Keshav
An impossible classical library: shelves rising out of sight, receding into shadow.

Keshav

Engineer & researcher

Enter the archiveOpen the FRONTIER volume
Frontierexp-009 · 27 October 2026

The question

Is the modelled saving real?

Hypothesis
Fail-fast verification saves wall time in proportion to the modelled charge.
Setup
1,257 candidates × 2 modes × 3 repetitions, timed; four campaigns of 700.
Result
6.4% measured against 92.9% modelled. About 92% of a verification's cost is fixed interpreter overhead (≈120.5 ms + 0.05 ms per case). Verdicts were identical.

Cost model refuted

On the 1,020 FAIL candidates: 92.9% modelled, 6.4% measured [5.1, 7.8%]

About 92% of a verification is fixed interpreter overhead.

Early stopping preserved correctness: identical verdicts on 1,257/1,257 candidates. What failed is the assumption that charged units track real execution time, on this Windows DSL harness.

Limitations One host (i5-1135G7, 8 threads); campaign-level timing was confounded by host load and is reported as uninformative.

Colophon

k, from receipts
VerifierReceipt schema 0.2 records verdict and cost_units, not wall-clock time. Each candidate’s k is cost_units / 0.005.
Timings, from measurements
exp-009 measurement records (M-measurements.jsonl.gz, evaluation.json): the median of three timed runs per mode, and the fit 120.53 ms + 0.0516 ms per case.
Source
Kesh3805/frontier · results/exp-009/README.md
An alignment chart199 cases per verification
Model promised
99.5%
Cases run, k
1 of 199
Fitted wall-time
7.8%
Fail-fast verification, per failing candidate. The charge counts cases run, so the model promises 1 − k/199. The clock pays 120.5 ms before the first case and 0.0516 ms per case after it.