The graded entry ladder is inert — entry_technique is computed, logged, and never acted on

[eaitbrahim]

Found while verifying #227 (the is_round_number fix). Recording it because it is the precondition for ever tuning the CTS tier boundaries.

The gap

keel/strategy/indicators_cts.py documents a graded entry ladder:

tier	technique	documented behaviour
Low (total < 5)	confirm_3bar	3-bar-reversal confirmation, smaller size, wider stop
Mid	signal_candle	—
High (total >= 8)	aggressive	larger size, tighter stop

engine.py:144 computes it and attaches it to Signal.entry_technique. That value then reaches exactly two destinations:

• engine.py:360 — serialized into the signals audit payload
• agent.py:1166 — a technique= field on an agent.enter_evaluated log line

Verified exhaustively: git grep -nE "technique (==|!=|in )" -- '*.py' returns nothing outside tests. sim/portfolio_sim.py carries it as a pass-through field and never reads it. No sizing code, no stop-placement code, and no rail consults it.

So none of the documented behaviour exists. Size and stop are identical whether CTS scores 2 or 12. The module promises a risk-graded ladder and delivers a label.

Why this matters beyond tidiness

1. The docstring asserts behaviour the code does not have. Anyone reading indicators_cts.py reasonably concludes that a high CTS produces a larger position — it does not. That is the kind of claim this repo normally refuses to leave standing.
2. It is the precondition for tuning low=5, high=8. Those boundaries currently cannot be validated, because changing them changes nothing measurable. fix(analysis): scale is_round_number to the price's own magnitude #227 showed 16.2% of bars shift tier with no order consequence — the tier is free to move precisely because it is inert.
3. It affects how fix(analysis): scale is_round_number to the price's own magnitude #227 should be read. fix(analysis): scale is_round_number to the price's own magnitude #227's "no order changes" conclusion is correct, and it is correct because of this defect. If the ladder were ever wired, a CTS shift of the size fix(analysis): scale is_round_number to the price's own magnitude #227 produces (BTC/ETH/PAXG each losing ≈0.96 points, aggressive roughly halving on those assets) would become a live change in position sizing.

Direction (not a decision)

Either wire the ladder — which is a live money-sizing change and needs the full gate: trials-ledger entry, backtest evidence that graded sizing beats flat sizing, and a rails review, since sizing interacts with the per-order and per-day caps — or delete the promise from the docstring and keep entry_technique as the audit label it actually is.

The second is much cheaper and is honest. The first should not happen without evidence that grading helps; "larger size on higher confluence" is exactly the kind of plausible rule that backtests well by accident.

Related: #227 (where this surfaced), #225 (the bug that prompted it).