You bring the strategy rules; I build and run the backtest on the GitHub-based quantitative backtesting harness I developed for my own macro research — Macro Backtest Runner v1.2 — and document exactly what the test does and does not prove.

This isn't a spreadsheet or a one-off script. The harness separates three questions that most backtests sloppily collapse into one:

COMMON_WINDOW_ARCHITECTURE — allocation and mechanics comparisons where every scenario runs over one identical sample (a ~23-24 month common window), so differences come from the architecture, not from cherry-picked periods

ANCHOR_FORWARD_THESIS — a portfolio/thesis frozen at a documented historical anchor date (first verified anchor: 2026-03-27) and measured forward, with no later information allowed to leak into the baseline

EVENT_DATED_SYSTEM — replays a documented decision ledger, where adds, cuts, and replacements enter only on the dates they actually happened

Eleven test harnesses cover allocation weighting, defense-mix sensitivity, dependency caps, satellite weight grids, orchestration splits, diversifier tests, replacement comparisons (FN allocation, RSP replacement), and correlation/brutality stress tests — 100+ documented allocation scenarios, including a 64-combination satellite weight grid (3 positions swept across 4 weight levels, backed by 27,514 simulated trades), plus 100-iteration stress loops, with tens of thousands of simulated trades persisted as structured result artifacts.

Completed-month rule (v1.2): the incomplete current month is excluded from every metric — no partial month dressed up as a finished one

Every compared scenario uses identical assumptions; only the variable under test changes

Costs, contributions, and drift thresholds are modeled explicitly

Requested vs. actual start dates are always reported, with the constraining series named

Every test carries a bias classification — common-window tests use a present-day universe retrospectively, so survivorship/selection bias is labeled, not hidden

A result counts only after the GitHub Actions run succeeds and the structured result artifact is persisted

We define your strategy in explicit, testable rules (entries, exits, sizing, filters) — if a rule can't be stated precisely, that's the first finding

I implement it in the harness and run it against historical data

You get the results with every assumption exposed: data source, costs/slippage treatment, look-ahead and survivorship controls, and parameter sensitivity

A backtest tests mechanics and behavior — it is not proof of future returns, and this service is research tooling, not investment advice. What it does give you is a disciplined way to find out whether a strategy idea survives contact with its own rules before real money is involved, and a record of exactly where it breaks.

Independent traders and small teams with a rule-based strategy idea who want it tested rigorously instead of vibes-tested.