Examine · How this work works

Every Seat

This account describes the mechanism, including the reveal. Read it before entering if you prefer to know how the work is constructed.

Five positions, one author

You move through five seats, choosing one of three positions at each. A safety lead informs a chief executive; another lab responds; two governments form positions. Your earlier words return as statements, constraints and costs in later briefings. The final trace makes those origins inspectable.

Fictional composites

The briefings are authored fiction informed by the dated record below. They are not quotations, reconstructions of private meetings, or claims about what named people would choose. The two labs and governments simplify a much wider field. Seat five combines several distinct voices; it does not represent a single Chinese decision-maker. Intelligence estimates, relative leads and private intentions inside the briefings are fictional premises, not verified measurements.

What changes

All three choices remain available at every seat. Choosing changes the inherited information and the cost of taking a position, rather than unlocking or disabling answers. Publication is determined by the second seat: withholding the findings prevents the first seat’s statement from reaching subsequent briefings. Each immediate predecessor materially changes the next seat; more distant choices may carry context instead.

Outcomes and pivots

There are 243 possible sequences. A fixed rule names four families, in priority order: a mechanism paired with reciprocal verification; two laboratory holds without a lead-first policy or rejection; a remaining laboratory hold or rejected mechanism; otherwise, race. These describe the constructed arrangement. They are not scores, forecasts or recommendations.

The pivot searches seats two to five in order. It changes one choice while holding the other four fixed and reports the earliest seat that could change the family. It does not replay later decisions, identify a morally responsible actor, or predict how real people would respond. The first seat shapes information but is not itself a family-classification input. When no single change alters the family, that limit is stated.

Reflection and another sitting

Reflection uses fixed templates: your five actions, the pivot and one remaining uncertainty. No model analyses your decisions. Sit Again starts another sequence; completed sittings can be compared while this page remains open. Nothing rates the visitor or names a correct route.

Privacy

Your choices and comparisons between sittings exist only in this page’s memory. They are not put in cookies, browser storage, URLs or history state, and are not sent to a server. Reloading, leaving or clearing the page removes them. We may measure structural stages reached, coarse active-time ranges, technical delivery signals, opening this mechanics/evidence page and aggregate transitions between pages. Choices, decision values, tuples, outcome families, pivots and reflection results are never collected. Ordinary delivery requests still reach the hosting service; following a source link contacts that source’s website.

Evidence and limits

The Record closes on 14 September 2026. Retrieval dates describe checking, not new events added after the cutoff. Each entry is limited to its primary source: an investigation documents its findings, a statement establishes a public position, and a forecast remains a forecast. A disputed item needs both sides. Missing verification is never filled from a secondary report. The Record cannot reveal private practice or establish a complete map of the debate.

AI assistance

This piece was built with AI assistance from systems made by Anthropic and OpenAI, two of the actors in the record. The runtime does not call an AI model. The mechanical verifier cannot establish whether readers experience the work as a trick, a leading question or a verdict; those judgments require real readers.

Enter Every Seat