Arman's TFSA Journey

Methodology

Complete workflow for reproducible portfolio tracking

What this site shows

I'm contributing to a TFSA every week with a linear ramp (week 1 = $10, then +$1 per week). I run a simple weekly process to shortlist candidates, ask an LLM for a plan, place trades on Wealthsimple, and record executed fills only. This page documents that process so anyone can reproduce it.

Streamlined Web Interface

Phase 1: LLM Planning (“LLM Workflow”)

When: Sunday evenings before market close

Purpose: Get AI-powered trading recommendations

  • • Shows candidate stocks/crypto from screener results
  • • Provides structured LLM prompts and JSON payloads
  • • Validates LLM responses for constraint compliance
  • • Seamlessly transfers recommendations to trade recording
  • • Visual workflow with step-by-step guidance

Access: Main navigation → “LLM Workflow”

Phase 2: Execution (“Record Trades”)

When: After executing trades on Wealthsimple

Purpose: Record actual execution details

  • • Auto-populates with LLM recommendations
  • • Enter actual fill prices and adjust quantities
  • • Record weekly deposit contributions
  • • Preview portfolio impact before submitting
  • • Updates holdings and creates timeline entries

Access: From LLM Workflow or Main navigation → “Record Trades”

Streamlined workflow: LLM Workflow gets AI recommendations → seamlessly transfers to Record Trades → fill in actual execution details → submit to update your portfolio. No more copy/paste between tools!

Data sources (local JSON/CSV)

  • data/entries.json — weekly deposits and executed trades (for the public timeline)
  • holdings.csv — stocks/ETFs (ticker, shares, avg_cost, currency)
  • crypto_holdings.csv — crypto (symbol, amount, avg_cost_cad)
  • benchmarks.json — S&P 500 closes & HISA rate for comparison
  • data/candidates/latest.json — pointer to current week's fresh candidates
  • data/candidates/YYYY-MM-DD/ — dated folders with weekly screener output (stocks.json, crypto.json)

Weekly process (high level)

1. Generate fresh candidates (Sunday)

  • Run weekly screeners: Fresh top-40 movers change every week. Using stale candidates defeats the whole point.
  • Stocks: TSX-60 + S&P 500 universe → compute 1-week movers → take top 20 up + top 20 down (volume-screened).
  • Crypto: CoinGecko markets (CAD), Wealthsimple-supported only → top 20 up + top 20 down (volume-screened).
  • Holdings are always appended to the candidate list so the LLM can rebalance/trim even if something isn't a weekly mover.
  • Crypto fee baked in: all buy/sell math uses FEE_RATE = 2% spread via effective prices.
  • File structure: Candidates saved to data/candidates/YYYY-MM-DD/ with latest.json pointer.

2. Planning Phase: Use “LLM Workflow” Page

  • Access: Go to “LLM Workflow” in the main navigation.
  • View candidates: See latest screener results with candidate summaries.
  • Copy prompts & payloads: Get structured JSON for stocks and/or crypto.
  • Ask LLM: Paste JSON into your LLM and get trading recommendations.
  • Validate: Paste LLM response back to verify it follows constraints.
  • Auto-transfer: Click “Proceed to Trade Recording” to automatically load recommendations.

3. Execute trades on Wealthsimple

  • Place orders based on LLM recommendations.
  • If a ticker/coin isn't available or fractional is blocked, skip it (or choose CAD equivalent) and re-ask LLM.
  • Wait for fills: Only record what actually executes.

4. Recording Phase: Use “Record Trades” Page

  • Auto-populate: If coming from LLM Workflow, trades are pre-filled with recommended actions and quantities.
  • Fill execution details: Enter actual fill prices and adjust quantities based on Wealthsimple results.
  • Add deposit: Record weekly contribution amount.
  • Preview impact: See how trades affect your portfolio before submitting.
  • Submit: Updates holdings*.csv (weighted average cost) and appends entry to data/entries.json.

5. Complete the week

  • Check workflow status: Use “Complete Week” component (localhost only) to review weekly progress.
  • Create snapshot: Click “Complete Week” button to capture portfolio state and create permanent record.
  • Automatic updates: Site renders portfolio value, contributions, and benchmarks from updated files.

Linear contribution schedule

Starting deposit a = $10, weekly increment d = $1, week index t = 1,2,...

This week's deposit

ct = a + (t-1)d = 10 + (t-1)·1 = t+9

Example: week 52 deposit c₅₂=61

Total contributed after N weeks

SN = N/2(2a + (N-1)d) = N/2(N+19)

Example: S₅₂ = 1846

Benchmarks

HISA (3% APY, daily comp)

Daily rate rd = (1+0.03)1/365-1. Each weekly deposit compounds daily from its date: V = Σ depositi · (1+rd)days_since_i.

S&P 500 DCA

Each week's deposit buys index units at that week's close; units are valued at the latest close.

Limitations

  • Prices are end-of-day snapshots; intraday fills may differ.
  • CAD/USD mixing is simplified (CAD-listed preferred; crypto priced in CAD).
  • Dividends/distributions are not modeled yet.
  • No background automation; only filled orders are recorded.
  • Candidates must be refreshed weekly; LLM Workflow warns if data is >36h old.

Reproducibility

All calculations live in /lib/math.ts and scripts in this repo. Anyone can replicate outcomes by:

  1. running the screeners,
  2. placing the same trades,
  3. logging fills via the TUI,
  4. rebuilding the site from the same JSON/CSV.