The paper candidate keeps accumulating untouched evidence while discovery, data work, and challenger training continue independently.
New theories become isolated challengers. They cannot alter the frozen paper candidate or its confirmation window mid-test.
No research lanes are currently published.
Held, mutated, and recombined counts come from the current public tournament artifact.
latest published tournament snapshot
held candidates · 64 prior-generation inputs
mutation-nursery parents · promotion blocked
mutated candidates
same-family crossovers
candidate-level state available on Tournament
source timestamp
Interpretation: Held currently means strict survivors retained from a prior generation; a future mutation-nursery hold must be identified explicitly. Mutated counts reproduction attempts, while same-family crossovers remain separate. Exact clones receive no implied novelty. This breeding state is research workflow telemetry, not a readiness gate.
Each goal has an accountable caretaker, a bounded evidence source, and an acceptance condition. Goal progress cannot authorize live money.
Current: 0 of 5 safety drills passed.
Acceptance: All five cold-start, fail-closed data, model abstention, paper-order lifecycle, and restart-recovery drills pass with retained evidence.
Caretaker: Readiness verifier
Current: 36 of 67 tracked research tasks complete.
Acceptance: Every tracked task has cited evidence and an independent verifier disposition; completion does not change a readiness gate by itself.
Caretaker: Core research caretaker
Current: 0 completed runs and 0 retained evaluations; latest never.
Acceptance: Scheduled bounded runs produce source-bound reports reproducibly, retain failures, and remain promotion-ineligible until separate correction and confirmation.
Caretaker: ML research caretaker
Current: 17289 tournament trades are published across 200 isolated candidates; this is not the frozen paper-candidate gate.
Acceptance: The selected frozen paper candidate accumulates at least 100 cost-reconciled closed trades and 90 consecutive paper days; zero-trade evidence cannot advance selection.
Caretaker: Tournament evidence caretaker
Current: 0 of 8 readiness gates passed; live activation remains blocked.
Acceptance: All eight gates pass and a separate human reviewer signs the final live-readiness decision.
Caretaker: Human readiness owner
The bot must demonstrate safe startup, abstention, a reconciled paper order, and recovery behavior before live trading can even be considered.
No working-start checks published.
Passing the working start is necessary, not sufficient. Real money remains unavailable until the longer paper record and human review are complete.
No later readiness checks published.
Positive expectancy after measured costs · profit factor at least 1.20 · drawdown at most 10% · zero safety violations · positive untouched evidence across three regimes · rule-only, no-martingale, and buy/hold baselines.
Readiness status is stored in the readiness_checks table of the local research database. The system transitions checks from pending to passed when automated validation drills or manual audit signatures verify compliance:
| Check ID | Database Field / Contract | Validation Method | Sanitization Boundary |
|---|---|---|---|
cold_start |
working_start = 1 |
Dry-run Freqtrade config syntax verification; zero-trade start test. | Excludes host IP, private paths, and API keys. |
fail_closed_data |
working_start = 1 |
Auto-blocks entry signals when OHLCV streams exceed max staleness. | Limits display to public candle freshness status. |
model_start |
working_start = 1 |
LightGBM model fits score checking; rejects out-of-envelope inputs. | Hides private model weights and feature definitions. |
paper_order_lifecycle |
working_start = 1 |
Reconciles the 0.20%-per-side configured fee and observed fill prices; the tournament separately models 0.10% slippage per side. | Publishes aggregate simulated performance only. |
restart_recovery |
working_start = 1 |
Simulates crash drills; checks state file lock and recovery files. | Hides OS process IDs and filesystem locks. |
paper_90_days |
working_start = 0 |
Tracks consecutive daily database ticks for at least 90 calendar days. | Displays total duration, not raw timestamps. |
paper_100_trades |
working_start = 0 |
Reconciles Freqtrade trades table; minimum 100 closed positions. | Excludes trade prices, volumes, and fill details. |
human_review |
working_start = 0 |
Manual signature verified against allowlisted PGP public key. | Publishes verification disposition only, hides key fingerprints. |
Readiness status is diagnostic, not a forecast. The live-money activation lane remains locked until all 8 gates change to passed and human authorization is signed.
