Islands of Stability, snapshot 1: pre-registration (r4)

This deposit fixes the rules of Islands of Stability snapshot 1 before any of its measurements are run. The snapshot measures how often a tool-using AI agent completes a fixed task (extracting fields from synthetic invoices) as the number of tools it can call grows from 1 to 6 and injected tool faults rise from 0 to 50 percent. Two propositions are tested, each by one pre-registered statistic. P1: success peaks at an intermediate number of tools, and the drop from that peak to six tools exceeds sampling error. P2: recovering from an injected fault costs measurable time compared with unperturbed controls. Every other number in the snapshot is descriptive. Contents of bundle.zip (SHA256SUMS lists a hash for every file):- configs/preregistration/r4.yaml: the decision rules the analysis reads; PREREGISTRATION.md states the same rules in prose.- configs/snapshot-1.yaml and configs/snapshot-1.lock.json: the snapshot configuration and the freeze lock, which hashes the rules, the configuration, the 100-document task set, the system prompt, the tool schemas, the grader, and the agent loop, fault, tool and provider code (harness v1.0.0).- docs/method-note.md: the statistical method, its simulation checks and the calibration.- results/s1/: calibration and determinism-gate records for the three local models (gpt-oss-20b, Qwen3-14B and Qwen3-4B, served by llama.cpp). Absolute paths on the author's machine were shortened in the gate records (see REDACTIONS.txt). The snapshot's results will be published separately and will cite this record. The bundle was replaced shortly after publication, before any snapshot run, to shorten local file paths in the three gate records; the rules, configuration and lock are byte-identical to the first upload (see REDACTIONS.txt).