Phase C · PRD §2.2, §7. Blocked by #577. A dated decision, not a renewal that happens by default.
Why this issue exists
keel's binding constraint is cost. The README's honest result is that no shipped rule family is net-positive at the taker fee actually paid, and that the viable parameter/fee intersection is empty. A paid analysis subscription does not move that number by a basis point. It is an operating expense against a system that is not yet profitable.
Nor can it pay for itself the way a reader might assume: it cannot improve entry timing, exits, or sizing without becoming a trading signal, which PRD §5 forbids outright. Its only sanctioned effect is to widen the candidate pool a human then attests.
This buys evidence and reach, not edge. If it has not improved the breadth or quality of attested candidates, it should be cancelled — and the default outcome of an unexamined subscription is that it renews forever.
What to measure
- candidates proposed
- candidates that passed the gate
- candidates a human actually attested — the only one of the three that matters
Compare against the same period's manual scouting.
Publish it whichever way it lands
docs/experiments/, in the same shape as the cost-fidelity restatements — including "this did not improve on manual scouting", if that is what the numbers say. A negative result published is the same asset as a positive one; a negative result unpublished is a subscription that renews.
The subscription is prepaid in fixed periods precisely so this decision has a natural date. Set that date when Desk is purchased (#570) and put it in this issue's title.
Done when
The three counts are measured over a full prepaid period, compared against manual scouting, published in docs/experiments/, and the subscription is either renewed on the evidence or cancelled.
Phase C · PRD §2.2, §7. Blocked by #577. A dated decision, not a renewal that happens by default.
Why this issue exists
keel's binding constraint is cost. The README's honest result is that no shipped rule family is net-positive at the taker fee actually paid, and that the viable parameter/fee intersection is empty. A paid analysis subscription does not move that number by a basis point. It is an operating expense against a system that is not yet profitable.
Nor can it pay for itself the way a reader might assume: it cannot improve entry timing, exits, or sizing without becoming a trading signal, which PRD §5 forbids outright. Its only sanctioned effect is to widen the candidate pool a human then attests.
This buys evidence and reach, not edge. If it has not improved the breadth or quality of attested candidates, it should be cancelled — and the default outcome of an unexamined subscription is that it renews forever.
What to measure
Compare against the same period's manual scouting.
Publish it whichever way it lands
docs/experiments/, in the same shape as the cost-fidelity restatements — including "this did not improve on manual scouting", if that is what the numbers say. A negative result published is the same asset as a positive one; a negative result unpublished is a subscription that renews.The subscription is prepaid in fixed periods precisely so this decision has a natural date. Set that date when Desk is purchased (#570) and put it in this issue's title.
Done when
The three counts are measured over a full prepaid period, compared against manual scouting, published in
docs/experiments/, and the subscription is either renewed on the evidence or cancelled.