Dashboard/Scout (Second-Brain)/log.md
Scout (Second-Brain)

log.md

Scout — Operation Log

Append-only. Most recent entries first.

2026-06-29 — Website logo, density, and scrollspy polish

  • Added Scout SVG mark and wordmark to the launch website.
  • Replaced the old block S brand treatment across static pages.
  • Tightened homepage density by roughly 10% while preserving the warm industrial direction.
  • Added section-aware homepage navigation and IntersectionObserver scrollspy.
  • Verified with website unit tests, route smoke, and Playwright desktop/mobile browser checks.

2026-06-29 — Unit economics pricing gate

  • Recorded Arijit's rejection of arbitrary $22 one-time and $9/month hosted pricing.
  • New gate: hosted pricing must be derived from unit economics before website, checkout, or launch plan claims a hosted price.
  • Current hypothesis: free local beta, invite-only hosted allowance, then pay-as-you-go or prepaid credits; subscriptions only after usage telemetry supports them.
  • Cost model must include hosting, LLM, firewall/security, support, maintenance, payment fees, storage/retention, browser minutes, screenshots, retries, target margin, and break-even volume.

2026-06-29 — Private beta launch checkpoint and relationship map

  • Updated Scout wiki index, overview, dev log, and relationship synthesis with current repo/product state.
  • Added ADR: Scout launch surface is controlled private beta with explicit limits, not public launch.
  • Captured downstream consumers: PRISM / Chowmes-PRISM, Competitive Intelligence, Algolia Search Audit skills, product/catalog workflows, job hunter workflows, and Scout website.
  • Recorded launch truth: private beta ready_with_limits; public launch blocked; public launch blockers require founder/legal/payment/risk decisions before further Codex execution.

2026-06-22 — UI wiring + PRISM endpoints + push

  • Wired /app/live-browser native-grab UI to show Crawl4AI-processed markdown + records (was raw blob). Download produces .md + optional .json.
  • Removed dead text/chars fields from NativeCaptureResult.
  • Built POST /structure (raw HTML → markdown + records via raw://) and POST /harvest (CDP URL → records from live tab). Both PRISM-callable, Pydantic contracts, API key auth.
  • Integration tested: 3 tests for /structure (real Crawl4AI), 2 for /harvest (real Chromium + marker-survival proof).
  • Total: 295 unit + 14 integration tests green, pyright 0 errors, ruff clean.
  • Branch codex/scout-platform-foundation pushed to GitHub (commit 353e26e).

2026-06-21 — Crawl4AI-over-CDP (core law enforced)

  • acquire_open_page() attaches Crawl4AI to human-cleared Chrome over CDP (browser_mode="cdp" + js_only=True). No re-navigation → no wall re-trigger.
  • Integration test: unique marker injected into live DOM survives extraction = proof of in-place read.
  • structure_capture() re-ingests held HTML through Crawl4AI via raw:// scheme. No network fetch.
  • 284 unit tests green.

2026-06-18 — Embedded browser + native-window fallback

  • CDP screencast engine (scout/core/live_browser.py) + WebSocket bridge (scout/api/routers/live_browser.py).
  • Embedded browser console UI at /app/live-browser.
  • Native-window open/capture with block detection and automatic UI notification.
  • Live finding: embedded canvas cannot clear PerimeterX press-and-hold (CDP input lacks OS fidelity). Native-window solve REQUIRED.

2026-06-17 — Human-assisted rung VALIDATED

  • scout unblock opened real Chrome → Arijit solved PerimeterX press-and-hold → Scout captured 615 real Zillow Roswell rentals (names, prices, beds).
  • First live proof the human-assisted rung beats a real anti-bot wall.
  • Block detector bug found + fixed (c7c8fbb): was matching vendor SDK scripts (persist on cleared pages) → false positive. Fixed to match challenge TEXT presence only.
  • Crawl-from-here finding: auto-nav to detail pages RE-CHALLENGES on PerimeterX. Per-page behavioral scoring → human must navigate each page (browse-and-harvest pattern).

2026-06-14 — T1 stealth rung (P2)

  • Fine-grained anti-detection: proxy, UA rotation, navigator patching, behavioral delays wired into scrape path.

2026-06-12 — Phase A complete + scope boundary ADR

  • UI audit + fixes + one-pager.
  • ADR: Scout = acquire + extract. PRISM = interpret. Scout never has an opinion.
  • ADR: Universal acquisition engine — consumers import Scout; scout.core never imports a consumer.

2026-07-06 — Rebuild shipped: neumorphic site + PLG backend live

  • Deployed twice: 70c7e07 (beta grant 10k + simple email-only /beta) and 908f6d9 (full rebuild).
  • LOCKED via HITL gates: mint-neumorphism design language (demo-first console IA, crisp dark data-screen, green accent, amber = evidence-only), reticle wordmark logo, pricing (Free 5k / Monthly $12·50k — never "unlimited" / demoted pay-go / dossier≈200), PLG funnel (anon demo 5/IP/day → free → paid Destinations), generic Destinations (Webhook+Algolia), support@ model (no reissue UI).
  • Dogfooded Scout to tear down firecrawl.dev (540+800 URLs mapped, 20 pages scraped, 5 full-page screenshots) → repo docs/competitor-crawl/firecrawl/.
  • Gates at deploy: 888 unit tests, pyright 0, ruff clean; live-verified all surfaces; PRISM unaffected.
  • Invites HELD until e2e ready. Next: support@ DNS (Arijit), branded key email wiring, Mintlify docs, GTM plan. Specs: repo docs/product/{design-system,pricing-model-2026-07-06,plg-playground-ux}.md