Functions
	_script ()
Any	_get (Dict[str, Any] d, str path)
Dict[str, Any]	baseline ()
Dict[str, Any]	burn (baseline)
	test_embedded_config_round_trips (baseline)
	test_layerx_defaults_unchanged (baseline)
	test_inputs_unchanged (baseline, burn)
	test_he_pad_matches_baseline (baseline, burn, key)

Variables
	ROOT = Path(__file__).resolve().parents[1]
	BASELINE = Path(os.environ.get("LAYERX_GOLDEN_BASELINE") or (ROOT / "docs" / "layerx" / "baseline-2026-10-07.json"))
str	CASE = "he_pad"
list	pytestmark
dict	TOLERANCES

Detailed Description

LE4 golden burn: Layer X's pad helium baseline must not move.

The reference is ``docs/layerx/baseline-2026-10-07.json``, written by
``scripts/layerx_baseline.py`` (case ``he_pad``: LE4 on the ``copv_study_he`` hot-fire drawing, on
the pad, erosion replay on, every other ``LayerXSettings`` field at its default). The overnight rule
is that a change moving the LE4 baseline by more than 1 % is reported; this test is what notices.

Slow: one helium burn with the replay loop is ~35-100 s depending on load, so it is skipped unless
``LAYERX_GOLDEN=1`` (the suite registers no ``slow`` marker). ``LAYERX_GOLDEN_BASELINE`` points it at
another baseline file.

What it holds fixed, and what it does not
-----------------------------------------
* **The design is the baseline's own.** The burn uses the config embedded in the baseline JSON,
not the live ``configs/ethalox_6800N.yaml``, so editing the design does not read as a physics
change. ``test_embedded_config_round_trips`` checks every field of it survives the schema
unchanged (a validator that rewrites a stored value *is* an input change).
* **The drawing, the CEA table and the DAQ tables are the checkout's**, and
``test_inputs_unchanged`` compares their hashes with the baseline's first. When one of them has
changed the figures are expected to move: re-run the script, review, and report the move.
* **Everything else is code**: lib/feedtwin, engine/layerx, the engine card, the replay.

Tolerances
----------
Measured: the burn is bitwise reproducible across processes on one machine (three runs, every
figure identical to the last digit), so on the baseline's own platform any non-zero move is a
change. The bands are wider than that on purpose:

* **0.5 % on the integral performance figures** (impulse, burn time, thrust, Pc, O/F): half the
1 % reporting threshold, so a change that must be reported cannot pass, with room for platform
drift (CI runs other numpy/CoolProp versions; that drift has not been measured here).
* **0.3 % on Isp**: Isp is a ratio of two integrals that move together; a 0.3 % move in delivered
Isp is already a material engine change (0.7 s).
* **1 % on the bottle** (pressure and gas used at burnout): the bottle integrates the whole burn's
gas demand and is a feed-side quantity the 1 % rule reads directly.
* **Absolute bands** where a relative one means nothing: tank pressures 3 psi (0.5 % of the
~618 psia peak), injector stiffness 0.003 (1.2 psi of the ~136 psi drop at a ~400 psia chamber,
~0.9 % of the value), the other tank's residual 0.02 kg (what an O/F move of ~0.5 % strands),
the time of the lowest chug margin one 50 ms step.
* **2 % on throat growth and recession**: erosion feeds thrust weakly (a 2 % change in a 4 % area
growth is ~0.08 % of area); tighter would flag erosion-model edits that cannot move a reported
figure by 1 %.
* **Exact** on the discrete outcomes: which tank runs dry, converged, no failed steps, no step
outside the engine card.

Checked against changes made in-process, not in the code: halving the regulator's supply-pressure
effect fails 8 of these (burn time +1.45 %, mean thrust -1.73 %, Pc -1.65 %, bottle +9.5 %, tank
peaks -24 psi) while total impulse moves only -0.30 %: the load is fixed, so impulse alone is a weak
witness to a feed-system change. Loosening the network solve from 1e-6 to 1e-4 passes all of them
(impulse -0.018 %, stiffness -0.15 % of its value). Doubling or shortening the step is a scheme change
and is held out by burning at the baseline's own ``dt`` (0.1 s: impulse +0.08 %; 0.02 s: -0.06 %,
lowest chug margin -0.58 % because its minimum is the first firing sample).

Function Documentation

◆ _get()

Any test_layerx_golden._get	(	Dict[str, Any]	d,
		str	path )

protected

Definition at line 113 of file test_layerx_golden.py.

Referenced by test_he_pad_matches_baseline().

◆ _script()

test_layerx_golden._script ( )

protected

Definition at line 106 of file test_layerx_golden.py.

Referenced by burn(), and test_embedded_config_round_trips().

◆ baseline()

Dict[str, Any] test_layerx_golden.baseline ( )

Definition at line 120 of file test_layerx_golden.py.

◆ burn()

Dict[str, Any] test_layerx_golden.burn ( baseline )

Definition at line 130 of file test_layerx_golden.py.

References _script().

◆ test_embedded_config_round_trips()

test_layerx_golden.test_embedded_config_round_trips ( baseline )

Every stored field of the baseline's design comes back out of the schema as it went in.

Definition at line 138 of file test_layerx_golden.py.

References _script().

◆ test_he_pad_matches_baseline()

test_layerx_golden.test_he_pad_matches_baseline	(	baseline,
		burn,
		key )

Definition at line 189 of file test_layerx_golden.py.

References _get().

◆ test_inputs_unchanged()

test_layerx_golden.test_inputs_unchanged	(		baseline,
			burn )

The drawing, CEA table and DAQ tables are the ones the baseline burned. If this fails, the
figures below are expected to move: re-baseline and report the move, do not widen a band.

Definition at line 177 of file test_layerx_golden.py.

◆ test_layerx_defaults_unchanged()

test_layerx_golden.test_layerx_defaults_unchanged ( baseline )

The baseline is "today's defaults". A changed default is a changed answer for every run the
tab starts, so it is reported like a physics change; the burn below still uses the baseline's
own settings, so the figures isolate the code.

Definition at line 162 of file test_layerx_golden.py.

Variable Documentation

◆ BASELINE

test_layerx_golden.BASELINE = Path(os.environ.get("LAYERX_GOLDEN_BASELINE") or (ROOT / "docs" / "layerx" / "baseline-2026-10-07.json"))

Definition at line 67 of file test_layerx_golden.py.

◆ CASE

str test_layerx_golden.CASE = "he_pad"

Definition at line 68 of file test_layerx_golden.py.

◆ pytestmark

list test_layerx_golden.pytestmark

Initial value:

=  [
    pytest.mark.skipif(os.environ.get("LAYERX_GOLDEN") != "1",
                       reason="slow LE4 golden burn: set LAYERX_GOLDEN=1 to run it"),
]

Definition at line 70 of file test_layerx_golden.py.

◆ ROOT

test_layerx_golden.ROOT = Path(__file__).resolve().parents[1]

Definition at line 66 of file test_layerx_golden.py.

◆ TOLERANCES

dict test_layerx_golden.TOLERANCES