Skip to content

A Civilization-style strategy game about governing Israel from 15 May 1948. Real ministers, real map, every fact sourced. Unity + deterministic C# sim.

77.5 KB
Raw Blame

Unit miniatures: the art slice

The owner's target for units (2026-10-11) is the promo: promo/final/03-gameplay-1936.png and the demo trailer (promo/demo/out/demo-16x9.mp4, 0:06–0:30). It shows Civ-style miniatures standing on hexes at symbolic scale: a trio of era-correct riflemen with a selection ring, a mounted rider, and aircraft with Israeli roundels flying low with a shadow. This page records the slice that matches it, the pipeline choice behind it, the scale, and the sources for every uniform, aircraft and ship.

The slice is a showcase only. It reads nothing from the sim and changes no game state. Normal play draws the units from the sim's state and sourced map-only data with the same models: see "Units in play" below. Where units move and fight is war mode's job (#104, with the sim side in #21). The golden hashes do not move.

How to see it

  • WebGL: add units to the page URL (index.html?units). Normal play never shows it, and neither does a plain ?shot: every track's art-checkpoint metrics stay comparable (controller ruling, 2026-10-11).
  • Art checkpoint: tools/art_checkpoint.sh --units <name> adds &units to the WebGL screenshots and -units to the editor renders. With shot the aircraft holds a fixed point, so the screenshots repeat.
  • Desktop player: run with -units.
  • Editor: MapCapture.Promo renders it when given -units.
  • Aircraft close-ups: MapCapture.Aircraft renders the S-199 and the Spitfire alone, from the top, the bottom, the left side and three quarters, into -captureDir (default captures/visual/aircraft), for checking the markings against the photographs. MapCapture.Promo -units -unitsDate YYYY-MM-DD draws the units in play as on that day.
  • Frame cost: ?vprofile&units prices it as the unitshowcase toggle (median(all) - median(-unitshowcase)). The units toggle is the units in play.

Code: unity/Assets/Scripts/Map/Units/. The materials are made by UnitMaterialAssets.Create in unity/Assets/Resources/Units/. Tests: UnitShowcaseTests.

Owner rulings (2026-10-11)

  1. Units are Civ-style tokens at symbolic scale (see "Scale"). The 11 October ruling said about 2,000x; the owner's decision of 14 October supersedes it: figures at 1.8 times a block, as T29 (1,400x).
  2. Promo-quality close-zoom models come from a separate track: a cloud-GPU AI-3D pipeline (TRELLIS, MIT). The code-built miniatures here stay as the far-zoom and fallback version.

The art hook. An authored model replaces the code-built one for its unit type when a prefab exists at unity/Assets/Resources/Units/Authored/<unit type>. The unit types are the *Type constants in UnitShowcase, for example rifleman-1948-steel-helmet and spitfire-mk9e-101sqn-1948.

An authored model follows the code-built convention:

  • real metres, facing +Z;
  • origin on the ground (an aircraft: its centre of gravity);
  • its own materials;
  • an authored aircraft brings its own propeller.

The showcase places and scales it as it does the code-built model. That is the whole hook: there is no registry beyond it, and there are no level-of-detail switches yet.

What is in the slice

Piece Model Where (PromoPose.Gameplay03)
Rifle trio three 1948 IDF riflemen: bush hat with rifle held ready, Mk II steel helmet with rifle grounded, Palmach stocking cap with rifle slung on the centre of the hex nearest 34.88 E, 32.01 N (east of Tel Aviv, clear of the settlement labels), inside a white selection ring that fits within the hex
Aircraft Supermarine Spitfire LF Mk IXe of 101 Squadron, from October 1948: clipped wing, red spinner and diagonal red-and-white rudder stripes, filled Star of David roundels with no outer ring (Efrati pp. 12 and 14) a 16 km-radius loop over Tel Aviv and the coast, 4.2 km above the highest ground under it, banked 28 degrees, its shadow cast on the ground by the sun
Navy modified Flower-class corvette INS Haganah (K-20), 1948, 63.4 m off Jaffa (34.63 E, 32.02 N)

Not in the slice:

  • The mounted rider. No 1948 IDF mounted unit was sourced for this slice. The promo's rider belongs to 1936, so it waits for the Yishuv era (#49) and a source.
  • Civilian boats. The owner ruled them out of scope; navy ships are units.
    • 2026-10-14: the owner reversed this for civilian boats ("Keep civilian boats"); they are drawn as map art in #536. Navy ships are still units.

Units in play

Normal play draws one token per unit from two inputs: the sim's state and content/map/units_1948.yaml.

  • The state gives the date, the security level and the reserve call-up (GameState.Date, Rung, Mobilization). Since Phase 5 PR 4 it also holds the starting forces (below), and since Phase 5 PR 14 the GDD loop's army tokens come from them (below, "Tokens from the state"; design: docs/superpowers/specs/2026-10-11-units-in-play-design.md).
  • The data file is read by both. The sim reads the rows a new game starts a unit from (StartingForces; Q-START-1, owner, 15 October 2026), so the file is in the content hash and moves the goldens (StartingForcesTests.TheUnitFileEntersTheContentHash). The WebGL manifest ships it.

Code:

  • UnitData: the file.
  • UnitRoster: which units show now.
  • UnitPlacer and UnitGround: each anchor's spot.
  • UnitLayer: the tokens and the ring.
  • UnitBadge: the badge.
  • UnitCard and InfoPanelPresenter: the card.

Tests: UnitsInPlayTests (EditMode) and tools/tests/test_units_1948.py (the data rules, in CI's Python job).

Rulings (2026-10-11)

The owner ruled:

  1. One token per brigade.
  2. The number of figures follows the call-up: 3 at or above the security-level-4 floor (rung_min_mobilization, 0.4), 2 at or above the level-3 floor (0.15), else 1.
  3. 101 Squadron shows an Avia S-199 until the Spitfire's sourced date, then the Spitfire.
  4. The four brigades with no sourced position (Carmeli, Alexandroni, the 8th, Oded) are listed in the file and hidden.
    • 2026-10-15 (refs #385): Carmeli (from 9 July 1948; coordinator's option (d)) and Alexandroni now have sourced positions (below) and are shown. Since #588 (owner, 18 October 2026) Carmeli also stands at Haifa before that. The 8th is sourced at Tel Litvinsky (24 May to 9 July 1948; owner, 2026-10-15: "Accept, verify later") but stays hidden: the map has no clear spot for it on its own side within one hex of the camp. Oded stays hidden (its sources conflict and name no place).

The controller adopted the design's defaults:

  • tokens stand only at dated, sourced positions and do not move with operation events;
  • after the armistices, only what a source says continued is shown;
  • there are no orders: the card links to the Defence screen.

After the staff review the controller ruled:

  • No invented end dates (C1). A window ends at the earliest of its sourced end, the unit's sourced disbandment, and the armistice of the front it stood on. Each anchor says which: end_basis: source with an end_note, or end_basis: front_armistice with its front. The armistices were signed with Egypt on 24 February 1949, Lebanon on 23 March, Jordan on 3 April and Syria on 20 July (1949 Armistice Agreements). No source yet gives a position after the armistices, so no unit shows after them.
  • "Since" only from a sourced start (I1). A window that opens before the game shows from 15 May 1948. The card says "since February 1948" from the anchor's sourced since, and gives no start where no source dates the arrival (Givati at Gedera, the ships).
  • Within one hex (I2). See "Placement".
  • The squadron's bases where two sources agree (I3), and the Spitfire from its first sourced service (I4). See the table.
  • Plain checkpoints stay comparable (I5). A screenshot run (?shot, -shot) switches the units off unless it also asks for them (&units, -units). The editor's promo renders draw them only with -units.
  • The promo's badge, a wider trio, sourced headgear; the ring stays selection-only (I6). See "Look".

The Arab forces: drawn with their own figures

The file also holds the Arab forces of 15 May and 11 June 1948 (#740, Phase 5 PR 1), one row per force, with side, echelon and one strength (owner's decision on #570). The source is research_notes/GDD phase 5/arab-forces-1948.md, section 10: a row stands only where two or more of the note's sources agree on the place and the date, or where the owner ruled (al-Malikiyya on 15 May is the ALA's, #226 ruling 2).

Row Side 15 May 1948 11 June 1948 (the first truce, to 8 July) Strength (15 May)
Egyptian Expeditionary Force egypt Gaza (to 21 May) Isdud, the forward base, and Majdal, the rear base 10,000
Arab Legion jordan Ramallah (to 11 June) Latrun 5,000
Syrian Army syria south of Lake Tiberias, before Samakh (to 18 May) the Mishmar HaYarden bridgehead 3,000
Iraqi Expeditionary Force iraq the high ground over Gesher (15 May only, by the owner's decision) Jenin 3,000
Lebanese Army lebanon no place two sources agree on: no window al-Malikiyya, 6 June to 8 July (owner, 28 October 2026: "6 Jun to 8 Jul (Recommended)"); a map-only token 1,000
Arab Liberation Army ala al-Malikiyya (to 29 May) 4,000
  • One value per strength (CONTENT_POLICY section 1; Q-START-3): from an Israeli source where one gives it, else the value two of the note's sources agree on. Every row gives the 15 May strength only: no source in the note dates one on 11 June (owner, 17 October 2026, #740: "15 May strength only for the rows that have no sourced 11 June figure ... Nothing is invented"). The Syrian, Iraqi and Lebanese figures rest on Glubb's figures alone (Rogan and Shlaim reproduce them), merged as data only and marked to_verify (owner, 17 October 2026, on #860).
  • Two places at once. At the first truce the Egyptian row stands at both its forward base (Isdud) and its rear base (Majdal) (owner, 17 October 2026, on #860). Its two anchors overlap in time, the only drawn row whose do. It is drawn as one token (owner, 27 October 2026: "One token"): in the GDD loop the row's one sim unit, from Gaza; in the legacy loop the first anchor whose window holds the date (UnitRoster.AnchorAt), so the forward base. Majdal stays in the row and in the card's sector text.
  • Drawn only with their own figures (owner, 17 October 2026, #740): "Data only. PR 1 adds the sourced Arab rows to units_1948.yaml and filters them off the map in UnitRoster until Phase 5 PR 17's figures land, after its sensitivity review. Nothing is drawn with IDF figures, and there is no placeholder art." Tal Heuberger's 1948 Collection has figures for two forces, and Tal Heuberger's unit models count as reviewed and approved (owner, 17 October 2026, #756), so those two are drawn (issue #665; see "The 1948 Collection on the tokens"): the Arab Legion and the Arab Liberation Army. UnitRoster.OnMap (not hidden, and side: israel or a side with a figure) keeps the roster and the placer to those rows. Tal Heuberger's generated figures have since joined them (owner, 27 October 2026, #756: "Sign off all 19"): the Syrian, Iraqi and Egyptian infantry on their forces' tokens, and the Lebanese regular on the Lebanese army's token. The Lebanese army has no sim unit, so its token is map-only (owner, 27 October 2026: "Map-only token (Recommended)"): in both loops it stands at its dated anchor, al-Malikiyya from 6 June to 8 July 1948, and never moves or fights (UnitRoster.MapOnly; it is the only Arab army row with no starting force). A drawn force's token has no code-built figure: it waits until its figure has arrived. If the figure fails to load (its bundle is missing or damaged), the token is drawn with the plain built-in figure instead (owner, 31 October 2026: "Yes, fallback in v0.3.2"; UnitFigures.Failed). In the legacy loop, three figures (Q-UI-3 at full strength: the rows give only the sourced 15 May strength; owner, 17 October 2026, #226); in the GDD loop, Q-UI-3 by the sim unit's strength ("Tokens from the state"); under the neutral pennant (Q-ART-1). Its spot never stands on ground Israel held on its window's first day (#226, ruling 9; control_hexes_1948_49.yaml). The card leaves out Israel's call-up line. The goldens do not move. Tests: UnitsInPlayTests.OnlyTheForcesWithAFigureAreDrawnAndOnlyOnceItHasArrived, UnitFiguresTests.TheArabForcesAreDrawnOnlyWithTheirOwnFiguresOnceTheyArrive.
  • Not in the file yet: the local Arab forces (the note places them only beside places its "Conventions" lists as sensitive and does not describe), the Legion's Jerusalem battalions, the Egyptian eastern force, the ALA on 11 June (al-Malikiyya's holder is contested; the Lebanese row stands there from 6 June by the owner's ruling of 28 October 2026), and the Saudi, Sudanese and Yemeni contingents (not units of their own, Q-START-4). The side local_arab_forces is known to the loader for them; its token figure is Tal Heuberger's local irregular (owner, 2 November 2026: "Use it instead of C09"), and its row waits for a sourced 15 May strength (owner, 1 November 2026, #1243: "No row until sourced").
  • The Python tests check the same rules in CI (tools/tests/test_units_1948.py: test_every_side_is_known, test_strength_has_one_value_and_a_source, test_no_anchor_rests_on_one_source_where_the_note_says_contested, test_the_arab_rows_name_no_sensitive_place).

The IDF brigades' 15 May strengths (#743)

No primary order of battle gives the IDF brigades' strengths on 15 May 1948, so a secondary source may give them (owner, 20 October 2026, #743): "Every row cites its source, carries status: to_verify, and is replaced when a primary source turns up." The figures are the brigade table of research_notes/M4 war mode/war-1948.md, section 1 (#906): en.wiki's 1948 war article, whose size column cites Ben-Gurion's War Diaries (Arabic edition, 1994, p. 303). The table gives no date; as_of is 15 May 1948, the day the ruling gives the strengths for.

Ben-Gurion's diary for 15 May 1948 (#1240). The owner's ruling of 21 October 2026 replaces seven of the table's figures with the diary's for the day, a primary source: the Ben-Gurion Archive's English transcription, item 221910 (Wayback copy, checked 2026-10-21): "Palmach – about 6,000, of whom 1,300 are in Harel [...], 1,300 in the Negev, 1,200 in the north, 1,000 women; Golani has 3,573 + 522 paid recruits; Carmeli 2,289; Alexandroni 3,588; Kiriati 2,504; Givati 3,229; Etsiyyoni 3,166 – the total for all brigades is about 24,400." Golani counts the 522 paid recruits (4 units either way); Harel and Negev, Palmach brigades, take the diary's Palmach in Harel and in the Negev. en.wiki's table stays as each row's second source. A person has checked the seven figures against the page, so the rows are fact-checked.

Row Strength (15 May)
Golani 4,095 (diary: 3,573 + 522 paid recruits; the table gave 4,500)
Carmeli 2,289 (diary; the table gave 2,000)
Alexandroni 3,588 (diary; the table gave 5,200)
Kiryati 2,504 (diary; the table gave 1,400)
Givati 3,229 (diary; the table gave 5,000)
Harel 1,300 (diary, the Palmach in Harel; the table gave 1,400)
Yiftach 4,500 (the table counts some Golani men in it)
Negev 1,300 (diary, the Palmach in the Negev; the table gave 2,400)
  • No figure, no strength: the 8th and Oded have no size in the note, so their rows give none and stay status: draft.
  • Etzioni and the 7th (#1241): Etzioni gives 3,166 on 15 May 1948 from Ben-Gurion's diary, the same page as the seven rows above, which are fact-checked; Etzioni stays status: to_verify until a person checks the figure. The 7th gives 1,650, its men in Operation Bin Nun Alef, dated 24 May 1948, the day its Hulda window opens: a secondary figure, so status: to_verify (research_notes/GDD phase 5/later-brigades-1241.md). Yiftach's 4,500 stays until a person settles the note's conflict (4,500, about 2,000 or 1,200).
  • The sim reads them since Phase 5 PR 4: see "The starting forces" below.
  • Tests: test_idf_brigade_strengths_are_secondary_and_to_verify (Python) and UnitsInPlayTests (EditMode).

The starting forces (Phase 5 PR 4, #743)

A Historical game on 15 May 1948 starts with one sim unit per row that is not hidden, is an army row, has a 15 May strength and has an anchor window covering the day (docs/rules/mvp.md R-UNIT-006; owner, 20 October 2026, #743: "all five defaults"). The unit stands on the hex holding the anchor's coordinates, by the map's own projection and rounding (HexPicker, tools/hexgrid.py), precomputed by tools/unit_anchor_hexes.py into content/map/unit_anchor_hexes.yaml (rerun it after changing an anchor); the token may still stand one hex away (UnitPlacer), which is presentation.

  • Types: every IDF brigade is a veteran_soldier. Each Arab side with a starting row has one infantry type, a labelled design value with 3 movement points: egyptian_army, arab_legion (Jordan), syrian_army, iraqi_army, arab_liberation_army. These ids are the sim's, not names shown to the player (Q-UI-2, Q-START-4).
  • Strength is the row's people in 1,000-person population units, the nearest unit with a half up, never below 1.
  • Who starts: Golani 4, Carmeli 2 (at Haifa, its sector seat), Alexandroni 4, Kiryati 3, Givati 3, Harel 1, Etzioni 3 (Jerusalem), Negev 1 (#1240, #1241); Egypt 10, the Arab Legion 5, Syria 3, Iraq 3, the ALA 4. Nothing starts for Yiftach (no window on 15 May), Oded, the 8th and the Lebanese row (no window on 15 May; a map-only token), or the squadron and the two ships (tokens only, Q-UNIT-5).
  • Arrivals (R-UNIT-009, #1241): a row whose strength is dated after 15 May, with an anchor window covering that day, arrives on that day: in the resolution of the monthly turn that reaches it, on that anchor's hex, with that strength. Its unit id is reserved at the start, after the starting rows, so unit i is still row i. Today that is the 7th: 2 units at Hulda from the 15 June 1948 turn.
  • Release gate: every row the sim starts a unit from must be fact-checked before a shareable build carries it (Q-START-3). Statecraft.Cli --release-gate, Build -shareable and tools/package_release.py refuse until then, and name the rows. A fact-checked row records the day of its person check in checked_on; a field whose sources were checked after that day (the strength, an anchor or an order target) counts as unchecked, and the gate names the row and the field (#2722). Today that is the order targets of Carmeli, Kiryati, Givati, the 7th, Harel and Negev (#2135), sourced on 1 November 2026, after the owner's checks of 21 October (#1240) and 22 October 2026 (#1241).
  • Tests: StartingForcesTests (dotnet) and UnitsInPlayTests.TheStartMatchesTheMap (EditMode).

Tokens from the state (Phase 5 PR 14, #753)

In the GDD loop (systems.loop: gdd, the real content) the army tokens come from GameState.Units, not from the rows' dated windows (UnitRoster.FromState; impact plan §2.6). The legacy loop keeps the rows and the call-up until D-E1 deletes it.

  • One token per military unit whose row is on the map, grouped per hex ((q, r) order). Unit i of a new game is the starting force Content.StartingForces[i] (R-UNIT-006), and that row gives the card, the flag, the headgear and the vehicles. A halutz draws no token. A raised Israeli recruit, brigade, artillery or armoured-car unit has no starting row; it gets a presentation-only token: a synthetic row with steel helmets for a recruit and the default headgear otherwise, and its type's vehicle, placed at the first Israeli starting brigade's start and sized by that brigade's sourced strength (UnitRoster.RaisedToken; nothing is added to the sim state). An Arab force's token still waits for its Collection figure or, if that fails, the built-in one.
  • The sim hex is authoritative. A unit on its starting hex stands on its starting anchor's spot (UnitPlacer's one-hex offset still clears the town kits and the roads). A unit that moved stands within one hex of its new hex, off the hexes the other tokens hold. When that search finds no clear spot, the unit stands compact on its own sim hex, apart from the other tokens where it can, and a spot that is not clear is accepted rather than draw nothing, so every unit can be selected (owner, v0.3.0 RC, 25 October 2026; UnitPlacer.OnOwnHex). The spot is placed once and cached, so a rebuild places only the units that moved. Its card names no place (no source does) and gives the row's sources.
  • Figures follow Q-UI-3 (owner, 15 October 2026; #226): 3 at two thirds or more of full strength, 2 at a third or more, 1 below that. Full strength is the sourced strength the unit started with (Q-UNIT-4), so a new game draws three figures on every token whatever the call-up, and losses take figures away. So the card leaves out the call-up: an army token's card has no national line, and the squadron's gives the security level alone ("Security level 5 of 5"), which still sets whether it flies.
  • The squadron and the ships are tokens only (Q-UNIT-5): they still come from their rows' windows.
  • So from 15 May 1948 the map draws the thirteen starting forces where the sim holds them, and the 7th from its arrival, on its arrival anchor. A row with no sim unit (Yiftach, whose window opens later) draws no token in the GDD loop.
  • Tests: UnitsFromStateTests (EditMode).

The 1948 Collection on the tokens (#665)

The tokens wear Tal Heuberger's figures from The 1948 Collection (#507, #680), streamed after the first frame (CollectionStream, the people and vehicles bundles); nothing is fetched before it, and nothing sits in Resources/. UnitFigures holds the table. Each model is the far (strategic-zoom) static mesh, put under the figure with the prefab's own root rotation kept (the FBX axis turn, -90 degrees about X), so the camera's yaw composes on top of it.

Model Where Source From
C04 sabra pioneer an Israeli rifleman in the bush hat (the uniform default's first figure; the showcase trio's first) #507's pairing always
C06 young recruit an Israeli rifleman in the Mk II steel helmet (every token's second figure) #507's pairing always
C07 veteran fighter an Israeli rifleman in the knitted stocking cap: the Negev Brigade's first, the showcase's third the Negev headgear photographs (Sources) always
C05 pioneer veteran the third figure of every Israeli brigade with the uniform default, except Harel and Yiftach a presentation choice: the default's third figure is not sourced always
#336's 1948 soldier the third figure of Yiftach and Negev, the Palmach brigades other than Harel: a rifleman in khaki in the Palmach's knitted stocking cap, so Negev's token shows C07, C06 and him and keeps its sourced caps. Drawn as his rigged copy, figure-soldier-1948-v1-rigged (#2384): Tal's mesh on the skeleton C06 and C07 share, with C07's idle, walk and fire takes (tools/rig_soldier_1948.py), so he walks and fires with the others; Tal's delivered file is unchanged and drawn nowhere owner, 21 October 2026 (#708, through the coordinator): "a third figure on some brigade unit tokens, next to C06 and C07"; the rows' numbers (11th and 12th Brigade (Palmach)) always
C12F Palmach scout Harel's third figure, at Sha'ar HaGai on the convoy road Harel's row (a Palmach brigade escorting the convoys, Izkor); en.wiki Netiva Ben-Yehuda (a Palmach woman's duties included "escorting convoys") always
C08 Arab Legion the Arab Legion's three figures the row; owner, #756 (Tal Heuberger's models approved) the row's windows
C09 irregular the Arab Liberation Army's first two figures en.wiki Arab Liberation Army: "an irregular army" of volunteers the row's windows
ALA infantry (Tal Heuberger, generated) the Arab Liberation Army's third figure, beside two C09; the token waits only for C09, and the third joins when its own bundle arrives the row; owner, 27 October 2026 (#756: "Sign off all 19"; "Third figure (Recommended)") the row's windows
Egyptian infantry (Tal Heuberger, generated) the Egyptian Expeditionary Force's three figures, one token (in the legacy loop the first anchor whose window holds the date) the row; owner, 27 October 2026 (#756: "Sign off all 19"; "One token") the row's windows
Syrian infantry (Tal Heuberger, generated) the Syrian Army's three figures the row; owner, 27 October 2026 (#756: "Sign off all 19") the row's windows
Iraqi infantry (Tal Heuberger, generated) the Iraqi Expeditionary Force's three figures the row; owner, 27 October 2026 (#756: "Sign off all 19") the row's windows
Lebanese regular (Tal Heuberger, generated) the Lebanese Army's three figures, on its map-only token at al-Malikiyya the row; owner, 27 October 2026 (#756: "Sign off all 19"; "Map-only token (Recommended)"; 28 October 2026: "6 Jun to 8 Jul (Recommended)", the window) the row's window, 6 June to 8 July 1948
H02 65 mm gun ("Napoleonchik") on the front of the base, before the trio: Carmeli, Givati, Harel, Yiftach, Negev Ben-Gurion's diary, 11 July 1948 (the cannon distribution) 11 July 1948
H01 sandwich armoured car on the front of the base, before the trio: Givati Wikimedia Commons, "Givati force" (GPO, Zoltan Kluger, 12 June 1948): a jeep leading an armoured unit of the Givati Brigade; the type is read from the picture (#1536) 12 June 1948
H03 Willys jeep on the front of the base, before the trio: Harel, Negev Harel: the Burma Road jeep convoy (Isseroff); Negev: the Negev Beasts' jeep unit at Negba (en.wiki Battles of Negba) 1 June and 2 June 1948
  • The vehicles are data: vehicles in units_1948.yaml (type, from, note, sources), checked by UnitData.Parse and tools/tests/test_units_1948.py. A vehicle stays for the rest of the row's windows.
  • Size: a figure takes the code-built figure's height; a vehicle is drawn 0.7 of a figure's height long, sized against the figures as the aircraft and the ship are, not to scale, so it fits on the base.
  • C05's pose: he had no idle clip and shipped in his rest pose, a T-pose. His static meshes are now posed at the first frame of his own walk (tools/slim_collection.py, POSE_CLIP).
  • Close zoom: below the town models' switch (TownModelLayer.CloseEnterM, 40 km; far again beyond CloseExitM, 45 km), each figure whose model has an idle clip (CollectionCatalog.Animated) shows its skinned near model playing the clip, as tall as its far model and under the same holder (UnitFigureDetail). The near models are asked for only once a camera first comes close; until one arrives, and for good if it is missing, the far mesh stays. C05 has no idle clip (walk and run only) and keeps his still far model (owner, 18 October 2026, #665: "Keep its still model"); #336's soldier, drawn as his rigged copy (#2384), plays C07's idle. The vehicles stay far.
  • Fail open: an Israeli figure keeps its code-built mesh until its model arrives, and for good if the bundle is missing or damaged; a vehicle is not drawn until its model arrives. An Arab force's token waits for its figure too, and is drawn with the built-in figure if that fails (UnitLayer.cs:379-381).
  • Tests: UnitFiguresTests (every one of the sixteen is placed on some day of the war; the table; the swap; arms down; fail open; the close-zoom swap and its idle clip, the missing near model and C05).

Weapons and the C06 point-fire aim (#2083, #2084)

The owner ruled on 2 November 2026: #2083, "Yes, on the soldier figures (Recommended)" ("Attach a period-correct weapon to the unit-token soldier figures (sourced per army and date), measured for load"), and #2084, "Yes, for firing units (Recommended)" ("Show the aiming pose on a unit token when it attacks"). Models by Tal Heuberger (CC BY-NC-ND 4.0).

Weapon Army From Figures Sources (agent-read; a person's check is pending)
K98k (Tal's weapon-k98k-v1) Israel (Haganah, IDF) 2 April 1948: the Nora's 4,500 Czechoslovak P-18 (K98k) rifles at Tel Aviv; the first air shipment landed the night of 31 March to 1 April C04, C05, C06, C07, the 1948 soldier's rigged copy (#2384) en.wiki Arms shipments from Czechoslovakia to Israel 1947–1949; Palyam, "Nora" (Ben-Tzur); en.wiki Karabiner 98k
Sten (Tal's weapon-sten-v1) Israel (the Palmach) in use 1945–1948 C12F, the Palmach scout (Harel's third figure) en.wiki Ayalon Institute ("the personal weapon of the Palmach"); en.wiki Sten ("clandestinely manufactured in Tel Aviv and on various kibbutzim in 1945–1948")

Every token's first day is 15 May 1948 or later (units_1948.yaml), after both rows' dates, so no figure is dated by hand (UnitWeaponsTests keeps that true). The rows live in UnitWeapons.Rows.

  • [GAP] The Arab forces. No source found places a K98k, an MG34 or a Sten with the Arab Legion, Egypt, Iraq, Syria, Lebanon, the ALA or local irregulars in 1948 (Syria's Czech rifles for the ALA sank in the Lino at Bari and were later captured). Their armies' British Lee-Enfield and Bren are not among Tal's models, so their figures stay as they are.
  • #336's 1948 soldier (Yiftach's and Negev's third figure) is drawn as his rigged copy (#2384), which has C07's 67 bones and hand bone and holds the K98k like the others; Tal's delivered file, with no rig, is drawn nowhere.
  • The MG34 stays a stored source. The "200 MG34s" is not one fact (research_notes/Map/mg34-1948-2288.md, #2310): for the January 1948 contract, one source names the MG34 for its 200 machine guns (Wikipedia, a sentence with no footnote of its own), Ben-Gurion's diary of 1 February 1948 gives 200 with no type, and a Russian magazine gives 20. The Nora's 200 machine guns have no type; the first flight, the night of 31 March to 1 April 1948, carried forty MG-34 (Morris, p. 117, as quoted on Wikipedia). Nothing places a machine gun on one token figure (an owner question).
  • The hand: each weapon hangs under the figure's Right_Hand bone. The socket is Tal's own fixed pair: in their C06 point-fire export the K98k does not move against C06's right hand over all 97 frames, and that relation, turned into the C06 rig's frame, is UnitWeapons.SocketPosition/SocketRotation/K98kSocketScale (the rifle about 1.1 m long on a 1.78 m man). C04, C05, C07 and C12F share C06's 67-bone rig and hold it the same way: in each one's bind pose the rifle's long axis is within 45 degrees of C06's (measured 2 November 2026: C04 34.4, C05 10.1, C07 13.5, C12F 37.8), along the hand. The Sten takes the same socket, scaled by the sourced lengths (Sten Mk II 762 mm, K98k 1,110 mm); its grip is not authored by Tal, and in C12F's idle pose it lies across her thighs, partly inside them (an owner question; Tal's own notes leave the Sten's handedness to a 3D review).
  • The C06 aim: while its token's attack plays (UnitTokenMotion), a C06 figure, near or far, holds one still frame of Tal's 97-frame point-fire clip (frame 37, inside the still aim of their contact sheet) for the whole attack, which lasts as long as the longest fire take on the token (C06's own is 37 frames); then every bone it moved is put back and its Animator runs again, also when the units are hidden mid-attack. Tal's export is one figure, so only C06 takes it; every other figure keeps its own fire take. The export's rig has the same bones and rest mesh as C06's but other local axes, so each sampled bone is turned into C06's frame by a correction computed from the two meshes' bind poses (UnitFigurePoseSource); our C06 in the aim lies on Tal's export and our K98k on their rifle (UnitWeaponsTests).
  • Borrowed fire take (#2375): C04 has no fire take of its own and uses a clip resampled per frame from C07's Armature|fire take (tangents rebuilt, near-static bones dropped); the two share Tal's 67-bone Armature (CollectionImport.BorrowedFireTakes). FireTakeRetarget bakes it beside C04's model as fire.anim, because the stream loads each figure's bundle alone; no FBX import changes. Tal Heuberger's take, CC BY-NC-ND 4.0; the owner confirmed Tal's permission for this adapted version on 3 November 2026. The Syrian, Iraqi, Egyptian, Lebanese and ALA infantry have no fire take yet and keep their idle or walk when their unit attacks: they hold no weapon, and they wait for #2399 (owner, 3 November 2026: "C04 only; enemies wait (Recommended)").
  • Load: the K98k, the Sten and the pose source each stream in a bundle of their own after the people's, never in the first download; the pose source carries only the clip, not the export's mesh or texture.
  • Fail open: until a weapon or the pose source arrives, and for good if its bundle is missing, the figure stands and fires as before.

What the player sees

Unit Positions (dated, from the sources) Ends
Golani (1st) Yavne'el (since February 1948); the Julis camp from December 1948 December 1948 (source); 24 Feb 1949 (Egypt)
Carmeli (2nd) Haifa, its sector's seat (since February 1948), in the town on Haifa's one unit clearing (#588); facing the Syrian bridgehead at Mishmar HaYarden, on the Rosh Pinna side, 9-18 July 1948 (Operation Brosh) 9 Jul 1948, Operation Brosh (source); 19 Jul 1948, the second truce (source)
Alexandroni (3rd) Netanya, its headquarters at Camp Dora (since March 1948) November 1948, the move to the Faluja pocket (source)
Kiryati (4th) Tel Aviv (since February 1948) 3 Apr 1949 (Jordan)
Givati (5th) Gedera, one of the three places its sector names (no start given) November 1948, the move to the centre (source)
Etzioni (6th) Jerusalem (since late 1947) 3 Apr 1949 (Jordan)
7th Brigade Hulda, its headquarters for the Latrun attacks, 24–31 May 1948; Ein Shemer, June 1948 sources
Harel (10th) Sha'ar HaGai, one place picked from the Jerusalem corridor (since the first half of May 1948) 22 Dec 1948, Operation Horev (source)
Yiftach (11th) Lydda, 11–19 July 1948 (Operation Dani; the IDF entered Lydda on 11 July) 20 Jul 1948 (source); disbanded May 1949
Negev (12th) the Mekorot camp near Nir Am (since March 1948) 24 Feb 1949 (Egypt)
101 Squadron Ekron, 20–29 May 1948; the Herzliya strip from mid-June; Hatzor from 21 November 1948 June 1949, the move to Ramat David (source)
INS Haganah, INS Wedgwood Haifa harbour (commissioned 1948; no start given) 24 Feb 1949 (Egypt: their sourced actions were against the Egyptian forces)

101 Squadron's bases. The map follows the dates two sources agree on:

  • en.wiki and he.wiki agree it was at Ekron from 20 to 30 May, and at Herzliya from mid-June.
  • Between 30 May and mid-June they disagree. en.wiki says Herzliya after the strafing of 30 May; he.wiki says Ekron, with Modi Alon taking off from there on 3 June. So the map shows no base in that gap.
  • Hatzor from 21 November 1948: Eduard p. 54 and he.wiki ("November"). en.wiki and Eduard p. 53's profile caption say October. he.wiki has Alon killed landing at Herzliya on 16 October, which fits November.

The aircraft. The S-199 until 15 October 1948, then the Spitfire.

  • The first Spitfires flew in, unarmed, on 27 September 1948 (Operation Velvetta, he.wiki).
  • The first sourced Spitfire sortie is 15 October: three Spitfires escorting Beaufighters to El Arish (he.wiki Supermarine Spitfire). The squadron flew both types (he.wiki 101 Squadron), and the S-199 flew on sporadically until mid-December (Avia S-199; Eduard p. 54). The ruling allows showing the S-199 until the switch and the Spitfire after.
  • en.wiki's Spitfire operational history calls a 101 Squadron claim of 21 October the IAF's "first ever aerial victory". That is wrong: the first victories were on 3 June 1948, by Modi Alon in an S-199. The file records this.

Placement (UnitPlacer.PlaceAll). Every anchor is placed once, from the data alone, so a token never moves because another unit appears. Its spot is at most one hex step from the sourced place (about 7.6 km between hex centres on 4.4 km hexes). It is the nearest such spot to the place, or to the anchor's sourced toward point when it has one (see below). In the GDD loop a simulated starting force can persist after its source anchor's window ends. A map-only army whose cached spot overlaps that starting spot gets a fixed display override, computed once from the same data before date or model-arrival filtering. It follows the same one-hex, source-offset, land and road constraints and must clear other cached spots. Overrides cannot stand in a town: the map's original spots and kit clearings stay unchanged. Only relocated display spots are reserved against later moved-token placement. Source anchors, their dates, sim units and legacy placement do not change. If no safe override exists, the token remains drawn at its original spot with an overlap warning.

The candidates are tried in four rounds:

  1. The centres of the anchor's own hex and its six neighbours, with the token at full size.
  2. If none is clear, those centres plus 18 points inside each hex (three rings of six, at 0.25, 0.45 and 0.65 of the circumradius: UnitPlacer.InsideOffsets), with the token compact: half the ring and half the trio's spread, to fit beside a town. Yiftach at Lydda and the squadron at Herzliya stand compact.
  3. If the whole neighbourhood is town, the compact point that reaches least deep into it (then the nearest; with a toward point, the nearest to that). The map then keeps a clearing of the compact ring's radius free of the kits' buildings and trees there (MapScene.UnitClearings, computed before the kits from the same data), so the token never stands in a building. The clearing may cut only into the anchor's own town, never a neighbour's. Kiryati (Tel Aviv), Carmeli (Haifa) and Etzioni (West Jerusalem) stand this way. Haifa and Tel Aviv each have a budget of one such clearing, and their census floors are measured with it (#588, owner, 18 October 2026; TownCensusTests).
  4. If none of those is clear, the anchor's point and rings round it every 500 m out to the one-hex limit, about 500 m apart on each ring, with the token compact and off the towns as in round 2. On 4.4 km hexes (#1183) the points of rounds 1-3 lie 1.1 to 2.9 km apart and can all miss the clear ground between villages: Majdal, the Egyptian rear base, found none among them (every one touched a village's disc or a route's corridor), while clear ground lies 3.5 km south-west of its point, west of Ni'ilya. The round runs last, so an anchor placed in rounds 1-3 keeps its spot (a round-4 spot holds its hex like any other; measured on the game map, no other anchor and no camp moved).

A spot is clear when:

  • the ring's disc stays off every settlement footprint (the kits' discs, times 1.1); for an air anchor the disc is the larger of that ring and the parked aircraft's span (4.7 km across, which does not compact), with no in-town exception;
  • an army token's three figure spots (always three, so the spot does not change with the call-up), or a parked aircraft's centre, are on land and at least 700 m beyond every railway and road corridor drawn while a token stands (a route opened after the last anchor's window, the 1949 rail link or the 1950 Tel Aviv–Netanya road, takes no spot; #786);
  • a ship has a sea hex, and its whole drawn hull (the corvette, 6.0 km by 0.96 km at its heading) lies in the drawn water; a ship tries the hex centres, then the points inside its sea hexes, and is never compact (#3038);
  • no other anchor whose window overlaps holds the hex.

Own side. The map holds no 1948 front line yet, and the one-hex limit alone does not keep a token on its own side: the nearest clear spot can be the other side's ground. So an anchor where that happens has a sourced toward, a place on its own side, and where needed a place_hex, the hex next to the anchor on that side. The token stands at the clear spot nearest that place, in that hex when one is given:

  • Etzioni stands in West Jerusalem, about 2 km west of the place the West Jerusalem article gives, not on the Hizma and Anata side, which Jordan held. It is in town, with a clearing. In every round the placer also tries rings round a toward point, every 500 m out to 2.5 km, so a spot can sit between a town's roads near the sourced place.
  • The 7th Brigade is anchored at Hulda, its headquarters for the Latrun attacks (Battles of Latrun), not at Latrun. The Arab Legion held Latrun and the Yalu–Beit Nuba ridge. The ground between was contested until the Beit Jiz and Beit Susin area was cleared on 28 May, and Sar'a was taken only on 13–14 July.
  • Harel stands by Saris on the corridor road, hex (-2,26), taken on 16–17 April 1948. It no longer stands by Sar'a and Artuf.

The test pins Etzioni's distance from the West Jerusalem point, Harel's hex, and the 7th within 2.6 km of Hulda; both the 7th and Harel stay more than 3 km from Sar'a. With the control layer (control_hexes_1948_49.yaml) the placer also keeps an Arab force's spot off the hexes Israel held on the anchor's first day (#226, ruling 9). On 4.4 km hexes (#1183) Isdud's nearest clear centre was the Israeli-held hex east of it. Israeli tokens keep the toward rule above, and a unit the sim moves is placed without the control layer. EveryAnchorStandsWithinOneHexClearOfTheTownsAndTheRoutes checks every anchor and logs each spot and its distance.

Look.

  • Models: the slice's models and the authored-prefab hook.
  • Army: a token is 1 to 3 riflemen, spread across the ring as in promo 03, about 2 km apart, so each figure reads on its own.
  • Headgear: only the Negev Brigade's is tied to it by a source: knitted stocking caps and Mk II steel helmets, from its two 1948 photographs. Every other brigade has the uniform default (bush hat, steel helmet, bush hat), from the 1948 photographs listed under "Sources".
  • Air: the squadron flies the slice's loop over its base at security level 4 or 5 and stands parked below that. A parked aircraft leaves a circular tree hole covering its drawn world-XZ footprint. Its camera-facing badge stays clear of the aircraft's tail (the rear 40%) as the camera sees it, moving down, left or right on screen by the least that clears it, never up, so it stays by the aircraft; a flying aircraft gets neither adjustment.
  • Navy: a ship is the K-20 corvette model; Wedgwood was the same modified Flower class.
  • Badge: each token has the promo's badge (UnitBadge): a dark rounded plate with an ivory glyph (two figures, an aircraft or a ship). It is turned to face each camera as it renders. On an army token it sits above the flag's top, a little behind the trio and to the side away from the flag's pole as that camera sees it (UnitLayer.FaceCamera; review of #559: it covered the middle figure's head once the trio stood shoulder to shoulder, and the flag now turns with the camera). The map's names keep clear of the badge and the flag where the camera draws them: GameRoot turns them to the labels' camera before the names are placed, and each token's keep-clear boxes follow (UnitLayer.KeepClear). A fixed box round every place the camera's yaw could take the badge, about 3.7 km across, pushed Tel Aviv's name off its kit at gameplay03 (MapLabelTests.PillsKeepClearOfTheUnitTokens; test TheKeepClearBoxesFollowTheBadgeAndTheFlagAsTheCameraTurns). Up close the badge is capped on the screen (owner, 31 October 2026: "Shrink up close (Recommended)"): from about 60 km out it keeps its 1.4 km width (#2430: about 30 km until then, where the plate was nearly as wide as the figures are tall), and a nearer camera draws it no taller than UnitBadge.MaxScreenShare of the screen's height, shrinking about its lower edge, which stays clear of the flag (test UnitBadgeScreenCapTests).
  • Ring: the white ring appears only on selection.
  • Far zoom (#2430): a token's figures and flag keep UnitShowcase.FigureScale up to 45 km (TownModelLayer.CloseExitM, where the close building models have gone), grow to 1.3 times it at 60 km and stay there (UnitShowcase.ZoomScale), so a figure stays near promo 03's size on the screen. The constant does not change, the figures keep their spots and the base keeps its size. Only the game camera sets it (UnitLayer.FaceCamera). The showcase trio does not grow.
  • Base and flag (demo2 T29, refs #197). An army token's figures stand shoulder to shoulder on one round earth base ringed by a bright white rim (UnitPlinth), and the side's flag flies on a pole at the base's back edge (UnitFlags). See "The base and the flag".
  • Slopes (owner question, 2026-10-11; rounds 3 and 4 of #559). The base is draped on the terrain's own triangles: every heightmap cell under it (79 x 211 m) is split along the diagonal the terrain itself uses there and raised by the lip (UnitPlinth.LipShare, a twentieth of its radius, about 100 m), the cells at the edge clipped to a 96-sided outline with heights from the terrain's planes. So the top follows the ground exactly wherever a token stands, with no grid density to tune, and an edge lip of the same height, sampled every 40 m along the outline, closes it to the ground. On a hill it neither stands up as a mesa nor leaves a gap downhill. A figure stands on the top at its own point (an uphill boot sets into the base, which hides it).
    • Tests: TheBaseFollowsTheGroundOnEveryHillHexOfTheMap (every hill hex of the map, not only today's spots: the top within half the lip of the ground plus the lip, and the 99th percentile within a quarter), TheBaseFollowsTheGroundOnEveryHillToken, EveryFigureStandsOnTheDrapedTop and NoRiflemanSinksIntoTheHillsOfTheMap. The level discs before stood up to 1,081 m (round 1) or 707 m (round 2) over Harel's valley side; round 3's 32 x 96 polar grid strayed up to 113 m on the map's hills.
  • Trees (#249). The kit trees whose crowns reach a drawn token's base are left out at draw time (VegetationLayer.Hide), so no orchard grows through a figure. Only the figures drawn that day clear trees, and they come back when the figure goes. With the units toggle off, as in a plain ?shot run, every tree is drawn. The planting, and so the kits' layout, never changes. Tests: NoKitTreeGrowsThroughAShownFiguresFeet and OnlyTheShownFiguresTreesGoAndTheToggleBringsThemBack.

Selection and the card.

  • Clicking a token's hex selects it. The ring goes round it and the hex outline is cleared, so it has one highlight.
  • A second click shows the hex and its map actions. A hidden layer (the units toggle off) selects nothing.
  • The card gives the name, the number and the service, where the unit stands (with "since" only from a sourced start), its formation and first commander where sourced, and the sector in the source's words.
  • In the Hebrew UI the place, the first commander and the sector each show their optional Hebrew counterpart in units_1948.yaml (#558): place_he on an anchor, first_commander_he and sector_he on the row, and for a date in prose since_he on an anchor and formed_he on the row (#1292; a plain day or month names its month through the date strings instead), and number_he on the row for the subtitle's number line (#1219 item 2), each with its *_he_sources (the sources of its proper names) and *_he_review: agent-checked until a native speaker has read it (#564). A field without one shows its English, independently of the others; the English card never changes. An AI-assisted contributor may also do that read and remove the line, when the pull request discloses the AI help (the owner, 4 November 2026: "AI-assisted can clear it"; docs/CONTENT_POLICY.md, section 2, rule 6). UnitData.Parse refuses a Hebrew text without its English (so a Hebrew commander never adds a line), a present but blank key, sources or a review flag without the text, and text without Hebrew letters; any other Hebrew key fails as every unknown key does. Their sources join the card's source hosts but, like name_he_sources, never clear wikipedia_only (#902). The shipped rows carry them since #1292 (all agent-checked, each name's source its he.wiki article), except the Iraqi force's sector, whose "Ma'ad" has no Hebrew spelling found. Tests: UnitsInPlayTests (TheHebrewProseShows..., EachHebrewFieldFallsBack..., AHebrewCommanderWithoutTheEnglishAddsNoLine, AProseDateShowsItsHebrewInTheHebrewCard, BadHebrewProseNamesTheProblem, HebrewUnitNumbersAreSourcedAndAgentChecked, TheHebrewUnitCardSubtitleShowsTheSourcedNumberAndEnglishKeepsItsNumber), on test-only fixtures.
  • It also gives the national values the token reflects on Israel's cards (legacy loop: "Reserve call-up 80% · security level 5 of 5"; in the GDD loop only the squadron's card has one, "Security level 5 of 5", see above) and the source hosts, with a draft note that says "Wikipedia only" when the position's own sources are.
  • A button opens the Defence screen. In the GDD loop the card also gives the Move order (GDD Phase 5 PR 15, slice 1, #754; Q-UI-2's name, "Move" / "תנועה"). The button shows only when Movement.Reachable has a hex for the unit (the token's sim unit is its row's starting force, UnitCard.SimUnit), and not once the game is over. Pressing it lights exactly those hexes (MapScene.ShowReachable, a band half the selection's width); a click on a lit hex moves the unit at once through SimBridge, and the move is final (R-TURN-004; owner, 23 October 2026, #1398: "No undo, like Civ"); a click on a farther land hex the unit has a route to gives it a go-to that continues on later turns (on touch, the first tap shows the route and a second tap confirms it); any other click, a second press, a turn or another selection ends the pick. The sim refuses a move that is not possible at the click, with its own reason. Slice 2 adds Fortify and Hold (Q-UI-2's "Fortify" / "ביצור", "Hold" / "החזקה"): each button shows only where the sim's Orders.Check takes the order, and where it does not, the card shows the sim's reason as the sim gives it (in English for now, as the End turn toast does). A press gives the sim's UnitOrder through SimBridge at once (one standing order per unit: Hold replaces a Fortify, and either replaces a go-to) and the card says when it takes effect. Slice 3 adds Attack (Q-UI-2's "Attack" / "התקפה"): the button shows only where the sim's Combat.Check takes some unit's hex, and the sim's reasons for the enemy hexes in reach it refuses (no Arms, no movement points left) show as the others do. Pressing it lights those hexes; for each the card gives the units there by name, the sim's strength ratio (Combat.Preview) and its combat_table row's outcome bands (Combat.Outcomes: how many faces of the draw give each outcome, and the people each side loses), as Q-UI-1 decided. A click on a lit hex gives the sim's UnitOrder(unit, attack, q, r) through SimBridge, and the attack is fought at once (owner, 23 October 2026, #1398: "Instant, like Civ"); its Why lines show as a toast. Orders.Give and Combat.Give run Orders.Check and Combat.Check first, so the card and the click refuse the same orders for the same reasons. Tests: UnitOrdersTests, InfoPanelLayoutTests.TheUnitCardWithItsOrdersFitsEveryWindow.

The Avia S-199 (UnitModels.AviaS199; type avia-s199-101sqn-1948):

  • The Bf 109G airframe, 8.94 m long with a 9.92 m span, with the Jumo 211F's wide three-blade paddle propeller and the MG 151 pods under the wings (Avia S-199 article).
  • One khaki green on all surfaces, about RAL 6013: Efrati pp. 8 and 11, confirmed from an Avia fuel cap. Eduard Info 11/2025, pp. 53–54, also has one green.
  • The identification band round the rear fuselage is white, blue, white (Efrati pp. 8 and 11).
  • Filled Star of David roundels sit on both fuselage sides behind the cockpit and under the wings (Efrati pp. 6, 8 and 11, and the 1948 photographs). None sit on top of the wings: over-wing roundels were "standard" on no IAF aircraft from May 1948 to January 1949 (p. 6) and were adopted on the Avias only after the Velvetta Spitfires arrived (pp. 11 and 14).
  • The red spinner, the rudder stripes and the upper-wing roundels of Operation Ten Plagues (from 15 October 1948, Eduard p. 53) are not drawn, because the map shows the Spitfire from that day.
  • The squadron's skull badge on the nose is too small to read at map scale and is left out.

Cost. The budget is 2.5 ms or less for the units toggle at gameplay03 (docs/visual-v1.md). The measurement was headless ?vprofile on the WebGL build of 943184c, 1920x1080 at DPR 1, 2026-10-11. It ran under the main checkout's machine_lock.py --exclusive --wait-unity, so no Unity job ran beside it.

Pose all -units units cost none
gameplay03 14.74 ms (spread 0.48, p95 17) 14.14 ms (spread 0.71, p95 15) 0.60 ms 11.09 ms
diorama02 17.66 ms (spread 0.53, p95 20) 17.39 ms (spread 0.17, p95 20) 0.27 ms 13.18 ms
  • The units cost well under the 2.5 ms budget at both poses. The new-game state draws 6 brigades (up to 18 riflemen, with their badges) and 2 ships.
  • The whole scene's p95 is over 16.7 ms at gameplay03 (17 ms), and at diorama02 even without units (20 ms). The base frame here (none 11.09 ms) is slower than in earlier runs. That is a scene-wide matter for the perf track, not this layer's.
  • The layer rebuilds only when the roster changes (a new turn, a call-up or security order). Between those, its only per-frame work is a flying aircraft's transform and turning about ten badges to the camera.

Left open.

  • The rows are drafts from Wikipedia (and Eduard and Commons) until a historian checks them against Lorch or the archives.
  • No dated front line: tokens keep within one hex of their sourced place, and three anchors carry a sourced toward point to stay on their own side (review MF1).
  • Kit baseline break: from 943184c the kits keep the unit clearings in Tel Aviv and West Jerusalem (and in Haifa since #588), in plain checkpoints too. So kit metrics before and after that commit are not comparable; this is marked in the art log.
  • Merges still to come (review I7). With promo-borderdata, the info panel's region label must be cleared when a unit card shows. With promo-screens, the wide-screen click guard must cover ShowUnit and UnitLayer.Select as well as Show, and the info panel must keep its info-defence button. Both branches add optional InfoPanelPresenter parameters, so the call in GameRoot should name its arguments.
  • Wedgwood's Israeli name and pennant number conflict between its two articles. The card shows "INS Wedgwood" and no number.
  • No source yet places any unit after the armistices.
  • The Spitfire's red spinner and diagonal rudder stripes date from early October 1948 on the Spitfires themselves (Efrati p. 12); IsraDecal attests them for January 1949.
  • In the historical playthrough the call-up stays at or above 0.4 until the armistices, so the tokens keep three figures.

Pipeline choice: code-built miniatures

Three pipelines were tried on a development Mac (Apple M5 Pro, 48 GB) on 2026-10-11. Each was placed on the map at the same scale and rendered at the gameplay zoom and close up, next to the promo's trio; the table summarises what each produced. The renders are not in the repository.

Pipeline Result Licence
(a) Image-to-3D, run locally: TripoSR (MIT code and weights), CPU, about 3.5 s per mesh Inputs were a promo crop and a 1948 photograph. Out came a blurry blob of about 45,700 triangles with baked vertex colours: the head and rifle were lost to the background cut-out, the side view was a flat card for the rider, and there was no rig. It would need clean, sharp turnaround references, which we cannot make locally, plus manual retopology. MIT; fine.
(a) Hunyuan3D-2, TRELLIS Not run locally. Hunyuan3D-2's community licence excludes the EU, the UK and South Korea, which is not compatible with an open project. TRELLIS needs an NVIDIA GPU with CUDA; the owner chose it for the cloud-GPU close-zoom track (rulings above). Hunyuan3D-2: not compatible. TRELLIS: MIT.
(b) CC0 base meshes: Quaternius Universal Base Characters (free "Standard" kit, CC0) A clean, rigged body, but a stylised "superhero" figure in underwear, in a T-pose. Every uniform, hat, belt and rifle would have to be modelled by hand in Blender. Kenney's characters are blocky, low-poly figures, which the spec rules out (§6, "no low-poly packs mixed in"). CC0; fine.
(c) Code-built from primitives (chosen) Lathed limbs and torsos, ellipsoids, boxes and plates, painted from one palette texture with a dark wash at the feet and a dry-brushed highlight on top. One URP Lit material, made at runtime, with no texture file. About 3,600 triangles per figure. Real silhouettes for each headgear and rifle pose. The figures cast and receive shadows. Our own code, MIT.

Why (c): at the gameplay zoom a figure is about 90 px tall. What reads at that size is the silhouette (hat brim, rifle, stance), the colour blocks (khaki shirt, dark trousers, pale webbing) and the shadow. In the gameplay03 checkpoint the code-built trio matches the promo's trio in size (87 px) and reads with all three. They are exact to the sources: each part is a sourced item, and a change is a reviewed line of code. Era variants (1956 helmets, 1967 Uzis, 1973 flak vests) are parameters. Builds are deterministic, there is no licence question, and the build-size cost is code only.

Their limit: close up (the 6 km hero zoom) they look like wooden mannequins, with faceted limbs and no faces or folds. The promo's close shots (0:21) show sculpted miniatures. The way past that limit is the owner's close-zoom track, which plugs authored models in through the art hook above.

Scale

Measured on the promo scaled to 1920 px wide: a rifleman is about 87 px tall, inside a ring about 190 px across. Our gameplay03 pose frames the same hex count as the promo (about 9 across: see PromoPose). The first checkpoint, at 2600x, drew our riflemen 113 px tall, so the first round used 2000x. The ring is at 0.66 of a hex's circumradius. The aircraft (x420, a 4.7 km span) and the ship (x95, 6.0 km) are sized against the figures, as Civ sizes them, not against each other.

Owner ruling (2026-10-11), superseded on 14 October (below): units stay Civ-style tokens at about 2,000x. This is far above the 10–25x ruling for buildings and trees. At 10–25x a man is 18–45 m tall, which is under one pixel at 60 km. The promo, which wins over strict scale by the owner's ruling, draws units as tokens. The scale is one constant.

Owner decision (2026-10-14): figures at 1.8 times a block, as T29; it supersedes the 11 October ~2,000x ruling. 1,400x (refs #197). T29's figures stand about 1.8 times a house. At 2000x ours stood about 2.5 times the White City blocks beside the showcase trio at close30 (the T54 audit's measure). The block is the kit's seven-storey heart block, 7 x 3.3 m at the city blocks' 60x, about 1.4 km; on the close30 render it measures about 1.3 km against the figure. So UnitShowcase.FigureScale = 1400: a man about 2.5 km tall, 1.8 times that block (UnitShowcase.HouseRatio, pinned to 1.6–2.0 by TheFiguresStandAbout1Point8TimesAHouse). A one- or two-storey village house (100x) is 0.33–0.66 km, so against those the figures still stand about 4–7 times as tall. The token's ring and its spot are unchanged, so no unit moved; the figures are drawn closer together (below), inside the ground the placer checked.

The base and the flag

Demo2 T29 (owner's reference) stands every unit on a round earth base ringed in bright white and flies the side's flag on a pole. Measured on 4x crops of three of its tokens (the 1:12 frame): the base (the ring's outer edge) is 1.41, 1.62 and 1.71 times the trio's width; the trio is about 1.15 figure heights wide (shoulder to shoulder); the ring is about 8% of the base's radius wide; the flag's cloth is 0.47–0.60 of a figure's height tall and 0.65–0.70 wide, on a pole about 1.5 figures tall standing at the base's back edge. UnitPlinth and UnitFlags draw the same from code:

  • Figures: drawn at UnitShowcase.DrawSpread (0.19 of a hex's circumradius between neighbours), shoulder to shoulder, the trio about 1.15 figure heights wide. The placer still checks a spot at TrioSpread, so no spot moved. The showcase trio stands the same way.
  • Base: one draped earth disc per army token (see "Slopes"), its ring 1.34 times the full trio's width across (UnitPlinth.BaseToTrio, 1.45 until #2430; pinned to 1.3–1.7 by TheBaseHugsTheTrioAsInT29). A compact token's figures stand at 0.6 of the spacing (UnitShowcase.CompactSpread; half until #2430 merged them into one shape), and its ring still fits its town's clearing (#588). The earth is the palette's Earth, dark and warm as T29's, mottled by smooth value noise over the world in its shade rows (no texture). Owner, 2026-10-15: this base (earth 0.45, 0.33, 0.22), the lighter of the two rendered, over a darker one (0.38, 0.28, 0.18) that brought the ring nearer T29's contrast but left the figures lighter than their base.
  • Ring: a band on the ground just outside the base, from 1.02 to 1.10 of its radius, each vertex over the highest ground across the band (the selection ring's rule), drawn unlit in white with the selection ring's shader. The selection ring is unchanged and still appears only on selection.
  • On renders (TheBaseRingAndFlagReadAsT29OnARender: a 30 km look at Harel's hill token and at the showcase trio, through the game's camera, measured as T29 was): see the numbers in PR #559. The bands are T29's own measures, set before measuring ours. The figures read darker than their base, as T29's (0.5–0.9 of its luma), at those looks and at gameplay03 (TheFiguresReadDarkerThanTheirBaseAsT29, #645: the shirts at 0.9 of #559's paint, read beside the chest brace, which #559's sample point hit).
  • Flag: by the unit's side in units_1948.yaml. Israel, Transjordan, Egypt, Iraq, Syria and Lebanon fly their sourced 1948 flags (below); the Arab Liberation Army, the one force the map draws with no sourced flag, flies the neutral pennant (Q-ART-1, owner, 15 October 2026: "a neutral pennant now, replaced by a sourced emblem when one is found"; never its dagger emblem, #226 ruling 8): a plain grey triangle, in no side's colours and not white (UnitFlags.BuildPennant). Every flag's cloth is 0.72 of a figure's height wide, its height its own law's proportion (Israel's 0.52, the others below; the pennant 0.36 at the pole), on a pole 1.5 tall at 0.9 of the base's radius, behind the trio and east of the badge (TheFlagIsAboutHalfAFigureTallOnTheBasesEdge). The cloth is drawn unlit (white, and a blue standing for the proclamation's "dark sky-blue"), so it reads white under the map's light and grade as T29's. Owner ruling (2026-10-15): the flag turns about its pole to face the camera (UnitFlagFacing), so it never shows edge-on as the camera orbits; the pole stays upright at the base's edge. The figures and the pole turn about the base's middle with the camera's yaw (UnitLayer.FaceCamera, #644), so from every side the trio stands side by side and the pole behind it (TheTrioReadsSideBySideAndThePoleStandsBehindItAtEveryYaw).
    • Use of the flag: the Flag and Emblem Law, 5709-1949, s. 2(a), forbids using the State flag for commerce, business or profession without a permit from the Minister of the Interior (the 1949 text, read on the cited scan; the amended text not checked). The game is free and non-commercial, and there is no paid release (owner ruling, 2026-10-22, #30). Only a side whose 1948 flag is sourced here gets one, so no flag is invented:
    • Israel: the Provisional Council of State's "Proclamation on the Flag of the State of Israel" of 25 Tishrei 5709 (28 October 1948), Iton Rishmi No. 32, 10 Heshvan 5709 (12 November 1948), p. 62 (https://fs.knesset.gov.il/0/law/0_lsr_340940.pdf, read 2026-10-14): 220 cm by 160 cm; white, with two dark sky-blue stripes 25 cm broad along its length, 15 cm from the top and the bottom edge; in the middle, equally far from both stripes, a Star of David of six sky-blue stripes 5.5 cm broad forming two equilateral triangles whose bases are parallel to the stripes. The Flag and Emblem Law, 5709-1949 (Sefer HaChukim No. 8, 24 May 1949, p. 37, s. 1(a); https://fs.knesset.gov.il/1/law/1_lsr_208225.pdf, read 2026-10-14) makes that proclaimed flag the State flag "in any size". Neither text sets the star's size: it is drawn 69 cm across, as the construction drawing at https://en.wikipedia.org/wiki/Flag_of_Israel (checked 2026-10-14), which fits the gazette's own drawing. The same design hung at the declaration on 14 May 1948 (https://en.wikipedia.org/wiki/Israeli_Declaration_of_Independence, checked 2026-10-14), so it flies from the game's first day. Pinned by TheIsraeliFlagFollowsTheSpecification.
    • Transjordan (#756): the Organic Law of 16 April 1928, Art. 3. Egypt: Law No. 47 of 1923 and its construction annex. Iraq: the Organic Law of 1925, Art. 4. Syria: the Constitution of 1930, Art. 4. Lebanon: the Constitution's Art. 5 as amended on 7 December 1943 (owner, 31 October 2026: "we need to have the flags of all arab forces like we have for the israeli ones"). Each flies its national flag (no source was found for a field war flag of any of them). The citations, what each law leaves open and the drawing each open point follows are in UnitFlags' class note; the research is .superpowers/coordination/arab-flags-1948-research.md. Every flag is 0.72 of a figure long with the same top; its height is its law's proportion (2:3 for Egypt and Lebanon, 0.48; 1:2 for Transjordan, Iraq and Syria, 0.36). Lebanon's cedar is traced from Wikimedia Commons' public-domain File:Flag_of_Lebanon.svg and drawn all green, the owner's choice where the sources on its 1948 colour conflict (31 October 2026, "All green (Recommended)"; UnitFlags.LebanonCedarColour). Pinned by EverySourcedFlagHasItsColoursAndItsLawsProportion and one test per flag; the local forces have Tal Heuberger's irregular as their figure (owner, 2 November 2026: "Use it instead of C09") and no sourced flag, so a local-forces token would fly the neutral pennant; no local-forces row exists yet, so none is drawn (TheFlagIsTheSidesAndOnlyASourcedFlagIsDrawn).
  • The showcase trio (?units, -units) stands at the land hex nearest its old spot whose base keeps off the railways and roads, and the kits keep a clearing there in such a run, as they do for a unit in a town (MapScene.ShowcaseClearing, UnitShowcase.Clearing), so no kit building or route crosses its base (NoKitBuildingOrRailCrossesTheShowcasePlinth). A plain screenshot keeps every kit.
  • Aircraft and ships have no base and no flag: they fly or float, and carry their own sourced markings.
  • Cost: per army token, a draped top (the terrain cells under it, see the PR), its lip and its ring (384), built once per roster change and freed with the tokens; one shared flag mesh (about 250). The base, lip and pole in the units' palette material; the ring and the cloth in three unlit colours. No texture, no new shader, no per-pixel cost (docs/visual-v1.md, #391), and no per-frame code.

Colour at map scale

The uniform's colours (shirt, faded shirt, trousers, bush hat in UnitPalette) are darker and browner than the khaki drill of the photographs. This is a readability choice at map scale, not a new historical claim: the 1948 photographs are black and white, and the kit itself (khaki drill, 37-pattern webbing, the three headgear) is unchanged and sourced below. Measured on editor renders of both tiers (gameplay03, the showcase trio at 20 km, the 7th Brigade at 15 km, Golani at 12 km, Harel at 15 km in the Judean hills and Etzioni at 30 km), a figure was 0.68–0.82 of the surrounding ground's luminance with the old, lighter khaki, and blended into it; it is now 0.50–0.57, and warm (red above green). Promo 03's figures are about 0.45 (figures round 1, refs #197). The webbing stays pale, as the promo's straps do. The token's base is painted earth (Earth), lighter than the uniforms (it replaced round 1's per-figure slope base). The aircraft and ship colours are untouched.

Sources (checked 2026-10-11)

Uniform and equipment, 1948:

Aircraft:

  • Israel bought 59 Spitfire Mk IXs from Czechoslovakia in 1948–49; IAF Spitfires fought in the 1948 war (the last air combat was on 7 January 1949). https://en.wikipedia.org/wiki/Supermarine_Spitfire_operational_history
  • Ferry flights of the Czech Spitfires to Israel (Operation Velvetta, 1948). https://en.wikipedia.org/wiki/Operation_Velvetta
  • The Avia S-199 flew first, on 29 May 1948 (Operation Pleshet). https://en.wikipedia.org/wiki/Avia_S-199 The Spitfire was chosen over it because the promo shows a Spitfire (promo beats era, and both are 1948-correct).
  • Spitfire Mk IX shape (a four-blade propeller): the IAF Museum airframe at Hatzerim. CC BY 2.5. https://commons.wikimedia.org/wiki/File:Spitfire-MkIX-hatzerim-1-1.jpg Clipped wing (corrected 2026-10-12): "All Spitfires ferry flown to Israel during the 1948 war were clipped wing Mk.IXe versions" (Efrati p. 14). The clipped tips cut the span by 3 ft 6 in (https://en.wikipedia.org/wiki/Supermarine_Spitfire). The model now has the clipped wing, not the promo's full elliptical one.
  • Variant LF.IXe, and 101 Squadron's red spinner and red-and-white striped rudder. On the Spitfires these are attested for January 1949 (IsraDecal), and from early October 1948 by Efrati (below).
    • IsraDecal Studio, "Spitfire Mk.IX, Israeli Air Force" (IAF-123, 1:24) marking instructions, May 2023. It lists "Spitfire LF.IXe 2011/26, 101 Squadron, January 1949", painted in RAF colours, with a red spinner and the rudder painted "in White. Paint Red stripes". https://www.isradecal.com/_files/ugd/ddb4b4_42f20e2bc83f4ba895c816e97ad03151.pdf
    • Eduard Info 11/2025: "In preparation for Operation Ten Plagues (October 15-22, 1948), later renamed Operation Yoav, 101 Tayeset fighters obtained high visibility markings consisting of red spinners and red and white stripes on the rudders". https://info.eduard.com/en/11-2025-1/strana-53 The page shows the squadron's Avia S-199s ("Messers"), not its Spitfires. So it dates the squadron's markings to October 1948. On the Spitfires, Efrati p. 12 confirms that date (below).
    • A Fundekals decal review also says 101 Squadron "had the white and red stripes on the rudder". https://mail.modelingmadness.com/scott/decals/fundekals/fd48idfspit.htm
    • Diagonal (corrected 2026-10-12): from early October 1948 the squadron painted "their spinners red and rudders with diagonal red and white stripes" (Efrati p. 12, with the photographs of Spitfires 14 and 17). This also dates the Spitfires' markings to October 1948.
  • Roundel of 1948–49: a filled blue Star of David on a white disc, the star clear of the disc's edge, with no outer blue ring.
    • Filled, not an outline (checked 2026-10-12, after the owner asked).
      • Efrati, p. 6, on the Bonanzas after 15 May 1948: "Ground visibility of the open star was poor so he [Boris Senior] suggested that the Star of David be filled resulting in the IAF insignia as applied on IAF aircraft ever since."
      • Every 1948–49 photograph checked shows the star filled: the S-199s D-120 and D-123 and the Spitfires 14 and 17 (Efrati pp. 11–12), and the public-domain S-199 photographs on Commons (below).
      • An open, outline star was used only briefly before that change. The model draws the filled star, and UnitsInPlayTests.TheRoundelIsAFilledStarOfDavidOnAWhiteDiscWithNoRing pins it.
    • Where the roundels sit:
      • Spitfire: both fuselage sides and both wing surfaces. The Velvetta Spitfires arrived with over-wing roundels painted over their Czechoslovak ones (Efrati p. 14).
      • S-199: the fuselage sides and under the wings (see "The Avia S-199" above).
      • On the fuselage a point is straight up, as in the photographs.
    • Flags of the World, citing a 2002 survey of Born in Battle: "almost all the roundels show the star within, but not reaching the edges of, the white disk ... This includes early planes from the 1948/49 war of independence (like Spitfire ...)". It adds: "the only aircraft showing a roundel with a thin blue ring around are Mosquitos and Meteors from the 1950ies". https://crwflags.com/fotw/flags/il%5Eairfo.html
    • The Commons "early variant" SVG with the ring is drawn from a Mosquito photograph, a 1950s type, so it is not used.
  • The camouflage, RAF Ocean Grey and Dark Green over Medium Sea Grey, is now confirmed for "all Spitfires ferry flown to Israel during the 1948 war" (Efrati p. 14; it was a single hobby source before). Alex Yofe's Spitfire Mk. IX in Israeli Air Force Service 1948–1956 has still not been checked.
  • Yoav Efrati, Colors & Markings of the Israeli Air Force (IAF, 2005), http://files.iaflibrary.org.il/DigitalLibrary/English/ColorsMarkings.pdf: the IAF's own library publishes this reference, which these marking corrections rest on (pages are the book's own numbers).

Navy:

Avia S-199 (units in play):

Brigades, the squadron's bases and the ships: every position, date and coordinate has its own sources and check date in content/map/units_1948.yaml.

The photographs are references only: no image file is copied into the repository.

Friendly destinations and drag onset (#1661, owner rulings 26 October 2026)

Selecting another friendly unit or a city while a movable unit is selected keeps the normal selection and offers "Send {unit} to {place}" for the previously selected mover. Activating that confirmation gives the existing Move or go-to command immediately; the confirmation itself also confirms a far route on touch. A new selection, state change (including a turn or loaded session), or screen opening discards the proposal. Right-click ordering stays immediate. The English and Hebrew confirmation share the same placeholders; the Hebrew is native-review-pending.

Map drags retain the 30 device-pixel threshold. Each press captures its own origin from input events. On the first crossing, only the part of that movement segment outside the threshold pans; subsequent motion pans normally, with no replay of held travel. The tap rule and complete between-frame touch swipes are ruling 3 (#1777).