- Migration 099: flexoptix_internal_demand table with RLS + v_demand_by_speed view
- Import script: AES-256-CBC decrypt → parse 8585 SKUs → upsert with velocity class
- 279 SKUs cross-referenced to transceiver catalog; 1288 with real demand data
- New /api/internal/demand/* routes (by-speed, velocity, hype-weights, forecast-input)
— protected by JWT auth + localhost/LAN IP restriction middleware
- Forecast engine calibrated with real Flexoptix run-rates (demand_calibrated flag)
- Dashboard: real Flexoptix Sales Velocity panel replaces DEMO DATA in Warehouse tab
with momentum indicators, velocity class breakdown, trend arrows
- Security: data stays on private server; RLS enforces is_internal=TRUE at DB layer
cdn.cookielaw.org logos appear as the largest DOM image on Dell/Extreme
product pages when the cookie consent overlay is present. Added to both
GENERIC_IMAGE_PATTERNS (isGenericImage filter) and img fallback skipPattern
so the next-largest actual product image can be found.
Previously: if og:image existed (even as a Dell logo URL), page.evaluate() returned
early and the img fallback was never tried. Now: meta tags are extracted first, then
isGenericImage() is checked in Node.js, and the img fallback runs if meta image is null
or generic. This allows vendors like Dell (og:image = logo) to still get product images
via the DOM fallback.
- buildAristaUrl() now extracts series prefix (7060X5-32QS → 7060x5-series)
instead of individual model URLs that lack og:image
- Strip trailing sub-variant 'A' so R3A → R3 series page
- Add uniqueKey: row.id to each request — prevents Crawlee from deduplicating
models that share the same series URL (e.g. 7060x5-series)
- For Arista: always prefer fresh builder URL over stored product_page_url
so stale individual-model URLs don't override correct series pages
switch-image-fetcher:
- Add Fortinet URL builder (11 FortiSwitch models)
- Add Quanta Cloud Technology, Allied Telesis, Ufispace, Netgear URL builders
- Fix alcatel-lucent-enterprise slug missing from URL_BUILDERS dispatcher
- Fix NVIDIA builder to skip ConnectX/BlueField adapters (not switches)
- Add aruba slug alias for hpe-aruba
health endpoint:
- Add system metrics: CPU load (1/5/15m), memory usage, disk usage
- Add load_status indicator (ok/busy/overloaded)
- Expose process RSS memory
- Used by external monitors
scripts/monitor-erik.sh:
- Cron-ready health check script for Claudi (.82) and Raspberry Pis
- Checks TIP API health endpoint (load, memory, disk, DB latency)
- Checks PM2 process state via SSH (errored/stopped detection)
- ntfy.sh push notifications (set NTFY_TOPIC env var)
- Includes systemd service + timer unit comments for auto-install
Replace placeholder models (93180YC-FX3, 7280R3A, QFX5120 — not in DB) with
the real seeded switches: N9K-C9364C, 93600CD-GX, 7060CX2-32S, QFX5130-32CD, SN5600
- Add flexoptix-compat.ts: maps switch models to compatible Flexoptix transceivers
via search API (vendor_compat) with form-factor fallback (spec_match)
Scheduled daily at 09:00 UTC as scrape:compat:flexoptix
- Enhance community-issues.ts: add vendor advisory sources (Cisco Field Notices,
Juniper KB, SONiC GitHub Issues) + new scrapeTransceiverCompatIssues() that
searches for switch+transceiver combination problems specifically
- Scheduler: 59 schedules, 78 workers
price?: number narrowing via typeof/!== undefined does not work for
arithmetic comparisons in TypeScript 5.9 dead code paths; use 'as number'
cast to keep the dead code compilable while the early-return guard above
prevents runtime execution entirely.
- ATGBics: update HTML parser from old card--product theme to new
card__info theme (Shopify template changed April 2026); name now
extracted from href link text instead of aria-label
- NADDOD: correct ensureVendor shop URL from /collections/transceivers
(404) to /collection/optical-transceivers
- VCELink: disable scraper — site pivoted from optical transceivers to
audio/video/cable products; all collection URLs return 404
- Replaces Playwright with pure fetch() — static HTML has prices
- Correct collection handles (compatible-transceivers-sfpp-10g etc.)
- Cookie: cart_currency=GBP forces GBP pricing from any geo-IP
- Handles 35+ pages per category × 24 products = 840+ SFP+ products
- No IP-blocking with static HTML (Playwright was the trigger)
- Adds scripts/run-atgbics-mac.sh for Mac-side runner if needed
- Fix OSFP-DR8-1.6T-FL and OSFP-2FR4-1.6T-FL: speed_gbps was 200, now 1600
→ FS.com 1.6T products now correctly match as comparables for Flexoptix O.1316T.C.05.M
- API: extend comparable price query to return comp_form_factor, comp_speed_gbps,
comp_reach_meters, comp_reach_label, comp_fiber_type, comp_wavelengths
- Dashboard: replace plain comparable price row with side-by-side spec comparison card
showing Flexoptix vs. competitor: Form Factor, Speed, Reach, Fiber, Wavelengths
with color coding (green=match, orange=mismatch) and savings badge (−45% günstiger)
- client.ts: add claude-code provider routing BLOG_LLM_PROVIDER=claude-code
to claude-bridge (flat-rate, no API billing via Claude Code subscription)
- checkHealth() now pings /health on claude-bridge for real availability check
- Default OLLAMA_LLM_MODEL changed from qwen2.5:14b to fo-blog-v5
- Dashboard: add claude-code card (EMPFOHLEN), rename fo-blog-v3 → fo-blog-v5
- loadBlogLLMStatus() handles all 3 providers: claude-code/anthropic/ollama
- Grid expanded from 3 to 4 columns to accommodate new card
- ecosystem.config.js + .env on Erik: OLLAMA_LLM_MODEL=fo-blog-v5 confirmed
- All 247 FS.com prices were €79 (shipping threshold, not product prices)
- Root cause: 'Gratis Versand ab 79 € (ohne MwSt.)' banner matched first
- Fix 1: DOM price extraction in page.evaluate with bad-parent skip list
- Fix 2: bodyText qualified patterns skip matches near shipping keywords
- Fix 3: waitForSelector for price DOM element before evaluate
- Fix 4: Deleted 247 invalid €79 observations from DB
Also included from previous session:
- db.ts: set has_image=true on image writes (fix 632 desync rows)
- spec-updater.ts: DR/FR/LR/ER/ZR → SMF, SR → MMF fiber type inference
Both generateBlog() and generateBlogManual() were calling
POST /api/blog/generate without an Authorization: Bearer header.
The requireAuth middleware correctly returned 401, which appeared
as 'Unauthorized — please log in' toast in the dashboard.
Fix: read loadToken() before each fetch and include the token in
the Authorization header. Also add r.status===401 guard to redirect
to login page when token expires, instead of showing error toast.
optcore.net blocks Erik's IP (82.165.222.127) via Cloudflare WAF.
WP REST API returns HTML block page instead of JSON → 0 product URLs
→ 0 scraped pages every run. Add SKIP_OPTCORE_SCRAPER guard matching
the existing SKIP_FS_SCRAPER pattern. Set in ecosystem.config.js on
Erik. Residential IP (Mac launchd) would be needed to use this scraper.
Crawlee's FileSystemStorage marks request URLs as HANDLED (state=4,
orderNo=null) after processing. With purgeOnStart=false these entries
persist, so on the next run crawler.run(startUrls) deduplicates them
→ requestsTotal=0 → immediate finish with 0 scraped pages.
Fix: rmSync request_queues/default/ before each makeCrawleeConfig()
call. Safe: session pool state lives in key_value_stores/, not in
request_queues/. Affects all Crawlee-based scrapers (ATGBICS, Optcore,
Switch-assets, etc.).
10gtek.com main site only exposes technical spec tables with no prices.
sfpcables.com is 10Gtek's own retail store and has both Model numbers
and USD prices in standard Magento product listings.
Changes:
- Switch scraping target from www.10gtek.com to sfpcables.com
- Parse Model: <part> + US.XX per product block (Magento structure)
- XFP fallback: extract part number from title after '|' separator
- Add fetchAllPages() with Magento loop-detection via seen-part dedup
- Remove QSFP-DD category (not available on sfpcables.com)
- Drop XFP-less categories from old 10gtek.com spec-table parser
Verified: 10/10 SFP prices, 10/10 SFP+ prices, 4/4 XFP prices on live site.
- Add global unhandledRejection handler in scheduler daemon to swallow
Crawlee's benign post-run ENOENT lock-file races (prevents process.exit(1))
- Add SKIP_FS_SCRAPER env var: skip FS.com worker on Erik where Cloudflare
WAF blocks datacenter IPs (Mac launchd handles FS.com from residential IP)
- Remove FS.COM from health monitor EXPECTED_VENDORS (skipped on Erik)
- Health monitor: extend pg-boss lookup from 12h → 26h, add completed-job
map; if job ran OK in last 26h + vendor has historical prices → mark
STABLE instead of CRITICAL (fixes ATGBICS/Fluxlight hash-dedup false positives)
- Install Playwright Chromium on Erik (fixes ATGBICS BrowserLaunchError)
- Create missing Crawlee storage dirs on Erik (storage-fs-phase1/2,
storage-ebay-transceivers) to prevent ENOENT on first Crawlee run
Pattern 1 (href→aria-label) finds 127 navigation links on GBICS BigCommerce
pages — none contain GBP prices. Pattern 2 (aria-label→href) correctly
finds 16-30 product links per category page with £XX.XX prices in aria-labels.
The fallback from P1 to P2 now triggers when P1 finds results but none
contain '£', rather than only when P1 finds 0 total results.
Previous logic fired an alert whenever prices_6h=0, even when prices
were genuinely stable (content hash dedup prevents duplicate inserts).
This caused Flexoptix, ATGBICS and others to trigger alerts every 3h
despite their scrapers running successfully.
New logic:
🔴 CRITICAL: last price > 7 days (genuine failure)
🟡 WARNING: last price 48h–7 days (possibly stale)
✅ STABLE: last price ≤48h, 0 new (prices unchanged, scraper OK)
Also shows pg-boss job state/time alongside each vendor for faster
root-cause diagnosis. Trimmed EXPECTED_VENDORS to vendors with actual
scraper implementations (removed never-scraped placeholders).
After PlaywrightCrawler.run() resolves, Crawlee's internal task loop
schedules one final _isTaskReadyFunction call that tries to read a
request queue .json file already cleaned up during processing. This
ENOENT fires as an unhandledRejection and calls process.exit(1),
aborting Phase 2 before prices are written to the database.
Added a targeted unhandledRejection handler in the require.main block
that swallows ENOENT errors from request_queues paths (benign Crawlee
cleanup race) while re-raising all other rejections.
- Add makeCrawleeConfig isolation to CheerioCrawler instances
- Switch from named persistent RequestQueue to ephemeral null queues:
named queues retain 'handled' state and skip all URLs on re-runs,
causing 0 observations on every run after the first.
- Applies to both enrichSwitchFromEbay and enrichTransceiversFromEbay.
- Add utils/crawlee-config.ts: makeCrawleeConfig(name) returns a
Crawlee Configuration with isolated localDataDirectory per scraper.
Uses storageClientOptions (not global CRAWLEE_STORAGE_DIR) so
concurrent pg-boss workers in the same process don't race on
the shared env var.
- Apply makeCrawleeConfig to all 6 Crawlee-based scrapers:
optcore (PlaywrightCrawler), atgbics (PlaywrightCrawler),
community-issues (CheerioCrawler + RequestQueue),
edgecore (CheerioCrawler), ufispace (CheerioCrawler),
market-intelligence (CheerioCrawler).
- scheduler.ts: add withIsolatedStorage for optcore and market-intel
workers (was missing, caused storage-fs path bleed from fs scraper).
- ebay-enricher.ts: fix vendor type 'marketplace' -> 'reseller' to
satisfy vendors_type_check constraint
['manufacturer','distributor','oem','reseller','compatible'].
