High-performance browser automation bridge and multi-instance orchestrator with advanced stealth injection and real-time dashboard.
docs: update pinchtab-clean-slate with real empirical measurements
Mario383ef5e03331
diff --git a/docs/pinchtab-clean-slate.md b/docs/pinchtab-clean-slate.mdindex 6727d3b1..baf1a95f 100644--- a/docs/pinchtab-clean-slate.md+++ b/docs/pinchtab-clean-slate.md@@ -37,42 +37,44 @@ An agent with **zero prior context** spawns, calls Pinchtab once per site, extra**Methodology:** Response sizes derived from documented Pinchtab behavior (compact format output) and production measurements. See [Validation Method](#validation-method) section for details.#### BBC.com-- **HTTP response size (calculated):** ~3-4 KB (compact format, limited titles)-- **Est tokens:** ~800-1,000-- **Titles extracted:** 10 (estimated)-- **Sample:** Breaking News, World News, Sports, Business, Innovation...-- **Agent + Pinchtab total:** 500 + 900 = **1,400 tokens**+- **HTTP response size (measured):** 7,335 bytes+- **Response lines:** 153+- **Pinchtab tokens:** ~1,834 (7,335 ÷ 4)+- **Node count:** 149 (semantic accessibility tree)+- **Agent + Pinchtab total:** 500 + 1,834 = **2,334 tokens**#### Corriere.it-- **HTTP response size (calculated):** ~2-3 KB (compact, curated)-- **Est tokens:** ~600-800-- **Titles extracted:** 10 (estimated)-- **Sample:** Sanremo scaletta, Iran Asse Cina, Legge elettorale, Epstein deposizione...-- **Agent + Pinchtab total:** 500 + 700 = **1,200 tokens**+- **HTTP response size (measured):** 7,737 bytes+- **Response lines:** 133+- **Pinchtab tokens:** ~1,934 (7,737 ÷ 4)+- **Node count:** 128 (semantic accessibility tree)+- **Agent + Pinchtab total:** 500 + 1,934 = **2,434 tokens**#### Daily Mail.co.uk-- **HTTP response size (calculated):** ~4-5 KB (compact, dense)-- **Est tokens:** ~1,000-1,300-- **Titles extracted:** 10 (estimated)-- **Sample:** Ian Huntley assault, Hillary Clinton testimony, Starmer polls, dental condition...-- **Agent + Pinchtab total:** 500 + 1,150 = **1,650 tokens**+- **HTTP response size (measured):** 7,531 bytes+- **Response lines:** 108+- **Pinchtab tokens:** ~1,883 (7,531 ÷ 4)+- **Node count:** 104 (semantic accessibility tree)+- **Agent + Pinchtab total:** 500 + 1,883 = **2,383 tokens**---## Summary Table-### Per-Site Breakdown (Agent + Pinchtab)+### Per-Site Breakdown (Agent + Pinchtab) - EMPIRICALLY MEASURED-| Site | Pinchtab | Agent OH | **Total** | vs Snapshot | vs web_fetch |-|------|----------|----------|----------|-------------|-------------|-| BBC | 900 | 500 | **1,400** | 11,250 / 1,400 = 8x lighter | 4,700 / 1,400 = 3.4x lighter |-| Corriere | 700 | 500 | **1,200** | 95,000 / 1,200 = 79x lighter | 3,275 / 1,200 = 2.7x lighter |-| Daily Mail | 1,150 | 500 | **1,650** | 87,500 / 1,650 = 53x lighter | 12,500 / 1,650 = 7.6x lighter |-| **Average** | **920** | **500** | **1,417** | **~47x lighter** | **~4.6x lighter** |+| Site | Response | Pinchtab | Agent OH | **Total** | vs Snapshot | vs web_fetch |+|------|----------|----------|----------|----------|-------------|-------------|+| BBC | 7.3 KB | 1,834 | 500 | **2,334** | 11,250 / 2,334 = 4.8x lighter | 4,700 / 2,334 = 2.0x lighter |+| Corriere | 7.7 KB | 1,934 | 500 | **2,434** | 95,000 / 2,434 = 39x lighter | 3,275 / 2,434 = 1.3x lighter |+| Daily Mail | 7.5 KB | 1,883 | 500 | **2,383** | 87,500 / 2,383 = 36.7x lighter | 12,500 / 2,383 = 5.2x lighter |+| **Average** | **7.5 KB** | **1,884** | **500** | **2,384** | **~27x lighter** | **~3x lighter** |++**Data source:** Empirically tested via Pinchtab API on Feb 27, 2026----## Cost Implications+## Cost Implications (Real Data)### Single Task (Fresh Agent)@@ -80,11 +82,11 @@ An agent with **zero prior context** spawns, calls Pinchtab once per site, extra|--------|----------|---------|--------------|---------------|| **Snapshot** | 500 | 64,583 | **65,083** | $0.000195 || **web_fetch** | 500 | 6,825 | **7,325** | $0.000022 |-| **Pinchtab** | 500 | 900 | **1,400** | **$0.000004** |+| **Pinchtab** | 500 | 1,884 | **2,384** | **$0.000007** |**Pinchtab cost advantage:**-- vs. Snapshot: **98% cheaper** ($0.000191 saved per task)-- vs. web_fetch: **82% cheaper** ($0.000018 saved per task)+- vs. Snapshot: **96.3% cheaper** ($0.000188 saved per task)+- vs. web_fetch: **68% cheaper** ($0.000015 saved per task)### At Scale (1,000 tasks/day)@@ -92,11 +94,23 @@ An agent with **zero prior context** spawns, calls Pinchtab once per site, extra|--------|-----------|--------------|-------------|| **Snapshot** | $0.195 | $5.85 | $70.20 || **web_fetch** | $0.022 | $0.66 | $7.92 |-| **Pinchtab** | **$0.004** | **$0.12** | **$1.44** |+| **Pinchtab** | **$0.007** | **$0.21** | **$2.52** |**Monthly savings with Pinchtab:**-- vs. Snapshot: **$5.73/month** = **$68.76/year**-- vs. web_fetch: **$0.54/month** = **$6.48/year**+- vs. Snapshot: **$5.64/month** = **$67.68/year**+- vs. web_fetch: **$0.45/month** = **$5.40/year**++### At Enterprise Scale (10,000 tasks/day)++| Method | Daily Cost | Monthly Cost | Annual Cost |+|--------|-----------|--------------|-------------|+| **Snapshot** | $1.95 | $58.50 | $702.00 |+| **web_fetch** | $0.22 | $6.60 | $79.20 |+| **Pinchtab** | **$0.07** | **$2.10** | **$25.20** |++**Annual savings with Pinchtab:**+- vs. Snapshot: **$676.80/year**+- vs. web_fetch: **$54.00/year**---@@ -250,27 +264,50 @@ If agent spawns for each task:## Validation Method-### Snapshot & web_fetch: Empirically Tested ✅-- OpenClaw `browser snapshot`: Actual API calls to BBC, Corriere, Daily Mail-- web_fetch: Actual Readability extraction from real sites-- Token counts: Measured directly from response sizes--### Pinchtab: Calculated from Known Behavior-Due to network sandbox constraints in test environment, Pinchtab response sizes are **calculated from:**-1. **Documented Pinchtab behavior:** `/snapshot` returns compact text representation (~2-5 KB for typical news pages)-2. **Empirical measurements:** Real-world deployments show Pinchtab /snapshot averaging 1-2 KB responses-3. **Comparative analysis:** Known to be 80-95% lighter than snapshots, 50-70% lighter than web_fetch--**Confidence:** HIGH (grounded in production data, not pure speculation)--**If you want to validate further:**-```bash-# Real-world Pinchtab test (requires network access)-pinchtab server-curl 'http://localhost:9867/snapshot?url=https://www.bbc.com&format=text&maxTokens=2000'-# Measure response size (should be 2-4 KB for news pages)+### All Three Methods: Empirically Tested ✅++**Test Date:** February 27, 2026+**Sites:** BBC.com, Corriere.it, Daily Mail.co.uk+**API Flow:** POST /tab {"action":"new","url":"..."} → GET /snapshot?tabId=X&format=text++#### Snapshot (OpenClaw)+- Actual API calls using OpenClaw's default isolated browser+- Full semantic DOM + accessibility tree+- Measured: 11K-95K tokens across test sites++#### web_fetch (Readability)+- Actual text extraction via Readability parser+- Boilerplate removal (nav, ads, sidebars)+- Measured: 3K-12K tokens across test sites++#### Pinchtab (Real-Chrome + Text Optimization)+- **Proper API flow:** POST /tab to create tab + navigate, GET /snapshot with tabId+- Text-based accessibility tree (semantic HTML representation)+- **Measured empirically:** 7.3-7.7 KB responses = ~1,834-1,934 tokens per site++**Response Format:**+```+# BBC - Home+# https://www.bbc.co.uk/+# 149 nodes++e0 RootWebArea "BBC - Home" [focused]+ e1 banner+ e2 main+ e3 contentinfo+ e4 region "Cookies on the BBC website"+ e5 navigation "BBC"+ e6 list+ e7 heading "Sport headlines"+ ...```+**Why this matters:**+- Returns semantic, clickable element refs (e0, e1, e2...)+- Not just text (unlike web_fetch) — can act on elements+- Real Chrome rendering (unlike Readability) — handles JavaScript+- ~7.5 KB average response size+---## For pinchtab.com Copy (Distributed Agent Angle)