The Feedstock-Interruption Observation Chains — Frozen Protocol, Results, and Registration Note

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

One registered observational record, deposited unaltered: a protocol frozen before the documentation pass began, the results of its single run, the registration note recording one piece of linkage housekeeping, and the ledger script that checks the record's internal consistency. It is the fourth census-class record of the programme, after the IBOR cessation census, the frozen-spread record, and the versioning census (deposited separately). **What the record is.** Three interruption-response chains — Stack Overflow–ChatGPT (primary), Wikipedia–LLM (parallel), and xz-utils (corroborating) — each scored as: a dated interruption documented in primary or first-tier sources; a ledger of dated responses coded under a taxonomy fixed in the frozen protocol (feedstock decline; quality margin; feedstock purchase; fencing/monetization; conversion; producer subsidy); a forbidden-outcome falsifier; and the manuscript's registered counterexample screen applied unchanged. What upgrades the observations from anecdote to structure is this pre-commitment, not statistics. **Honesty architecture, stated up front.** The protocol's analyst-disclosure section records that **selection on outcome is designed in**: the three chains were chosen because their headline events are publicly known, so they illustrate and discipline — they are not a sample, and no frequency claim may be built on them. The protocol also fixes, ex ante, that the feedstock node is **contribution flow, never pageviews** (excluding the demand-side AI-summary confound), and that all three chains are **approximate** to the manuscript's interruption class — corroboration and a falsifier screen, not executed identification; the manuscript's sentence that it "defines the experiment and does not claim to have executed it" stands. **Headline result.** Falsifier empty; predictions P-C1 and P-C2 met in all three chains — every chain documents feedstock-side decline or thinning, and every chain documents at least one predicted feedstock-preserving response (purchase, fencing, or subsidy), each dated and sourced. **The quality-margin prediction (P-C3) is met by reporting, and the reading is a null-to-date:** a measured decline in venue output quality is documented for neither live venue (Stack Overflow carries a documented six-month null; Wikipedia documents contamination pressure and provenance screens, not a measured decline), while the xz-utils failure event is itself the realized quality failure at a documented feedstock thinning. One deviation is disclosed (D-1): the frozen taxonomy lacked a code for provenance-screening responses, so four such events are recorded as uncoded context rather than forced into a class. This record disciplines the applications of the manuscript *The Causal Partition of Information* (Yip, 2026); no proposition of that paper depends on it, and the chains are deliberately not yet integrated into the manuscript body.

Links and sources

Comments