[thejustinwalsh]

Draft frontier for #247: full assignments prepare target-owned text/layout in Rust, preserve sparse dirt through the existing retained rope/gather/publication owners, and reuse accepted Three topology through #291. Existing public setter ergonomics, immediate measurements, batching and explicit publisher lifetime remain intact. Replaced mutation scans and measurement reconciliation helpers are retired; this does not add a second preparation pipeline.

The latest gather change streams dirty owners through one traversal of the existing source rope and glyph callback. Count changes and early Stop resume the existing suffix builder; clean decoration records, recordless owners and failure recovery remain covered by independent cold-output controls.

This PR depends on #291 and remains draft. It is not required for the small publication release cut. Original development history is preserved in local backup refs; the final source delta was consolidated before rebasing to avoid replaying obsolete settlement/error implementations already retired on main.

Current combined native shaper suite passes, including 399 unit tests and independent outline/public API controls. The integrated Wasm/TypeScript build is running; combined Node/fuzz/static, exact-artifact size, performance and live browser validation remain pending. Earlier rope timings do not clear this head. Independent reviews of the host integration and authenticated gather-stream patch found no actionable defect; these reviews are scoped and are not a new audit of every historical change.

Remaining gates are tracked in .agents/docs/planning/publication-frontier.md: bulk/edit-sized publication performance, exact draw map/reduce, cumulative retirement/maintainability/size, exact-head CI and release-readiness. #247 is not closed. CI uses benchmark:stress; full combined release comparison follows only after correctness and artifact gates pass. Do not merge this draft based solely on the standalone #291 report.

[github-actions]

Docs report 📚

Every changed package is attested at this head. A reviewer verifies each claim after merge.

Package	Concept	Attestation	Status
@pmndrs/glyph-benchmarks	packages/benchmarks.md · edited	Promoted the existing offline packed-core-consumer peer isolation to the lowest held layer; package concept remains acc…	✅
@pmndrs/glyph	packages/glyph.md · edited	Preserved prepared anchor measurements and explicit renderer invalidation in the lowest held layer; corrected existing …	✅

✅ attested · ⚠️ stale · ❌ missing

[github-actions]

size-limit report 📦

Surface	gzip	Surface	gzip
Core JS	91.9 kB (+0.82% )	Inter font · Bitmap	558.2 kB (0%)
Shaper Wasm	592.9 kB (+9.47% )	Inter font · MTSDF	8.17 MB (0%)
Three.js adapter JS	143.4 kB (+0.67% )	Inter font · Slug	618.3 kB (0%)
React adapter JS	143.1 kB (+0.62% )	Font Awesome icons · Bitmap	449.9 kB (0%)
Vue adapter JS	142.9 kB (+0.62% )	Font Awesome icons · MTSDF	8.76 MB (0%)
R3F hello-world app JS	747.9 kB (+0.12% )	Font Awesome icons · Slug	657.9 kB (0%)
Tres playground app JS	572.2 kB (+0.16% )	Bitmap baker JS	4.9 kB (0%)
Font validator JS	144.8 kB (0%)	Bitmap baker Wasm	231.1 kB (0%)
Runtime bake host JS	6.1 kB (0%)	MTSDF baker JS	5.6 kB (0%)
Runtime bake Worker JS	13.5 kB (0%)	MTSDF baker Wasm	215.5 kB (0%)
Font baker JS	2.6 kB (0%)	Slug baker JS	4.5 kB (0%)
Font baker Wasm	389.0 kB (0%)	Slug baker Wasm	183.7 kB (0%)

[github-actions]

Package performance: stress suite

4 faster · 1 slower · 3 neutral · 0 skipped

Baseline 0.1.0-canary-4e25a96a-20261015 → candidate 0.1.0 (f07e2120)

Latency distribution across 8 comparable workload p50s. Each workload counts once; these are not frame times.

statistic	baseline	candidate
Median workload p50	3.660ms	3.825ms
Workloads ≤1 ms	3	3
Workloads ≤2 ms	4	4
Workloads ≤3 ms	4	4
Workloads >3 ms	4	4

Largest candidate workloads (including neutral results):

workload	baseline	candidate	Δ time
publish 1024 retained Text instances	40.40ms	79.32ms	+38.920ms
reorder then edit one of 1000 retained labels	14.34ms	12.68ms	-1.660ms
borrow glyphs from 1000 steadily promoted retained labels	5.62ms	6.12ms	+0.500ms
reorder 1000 unchanged retained labels	6.56ms	5.98ms	-0.580ms
toggle one group containing 1000 retained labels	1.70ms	1.67ms	-0.030ms

                                                             faster ◀       ┊       ▶ slower    Δ p50      p  baseline   candidate
- publish 1024 retained Text instances                                      ┊              ▶   +96.3%  <.001  ▅█▁▁▁▁▁▁▁▁ ▁▁▁▁▁██▇▇▂
+ reorder 1000 unchanged retained labels                                ├●─┤┊                   -8.9%   .015  ▁▂▇█▃▃▂▁▁▁ ▄▆█▅▄▁▁▁▁▁
+ reorder then edit one of 1000 retained labels                        ├─●  ┊                  -11.6%  <.001  ▁▁▃█▇▅▃▂▃▁ ▃█▅▃▂▂▁▁▁▁
+ edit and sparsely borrow one of 1000 retained labels              ├●─┤    ┊                  -23.4%  <.001  ▁▁▁▇▇█▆▄▃▁ ▂██▅▂▁▁▁▁▁
+ edit and measure one of 1000 retained labels                    ●┤        ┊                  -32.2%  <.001  ▁▁▁▅▇█▅▅▂▃ ▃█▆▂▁▁▁▁▁▁
  borrow glyphs from 1000 steadily promoted retained labels              ├──┼──●───────────▶    +8.8%  1.000  ▅▄▆▃▁▁▁▆▅█ ▁▁▂▃▃▃██▃▁
  measure 1000 unchanged retained labels                                ├───●───┤               +1.3%   .721  ▁▅▆▆█▆▄▄▂▁ ▂▅▆▇█▆▄▄▂▂
  toggle one group containing 1000 retained labels                        ├●┼─┤                 -2.0%   .959  ▁▃▇▆█▄▃▂▁▁ ▁▃█▆█▅▃▂▂▁
                                                             └──────────────┴──────────────┘
                                                             -50%           0           +50%

status	bench	baseline	candidate	Δ time	Δ p50	p	95% CI
🔴 slower	publish 1024 retained Text instances	40.40ms	79.32ms	+38.920ms	+96.3%	<.001	+92.0..+99.7%
🟢 faster	reorder 1000 unchanged retained labels	6.56ms	5.98ms	-0.580ms	-8.9%	.015	-11.8..-2.1%
🟢 faster	reorder then edit one of 1000 retained labels	14.34ms	12.68ms	-1.660ms	-11.6%	<.001	-17.5..-10.6%
🟢 faster	edit and sparsely borrow one of 1000 retained labels	825.42µs	632.28µs	-0.193ms	-23.4%	<.001	-26.8..-17.8%
🟢 faster	edit and measure one of 1000 retained labels	709.92µs	481.54µs	-0.228ms	-32.2%	<.001	-34.9..-28.5%

3 neutral, 0 skipped

status	bench	baseline	candidate	Δ time	Δ p50	p	95% CI
neutral	toggle one group containing 1000 retained labels	1.70ms	1.67ms	-0.030ms	-2.0%	.959	-6.5..+6.3%
neutral	measure 1000 unchanged retained labels	509.91µs	516.39µs	+0.006ms	+1.3%	.721	-12.0..+14.2%
neutral	borrow glyphs from 1000 steadily promoted retained labels	5.62ms	6.12ms	+0.500ms	+8.8%	1.000	-11.6..+54.1%