FACEHARD — native Python port, verbatim clone + CLI

A faithful Python reimplementation of FACEHARD 8.0, Nathan Okun's face-hardened naval armour penetration program (QuickBASIC, 2018). It reproduces both the program's numbers and its exact text output, validated against the generated BASIC working oracle (fh_oracle) by differential fuzzing. Intentional oracle corrections are documented separately.

Two ways to use it: a modern CLI (facehard run/pen/list) and a byte-faithful clone of the original interactive program (facehard emulate). Stdlib-only, pipx-installable.

Versions. The package is versioned independently of the model: this is package 1.0.0, reproducing Okun's FACEHARD 8.0 (8 August 2018). facehard.MODEL_VERSION and facehard --version both report the model version alongside the package version.

Source. Okun's original programs, papers and armour data are published at NavWeaps — The Nathan Okun Collection.

Install

pipx install facehard
facehard --version

CLI

facehard                     interactive wizard (modern menus)
facehard emulate             the ORIGINAL FACEHARD interactive session, verbatim
facehard run   [options]     one impact; --output narrative|limits|plug|metrics|json|all
facehard pen   [options]     penetration thickness at a striking velocity
facehard list  armors|nations|projectiles [--nation N]

run/pen scenario flags: --armor 1-25 --nation 1-8 --proj N -d/--diameter -w/--weight -b/--body-weight -v/--velocity -o/--obliquity; backing --wood --cement --metal --metal-type --metal-plates; nose loss --remove {none,cap,windscreen,caphead} --windscreen-wt --caphead-wt. run also takes -t/--thickness (the plate to shoot); pen omits it because it computes the thickness defeated.

Library API

from facehard import calc, results, render_results, penetration, names

calc(armor=14, nation=1, proj=18, TA=16, D=16, WT=2700, WB=2048, OB=30).vltru
results(armor=14, nation=1, proj=18, TA=12, D=16, WT=2700, WB=2048, OB=30, VS=2200)
render_results(armor=14, nation=1, proj=18, TA=12, D=16, WT=2700, WB=2048,
               OB=30, VS=2200, show_limits=True)          # verbatim BASIC text
penetration(armor=14, nation=1, proj=18, D=16, WT=2700, WB=2048, V=2500, OB=0)  # 29.5 in

Every numeric menu choice also has a named IntEnum — drop-in for the plain numbers, which stay valid everywhere:

from facehard import penetration, Armor, Nation, USProjectile, projectiles

Nation.UK == 2                                            # original menu numbers, unchanged
penetration(armor=Armor.BRITISH_CA, nation=Nation.USA,
            proj=USProjectile.APC_6IN_MK35_16IN_MK8,      # 16-in Mk 8 (IOWA)
            D=16, WT=2700, WB=2048, V=2500, OB=0)

projectiles(Nation.JAPAN).APC_TYPE91                      # per-nation projectile menus
Armor.JAPANESE_VH.menu_text                               # the original menu line

Nation, Armor, BackingMetal (the metal_type parameter), and one projectile enum per nation (USProjectile, BritishProjectile, GermanProjectile, FrenchProjectile, ItalianProjectile, JapaneseProjectile, AustroHungarianProjectile, RussianProjectile). The short names summarise menu lines that often cover several marks — member.menu_text is authoritative.

All of calc/results/render_results accept backing, nose-covering loss (remove, windscreen_wt, caphead_wt), and manual armour overrides (armor_overrides={"Q": 0.8, "UB": 60, ...} for UB/Q/QDAM/CARTWL/CMPND/THNCHL/SOFTSHAT/THKTHN).

Coverage — the whole program

Not reproduced: nothing functional — only the DOS graphics-mode screen clears (cosmetic) differ.

Fidelity — how close to the BASIC?

precision_audit.py (1500 fuzzed examples, deterministic):

• Zero velocity deviations across 6,818 ballistic-limit and post-impact velocity comparisons.
• Exit angles agree exactly in the retained 10,000-case differential campaign.
• The remaining differences are BASIC display rounding of plug weights to one decimal place (maximum 0.05 lb).

The generated working oracle uses double precision and contains the documented post-impact-limit correction in ORACLE_PATCHES.md.

Verbatim-text fuzzers confirm the same at the string level: fuzz_narrative.py 99.97% of result lines, fuzz_emulate.py 99.98% of full-session lines, fuzz_override.py 100% of override sessions — residual diffs are those same sub-unit display roundings.

Layout

Licence

Not open-source. The penetration mechanics and associated IP — including the formulas, decision logic, data tables, and original FACEHARD program flow and text — belong to the estate of Nathan Okun and, per NavWeaps' terms, may not be used commercially without prior written permission. The Python implementation and its original additions are Copyright © 2026 Adrian Beale and are released under the PolyForm Noncommercial License 1.0.0. The combined distribution is for noncommercial use only; commercial use may require permission from both rights holders. See LICENSE and NOTICE.