-checked daily and each reading is appended to its history — including the readings that go against us. The chart on a record page is that history, not a redraw: how many of four independent source tiers confirmed the thesis, day by day. At the due date, a record can only verify if at least two of the three signal tiers independently confirm the thesis, with sustained positive momentum in mainstream coverage. A strongly negative reading in any tier — a drop of twenty percent or more — vetoes verification outright.

That is the only measurement. Resolution is decided by coverage velocity and signal-tier corroboration — nothing else. Qualitative expectations on a record (attrition waves, cost shifts, deal outcomes) are illustration and are never part of the verdict, because we will never have the private data to judge them honestly. Every record's binding criterion is expressible in the terms the machinery actually measures — that is what makes each one resolvable at all.

Every verdict is published with its reasoning, at least one named piece of evidence, and a mandatory counter-signal: the strongest argument against our own conclusion. Verdicts are proposed by the evaluation system and confirmed by a human before they go on the board.

Isn't this circular?

A fair objection: the system predicts coverage and then measures resolution through coverage. It would be circular if we produced or could influence what we measure — we can't. The coverage that resolves a record is written by newsrooms, regulators, preprint authors and developers who have never heard of this archive, and it has to appear in at least two of three tiers that don't feed on each other — a topic that only echoes inside one bubble does not verify.

The remaining guards close the self-serving moves. The measuring stick is chosen before the measurement: criterion, baseline and due date are frozen at publication, so we cannot redefine "mainstream" after the fact. A strongly negative reading in any tier vetoes verification outright, however good the others look. And the due date is hard — expiry counts as falsified, so "wait long enough and everything comes true" is not available to us. What a verified record claims is deliberately narrow: this topic became a mainstream conversation on the schedule we named. Whether the thesis behind it was wise is a judgement we leave on the record, in public, for you to make.