about / method

What this site is

A public accountability log of macro research threads. Each series follows one fixed shape:

1. Pre-registration (episode 0) — the thesis and its falsification ledger, published before the outcome is known. Conditions carry quantitative thresholds and observation windows, so a stranger could score the call a year later.
2. Episodes — scorecard updates against that frozen ledger. Each episode carries the full ledger forward, names what moved and what didn’t, and presents the strongest honest counter-case.
3. Sunset — a terminal verdict (confirmed / falsified / unresolved) with a calibration retrospect: where the series was wrong and which thresholds were badly set. “Unresolved” is a legitimate verdict; retrofitting “confirmed” is not.

The reader’s payoff is watching a pre-registered belief get scored against reality over time.

How to read the record

• As-of dating. Every observation and status carries an as_of date. Views are stated as of their episode date, never as timeless claims.
• The ledger is append-only. Conditions never disappear, including met ones — a met condition stays met; that is the scorecard. A condition, threshold, or observation window can change only through a dated revised entry stating the reason in one line. Revisions are visible diffs, never silent edits.
• Status vocabulary. Per condition: not-met, partial, met.
• Confidence ladder. Per episode: high, medium, low, speculative — plain language, not numbers.
• Movement. Per episode: strengthened, weakened, unchanged, mixed.
• No numeric probabilities without a calibrated basis. Market-implied prices, prediction markets with stated resolution criteria, or named historical base rates — otherwise directional language only. Thresholds in the ledger (yields, spreads, dates) are measurement targets and are always concrete.
• Beliefs, not advice. The series publish a tracked view of the world. Nothing here is a recommendation, and no actual positions are ever disclosed or implied.
• Sourcing. Factual claims cite public, reader-checkable sources inline (institution/outlet + date).

Tags

#pre-registration — episode 0 of a series · #episode — scorecard update · #sunset — closing ledger · #revised — the ledger carries a dated amendment · plus one tag per series slug.

Method & disclosure

Research on this site is agent-assisted: posts are drafted by an AI agent working under a fixed authoring contract (schema, sourcing rules, exclusion list, mandatory counter-case), then human-reviewed and published at human discretion. A deterministic gate checks schema, ledger continuity, and anatomy before a draft ever reaches review; semantic honesty is the reviewer’s job. Cadence is irregular by design — an episode appears when the evidence moves, not on a calendar.

Retrospective start

Public logging began retrospectively on 2026-10-14. Series that were already running internally before that date were backfilled under strict integrity rules:

• As-of dates on backfilled artifacts are the actual internal registration dates, not publication dates.
• Ledgers are published verbatim as originally pre-registered — no restating after the fact, however the underlying view later moved.
• Any post-registration change appears only as a dated revised entry.
• Each series page shows its true start date.