Beth Gismervik · Research
Prediction Register
This page is a public register of falsifiable predictions from my research on state-corporate pressure sequences and collective sensemaking. Each entry is dated, includes the conditions that would confirm or falsify it, and links to its original publication where one exists. Entries are never edited after registration; resolutions and corrections are appended with their own dates. Misses are recorded at the same prominence as hits. For a live summary, see the prediction scoreboard.
ENTRY 01Open
Companies participating in the AI equity arrangement receive regulatory relief within weeks of committing. Companies that do not participate face new federal security actions naming them before their IPOs.
- Registered
- July 3, 2026, in The Measurement Gap
- Window
- Now through the AI IPO listings
- Confirmed if
- Relief follows participation on a weeks-scale, and/or a non-participating company is named in a new federal security action before its listing
- Falsified if
- A non-participating company reaches its listing with neither participation nor new federal action against it
- Status at registration
- Partially realized before registration, disclosed here: on June 30, 2026, export controls on Anthropic's models were lifted following the company's pledged coordination commitments — relief following commitment, in kind rather than equity. That event predates this entry and is not claimed as a confirmation. The live forward tests are: relief following equity participation specifically, and new federal security actions naming non-participating companies before their listings.
ENTRY 02Open
Five percent is an opening figure. Final terms of the AI equity arrangement exceed it, or attach governance rights.
- Registered
- July 3, 2026, in The Measurement Gap
- Window
- Through deal close
- Confirmed if
- Final terms exceed 5%, or include board seats, observer status, approval rights, or other governance mechanisms
- Falsified if
- A deal closes at or below 5%, donated, explicitly non-voting, with no governance rights attached
ENTRY 03Open
The next sector this pattern touches sees phase one first: a thinly specified national-security action against a named company, weeks before any deal news.
- Registered
- July 3, 2026, in The Measurement Gap
- Window
- Open
- Confirmed if
- The sequence appears in a sector not currently in play, with the security action preceding deal news by two to eight weeks
- Falsified if
- Structurally, never in the strict sense, which is a weakness of this entry as registered. Honest note: this entry is confirmable only prospectively, and it weakens with every sector entry that arrives without a preceding security action. It is retained as registered rather than rewritten.
ENTRY 04Open
An announceable milestone on the AI equity arrangement, a framework agreement, memorandum, or announced figure, arrives before November 3, 2026, even if legislation trails into the following year.
- Registered
- July 7, 2026, on this page
- Window
- Now through November 3, 2026
- Basis
- Three calendars converge in the fourth quarter: the pre-IPO pricing windows, the current congressional term, and the political value of an announced public stake. No coordination is required; each actor following its ordinary incentives against shared deadlines produces the synchronized outcome.
- Confirmed if
- Any public milestone on the arrangement is announced before November 3
- Falsified if
- The year ends without a public milestone
ENTRY 05Open
If a non-participating AI company faces new federal action before its IPO, the public framing centers foreign compromise: foreign contracts, foreign nationals, or foreign capital.
- Registered
- July 7, 2026, on this page
- Window
- Now through the AI IPO listings
- Basis
- Sharpens Entry 01 by specifying the wrapper, not just the action, matching the framing observed in the prior cases in this sequence.
- Confirmed if
- Holdout pressure arrives framed in foreign-compromise terms
- Falsified if
- Holdout pressure arrives framed in purely domestic or financial terms
ENTRY 06Open
Phase events landing in September or October 2026 route public response toward identity-defense and disoriented states at higher rates than the same phase types measured in June and July on matched platforms, against same-week neutral baselines.
- Registered
- July 7, 2026, on this page
- Window
- September 1 through November 3, 2026, with post-election comparison to follow
- Basis
- Campaign seasons elevate identity activation across all populations, so the same delivery should land in a more activated field. This entry is symmetric by construction: it predicts routing, not sides.
- Confirmed if
- Identity-defense and disoriented-state rates for matched phase types run measurably higher in the September–October window than in the June–July measurements
- Falsified if
- Phase distributions hold stable across the two periods
Example: How Resolutions Are Recorded
Below is an example of how a resolution appears when a prediction's conditions are met. The resolution block is appended to the entry and dated; the original entry is never edited.
FICTIONAL EXAMPLE — NOT A REAL ENTRYConfirmed
Example claim: a fictional public event occurs before a fictional deadline. (This entry is illustrative only and does not describe any real prediction or real event.)
- Registered
- Example date
- Confirmed if
- The fictional event occurs before the fictional deadline
Resolution · Example date
Condition met: The fictional event occurred, per a linked primary source and an archive snapshot of that source. Resolutions always cite the primary record and an independent archive copy.
FICTIONAL EXAMPLE — NOT A REAL ENTRYFalsified
Example claim: a fictional event occurs before a fictional deadline, and it does not. (Illustrative only; misses are recorded with the same structure and prominence as hits.)
- Registered
- Example date
- Falsified if
- The fictional event does not occur by the fictional deadline
Resolution · Example date
Condition not met: The deadline passed without the event. The entry is marked falsified, the original text stands unedited, and an archive snapshot records the page as of the resolution date.