margin_long/margin_short liquidation appears to be gated on percent_of_equity sizing only — default_qty_type=fixed strategies never get margin-called, even at 34x initial capital in losses

[h1rot]

The observation

Running the same etpl-1.0 strategy template (initial_capital=100000, margin_long=margin_short=5.0,
default_qty_type=strategy.fixed, market-only entries) through PineForge engine0.12.2-codegen0.10.1
and through PyneCore v6.7.0 on identical OHLCV input produces wildly different trade counts and P&L:

	PyneCore	PineForge
trades	1,683	22,426
net PnL	-92,048.49 (≈ -92% of capital)	-3,411,226.05 (≈ -34x capital)
last trade	2021-03-05 (then silent for 5.3 years / 82% of the data range)	continues to the data end

PyneCore's trade CSV shows explicit "Margin call" exit labels once equity approaches the
~91–92% drawdown mark, after which it stops opening new positions in any meaningful size.
PineForge shows no equivalent — losses run to 34x the initial capital, which is not something
a real margin account could ever reach.

Root cause (as far as I can tell from the shipped image)

include/pineforge/engine.hpp:436-461 documents the entry-admission margin re-check as valid
only under the invariant percent_of_equity sizing && pct<=100 && margin<=100 && sizing_equity>0.
Outside that — including plain default_qty_type=fixed, which is what etpl-1.0 uses for
cross-engine parity testing — the re-check is described as out of scope and orders are admitted
regardless of equity.

I confirmed this experimentally: compiling the same strategy with margin_long=1000 (i.e. >100)
baked into the .pine source did not change trade count or qty versus margin_long=5.0 — consistent
with "margin>100 is outside the re-check range," but also meaning there does not appear to be any
configuration of margin_long/margin_short under fixed sizing that causes PineForge to actually
liquidate a losing position or stop opening new ones.

Confirmation via neutralization

Setting margin_long=margin_short=0 on both engines (which both source trees early-return out of the
margin-call path entirely) makes the two engines converge:

	PyneCore (margin=0)	PineForge (margin=0)
trades	15,944	15,827 (Δ 0.7%)
net PnL	-593,524	-583,679 (Δ 1.7%)
trades to data end	yes	yes

This isolates the divergence to the margin-call path specifically, not some other engine difference.

The question

1. Is "margin re-check applies only to percent_of_equity sizing" intentional, or is fixed-sizing
   margin enforcement simply not implemented yet?
2. I noticed the git history (via the public repo, not the shipped image) has
   2026-07-25 fix(margin): close one whole contract on every floor-zero liquidation (#136),
   which postdates the engine0.12.2 release (2026-07-23) and isn't in any tagged
   pineforge-release build yet. Does that fix address fixed-sizing accounts at all, or is it
   scoped to percent_of_equity? Is there a release timeline for it?
3. If fixed-sizing margin enforcement is out of scope by design, is there a recommended
   workaround for strategies that need both fixed position sizing and realistic margin-call
   behavior (short of switching to percent_of_equity, which changes the sizing semantics we
   need for cross-engine parity testing)?

Reproduction

Minimal repro available on request (strategy template + OHLCV slice + exact docker run invocation
used above). Happy to share if useful.