Log

Gardener’s Log

[2026-05-14] ingest | trade-data-sources-map — replicating ImportGenius

Ingested comprehensive trade-data sources catalog covering the full stack required to replicate ImportGenius’s capabilities. Durable cross-project knowledge for Renso engagement, Renso Hub vision, and general company diligence work.

  • CREATED [[trade-data-sources-map]] — 12-section canonical reference: (1) core insight — ImportGenius is US CBP AMS + ~12 Latin American customs systems + broker-network Asian data + entity resolution; (2) pricing tier table for 13 commercial aggregators; (3) foreign customs systems by country with gold-tier open sources (Peru SUNAT, Colombia DIAN, Ecuador SENAE, Bolivia, Costa Rica, Panama, Russia FCS, Ukraine, Turkey) vs aggregate-only (Vietnam, China, Pakistan, Bangladesh, Indonesia, Thailand, Philippines, India, Mexico, Chile, Argentina, Brazil Comex Stat); (4) vessel/container tracking free and paid tiers; (5) customs aggregate free APIs (UN Comtrade, US Census, Eurostat, StatCan, USITC, WITS, WTO, IMF DOTS, OECD); (6) entity resolution sources (GLEIF, OpenCorporates, MRAS, OrgBook BC, Companies House, OFAC/EU/UN sanctions, FCC ID, Wikidata); (7) HS classification sources including LLM batch classifier (85–92% HS-4 accuracy); (8) adjacent intelligence (FDA, USDA, CFIA, EU CBAM, trade press); (9) mirror data technique for closed-customs countries with Vietnam factory fingerprinting recipe; (10) $0/month replication architecture (ingestion → resolution → classification → storage → query layers); (11) coverage gaps (Vietnam/China/APAC B/L, satellite AIS, pre-2015 history, entity resolution scale); (12) open-source prior art (frnsys/trade, marcdacosta/ambient-shipping)

Pages linked:

  • LINKED [[renso-trade-intelligence-stack]] — where this catalog gets applied
  • LINKED [[renso-engagement-frank-cto]] — Phase 3 deliverables this powers
  • LINKED [[renso-strategic-vision]] — Renso Hub data pipelines
  • LINKED [[Financial Data Sources]] — complementary macro data context
  • LINKED [[Canadian Government Data Indexing]] — StatCan and Canadian API patterns

Index updated:

  • UPDATED [[index]] — added new “Trade Data Intelligence” section with [[trade-data-sources-map]]; added to Recent Additions

[2026-05-14] ingest | Renso Foods Canada engagement intelligence

Ingested consolidated Renso Foods Canada engagement intelligence from internal proposal docs and open-web research. Sources referenced by path (not copied into mybrain/sources/):

  • /Users/franknguyen/renso/Renso-Engagement-Proposal.md — engagement structure, phased roadmap, compensation options
  • /Users/franknguyen/renso/Renso-Strategic-Vision.md — 10-year platform thesis, financial projections, sister entities
  • /Users/franknguyen/renso/Renso-Deal-Strategy.md — negotiation context, valuation analysis, structural risks, no-holdco confirmation
  • /Users/franknguyen/renso/Product_Intelligence_Engine.py — trade intelligence tooling
  • /Users/franknguyen/renso/ad_library_search.py — Meta Ad Library competitive intelligence
  • /Users/franknguyen/renso/Master_Trade_Data_Directory.md — macro/micro trade data source directory
  • External press: vietnamplus.vn post308185, post321007; vietnamnet.vn model-helps-boost-vietnamese-exports-to-canada-2411838
  • Company-controlled: rensofoods.com/about, rensofoods.vn/en/introduction, rensofoods.ca, ca.linkedin.com/company/renso-foods
  • Open-web research pass: 2026-05-14

Pages created:

  • CREATED [[renso-foods-canada]] — company profile: Vietnam parent (Renso Group JSC), HHM = Huu Hong Manufacturing reconciliation (most critical cross-source finding), Canadian footprint (Etobicoke/Richmond), key people (Linda Vu CEO, Faisal Rahman, Operations Director as counterparty), business profile ($10M claimed, 70–100+ SKUs), press footprint, competitive context (AFOD, Young & Young, Wismettac), risk signals (no holdco, trademark drift, pho restaurant unverified)
  • CREATED [[renso-engagement-frank-cto]] — Frank’s fractional CTO role: Tier 3 of 5-tier ladder, 4 compensation options (A pure equity 2.0% / B 1.5%+$4-5K / C 1.0%+$7-8K / D $10-12K cash), vesting (normal 25/25/25/25 vs back-loaded 10/20/30/40), minority protections, phased roadmap (Phase 1 operational stabilization Mississauga / Phase 2 growth automation / Phase 3 AI differentiation), negotiation context (no signed agreement, Operations Director counterparty, CEO signature required, opening offer $17K paper value after discount stack)
  • CREATED [[renso-strategic-vision]] — 10-year bet: Renso Hub platform thesis (rensohub.com currently offline), Year 10 base case $273M revenue / $64.5M EBITDA / 49% GM / $1.2B EV, sister entities (Ko Bites, Golden Bear, Mart, Foundation, HHM), SR&ED leverage ($21.2M cumulative projected), Siete Foods comparable, no-holdco mismatch between vision and Frank’s actual equity scope
  • CREATED [[renso-trade-intelligence-stack]] — Frank’s side-projects: macro/micro/joining triangulation architecture (StatCan 34-10-0012-01, UN Comtrade, Eurostat, SUNAT mirror data, GLEIF LEI resolution), Product Intelligence Engine (brand→owner→factory→logistics chain tested against Renso’s own supply chain), Meta Ad Library tooling (Vietnamese-food advertisers targeting Canada), connection to Phase 3 engagement deliverables

Index updated:

  • UPDATED [[index]] — added “Engagements & Projects” section with all four new pages; added all four to Recent Additions

[2026-05-04] ingest | the-autobots production fix, commit 70d68b1

  • CREATED [[Agent Fleet Retry Loops]] — new page capturing the reviewer→coder retry-cascade pathology: three compounding failure modes (critique buried at bottom, no structured required_change field, LLM-call failure silently treated as rejection), the complete fix pattern, and prior-art anchors (Reflexion, MetaGPT, Live-SWE-agent, LLM-as-judge ECE study, METR time-horizon)
  • UPDATED [[index]] — added new top-level category “Agent Systems & Orchestration” with [[Agent Fleet Retry Loops]] as first entry

2026-04-25

  • MAINTENANCE: Corrected the Point2 Homes link in the master list to point specifically to commercial/retail listings rather than residential rentals.
  • MAINTENANCE: Removed Zonado.com from the master list due to user safety concerns and replaced it with Zolo.ca (Commercial).
  • RESEARCH: Expanded research to identify exactly 30 distinct commercial leasing sources in Toronto, including ethnic-market portals (RealMaster) and niche property managers.
  • SYNTHESIS: Finalized [[The Toronto Cafe Leasing Master List: 30 Essential Sources]] as a comprehensive multi-layered search directory.
  • RESEARCH: Conducted exhaustive deep-dive into Toronto cafe leasing market, identifying 5 layers of sources (Institutional, Boutique, Niche Brokers, Public, and Hyper-Local).
  • SYNTHESIS: Created [[Toronto Cafe Leasing: Master Source List]] as a comprehensive guide for independent cafe operators.
  • RESEARCH: Clarified legal requirements for Ontario commercial leases (Sole Proprietor vs. Corp) and “Personal Guarantee” standards.
  • SYNTHESIS: Defined the “Tenant Package” (Business Plan, Credit, Funds) in [[Commercial Real Estate Leasing (Toronto)]] to improve user outreach success.
  • RESEARCH: Identified major Toronto retail landlords (Allied REIT, Hullmark) and alternative listing sources (Kijiji, BIA lists).
  • SYNTHESIS: Updated [[Commercial Real Estate Leasing (Toronto)]] with direct landlord portals and neighborhood-specific search strategies.
  • RESEARCH: Refined search for Toronto-specific commercial leasing (Spacelist.ca, OMG Real Estate, TMI terminology).
  • SYNTHESIS: Updated [[Commercial Real Estate Leasing (Toronto)]] with GTA-specific platforms and “Vented space” requirements.
  • RESEARCH: Researched specific open-source tools, API endpoints, and data volumes for Canadian government RAG pipelines.
  • SYNTHESIS: Authored [[Canadian Government Data RAG Roadmap]], detailing a multi-stage Hybrid RAG implementation strategy.
  • INDEX: Updated [[index.md]] with the new roadmap.

2026-04-24

  • PROPOSAL: Drafted [[Project Proposal - Canadian Government Data Intelligence System]] based on the researched CGDIS architecture.
  • ROADMAP: Documented the technical implementation details in [[Canadian Government Data RAG Roadmap]].
  • RESEARCH: Completed deep research on Canadian Government data ingestion and LLM sectorization using @deep-researcher.
  • SYNTHESIS: Created [[Canadian Government Data Indexing]] detailing HybridRAG strategies, Open Data APIs, and scraping constraints.
  • INDEX: Updated [[index.md]] with the new research and proposal.

2026-04-23

  • RESEARCH: Delegated deep research on best GitHub repositories for AI, ML, and Machine Vision to @researcher.
  • SOURCE: Saved research report to sources/github_repos_ml_research_2026.md.
  • SYNTHESIS: Created [[AI and ML Learning Resources]] with categorized repository lists (Beginner to Advanced).
  • SYLLABUS: Integrated LLMs-from-scratch and MLOps milestones into Phase 04 and Phase 05 of the [[Syllabus]].
  • LEARNING PATH: Updated [[Learning Path - ML Development]] with new resources and related paths for CV and MLOps.
  • INDEX: Updated [[index.md]] with the new resources page.

2026-05-23

  • RESEARCH: Completed exhaustive technical report on Apple Foundation Models (AFM) metrics and benchmarks.
  • METRICS: Extracted MMLU (80.0), GSM8K (72.4), and IFEval (89.1) for AFM-server; documented 3B on-device parity.
  • QUANTIZATION: Detailed “Low-Bit Palettization” (3.5-3.7 bpw) and LoRA-based accuracy recovery.
  • HARDWARE: Benchmarked AFM-on-device across M1-M4 Max chips; documented 58.7 TPS on M4 Max.
  • CONTRADICTION: Analyzed the “Helpfulness Gap” where AFM-on-device matches Llama-3-8B in human preference despite lower raw parameters.
  • UPDATE: Created [[Apple Foundation Models (AFM) Metrics]]. Updated [[index]].

2026-05-22

  • STRESS TEST: Performed technical deep-dive into Arthur-Ficial/apfel and Apple Foundation Models (AFM).
  • RESEARCH: Verified FoundationModels.framework architectural bottlenecks (signed-binaries only, 2-bit ANE quantization).
  • REVISION: Corrected “macOS 26 Tahoe” hallucination to macOS 15.1+ (Sequoia) with SDK 26.4 context.
  • BENCHMARK: Documented inference latency comparison (APFEL/ANE vs MLX/GPU) on M3 chips.
  • SOURCE: Created sources/apfel_deep_dive_raw.md.
  • UPDATE: Synthesized findings into [[APFEL & Apple Local Inference]].

2026-04-22

  • INGEST: Performed deep research on Arthur-Ficial/apfel and Apple Foundations. Created [[APFEL & Apple Local Inference]]. Updated [[index]].

2026-04-15

  • INGEST: Added research on 2028 Global Intelligence Crisis (GIC) and Citrini unlearning. Created [[2028 Global Intelligence Crisis (GIC)]] and [[Model Unlearning & Editing]].
  • UPDATE: Expanded [[index]] with new crisis taxonomy.

2026-04-10

  • PROJECT: Initialized [[Learning Path - ML Development]] and [[Syllabus]].
  • FOUNDATIONS: Created [[Backpropagation]].