URGENT: remove two fabricated 4C input keys from the released knowledge

[alhermann]

One file, 21 insertions, 11 deletions. Deliberately minimal so it can land ahead of the large consolidation in #50.

What main serves today

"needs":    [... DYN_VISCOSITY, BULK_MODULUS, SOUNDSPEED]
"pitfalls": ["SOUNDSPEED too low → fluid compresses unrealistically;
              rule of thumb c >= 10 * v_max",
             "SMOOTHING_LENGTH too small → spurious tensile-instability
              voids; rule of thumb h ~ 1.3 * particle spacing"]

Measured against the 4C checkout:

key	in 4C source	in 4C's 2171 upstream decks
SOUNDSPEED	0 files	0 decks
SMOOTHING_LENGTH	0 files	0 decks
MAT_ParticleSPHFluid (the real material)	2 files	60 decks

Two invented keys, listed beside two real ones, one carrying a numeric tuning rule — a rule about a parameter that does not exist. A user who follows it writes a MATERIALS block 4C refuses to parse.

And the diagnostic they get, Failed to match specification in section 'MATERIALS', is the same message a mis-cased real key produces. So the error does not tell them which mistake they made. That is why the corrected entry points at 4C -p, which dumps the accepted grammar from the binary and is the only reliable way to tell a typo from a fabrication.

Why this is separate from #50

#50 fixes this among ~2,900 other files and cannot be reviewed quickly — Copilot declined it for exceeding 300 files. Meanwhile these fabrications are live in the DOI-badged release (10.5281/zenodo.20543501), reachable by anyone who clones or cites it, and have been for the whole time the fix sat on a branch.

Where it was hiding, which is the interesting part

The text lives in the reference-stub catalog, not in a get_knowledge() row. A knowledge-level scan does not reach it — which is why it survived earlier audits — but prepare_simulation serves it, so a user does. Any future audit of served content needs to cover this path.

Verified

• fourc backend still loads, 48 physics rows intact.
• The fabricated tuning rule is gone from the source; the correction is present.
• Every factual claim in the replacement text was checked against /home/alexander/4C directly.

Still outstanding on main after this

src/backends/kratos/generators/dem.py writes PARTICLE_FRICTION into every generated DEM deck. That key is absent even from the full 28-application Kratos build; the real keys are STATIC_FRICTION/DYNAMIC_FRICTION, per contact pair. It is not fixed by #50 either — #50 documents it in a warning while the template still emits it, the same contradiction this PR fixes for SOUNDSPEED. A proper generator rewrite is in progress; it needs a schema change, not a text edit, which is why it is not bundled here.

[Copilot]

Pull request overview

This PR corrects the 4C backend’s reference-stub catalog entry for the SPH dam-break example by removing two fabricated input keys and replacing the guidance with verifiable, solver-grounded instructions. This aligns the served “prepare_simulation” help text with 4C’s actual accepted MATERIALS grammar, reducing the risk that users generate non-parsable decks.

Changes:

• Replaces the SPH dam-break needs entry to reference the real 4C SPH fluid material (MAT_ParticleSPHFluid) instead of the previously described fabricated parameter set.
• Removes the fabricated SOUNDSPEED / SMOOTHING_LENGTH pitfalls (including the tuning rule) and replaces them with a warning explaining they are not valid 4C keys and how to verify accepted names via 4C -p.