learning curves: user evolution analysis

analysis of 4656 threads across 9 months (may 2025 – jan 2026).

aggregate trends by month

month	threads	avg turns	avg steering	unique users
2025-05	24	75.1	0.33	4
2025-06	297	57.0	0.37	3
2025-07	344	46.2	0.20	5
2025-08	288	61.4	0.39	4
2025-09	254	56.3	0.35	5
2025-10	296	38.1	0.38	6
2025-11	496	34.0	0.31	5
2025-12	1620	43.1	0.34	9
2026-01	1037	42.6	0.26	16

key observation: thread length decreased significantly from early months (75 avg turns in may) to stabilizing around 35-43 turns by late 2025. steering frequency remains relatively consistent (0.26-0.39), suggesting users maintain similar correction patterns regardless of experience.

top user longitudinal analysis

@concise_commander (1219 threads, power user)

most active user with consistent high engagement
avg steering: 0.81 (highest among power users)
thread length: 86.5 avg turns (longest threads)
jan 2026: steering dropped to 0.58 from 0.85+ earlier months
pattern: uses longer sessions with more intervention; recent improvement

@steady_navigator (1171 threads)

second most active, LOW steering user
avg steering: 0.10 (minimal corrections needed)
avg turns: 36.5 (efficient sessions)
notable: steering stayed consistently under 0.15 across all months
pattern: writes precise prompts that rarely need correction

@verbose_explorer (875 threads)

high variance early, now stabilized
june 2025 outlier: 197 avg turns, 1.13 steering (early adoption friction)
jan 2026: 22.7 avg turns, 0.09 steering (dramatic improvement)
pattern: steep learning curve visible — 68% reduction in turn count

learning curve patterns

pattern 1: efficiency gains (@verbose_explorer)

first month:  68 avg turns, 0.22 steering
latest month: 23 avg turns, 0.09 steering
improvement:  66% fewer turns, 59% less steering

pattern 2: stable expert (@steady_navigator)

first month:  32 avg turns, 0.0 steering  
latest month: 28 avg turns, 0.08 steering
pattern:      consistently efficient from start

pattern 3: high-touch workflow (@concise_commander)

first month:  102 avg turns, 0.5 steering
latest month: 86 avg turns, 0.58 steering
pattern:      uses agent for complex/long tasks, steering is intentional style

steering as % of turns

user	2025-05	2025-12	2026-01	trend
@concise_commander	0.49%	0.97%	0.67%	stable
@steady_navigator	0.0%	0.21%	0.29%	minimal
@verbose_explorer	0.32%	0.74%	0.40%	improving

findings

learning is real: @verbose_explorer demonstrates clearest learning curve — 66% reduction in session length over 8 months
prompt style matters more than experience: @steady_navigator started with low steering and maintained it; this is prompt craft, not just time
power users plateau differently: @concise_commander uses longer sessions intentionally — high steering isn’t failure, it’s workflow choice
aggregate hides individual: overall steering is flat, but individual users show distinct trajectories

resolution rate caveat

completion_status shows 0% resolution across all months — this field appears unpopulated or uses different semantics. recommend checking if status is tracked elsewhere.

correction note: earlier analysis miscounted @verbose_explorer’s spawn threads as HANDOFF. his resolution rate is 83% (not 33.8%). turn/steering metrics above were unaffected — learning curve observations remain valid.

generated: 2026-01-09