# Garbage collection and frame gaps

## Web

This instrumented diagnostic measures correlation. It does not prove that GC caused a delivery gap.
Role A is the source. Role B is the experimental direct-RGBA receiver. Both are research fixtures.

GC and gap intervals share each role’s proved page clock. The first and last anchors bound the interval analysis.
Memory samples use a separate controller clock. No interpolation aligns memory samples with GC.
Interval marks have a minimum visual width of 0.35 pixels, about 63 ms on a 180-second axis. Short marks are expanded. Their drawn width does not show their exact duration.

### Role A

The trace contains 127 merged main-thread pause intervals, with 93.217 ms of total duration.
The journal records 1 interval of at least 25 ms. The scoped table includes 0 intervals and excludes 1.
Excluded interval counts: 0 cross an anchor, and 1 lie outside the anchor interval.
0 scoped gaps overlap a mapped pause. 0 scoped gaps have no mapped pause overlap.
0 pause intervals overlap a recorded gap. 127 fully covered pause intervals have no recorded gap overlap.
This pause classification uses all recorded gaps, including those that cross an anchor. 0 pause intervals overlap a boundary-crossing gap.
Other pause intervals without complete telemetry coverage: 0.
Telemetry covers the first trace anchor.
Telemetry covers the last trace anchor.

| Stage | Gaps | With GC | Without GC | Gap duration ms | GC overlap ms |
|---|---:|---:|---:|---:|---:|
| packetArrival | 0 | 0 | 0 | 0.000 | 0.000 |
| arrivalToConsume | 0 | 0 | 0 | 0.000 | 0.000 |
| renderEntry | 0 | 0 | 0 | 0.000 | 0.000 |
| uploadAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| renderAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| finalReadback | 0 | 0 | 0 | 0.000 | 0.000 |
| sourceRenderEntry | 0 | 0 | 0 | 0.000 | 0.000 |
| sourceRenderAPI | 0 | 0 | 0 | 0.000 | 0.000 |

Median pause duration is 0.649 ms. The nearest-rank 95th percentile is 1.111 ms.
Median pause start spacing is 1366.514 ms. No periodicity claim follows from that statistic.
The clock offset spread is 0.000 microseconds. The accepted alignment uncertainty is 100 microseconds.

### Role B

The trace contains 197 merged main-thread pause intervals, with 201.498 ms of total duration.
The journal records 404 intervals of at least 25 ms. The scoped table includes 398 intervals and excludes 6.
Excluded interval counts: 0 cross an anchor, and 6 lie outside the anchor interval.
150 scoped gaps overlap a mapped pause. 248 scoped gaps have no mapped pause overlap.
124 pause intervals overlap a recorded gap. 73 fully covered pause intervals have no recorded gap overlap.
This pause classification uses all recorded gaps, including those that cross an anchor. 0 pause intervals overlap a boundary-crossing gap.
Other pause intervals without complete telemetry coverage: 0.
Telemetry covers the first trace anchor.
Telemetry ends 6.600 ms before the last trace anchor.

| Stage | Gaps | With GC | Without GC | Gap duration ms | GC overlap ms |
|---|---:|---:|---:|---:|---:|
| packetArrival | 67 | 36 | 31 | 2974.100 | 58.376 |
| arrivalToConsume | 276 | 68 | 208 | 7542.600 | 95.805 |
| renderEntry | 51 | 46 | 5 | 1534.200 | 70.679 |
| uploadAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| renderAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| finalReadback | 4 | 0 | 4 | 118.100 | 0.000 |
| sourceRenderEntry | 0 | 0 | 0 | 0.000 | 0.000 |
| sourceRenderAPI | 0 | 0 | 0 | 0.000 | 0.000 |

Median pause duration is 0.960 ms. The nearest-rank 95th percentile is 1.857 ms.
Median pause start spacing is 571.600 ms. No periodicity claim follows from that statistic.
The clock offset spread is 0.002 microseconds. The accepted alignment uncertainty is 100 microseconds.

### Scope

Only source-pinned MinorGC and MajorGC phases on the proved renderer main thread contribute pause intervals.
Nested phases are not added to their enclosing pause. Background and unmapped events remain separate counts.
The analysis includes retained telemetry intervals of at least 25 ms. Stage intervals can overlap, so their totals must not be added.
Zero mapped overlap does not exclude other GC work, scheduling delays, native work or driver work.
A brief pause can coincide with a missed frame deadline. Its duration fraction does not establish or exclude a causal effect.
The heap samples are stocks. Backing storage includes ArrayBuffers and external strings, not only received packets.
No RSS, allocation rate, physical display result or throughput acceptance is claimed.

### Evidence

Result SHA-256: 38c082cc8f56dcc4321b04423070e996d198d4fc4e4b45b6d4865aa442e0de26

[Performance characteristics](web.svg) · [Structured analysis](analysis.json) · [Checksums](SHA256SUMS)

## Desktop

This instrumented diagnostic measures correlation. It does not prove that GC caused a delivery gap.
Role A is the source. Role B is the experimental direct-RGBA receiver. Both are research fixtures.

GC and gap intervals share each role’s proved page clock. The first and last anchors bound the interval analysis.
Memory samples use a separate controller clock. No interpolation aligns memory samples with GC.
Interval marks have a minimum visual width of 0.35 pixels, about 63 ms on a 180-second axis. Short marks are expanded. Their drawn width does not show their exact duration.

### Role A

The trace contains 129 merged main-thread pause intervals, with 99.715 ms of total duration.
The journal records 0 intervals of at least 25 ms. The scoped table includes 0 intervals and excludes 0.
Excluded interval counts: 0 cross an anchor, and 0 lie outside the anchor interval.
0 scoped gaps overlap a mapped pause. 0 scoped gaps have no mapped pause overlap.
0 pause intervals overlap a recorded gap. 129 fully covered pause intervals have no recorded gap overlap.
This pause classification uses all recorded gaps, including those that cross an anchor. 0 pause intervals overlap a boundary-crossing gap.
Other pause intervals without complete telemetry coverage: 0.
Telemetry covers the first trace anchor.
Telemetry covers the last trace anchor.

| Stage | Gaps | With GC | Without GC | Gap duration ms | GC overlap ms |
|---|---:|---:|---:|---:|---:|
| packetArrival | 0 | 0 | 0 | 0.000 | 0.000 |
| arrivalToConsume | 0 | 0 | 0 | 0.000 | 0.000 |
| renderEntry | 0 | 0 | 0 | 0.000 | 0.000 |
| uploadAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| renderAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| finalReadback | 0 | 0 | 0 | 0.000 | 0.000 |
| sourceRenderEntry | 0 | 0 | 0 | 0.000 | 0.000 |
| sourceRenderAPI | 0 | 0 | 0 | 0.000 | 0.000 |

Median pause duration is 0.716 ms. The nearest-rank 95th percentile is 1.321 ms.
Median pause start spacing is 1366.488 ms. No periodicity claim follows from that statistic.
The clock offset spread is 1.000 microseconds. The accepted alignment uncertainty is 100 microseconds.

### Role B

The trace contains 167 merged main-thread pause intervals, with 174.091 ms of total duration.
The journal records 430 intervals of at least 25 ms. The scoped table includes 428 intervals and excludes 2.
Excluded interval counts: 0 cross an anchor, and 2 lie outside the anchor interval.
143 scoped gaps overlap a mapped pause. 285 scoped gaps have no mapped pause overlap.
114 pause intervals overlap a recorded gap. 52 fully covered pause intervals have no recorded gap overlap.
This pause classification uses all recorded gaps, including those that cross an anchor. 0 pause intervals overlap a boundary-crossing gap.
Other pause intervals without complete telemetry coverage: 1.
Telemetry covers the first trace anchor.
Telemetry ends 8.900 ms before the last trace anchor.

| Stage | Gaps | With GC | Without GC | Gap duration ms | GC overlap ms |
|---|---:|---:|---:|---:|---:|
| packetArrival | 86 | 35 | 51 | 3525.300 | 55.159 |
| arrivalToConsume | 293 | 67 | 226 | 7997.500 | 93.590 |
| renderEntry | 46 | 41 | 5 | 1410.200 | 57.719 |
| uploadAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| renderAPI | 0 | 0 | 0 | 0.000 | 0.000 |
| finalReadback | 3 | 0 | 3 | 107.000 | 0.000 |
| sourceRenderEntry | 0 | 0 | 0 | 0.000 | 0.000 |
| sourceRenderAPI | 0 | 0 | 0 | 0.000 | 0.000 |

Median pause duration is 0.957 ms. The nearest-rank 95th percentile is 1.670 ms.
Median pause start spacing is 921.904 ms. No periodicity claim follows from that statistic.
The clock offset spread is 0.000 microseconds. The accepted alignment uncertainty is 100 microseconds.

### Scope

Only source-pinned MinorGC and MajorGC phases on the proved renderer main thread contribute pause intervals.
Nested phases are not added to their enclosing pause. Background and unmapped events remain separate counts.
The analysis includes retained telemetry intervals of at least 25 ms. Stage intervals can overlap, so their totals must not be added.
Zero mapped overlap does not exclude other GC work, scheduling delays, native work or driver work.
A brief pause can coincide with a missed frame deadline. Its duration fraction does not establish or exclude a causal effect.
The heap samples are stocks. Backing storage includes ArrayBuffers and external strings, not only received packets.
No RSS, allocation rate, physical display result or throughput acceptance is claimed.

### Evidence

Result SHA-256: 49a2e8da3427c03805039a919b20581d412aa18a047d7992ef3e23ca1461611e

[Performance characteristics](desktop.svg) · [Structured analysis](analysis.json) · [Checksums](SHA256SUMS)