[joseedson18jc]

Fixes #

Describe the changes you have made in this PR -

Demo/Screenshot for feature changes and bug fixes -

---

Code Understanding and AI Usage

Did you use AI assistance (ChatGPT, Claude, Copilot, etc.) to write any part of this code?

• No, I wrote all the code myself
• Yes, I used AI assistance (continue below)

If you used AI assistance:

• I have reviewed every single line of the AI-generated code
• I can explain the purpose and logic of each function/component I added
• I have tested edge cases and understand how the code handles them
• I have modified the AI output to follow this project's coding standards and conventions

Explain your implementation approach:

---

Checklist before requesting a review

• I have added proper PR title and linked to the issue
• I have performed a self-review of my code
• I can explain the purpose of every function, class, and logic block I added
• I understand why my changes work and have tested them thoroughly
• I have considered potential edge cases and how my code handles them
• If it is a core feature, I have added thorough tests
• My code follows the project's style guidelines and conventions

---

Note: Please check Allow edits from maintainers if you would like us to assist in the PR.

[github-actions]

Greptile code review

This repo uses Greptile for automated review. Before merge, aim for Confidence Score: 5/5 with zero unresolved review threads — see CONTRIBUTING.md.

Run a review — add a PR comment with:

@greptile review

Give it ~5-10 minutes (sometimes longer) for results, then fix feedback and re-trigger until you reach Confidence Score: 5/5.

Optional: automate with the greploop skill.

[greptile-apps]

Confidence Score: 0/5

[Medium impact] The PR adds local crypto research and simulated paper-trading workflows.

Do not merge yet: paper trading and universe calculations can report results that do not match the trades or members described.

Findings

1.  Paper fills happen too early ▶
2.  Old paper decisions return ▶
3.  Unfinished bars stay saved ▶
4.  Relisted coin history disappears ▶
5.  Portfolio returns are wrong ▶
6.  Dead coins count as survivors ▶
7.  Books hold different coins ▶
8.  Unqualified strategies still win ▶
9.  Missing venue stops the report ▶
10.  Offline reports still need network ▶
11.  No casualties stops the report ▶
12.  Walk-forward skips trading fees ▶
13.  Charts show the wrong dates ▶
14.  Background trader never starts ▶

Summary

Adds multi-year strategy studies, whole-grid books, cost and execution comparisons, cross-venue checks, a monthly coin universe, and paper trading.

• Researchers can run a BTC strategy sweep and get charts and a report.
• The history runner compares strategy books across years and reports their results.
• Historical strategy comparisons now cover more coins, exchanges, and delisted pairs.
• A paper trader records new strategy decisions in a saved ledger.

Previous unnumbered threads: joseedson18jc acknowledged the incomplete-bar finding with “ok.” The no-trade chart and shared-constants threads were manually resolved without explanation. The documented Python minimum is now corrected.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Exchanges[Exchange bars] --> Store[ArcticDB]
  Store --> Strategies[Strategy grids]
  Strategies --> Books[Combined exposures]
  Books --> Returns[Execution and costs]
  Returns --> Studies[Historical comparisons]
  Studies --> Reports[Reports and charts]
  Strategies --> Decisions[Paper decisions]
  Decisions --> Ledger[Saved ledger]
  Ledger --> PaperReturns[Paper returns]

Loading

_{Reviews (2) · Last reviewed commit: "Merge pull request #3 from joseedson18jc..." · Reviewed by Greptile}

[greptile-apps]

Walk-forward skips trading fees

walk_forward copies returns from an already-running strategy without paying to enter its position. This skips the initial entry fee and any fee needed when a later pick changes the held position. The report and chart therefore overstate returns after fees. Track the position actually held across windows and charge for changes when each pick takes over.

[joseedson18jc]

gggg

[joseedson18jc]

apptoved

[joseedson18jc]

approved

[github-actions]

Greptile code review

This repo uses Greptile for automated review. Before merge, aim for Confidence Score: 5/5 with zero unresolved review threads — see CONTRIBUTING.md.

Run a review — add a PR comment with:

@greptile review

Give it ~5-10 minutes (sometimes longer) for results, then fix feedback and re-trigger until you reach Confidence Score: 5/5.

Optional: automate with the greploop skill.

[joseedson18jc]

But the right correction is clear:

In quant_desk/desk/strategy.py, the entry logic should allow the first valid non-zero EMA signal to open a position when the strategy is flat.
The condition should not require a prior non-zero s_prev_sign before the first trade.
Also, generated output under quant_desk/output/ should be ignored, not committed.
The corrected logic should be equivalent to:

If s_position == 0 and sign != 0 after warmup, open at sign
Else if already in a position and the sign flips, close/reverse with fee
Keep the fee logic tied to the position change magnitude

[greptile-apps]

Paper fills happen too early

effective_exposure backdates ordinary paper decisions to the bar close. The loop normally runs 90 seconds afterward, but late=False makes next_open pricing use the opening price from 90 seconds earlier. A run made hours into a 4h bar gets the same treatment. Paper returns therefore include moves before the bot made its decision.

Use the actual decision time and a price available afterward.

[greptile-apps]

Old paper decisions return

Catch-up rows can replace a newer paper decision later. After reconnecting at 16:02, older late rows map to 20:00 while the newest 16:00 row maps to 16:00. If the next run is missed, rebuilding performance makes the old catch-up exposure take over at 20:00, even though the bot was offline.

Discard superseded catch-up decisions or ensure older bars cannot replace newer decisions.

[greptile-apps]

Unfinished bars stay saved

For supported 1h, 8h, and 1d books, fetch_recent can return an unfinished bar. An 8h book fetched at 04:02, for example, gets a partial bar stamped 08:00. record saves that decision, excludes the same timestamp on later runs, and keeps the partial prices through combine_first. The completed bar never replaces it.

Drop unfinished resampled bars before deciding or storing them.

[greptile-apps]

Relisted coin history disappears

_segment_4h applies a segment number from the full daily history to a shortened 4h download. If a ticker's second incarnation first joins the universe more than 200 days after the first incarnation ended, the download contains only one segment. The _2 lookup then returns empty and drops that member from the study.

Preserve each instrument's start and end dates instead of renumbering segments after shortening history.

[greptile-apps]

Portfolio returns are wrong

portfolio_daily compounds each already-weighted coin separately and then adds the daily returns. That is not the return of the combined portfolio. With two equally weighted coins each gaining 10% on two bars, this gives 20.5% instead of 21%. The report then calculates growth and drawdowns from the wrong series.

Add the weighted returns at each bar first, then compound the combined series into days.

[greptile-apps]

Missing venue stops the report

holdouts_chart requires Bitstamp even though venue ingestion explicitly skips failed downloads. If Bitstamp fails or is absent from the cached data, this lookup raises KeyError after the other studies finish, before the final report is written.

Skip the unavailable panel or check required venues before starting the study.

[greptile-apps]

Offline reports still need network

--no-fetch still downloads missing universe members and always runs live paper fetching. Even with every study dataset cached, an offline run reaches paper.run_once() and fails before writing the final report.

Keep network work behind the fetch flag and make the paper step separately optional.

[greptile-apps]

No casualties stops the report

casualties crashes when no eligible member has stopped trading. With no rows, the returned DataFrame has no buy_and_hold_while_member column, so sorting raises KeyError and prevents the report from finishing.

Return an empty frame with the expected columns, matching the report's existing support for an empty casualties table.

[greptile-apps]

Charts show the wrong dates

The heatmap title says January 2022 through November 2026, but the plotted Sharpe values cover the whole history starting in August 2017. The fixed end date also becomes wrong on later runs. Readers therefore see the wrong period for these results.

Build the label from the study's actual dates.

[greptile-apps]

Background trader never starts

The documented background command fails on a fresh checkout because output/paper/ does not exist. The shell opens the redirected log before Python starts, so run_once never gets a chance to create the directory.

Add mkdir -p output/paper before this command so a new user can start the trader.