Episode 5: From Frontier Search To Portfolio Thinking

Read this article with Options Backtesting API, Backtesting Framework, Backtesting Data Quality Checklist, Backtesting Execution Realism, Quote-Aware Options Backtests, and Backtest to Paper Trading Parity Checklist.

Scope

This episode covers the strategic pivot that became explicit in early April.

The key evidence lives in:

• RUNS.md, especially the April 6 audit entries
• roadmap.json
• PAPER_BOTS.md

Result Snapshot

The decisive phrase from the repo is not a model name. It is the audit verdict:

• framework_sound_strategy_mismatch

And the recommended next action:

• move_to_new_strategy_ideation

That is the point where the project stopped treating the problem as "optimize ORB harder" and started treating it as "assemble a diversified set of strategies that actually survive the deployability bar."

The roadmap then made the portfolio goal explicit:

• build a small diversified paper-bot portfolio
• stop broad standalone ORB frontier search as the main path

Scientifically, this is a change in objective function. The project stopped optimizing for the best-looking isolated branch and started optimizing for a candidate set that could plausibly coexist under overlap, robustness, and deployability constraints. That is a much harder problem, but it is also a much more realistic one. A live portfolio does not care whether one backtest looked beautiful if that sleeve adds no diversification or fails under operational scrutiny.

Why The Pivot Happened

The repo had accumulated enough evidence by this point to justify a strategic change:

• March realism fixes made results harsher
• ORB audit narrowed what was still defensible
• complement branches were beginning to fail quickly
• the remaining interesting lanes did not all look like ORB

That creates a classic systematic trading choice:

• keep optimizing the incumbent family because it is familiar
• or switch from single-family frontier search to portfolio assembly

This repo chose the second path.

What Worked

What worked was the willingness to name the problem correctly.

The portfolio roadmap did not say "best ORB." It said:

• c66 lead paper bot
• c4 next candidate
• c36 backup candidate
• QQQ dispersion remains research-only

That is already a portfolio mindset. These models are not clones. They are being evaluated partly on their ability to coexist.

The repo's later gates use overlap days, correlation context, sample density, quote realism, and operational parity as part of the promotion logic. Once those dimensions matter, a branch can be economically interesting and still not deserve a slot. The public series should keep returning to that point because it is central to the research philosophy the repository eventually converged on.

This is one of the strongest scientific signals in the repo because real trading systems do not get paid for having one beautiful backtest. They get paid for having a basket whose members are credible together.

What Did Not Work

What did not work was the implicit earlier hope that the project would discover one dominant ORB configuration and then merely scale it.

The audit and roadmap jointly reject that.

The practical negative result is:

• strategy families were failing under a legitimate deployment bar
• more sweeps inside the same family were no longer the highest-value use of time

That is a hard decision to make in research because it often feels like giving up on sunk effort. In fact, it is the opposite. It is refusing to protect sunk effort from evidence.

Why This Week Matters

This was the week the repo got a portfolio objective, more than a model objective.

In restrained One Piece language, this is where the journey stops being about the strongest fighter and starts being about the crew.

One strong but highly correlated or fragile strategy is not the One Piece of Sharpe. A low-overlap basket of believable models is much closer.

Public Build Takeaway

The public lesson from Episode 5 is worth stating directly:

• when the evidence says the family is mismatched to the bar, pivot
• define the portfolio before you define the victory speech
• treat negative model evidence as a budgeting signal

This repo changed more than direction. It changed the question. That is often the highest-leverage move in systematic research.

Related workflow

For the Episode 5: From Frontier Search To Portfolio Thinking workflow, continue through Options Backtesting API, Backtesting Framework, Backtesting Execution Realism, Backtesting Data Quality Checklist, Quote-Aware Options Backtests, and Backtest to Paper Trading Parity Checklist.

How the terminology applies

For Episode 5: From Frontier Search To Portfolio Thinking, the backtesting workflow should treat Point-in-time contracts, Quote-aware fills, Reject reasons, Replay artifact, Cache key, and Signal timestamp as operational state rather than glossary decoration. That framing keeps the research claim causal: the strategy can only select instruments, prices, and labels that existed at the decision time.

A developer implementing this research idea should persist Look-ahead leakage, Walk-forward test, Slippage model, Same-bar fill, Promotion gate, and Options data API beside the result, instead of leaving those words in a term card. It also turns attractive performance into an auditable record where fills, skips, thresholds, and replay inputs can be challenged independently.

The review artifact for Episode 5: From Frontier Search To Portfolio Thinking becomes more useful when OPRA-originating data, OCC option symbol, Bid/ask spread, Midpoint, Quote/trade condition, and Quote vs trade semantics appear in the same body of evidence as the selected rows. When a result is promoted, these fields should appear in the run manifest, rather than a prose summary or final equity curve.

In production notes for this backtesting workflow, REST snapshot, WebSocket stream, Entitlement gate, Quote freshness, Timestamp semantics, and Pagination cursor define the checks that decide whether the workflow is reproducible. The result is a backtest that can be rerun, compared across threshold families, and rejected when the evidence is not strong enough.

For Episode 5: From Frontier Search To Portfolio Thinking, the practical acceptance test is simple: another developer should be able to read the body, identify the exact inputs, reproduce the request sequence, and explain the accepted and rejected rows without relying on the bottom terminology grid. If a phrase appears in the page vocabulary, it should correspond to a stored field, a validation check, a replay step, or an implementation decision in the backtesting workflow.

This is also the reason the article should not measure success only by the final chart, table, or headline metric. The better standard is whether the data path, timing model, entitlement state, and evidence trail survive review. When those pieces are written directly into the body, the terminology becomes part of the workflow readers can implement.

Portfolio candidates still start as rows

Episode 5 moved the research from isolated winners toward portfolio thinking, but the unit of evidence was still a market-data row. Each candidate sleeve should keep its signal timestamp, selected contract, quote window, trade window, cache key, and reject history. Portfolio construction can compare sleeves only after the underlying artifacts agree on schema version, contract symbology, entitlement state, and fill policy.

That requirement matters when two sleeves look complementary. If one sleeve uses clean NBBO quotes and another relies on aggregate bars without top-of-book evidence, the correlation estimate is not comparing like with like. If one branch counts no-bid exits and another silently drops them, the drawdown comparison is also distorted. Portfolio review should therefore include spread percent, stale quote rejects, quote condition filters, trade condition filters, rate-limit assumptions, and the exact replay manifest.

The practical standard is simple: a sleeve should not enter a portfolio because its curve looks different. It should enter only after the evidence record proves that its difference survives the same contract selection, quote-aware fill model, and operational rejects used by the rest of the book.