PRE-GHR Position G: When Pacing Is Also Leverage — How Recording Treats "Lead First, Negotiate Later" (当定速也是杠杆:记录如何对待「先领先、再谈判」)

**Position G v1.3 (PRE-GHR 独立立场篇 / Position track, third record).** *When pacing is also leverage: how recording treats "lead first, negotiate later".* F answered what recording means when you cannot constrain; G answers the adjacent cell — *which conflicts must not be synthesized into the superintelligence question*. Argumentation is unchanged from v1.1/v1.2: v1.3 corrects the accounting, attribution and declaration layers only. **1. The second text of "persuasion".** D. Amodei, *We Must Pace the Frontier* (September 2026) is read here as a buildable pacing roadmap, not as consensus: a three-step plan (embedded third-party evaluators with employee-like access and the right to publish findings without editorial control → coordination among democratic states → global coordination), a lead-defense package (chip export controls, crackdown on unauthorized distillation, model-weight security), and a four-level negotiation ladder (L1 narrow obviously-dangerous uses such as biological weapons; L2 pre-release acute-risk testing; L3 an RSI "speed limit"; L4 full pacing/pause, which the original itself judges unlikely any time soon). Its structural signature is *leverage first, negotiation later* — pacing is not a concession, it is a lever. **2. The negative-sum zone.** What can be negotiated across borders is the set of losses common to all sides: runaway spillover, biological weapons, incident disclosure. Disclosure is treated not as goodwill but as the cross-border interface of recording: the ledger structure is the *important condition* for when a failure becomes known — not intent (the series' own two-layer discovery period is cited from Position F). **3. Prohibition of synthesis (the paper's defining move).** *Muddling through* pretends the seam does not exist; *prohibition of synthesis* admits the seam, records the cut — and refuses to map that cut into a superintelligence coordinate system. The admissible domain may shrink; it must not be averaged between two anchors (a representation-theoretic sentence from companion paper A is cited by concept DOI, without arguing which anchor is true). Two test questions are supplied: who has the authority to declare a question frozen in the atmosphere, and does the rule bind its own designer? **4. Stances (marked as stances, not predictions, not theorems):** a single signatory interlocutor — where the open question is the absence of an *independently verifiable representation record*, not the absence of any ledger; ideology frozen at the atmospheric layer, not entering the alignment function; the superintelligence question escalated in domain rather than settled inside the two-anchor frame. **5. The third kind of winner:** a *mechanism position* defined by two criteria — its stock consists of the capacity to veto others' interventions plus the legitimacy accumulated by staying out, and it occupies no station on the intervention axis — with an explicit refusal to map any real-world actor onto it. Five open questions follow, with judgments deliberately left blank. **External-name and source declaration (series standard).** The only first-hand source is D. Amodei's essay: the local archive holds the raw HTML and an extracted text (md5 3aa4a1d270fad6eb95f918477881ccdd / 4eb71036aaaa90bd8f560b80d57baf9f), and all verbatim quotes are anchored to line numbers of that extraction; the page states only "September 2026", so no day-level date is asserted in the text. Position F is cited by DOI; companion paper A is cited by concept DOI 10.5281/zenodo.22058663 (quoted sentence verified against A's v4 text, file md5 dcbf2c50470766500823e70ef9f935d3). All library material is cited by line-number anchors. **Attribution / contributor ledger (platform + model double-write rule, series standard).** Author: Wang Miaosheng (ORCID 0009-0003-2767-2421). The conversational organizer that supplied the working notes is *not* an author. Structure observation and the two closing lines on the ascetic position, together with the three external tickets filed on v1.1 = Grok — **its platform and model version are not recorded in this ledger: declared, not hidden**. Off-axis placement (a veto on the axis itself, not a fourth voice) and the counterexample set, plus red-pen review and editorial rulings = Kimi. Foundation audit = GPT-5.6 Luna. Post-publication boundary patrol on v1.2 (three tickets: two accepted, one redirected to N) = GLM (Zhipu) — **model version likewise not recorded in the ledger: declared, not hidden**. Drafting and engineering execution = Hermes Agent (Nous Research; API model DeepSeek v4 Flash). Data: EntropyRuntime ledger. **Version note (v1.3).** v1 published 2026-09-13; v1.1 corrected a stale status line, the title note and the companion-A citation form; v1.2 applied four wording-level corrections raised by an external foundation audit and ruled on by the series editor. **v1.3 corrects the accounting, attribution and declaration layers with no change to the argument:** (1) the audit envelope and the editor's re-graded disposition are now stated side by side (2 P1 + 3 P2 + 2 N as filed → 1 P1 + 2 P2 + 2 P3 + 2 N as ruled; all seven tickets accounted for), removing an envelope that could not be reconciled with the paper's own disposition list on any of the three layers that carried it; (2) the version DOI of this record is written in three places — the manuscript's status line, this metadata note, and the *isNewVersionOf* edge — instead of a pointer to "see the Zenodo record"; (3) contributor attribution now carries the platform+model double-write for every AI contributor, and where a platform or model version is not recorded, that absence is declared rather than left blank. **Outstanding obligations (declared, not hidden).** (1) The author's section-by-section reading of the Amodei original was completed on 2026-09-13. (2) The mechanism paragraph concerning a written normative anchor versus current operating standard is *not* in this paper: its gate is a preregistered sampling, not yet cleared — a deliberate omission, not a gap. (3) Position-track registration in the Series Map ships with the next map update. (4) One archived material item's candidate slot is retained as an independent-paper candidate. License CC-BY-4.0.