init commit

README.md

@@ -0,0 +1,149 @@
+# Career Permutation Generator — README
+
+Models 2,535 different combinations of career path, property location/strategy,
+deposit size, savings rate, and purchase timing for Breadway (solo, age 18→50),
+running each combination 200 times with randomised income shocks, interest
+rate movements, and property crash risk to produce a realistic RANGE of
+outcomes rather than one fragile prediction.
+
+---
+
+## Files
+
+| File | What it is |
+|---|---|
+| `income_schedules.py` | The 13 career path income curves + HECS debt totals. Edit this to change/add careers. |
+| `generator_v2.py` | The simulation engine. Run this file to regenerate everything. |
+| `career_permutations_v3.csv` | Full output — one row per scenario combination (2,535 rows). |
+| `leaderboard_v3.csv` | Top 10 scenarios in four categories, pulled from the full CSV. |
+| `super_cgt_notes.txt` | Plain-English notes on the superannuation and capital gains tax rules actually applied. |
+
+**To regenerate:** put `income_schedules.py` and `generator_v2.py` in the same
+folder, run `python3 generator_v2.py`. Takes about 75 seconds. It overwrites
+`career_permutations_v2.csv` and `leaderboard.csv` in that folder — copy them
+out afterwards.
+
+---
+
+## How it works, in order
+
+**1. Pick a career path.** 13 options, each with a year-by-year income
+schedule from age 18 to 50 (gap year → study → graduate → career progression).
+All salaries are pulled to the conservative end of researched ranges, not the
+midpoint.
+
+**2. Pick a property approach.** Location (Geraldton house, Geraldton land,
+Perth, Canberra, or none/rent-forever) × strategy (buy and hold, buy land then
+build, or rent forever) × deposit size (20% or 50%) × the age you attempt to
+buy (23 or 28) × what age you might sell and rebuild in Geraldton instead
+(never, 35, 40, 45, or 50). Invalid combinations are automatically skipped
+(e.g. you can't "buy land then build" in Perth, you can't set a sell age if
+you're renting forever).
+
+**3. Pick a savings rate.** How much of your leftover income after costs
+actually gets saved rather than spent — 65%, 75%, or 85%.
+
+**4. Run that exact combination 200 times.** Each run is identical in its
+setup but different in its luck: random job-loss events, random year-to-year
+income variation, a randomly generated 33-year interest rate path, and random
+property market crashes. This produces 200 different final net-worth numbers
+for the same scenario. The CSV reports the 10th percentile (bad luck), 50th
+percentile (typical), and 90th percentile (good luck) of those 200 outcomes.
+
+**5. Repeat for all 2,535 valid combinations**, producing one summary row per
+combination — so 507,000 individual 33-year simulations in total.
+
+---
+
+## What gets randomised, and why
+
+**Income jitter** — every year's "scripted" salary gets ±4–7% random noise
+(varies by career; software/data jitters more, RAAF jitters least) to reflect
+that real raises and bonuses aren't perfectly smooth.
+
+**Job loss / income gap risk** — from age 23 onward, each career has an
+annual probability of a bad year (redundancy, project ending, injury-equivalent
+income hit), calibrated by how stable that sector actually is. FIFO and
+trades sit at 3–5%/yr (cyclical, project-based work). Engineering and defence
+sit at 1.5–2%/yr. RAAF sits at 0.5%/yr (military job security is real).
+**These probabilities are my own calibrated estimates, not sourced unemployment
+data** — directionally sensible, not empirically precise.
+
+**Interest rate path** — instead of a flat 5.9% for 33 years, each run
+generates its own random 33-year rate path: a mean-reverting walk anchored to
+a 5.5% "neutral" mortgage rate, with normal year-to-year drift plus an
+occasional (~1-in-12-year chance) sharp move up or down, mimicking real RBA
+cycles (confirmed history: cash rate has ranged from 0.10% in 2020 to 17% in
+1989; mortgage rates track roughly 1.5–2.5% above cash rate). Mortgage
+repayments are recalculated every year against the rate that actually applied
+that year, the way a real variable-rate loan behaves.
+
+**Property crash risk** — each location has its own annual probability of a
+market downturn and how severe it is if one hits, based on confirmed history:
+Perth fell ~15% peak-to-trough 2014–2019 after the mining boom ended; Pilbara
+mining towns (Karratha, Port Hedland) fell close to 80% in the same period.
+Geraldton sits between the two — more diversified than a pure mining town,
+less stable than Perth. Canberra (government town) gets the smallest crash
+risk of anywhere modelled. A crash applies a temporary value discount that
+recovers in a straight line over 4–6 years.
+
+**Cost inflation** — rent and living costs inflate at 3%/yr from age 26
+onward, rather than sitting flat for three decades.
+
+---
+
+## What's deliberately NOT in this model
+
+- Addi's income, a shared household, marriage, or splitting up — this is
+  Breadway solo only, by request.
+- Children — explicitly parked, not modelled at all.
+- Health/injury shocks that take someone out of work for an extended period
+  (the job-loss risk models *economic* job loss, not a serious injury).
+- Car ownership and replacement costs.
+- Lifestyle inflation — the model assumes the same fixed living costs whether
+  you're earning $80k or $200k, which isn't how real spending works.
+- Correlation between bad property markets and job loss — in reality a
+  commodity price crash often causes both at once (that's exactly what
+  happened in the Pilbara). This model treats them as independent risks,
+  which understates how bad the worst-case scenarios could really get.
+- Anything past age 50.
+
+---
+
+## Column reference (career_permutations_v3.csv)
+
+| Column | Meaning |
+|---|---|
+| `career_path` | Which of the 13 careers |
+| `location` | Property location (or NONE if renting forever) |
+| `strategy` | BUY_HOLD / BUY_LAND_BUILD / RENT_FOREVER |
+| `sell_age` | Age at which the Perth/Canberra house is sold and Geraldton is built instead (N/A if never) |
+| `deposit_pct` | Deposit size attempted (0.20 or 0.50) |
+| `savings_rate` | Share of leftover income actually saved (0.65 / 0.75 / 0.85) |
+| `requested_buy_age` | The age the purchase was *attempted* |
+| `pct_scenarios_bought_property` | Of the 200 runs, what % actually managed to buy (some never afford the deposit) |
+| `median_actual_buy_age` | Of the runs that did buy, the typical age it actually happened (may be later than requested if savings took time to catch up) |
+| `pct_debt_free_by_50` | Of the 200 runs, what % had zero mortgage/loan balance by age 50 |
+| `median_debt_free_age` | Typical age debt-free was reached, across runs that got there |
+| `net_worth_p10_age50` | 10th percentile outcome — roughly "if things went badly" |
+| `net_worth_p50_age50` | 50th percentile — the typical/median outcome |
+| `net_worth_p90_age50` | 90th percentile — roughly "if things went well" |
+| `net_worth_range_age50` | P90 minus P10 — how much luck matters for this combination |
+| `median_liquid_age50` / `median_super_age50` / `median_property_value_age50` / `median_mortgage_remaining_age50` | Median breakdown of the net worth figure across the 200 runs |
+
+`leaderboard_v3.csv` uses the same columns with one extra: `leaderboard_category`,
+which tags each row as one of: Max Wealth (highest P50), Fastest Debt Free
+(lowest median debt-free age), Most Reliable (smallest P90–P10 spread among
+above-median-wealth scenarios), or Best Worst-Case (highest P10 — the safest
+floor if things go badly).
+
+---
+
+## Honest framing
+
+This is a wealth-accumulation model. It answers "how much money might I have
+at 50 under this combination of choices" with a realistic spread rather than
+a single confident number. It does not answer whether the work itself is
+enjoyable, sustainable, or compatible with the rest of a life — FIFO tops
+almost every wealth ranking here because the model only measures money, and
+real people don't only optimise for that.