docs: add Trimmed Build (ENT Tier 3) product page and index it across repo docs and llms.txt

Author: MSTU0

BENCHMARK.md

@@ -9,6 +9,7 @@ Performance data comparing BotBrowser fingerprint protection overhead against st
 | **How much overhead does BotBrowser add?** | Near-zero. Speedometer 3.0 shows **<1% difference** in both headed and headless modes, within run-to-run variance. |
 | **Are fingerprint APIs slower?** | No. Canvas, WebGL, Navigator, Screen, and Font APIs show **identical latency** with or without fingerprint profiles on macOS, Linux, and Windows. |
 | **How does Per-Context Fingerprint scale?** | At 50 concurrent profiles: **29% less memory**, **57% fewer processes**, **2x faster creation** vs launching 50 separate browser instances. Full fingerprint isolation verified. |
+| **How does Trimmed Build compare to Standard?** | On Linux x64 (400 official samples): **62% lower wall time**, **85% faster per-context creation**, **68% lower CPU peak**, **31% lower PSS peak** versus Standard, with **100% success rate** and **0 residual processes**. See [Trimmed Build](#trimmed-build). |
 
 ## Single Instance Performance
 
@@ -140,6 +141,98 @@ Continuous create/destroy cycle test (200 iterations):
 
 ---
 
+<a id="trimmed-build"></a>
+## Trimmed Build: Linux x64 Short-Session Performance
+
+Controlled comparison of **BotBrowser 148 Standard Build** vs **BotBrowser 148 Trimmed Build** on the same Linux x64 host using the same benchmark matrix. For positioning, engineering design, when-to-choose guidance, and FAQ, see the [Trimmed Build product page](TRIMMED_BUILD.md). This section is the performance evidence.
+
+### Test Inputs
+
+| Build | Configuration | Chromium version | Profile |
+|---|---|---|---|
+| Standard | BotBrowser 148 x64 standard build | 148.0.7778.168 | Windows x64 demo profile, standard-compatible |
+| Trimmed (ENT Tier3) | BotBrowser 148 x64 trimmed build (May 2026) | 148.0.7778.168 | Windows x64 demo profile, trimmed-compatible |
+
+Matrix: `1..20 contexts × 10 repeats × 2 builds = 400 official samples`. Both profiles were verified to support per-context flags, per-context proxy settings, proxy auth, and custom headers before the matrix started.
+
+### Method
+
+- Each repeat created `N` independent BrowserContexts, applied `BotBrowser.setBrowserContextFlags`, assigned per-context external proxy settings, loaded a lightweight external page, and ran a minimal fingerprint and proxy probe.
+- Process metrics were sampled every `250ms`. CPU came from Linux process accounting; RSS/PSS/USS came from process memory rollups.
+- Each repeat closed the browser and checked for residual Chromium processes.
+- Both builds used identical launch flags, identical context counts, identical repeat counts, identical warmup shape, identical external target, and identical sampling interval.
+
+### Overall Result
+
+Negative delta means **Trimmed is faster/lighter than Standard**.
+
+| Metric | Standard mean / median / p95 | Trimmed mean / median / p95 | Trimmed delta mean / median / p95 |
+|---|---:|---:|---:|
+| Wall time ms | 7,262.91 / 7,191.04 / 11,666.96 | 2,771.09 / 2,818.88 / 3,758.50 | **-61.85% / -60.80% / -67.79%** |
+| Avg create ms/context | 443.44 / 418.73 / 604.90 | 65.10 / 64.75 / 75.54 | **-85.32% / -84.54% / -87.51%** |
+| Avg first nav ms/context | 2,456.19 / 2,417.16 / 3,072.68 | 1,505.45 / 1,515.68 / 1,849.09 | **-38.71% / -37.30% / -39.82%** |
+| CPU avg % | 4.12 / 4.22 / 4.78 | 1.83 / 2.02 / 2.89 | **-55.42% / -52.06% / -39.51%** |
+| CPU peak % | 25.77 / 27.82 / 40.38 | 8.29 / 9.11 / 12.10 | **-67.81% / -67.25% / -70.03%** |
+| RSS peak KB | 2,591,811 / 2,585,150 / 3,762,505 | 2,249,774 / 2,236,104 / 3,303,095 | -13.20% / -13.50% / -12.21% |
+| PSS peak KB | 960,327 / 959,095 / 1,309,422 | 660,188 / 656,149 / 895,552 | **-31.25% / -31.59% / -31.61%** |
+| USS peak KB | 800,760 / 793,448 / 1,128,869 | 497,876 / 497,038 / 718,904 | **-37.82% / -37.36% / -36.32%** |
+| RSS stable KB | 1,499,234 / 1,491,875 / 1,819,937 | 1,377,981 / 1,360,862 / 1,758,705 | -8.09% / -8.78% / -3.36% |
+| PSS stable KB | 637,313 / 635,427 / 747,051 | 492,405 / 490,987 / 591,358 | -22.74% / -22.73% / -20.84% |
+| USS stable KB | 489,925 / 489,912 / 592,544 | 346,048 / 343,908 / 441,309 | -29.37% / -29.80% / -25.52% |
+| Process peak | 26.27 / 27.00 / 40.10 | 19.50 / 19.50 / 28.05 | -25.76% / -27.78% / -30.05% |
+| Success rate | 100% | 100% | unchanged |
+| Residual processes | 0 | 0 | unchanged |
+
+### Per-Context Matrix
+
+Wall time, context creation, and first navigation at every context count from `1` to `20`. Negative percentage means Trimmed is faster than Standard at that context count.
+
+| N | Wall Standard ms | Wall Trimmed ms | Wall delta | Create delta | First nav delta | CPU avg delta | CPU peak delta | PSS peak delta |
+|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| 1  | 1,564.92  | 748.41    | -52.18% | -85.49% | -52.51% | -95.52% | -97.09% | -27.89% |
+| 2  | 3,507.80  | 1,948.97  | -44.44% | -88.33% | -43.61% | -92.05% | -89.86% | -28.94% |
+| 3  | 3,988.82  | 1,992.15  | -50.06% | -87.99% | -43.78% | -80.93% | -63.42% | -29.17% |
+| 4  | 3,955.15  | 2,136.14  | -45.99% | -84.01% | -37.29% | -74.09% | -52.54% | -30.15% |
+| 5  | 4,432.40  | 2,204.01  | -50.28% | -85.08% | -37.87% | -68.19% | -51.68% | -30.41% |
+| 6  | 5,592.26  | 2,316.21  | -58.58% | -87.36% | -40.32% | -63.93% | -66.42% | -29.53% |
+| 7  | 5,141.93  | 2,342.06  | -54.45% | -83.81% | -35.61% | -59.40% | -61.96% | -31.03% |
+| 8  | 6,149.22  | 2,556.26  | -58.43% | -84.29% | -37.96% | -54.73% | -59.51% | -30.61% |
+| 9  | 6,147.59  | 2,701.18  | -56.06% | -83.77% | -31.58% | -53.26% | -66.09% | -31.56% |
+| 10 | 6,985.46  | 2,735.62  | -60.84% | -85.10% | -36.79% | -54.62% | -68.59% | -32.06% |
+| 11 | 7,258.48  | 2,891.23  | -60.17% | -84.34% | -36.46% | -52.11% | -61.73% | -31.18% |
+| 12 | 7,628.21  | 2,971.22  | -61.05% | -83.98% | -33.97% | -48.32% | -68.29% | -31.84% |
+| 13 | 9,976.22  | 3,009.45  | -69.83% | -87.65% | -45.89% | -49.07% | -71.87% | -31.06% |
+| 14 | 9,262.40  | 3,222.77  | -65.21% | -85.47% | -37.76% | -47.50% | -67.77% | -32.02% |
+| 15 | 9,518.44  | 3,271.86  | -65.63% | -85.19% | -38.25% | -44.81% | -65.17% | -32.83% |
+| 16 | 9,709.20  | 3,935.21  | -59.47% | -79.76% | -29.05% | -34.23% | -65.82% | -32.73% |
+| 17 | 10,223.93 | 3,619.99  | -64.59% | -84.21% | -37.29% | -41.66% | -68.99% | -31.83% |
+| 18 | 11,073.62 | 3,520.06  | -68.21% | -85.94% | -39.39% | -43.13% | -70.31% | -32.56% |
+| 19 | 11,321.90 | 3,572.46  | -68.45% | -86.59% | -40.62% | -43.81% | -72.06% | -30.90% |
+| 20 | 11,820.21 | 3,726.52  | -68.47% | -84.57% | -41.15% | -39.52% | -71.87% | -31.44% |
+
+> Trimmed is faster than Standard at **every** context count for wall time, average context creation, average first navigation, CPU avg, CPU peak, and PSS peak. The gap is largest in CPU peak (`-51% to -97%`) and per-context creation (`-79% to -88%`), which directly drive throughput for short-session privacy workloads that rotate identities quickly.
+
+### Stability
+
+- Browser main process, network service, GPU, and utility process counts are unchanged across builds.
+- Renderer count scales with context count in both builds.
+- Trimmed reduces peak process count overall: mean `26.27 → 19.50`, median `27.00 → 19.50`, p95 `40.10 → 28.05`.
+- All official runs exited cleanly: `400/400` successful samples, `0` errors, `0` residual Chromium processes.
+
+### Scope Notes
+
+- Linux x64 is the deployment shape Trimmed Build is built for: short-session, high-concurrency Ubuntu fleets are where the gap is largest.
+- The benchmark includes external navigation and uses a fixed matrix for both builds.
+- Profile bundles, host class, launch flags, and target endpoint were held identical between builds; runtime credentials and absolute paths are intentionally omitted from this customer-facing report.
+
+### Next Steps
+
+- Product overview, engineering design, and FAQ: [TRIMMED_BUILD.md](TRIMMED_BUILD.md)
+- Access: [Enterprise](https://botbrowser.io/enterprise/) or [Pricing](https://botbrowser.io/pricing/)
+- Public mirror with extended commentary: [Trimmed Build on botbrowser.io](https://botbrowser.io/docs/deployment/trimmed-build/)
+
+---
+
 ## Cross-Platform Verification
 
 BotBrowser supports applying any platform's fingerprint on any host OS. Tested on macOS, Linux, and Windows.

INSTALLATION.md

@@ -2,13 +2,24 @@
 
 This guide provides step-by-step installation, troubleshooting, and deployment options across operating systems. Usage remains subject to the project [Legal Disclaimer](DISCLAIMER.md) and [Responsible Use Guidelines](RESPONSIBLE_USE.md).
 
+<a id="choosing-a-build"></a>
+
+## Choosing a Build
+
+BotBrowser ships in two builds. Both share the same fingerprint protection model, profile format, CLI flag surface, and CDP commands.
+
+- **Standard Build** is the default public release. Use this build for long-running sessions, interactive workflows, and scenarios that exercise the full browser feature surface. Available on the [Releases](https://github.com/botswin/BotBrowser/releases) page; the instructions on this page describe Standard Build installation.
+- **Trimmed Build** (ENT Tier3) is a separately built distribution tuned for short-session, high-concurrency automation. Linux x64 benchmark: **62% lower wall time**, **85% faster per-context creation**, **68% lower CPU peak** versus Standard. Trimmed Build is not on the public Releases page; it is distributed through the enterprise channel. Product overview: [TRIMMED_BUILD.md](TRIMMED_BUILD.md). Full performance table: [BENCHMARK.md#trimmed-build](BENCHMARK.md#trimmed-build). Access: [Enterprise](https://botbrowser.io/enterprise/) or [Pricing](https://botbrowser.io/pricing/).
+
+Profiles, automation code, and CLI flags from Standard Build work unchanged on Trimmed Build, so you can build against Standard locally and switch to Trimmed in production.
+
 <a id="download--installation"></a>
 
 ## Download & Installation
 
 ### 1. Download Installer
 
-Get the installer for your OS from the [Releases](https://github.com/botswin/BotBrowser/releases) page.
+Get the Standard Build installer for your OS from the [Releases](https://github.com/botswin/BotBrowser/releases) page. (Trimmed Build is distributed through the enterprise channel; see [Choosing a Build](#choosing-a-build).)
 
 ### 2. Windows Installation
 

PER_CONTEXT_FINGERPRINT.md

@@ -2,6 +2,8 @@
 
 Run multiple independent fingerprint identities within a single browser instance. Shared GPU/Network/Browser processes. Reduced resource overhead.
 
+> **Pairs with Trimmed Build (ENT Tier3).** Per-Context Fingerprint and [Trimmed Build](TRIMMED_BUILD.md) compose. Per-Context Fingerprint cuts memory and process count at scale; Trimmed Build cuts per-context spin-up by 85% and CPU peak by 68% on Linux x64. Together they form the densest deployment shape BotBrowser ships.
+
 ## The Problem
 
 Traditional approaches require launching a separate browser instance for each fingerprint profile. This architecture has significant overhead due to Chromium's multi-process design.

README.md

@@ -55,7 +55,7 @@ All engineering focuses on privacy research, cross-platform tracking-resistance
   </tr>
   <tr>
     <td width="50%"><strong>Per-Context Fingerprint</strong> enables <a href="PER_CONTEXT_FINGERPRINT.md">independent fingerprint bundles per BrowserContext</a> without spawning new processes, with millisecond-level switching and reduced memory overhead</td>
-    <td width="50%"><strong>Zero-Overhead Performance</strong> adds <a href="BENCHMARK.md">no measurable latency</a>: Speedometer 3.0 within &lt;1% of stock Chrome, zero fingerprint API overhead across macOS/Linux/Windows, and 29% memory savings at scale with Per-Context Fingerprint</td>
+    <td width="50%"><strong>Zero-Overhead Performance</strong> adds <a href="BENCHMARK.md">no measurable latency</a>: Speedometer 3.0 within &lt;1% of stock Chrome, zero fingerprint API overhead across macOS/Linux/Windows, 29% memory savings at scale with Per-Context Fingerprint, and a <a href="BENCHMARK.md#trimmed-build">Trimmed Build</a> (ENT Tier3) delivering 62% lower wall time and 85% faster per-context spin-up on Linux x64</td>
   </tr>
 </table>
 
@@ -116,6 +116,26 @@ Examples: [Playwright](examples/playwright/) • [Puppeteer](examples/puppeteer/
 - Docker: [docker/README.md](docker/)
 - Full flags: [CLI_FLAGS.md](CLI_FLAGS.md)
 
+<a id="builds"></a>
+## Builds
+
+BotBrowser ships in two builds. Both share the same fingerprint protection model, the same profile format, the same CLI flag surface, and the same CDP commands.
+
+| | Standard Build | Trimmed Build (ENT Tier3) |
+|---|---|---|
+| **Distribution** | Public [releases](https://github.com/botswin/BotBrowser/releases) | Enterprise channel only |
+| **Built for** | Long-running and interactive sessions | Short-session, high-concurrency automation |
+| **Browser feature surface** | Full | Tuned for short-session workloads |
+| **Fingerprint protection** | Same | Same |
+| **Per-Context Fingerprint** | Same | Same |
+| **Profile compatibility** | Same | Same |
+
+**Linux x64 benchmark (400 official samples, `1..20 contexts × 10 repeats × 2 builds`)**: Trimmed Build cuts wall time by **62%**, per-context creation by **85%**, first navigation by **38%**, CPU peak by **68%**, PSS peak by **31%** versus Standard, with **100% success rate** and **0 residual processes** for both builds.
+
+Trimmed Build is the right choice when context spin-up dominates total wall time, when CPU peak per context limits density, or when shared memory is the binding constraint. Standard Build remains the right choice for interactive workflows and scenarios that exercise the full browser feature surface.
+
+Product overview, engineering design, FAQ: [TRIMMED_BUILD.md](TRIMMED_BUILD.md). Full performance table: [BENCHMARK.md#trimmed-build](BENCHMARK.md#trimmed-build). Access: [Enterprise](https://botbrowser.io/enterprise/) or [Pricing](https://botbrowser.io/pricing/).
+
 ## Feature Reference
 
 > Configuration priority: [CLI flags](CLI_FLAGS.md) (highest) > [Profile configs](profiles/PROFILE_CONFIGS.md) > defaults. Timezone, locale, and language auto-derive from your proxy IP.