Paper-First Promotion Gates, Earning the Right to Trade

Pull-quote: “A strategy earns live capital the way code earns production: by passing gates somebody wrote down before the run started.”

Why this matters

The most dangerous moment in a strategy’s life is the transition from research to live trading, because that is where enthusiasm does the sizing. A good backtest feels like evidence, and the feeling arrives precisely when the least is actually known: nothing has yet been proven about the strategy’s behavior against live feeds, live order handling, and live operational reality. Software engineering solved the equivalent problem years ago with staged deployment and promotion criteria. A systematic desk needs the same machinery: a pipeline where strategies advance only by meeting written gates, and where demotion is as mechanical as promotion.

The pipeline

Nothing should trade live without passing paper-first promotion gates. The pipeline looks like this:

 research ──► backtest ──► PAPER ──────────► LIMITED LIVE ──► FULL ALLOCATION
              (parity-      live data,        real orders,     strategy caps
               grade        real decisions,   small caps       still apply
               replay)      no orders
                 │             │                  │                  │
              gate 1        gate 2             gate 3          continuous
              written       written            written         review, with
              criteria      criteria           criteria        demotion rules

Paper trading is not a victory lap after the backtest. It answers a different question. The backtest asks whether the logic would have worked against history. Paper asks whether the system works against the present: does the strategy behave on live data with live latencies, do its decisions match what replay says they should be, does the operational tooling hold up. Paper is where plumbing fails, and plumbing failures are the cheap kind only while no orders are real.

Decision parity is the paper-stage measurement that matters most. Run the same session through the replay harness afterward and compare: did the paper system decide what the research code says it should have decided, tick for tick. Divergence means the research environment and the production environment disagree about something, and finding that something now is the entire point of the stage.

What each stage can and cannot prove

The table explains why skipping stages fails: each stage retires a risk the previous one structurally cannot see. Paper cannot observe your market impact because paper orders touch no book. Limited live exists precisely to measure what paper cannot, at a size where the tuition is affordable.

Gates are written before the run

The gate criteria are pre-registered: defined before the stage begins, recorded in the append-only audit log, and evaluated against what was written rather than renegotiated after the data arrives. Thresholds chosen after results are in drift toward whatever the most enthusiastic person wanted. Promotion decisions, and the evidence behind them, land in the same log as everything else, so a later review can reconstruct why a strategy was trusted and whether the process was followed.

Demotion is the half that makes the system credible. The same written criteria run in reverse: a live strategy that violates its expectations returns to paper, mechanically, without a meeting in which sunk cost gets a vote. Strategy caps and the rest of the four risk layers still apply at every stage, so even a fully promoted strategy operates inside limits it cannot loosen.

Closing

Promotion gates are how a desk separates evidence from enthusiasm. Backtest, paper, limited live, full allocation: each stage retires a specific risk, each gate is written before the run, and demotion is as mechanical as promotion. A strategy that cannot pass this pipeline was never an edge. It was a feeling with a Sharpe ratio attached.