# Optimizer under constraints -- the most return inside a volatility budget
#
# Idea: fit the weights on part of the history and judge them on the rest,
#   under a volatility cap and per-line bounds. Three things separate this
#   from a plain `#meta optimize:max-sharpe` line, and each answers a way a
#   plain optimizer misleads:
#
# 1. THE OBJECTIVE IS THE REAL QUESTION. Not "the best Sharpe" but "the most
#    return that stays under 12% volatility". Sharpe is also a poor proxy
#    here specifically: this toolkit computes it at a ZERO risk-free rate, so
#    a cash-like sleeve buys ratio for free. A volatility cap cannot be gamed
#    that way.
# 2. BOUNDS PER LINE. An unconstrained optimum is a corner solution, and a
#    corner fitted on one window is the least durable thing an optimizer
#    produces. The floor (min-weight:5) keeps every engine in the book; the
#    ranges are the ones a holder could defend for reasons the backtest
#    cannot see.
# 3. A HELD-OUT WINDOW. train:..2015 fits the weights on 1996-2015 only,
#    while the report measures them over the WHOLE period, so the column you
#    read is out of sample. The note prints both halves, and the gap between
#    them is the honest part of the exercise: it is routinely wide.
#
# Read the two columns as "the weights as written" against "what the search
# would have chosen", and the note under them as what that choice was worth
# once the data ran out.
#
# Set the cap below what these five lines can reach together (try max-vol:9)
# and the note says so instead of returning a plausible-looking answer: an
# infeasible problem has no optimum, only a least-bad violation.
#meta rebalance:90
#meta sim:on
#meta optimize:max-return,max-vol:12,min-weight:5,bounds:NTSG:10-35,bounds:GOLD:5-25,train:..2015

30 NTSG     # equity + duration in one line (90/60)
15 ZPRV     # US small-cap value
25 DBMFE    # managed-futures trend
20 GOLD     # gold
10 IDTL     # long-duration Treasuries