CVN-N001-EK — Tradability program charter (Phase 0 draft)¶
DRAFT — UNLOCKED. This charter instantiates ADR-0102 for one program. It contains no quantitative value: every value is a placeholder with a named source Story and lock authority. It is not locked — locking is S03 (D3) under joint sign-off (operator + methodology reviewer + risk owner), with role separation. Filling values is S02 (D2).
1. Program thesis (named — a hypothesis, not a claim)¶
Candidate thesis under test: whether a long-only directional ML entry edge on
defi_top5, with an ATR-H4 triple-barrier label and a GBT model class, can be shown economically tradable at conservative P90 cost and target capacity.
The charter tests this thesis; it does not assert it. A switch to a different hypothesis family (e.g. cross-sectional / market-neutral) is a separate Epic (ADR-0102 Invariant 2).
2. Placeholders (no value in S01)¶
| Placeholder | Symbol / field | Source Story | Lock authority |
|---|---|---|---|
| Conservative cost at target capacity | P90_cost |
S02 | S03 |
| Economic break-even threshold | E_econ_min |
S02 | S03 |
| Predictive→economic mapping | map(E_pred→E_econ) |
S02 | S03 |
| Minimum predictive effect | E_pred_min |
S02 | S03 |
| Minimum detectable effect / minimum universe | MDE, N_min |
S02 | S03 |
| Principal null-gate (purged/embargoed, conservatism-wins) | null_gate |
S02 | S03 |
| Family budget | B_family |
S02 | S03 |
ONE-ITERATION slot budget |
B_slot |
S02 | S03 |
| FDR / multiple-testing budget | B_fdr, alpha |
S02 | S03 |
| R1 — Sortino definition (periodicity / MAR / annualisation + floor basis) | sortino_def |
S02 | S03 |
| Minimum trade count + power thresholds | N_trades_min, power_min |
S02 | S03 |
| Calibration method + validation metric | calib, brier_max |
S02 | S03 |
| Stress / regime-sensitivity set | stress_set |
S02 | S03 |
| Execution-realism (slippage, fees, funding, venue liquidity) | exec_realism |
S02 | S03 |
| Position-sizing guardrails + concentration limits | sizing_limits |
S08 | S03 |
economic_action_intent controlled vocabulary |
action_intent_vocab |
S02 | S03 |
No cell above carries a number in S01. Any value committed here before S03 is a lock violation.
3. Closed list of structural coordinates¶
Changing any of these creates a new tuple (ADR-0102 Invariant 5). This list is closed; additions are a charter amendment (re-lock under S03):
label_family · feature_family · model_class · universe · horizon · null_gate · cost_model ·
split_policy · purge_embargo_policy.
Everything not on this list that is tunable (thresholds, top-k, calibration) is action policy — tunable inside a tuple only under a pre-registered, budgeted search (Invariant 5), never on the final holdout.
4. Registered-tuple schema (this program)¶
The tuple fields are defined in architecture §4.1;
this program additionally fixes economic_action_intent as a controlled-vocabulary field (placeholder §2) to
make material-equivalence comparison objective.
5. Material-equivalence rule (killed-tuple gate)¶
Per architecture §4.2: a new tuple is
materially equivalent to a killed one if it preserves the same program_thesis, label_family,
feature_family, model_class, and economic_action_intent, regardless of cosmetic changes. Disputes →
methodology reviewer + risk owner before registration. Worked examples are a deliverable of the ADR
follow-up (committee recommendation).
6. Lock procedure (S03 — not performed in S01)¶
- S02 fills every §2 placeholder with a signed, versioned derivation (notebook / MLflow run) — in
particular the
E_econ_min/E_pred_minderivation (committee recommendation). - Joint sign-off: operator + methodology reviewer + risk owner (blocking veto, architecture §7).
- Freeze with immutability (hash / write-once). Role separation: locker ≠ run launcher.
- Any later change = a recorded re-lock (the prior locked version is retained).
7. Status¶
DRAFT — UNLOCKED. Next: S02 derives values; S03 locks. S04 is the first Story permitted to run a
Phase-2 predictivity test under this charter.