thread length analysis by outcome

summary stats

status	count	avg turns	min	max
RESOLVED	2745	67.7	3	988
FRUSTRATED	14	84.3	1	160
COMMITTED	305	57.0	2	506
HANDOFF	75	38.9	1	339
EXPLORATORY	124	5.8	1	49
PENDING	8	240.1	5	1623
UNKNOWN	1560	16.0	0	397

histogram: turns by outcome

                   1-10   11-25  26-50  51-75  76-100  101-150  151-200  200+
RESOLVED          ███     █████████  ████████████  █████████  █████████  ██████████  ████   ██
                   195      400       473      275     262       287       114     64

FRUSTRATED          █       █       ███            ███       █████        █
                    1       1        3              3         5          1

COMMITTED         ████   ██████     ██████   █████   ███      ████       ██        █
                   45      77        56      49      27        32        13        6

HANDOFF           ██████████████████   ██████   █████   ████   █████    ██████      ██       █
                   260      90        55      40      51        63        12        3

findings

sweet spots by outcome

RESOLVED threads: bimodal distribution

peak 1: 26-50 turns (473 threads, 22.8%) — quick resolutions
peak 2: 101-150 turns (287 threads, 13.9%) — complex but successful
healthy distribution across all ranges; long threads CAN succeed

FRUSTRATED threads: skew toward longer

64% occur at 76+ turns (9/14)
avg 84.3 turns vs RESOLVED avg 67.7
suggests frustration correlates with thread length, though n=14 is small

COMMITTED threads: front-loaded

57% finish in ≤50 turns (178/305)
lower avg (57.0) than RESOLVED (67.7)
commits happen faster than resolutions — hunch: exploratory work before committing

HANDOFF threads: very front-loaded

corrected: HANDOFF count was inflated from 574 to 75 after fixing subagent miscount
avg 38.9 turns
early handoffs likely: task confusion, scope mismatch, or “not amp’s job”

per-user patterns

user	resolved_avg	frustrated_avg	resolved_n	frustrated_n
@concise_commander	92.9	84.7	738	6
@steady_navigator	47.2	64.3	764	4
@verbose_explorer	73.5	103.5	296	2
@patient_pathfinder	26.3	—	79	0
@precision_pilot	80.1	113.0	74	1
@feature_lead	34.9	—	25	0

@steady_navigator has shortest avg resolved (47.2 turns) — efficient or smaller scope tasks?
@concise_commander longer avg resolved (92.9) but still resolves successfully at scale (738)
when users DO get frustrated, their frustrated threads are 16-38% longer than their resolved avg

takeaways

26-50 turns is the sweet spot for resolutions — most common successful outcome
frustration warning: threads approaching 100+ turns without resolution merit intervention
handoff pattern: early (≤10 turns) handoffs suggest task/tool mismatch, not failure
user variation matters: some users naturally work longer threads successfully (@concise_commander), others prefer quick hits (@steady_navigator, @patient_pathfinder)