Related: Project overview · Numbers ledger · OOS dashboard
What the case study abstracts, what it keeps, and how this public repository enforces the redaction.
| Detail | How it appears publicly | Why |
|---|---|---|
| Exact indicator parameters (MA spans, channel lengths, oscillator levels/thresholds) | “moving-average crosses”, “channel breakouts”, “oscillator and band signals” — family level only | Parameter sets are the replicable part of the work product |
| The full 24-config zoo listing | “a fixed zoo of 24 simple configurations, four families” | Same |
| Weekly winner identities | Family-level ribbon only (trend / breakout / mean-reversion / cash / mixed); no config names in any chart label | Same |
| Current live stance (which configs, what size, what direction) | Not in the case study at all | Operational |
| Tie-breaking rule details in selection | Described as “how exact score ties break”, no implementation detail | Minor, but parameter-adjacent |
| Live dashboard URL / port / machine details | Not mentioned | Operational security |
| Detail | Rationale |
|---|---|
| Structural choices: weekly cadence, 30-day training window, 4h bars, top-3 averaging, 24-candidate pool size | These are the case study’s subject matter; without them the graveyard is meaningless. They are method, not parameters. |
| Sizing: 50% vol target, 1.5× cap (and the 35%/1× reference variant) | The risk-dial discussion is a core honesty point; constants are not the secret sauce (the operator brief allowed naming the vol target). |
| Cost model (0.15%/side) and funding methodology | Required for any reader to assess realism. |
| All aggregate results, drawdowns, Sharpe estimates and their errors | The point of the document. |
None exist in the repo: the system uses only public, keyless APIs. The secret scan covers api-key patterns, private keys, wallet addresses and .env content across all published files. Scan result: no findings in README.md or docs/. The live trade record and the fit-watchdog status file live only in the private repository and are absent from this mirror.
The full strategy source — which defines the 24-config zoo and every indicator parameter the case study abstracts — lives in a separate, private research repository. The redaction is therefore enforced structurally: this public repository contains only the case study (README.md, docs/assets, docs/interactive, and these two docs). None of the strategy source, parameters, or config listings are present here to leak.
This is the mirror model: a standalone public case study, with the code repository kept private. Casual and determined readers alike see the same family-level abstraction, because the parameters simply are not in this repo. The case study text stays honest either way.