Three weeks ago, viewers per model hit a low of 25.95. This week it reached 36.50 — the highest point in the entire dataset. The supply-demand crisis this series documented appears to be reversing. But the data suggests the recovery is more fragile than it looks.

The first installment of this series opened with a finding that framed every subsequent analysis: viewers per model had fallen 28% in 13 days, from 36.95 on March 27 to 26.58 on April 9. We called it the attention squeeze. The question this week is whether the squeeze is easing — and if so, why.

The 30-day dataset now available gives us enough history to answer that question with more confidence than a single week's snapshot allows.

The Full 30-Day Picture

Date Avg streamers Avg viewers Viewers per model
Apr 2 7,813 204,824 26.22
Apr 5 6,399 214,835 33.57
Apr 7 8,054 227,312 28.22
Apr 9 8,559 227,489 26.58 ← series low
Apr 12 6,771 231,459 34.18
Apr 15 8,478 220,020 25.95 ← second low
Apr 19 6,682 236,807 35.44
Apr 21 8,509 288,601 33.92
Apr 23 6,937 214,222 30.88
Apr 24 5,980 218,253 36.50 ← new high

The 30-day data reveals a pattern that the week-by-week analysis obscured: viewers per model oscillates between roughly 25 and 36, driven primarily by day-of-week effects rather than a directional trend. The series low of 25.95 on April 15 (a Wednesday) and the new high of 36.50 on April 24 (a Thursday, but with unusually low streamer counts) both reflect this weekly rhythm more than any structural change.

The April 21 data point is worth noting separately: 288,601 total viewers — the highest total in the dataset — combined with 8,509 streamers, producing a viewers-per-model of 33.92. This was the first time total viewership crossed 280,000 in the observed period. Whether this represents a genuine demand expansion or a one-time event is not determinable from this data.

The Weekly Rhythm Is the Story

The oscillation in viewers per model is not random. It follows a clear weekly pattern.

Day Avg streamers Avg viewers Viewers per model
Sunday 6,682 236,807 35.44
Monday 7,272 222,021 30.53
Tuesday 8,240 220,530 26.76
Wednesday 8,478 220,020 25.95
Thursday 8,044 221,773 27.57
Friday 8,388 231,891 27.64
Saturday 8,188 236,752 28.92

The Sunday-to-Wednesday swing — from 35.44 to 25.95 — represents a 27% compression in viewers per model driven almost entirely by streamer supply. Total viewers on Sunday (236,807) and Wednesday (220,020) differ by only 7.5%, but streamer counts differ by 27% (6,682 vs 8,478).

This weekly rhythm means that any single-day reading of viewers per model is heavily influenced by which day of the week it falls on. The apparent "recovery" in the April 24 data (36.50) coincides with a day when only 5,980 streamers were online — the lowest single-day count in the dataset. The total viewers of 218,253 were actually below several other days. The high viewers-per-model is a supply-side phenomenon, not a demand-side improvement.

Gender Supply Trends

The gender breakdown over the past two weeks shows whether supply growth is uniform or concentrated in specific categories.

Date Female Male Couple Trans Total viewers
Apr 10 6,285 1,407 388 709 244,011
Apr 15 6,030 1,401 360 686 220,020
Apr 19 4,514 1,246 332 589 236,807
Apr 21 6,029 1,396 365 719 288,601
Apr 24 4,007 1,178 287 508 218,253

Female streamer counts show the most variation — ranging from 4,007 (Sunday April 24) to 6,285 (Friday April 10). This range of 57% within two weeks confirms that the day-of-week supply effect is primarily driven by female streamers choosing when to broadcast. Male, couple, and trans counts are more stable across days.

The couple category shows a consistent range of 287–388 — a narrow band relative to the female category's volatility. Couple streamers appear to have more consistent broadcasting schedules, which may partially explain the category's stable audience advantage documented in earlier analyses.

The April 21 spike in total viewers (288,601) is notable because it occurred on a day when all gender categories were at or near their weekly highs. It was not driven by any single category.

The Market Clearing Signal

The most practically significant data in this week's analysis involves whether low-quality streamers are exiting the platform — the natural market correction mechanism we identified as a possibility in the first installment.

Date Total streamers Low quality (<5 median) Low quality % Platform avg median
Apr 10 39,590 16,690 42.2% 17.1
Apr 12 44,336 18,152 40.9% 19.6
Apr 15 53,226 24,038 45.2% 16.6
Apr 19 44,416 17,798 40.1% 20.6
Apr 21 51,781 21,899 42.3% 17.8
Apr 23 53,974 23,944 44.4% 20.5
Apr 24 20,279 8,937 44.1% 25.4

The low-quality streamer percentage — those with median audiences below 5 — has remained stubbornly in the 40–45% range throughout the observation period. There is no trend toward market clearing. The proportion of streamers achieving very low audiences has not declined.

The April 24 data shows a platform average median of 25.4 — substantially higher than the 16–20 range seen on other days. This is a Sunday effect: with only 20,279 streamers active, the low-quality tail is smaller in absolute terms, and the average is pulled upward by the established streamers who broadcast consistently regardless of day of week.

This is the clearest evidence in the dataset that the supply-demand imbalance is structural rather than self-correcting. The proportion of low-quality streamers has not changed materially over two weeks of observation. The market is not clearing. Streamers who achieve minimal audiences are continuing to broadcast at similar rates.

Where the Platform Is Likely Heading

Three scenarios for the supply-demand trajectory, updated with the current data.

Scenario 1 — Demand expansion (now less likely) The April 21 total viewer spike to 288,601 offers mild evidence of demand growth. If this represents a genuine expansion rather than a one-off event, viewers per model could stabilize above 30 even as streamer counts remain elevated. The data is too limited to confirm this.

Scenario 2 — Weekly cycling without trend The 30-day data is more consistent with this scenario than any directional change. Viewers per model oscillates between 25 and 36 based on day-of-week supply dynamics, with no sustained compression or expansion. The platform exists in a holding pattern where supply and demand fluctuate but neither drives a decisive shift.

Scenario 3 — Continued structural pressure The low-quality streamer proportion remaining at 40–45% with no reduction confirms that supply is not self-correcting. If demand does not expand to absorb the existing supply — and the 30-day trend does not show consistent demand growth — the baseline viewers-per-model on high-supply weekdays may gradually compress further.

The current data best supports Scenario 2, with elements of both Scenario 1 (the April 21 viewer spike) and Scenario 3 (the stable low-quality proportion). The attention squeeze has not resolved. It has stabilized at a level significantly below where the series began.

What This Means in Practice

The Sunday structural advantage — documented in earlier analyses and confirmed in the 30-day data — remains the most actionable single finding for streamers reading this series. A Sunday broadcast faces 2,500 fewer competitors than a Wednesday broadcast for a similar total viewer pool. This advantage is not narrowing in the current data.

The April 21 viewer demand spike, if it recurs, suggests that certain external conditions can expand total viewing substantially above baseline. Whether those conditions can be identified or anticipated from current data is not determinable. What is clear is that demand is not uniformly flat — it has significant single-day variance that supply-side analysis alone cannot explain.

For streamers evaluating whether the platform's competitive environment is improving or worsening: the honest answer is that it depends which day of the week you are measuring. The structural condition of 40-45% of sessions achieving fewer than 5 median viewers has not changed. The weekly rhythm that produces both 25.95 and 36.50 viewers-per-model within the same two weeks has not changed either.

Data sourced from site_sync_stats covering April 2–24, 2026 (approximately 3,200 sync events). Low-quality streamer analysis from model_life_cycles covering completed sessions with median_users recorded. Gender breakdown from site_sync_stats gender-specific online counts.