Results

Read from the timestamps the paired agent runs already left behind, so this costs nothing and measures the same runs the token numbers came from.

	baseline	with jev
find (grep-only vs jev-first), 7 pairs	57 s	39 s	−32% mean
ask (grep-only vs ask-first), 3 pairs	52 s	51 s	−3%
Read hook (whole file vs window), 3 pairs	9 s	9 s	−6%

The mean is a lie; the median is the answer

For find the per-pair ratios are 0.16, 0.90, 1.02, 1.03, 1.14, 1.73, 1.74. The median is 1.03 — a 3% slowdown. The −32% mean comes entirely from one task, repo C (Go + Flutter)/password-reset, where the grep-only agent spent 215 s and twenty turns establishing that a password-reset flow does not exist, against 34 s for the agent that asked once. Drop that one case and the mean is +1.09.

So the honest statement is:

These tools save tokens, not time. Wall-clock is a wash, except when the baseline would otherwise flounder — and then the difference is large.

That follows from the mechanics rather than contradicting them. A jev find call costs ~13 s and removes two or three turns; a turn takes about that long. The tokens those turns would have parked in context are gone for good, but the clock barely notices.

Per-call latency, for reference

jev find	~13 s (one request per file, plus a verify pass)
jev ask	~2 s per question over a directory
Read hook	~1.25 s, synchronous, added to a narrowed read
lint	~0.73 s, async, so nothing waits on it

Limits

• The two arms of each pair ran concurrently, so absolute seconds include whatever contention there was. The ratio within a pair is the trustworthy number; the absolute times are not.
• Seven, three and three pairs. One outlier moved the find mean by 33 points, which is exactly how little it takes at this sample size.
• Latency is the one axis where the Read hook is a pure cost: it adds 1.25 s to a read and gives back no turns, because the agent was going to read the file either way.