List of Routines | Shun Games

The entries below are those shown on the game’s “Thinking Routine Fine Tuning” screen, with the same names and descriptions as in the game. The list is generated from the game’s data, so it changes whenever the game is updated.

Shelf	Entries	Default
Default On	59	On
Rulings	69	On
Candidates	5	Off
Dormant	69	Off

Every entry can be switched on or off separately for each side.

How to read the measurements

Each entry carries the same record of measurement that the game’s settings screen shows. The tags mean the following.

better on no difference worse on not measured

Three figures in a row, such as “+3.3/+8.3/+1.7”, are the difference in win rate, in percentage points, between the side with the mechanism and the side without it, for equal forces, superior forces and inferior forces, in that order. “N=30” is the number of games played per condition.
“No difference” does not mean “has no effect”. Where several mechanisms can stand in for one another, switching off only one leaves the others to take up its work, so a comparison of one at a time shows nothing. In one measurement, switching off seven of the entries carrying this tag together lowered the win rate by 16.7 points with equal forces (30 games per condition).
“Knobbite” is a check, board by board, of whether switching a mechanism off changes the result of the game at all (“bites” means that it does).
On the Rulings shelf, the measurements are reference material and not the grounds for adoption, so no tag is shown.
Where no measurement is recorded, nothing is written.

Default On

Mechanisms that are active by default. Those belonging to the same idea of how to fight are grouped into a single entry. Disabled items are dropped for this side's PC only. Default is all on.

Strategy 7

Watch The Clock not measured: Raises or lowers aggression with the score and turns leftMeasured8-23 knobbite: tempohold 5/d3109, tempo 4/d2164; tempowin Guandu only d6
Never Give Up not measured: Keeps fighting in character once the game is lostMeasured8-23 knobbite: Gettysburg and Guandu only, d4-9 = practically inert
Work Backward From The Win better on: Works backward from the score and turns left to decide whether to hold or push for objectivesMeasuredN=10 passive while behind 82.5%
Step On Victory Points better on: Rewards moves that step onto a neutral or enemy-held victory point in proportion to its valueMeasured8-23 knobbite: bites on 6 boards (max d1824 Gettysburg); verdict not measured
The map the high command reads not measured: The strategic map reads what a briefing tells a human player: how many points each location is worth, where the supply depots are, and when reinforcements arrive. Until now it saw only where the victory locations sat - a fifteen-point location and a five-point one pulled equally hard, no depot counted at all, and troops still waiting off the board were measured as though they already stood on it
A plan before the first move not measured: Before the fighting starts, the board is read through once and a single answer is settled on: how is this battle won? By taking ground, by holding it, by one decisive stroke, by cutting the enemy's supply, or by wearing them down. Along with it comes a decision about haste, and that decision governs which routines the army uses. A side in a hurry pushes forward even where the line frays; a side with time keeps every unit that has taken its post on the line. Until now nothing read the board through at the start - every judgement was made afresh from whatever the current turn happened to look like
Headquarters (Concentrate, Break Through, Envelop) not measured: At the outset the AI reads the map, divides the field into three sectors perpendicular to the line joining both armies' centres of mass, and decides for each side which sector to hold and which to make the main effort. The main effort is chosen by the race to concentrate (own strength that can be brought to the front within two turns against enemy strength that can arrive within one) together with the victory points obtainable there; close alternatives are chosen by lot. The battle is divided into four phases: preparation, breakthrough, envelopment and closing. In the first two, all units except those left to hold are moved together to the main sector, and holding units receive only the remaining activations. In the last two, enemies within the main sector are surrounded and struck. A side that wins by holding stands on the defensive everywhere; a side able to envelop (wider and stronger) attacks along the whole line.MeasuredUnder measurement.

Operations 35

Anchor On Rivers not measured: Forms up behind a river before contact and refuses flanks on itMeasured8-23 knobbite: only on boards with rivers (precontact 6/d2354, river 5, rivergate 4, riverratio & rivercomplete 3)
Read Enemy Intent better on: Estimates enemy reach and stance, then adjusts its ownMeasuredN=100 +3 to +4.5pp
Cut Escape Routes better on: Seals retreat paths with ZOC and focuses fire on trapped unitsMeasuredN=100 kill rate +0.65pp
Concentrate On Main Effort no difference: Picks a decisive sector and pulls troops along that axisMeasuredN=60 neutral; 8-23 Gettysburg AB axis score -34, strategyaxis +5.0 = both noise
Shift Force Between Sectors better on: Thins out quiet sectors and sends the surplus to the decisive pointMeasuredN=100 +8pp
Keep A Reserve better on: Holds units back and feeds them into collapsing sectorsMeasuredN=30 +1.7 when outnumbered
March To Distant Ground no difference: Sends idle units toward thinly held enemy areasMeasuredadopted on qualitative grounds; 8-23 knobbite: march 4/d2292, stratmarch 2, marchlock Guandu only
Hold A Continuous Line better on: When forces are near even, receives the attack on a gapless lineMeasuredN=30 +10.0 / -6.7 / +10.0
Shave The Shoulder better on: When stronger, widens a breach by chipping at its edgesMeasuredN=30 +3.3 / +8.3 / +1.7
Refuse Or Extend Wings better on: Folds weak wings back and stretches strong ones to envelopMeasuredN=100 +5 extending; 8-23 Gettysburg AB wingfold score -16.6 = noise
Close Gaps In The Line not measured: Keeps spacing and pulls stragglers back to the frontMeasured8-23 knobbite: slot all 8, linereturn 6, line 5; *frontslots inert on all 8 boards
Defend Victory Points better on: Sends the nearest spare unit back when a held VP is threatenedMeasuredN=30 +10.0 at parity
Prevent Sudden Death better on: Plugs the instant-loss hex with ZOC and keeps the marked unit clearMeasuredN=60 enemy reached 77→48%
Strike Undefended VPs better on: Sends spare strength to points the enemy has left openMeasuredN=100 +2 to +4pp
Hold Key Ground no difference: Refuses to give up high value ground even at bad oddsMeasuredN=100 +2pt (neutral); *8-23 Gettysburg AB score -213 t=-2.21 (uncorrected) = default-ON but sign may be wrong; needs checking
Fall Back In Stages not measured: Withdraws a broken sector line by line instead of routingMeasured8-23 Gettysburg AB retreat score +180 t=+1.67, rearguard +13.7 = both noise
Consolidate When Few no difference: Massing up to defeat the enemy in detail once thinned outMeasuredN=100 neutral (no regression)
Plan The Attack not measured: Converges several units on one point and chains attacksMeasured8-23 knobbite: attackplan/chainorder all 8 boards, forcedcombat 5; verdict not measured
Avoid Bad Attacks no difference: Declines low odds attacks and minimises losses when forcedMeasuredN=50 within noise
Work Out Must Attack better on: Repairs only the offending units when the duty cannot be metMeasuredadopted; 8-23 knobbite: bites on all 8 boards (max d1255 Gettysburg)
Advance To Close The Ring not measured: Prefers post combat advances that complete an encirclementMeasured8-23 knobbite: advencircle bites on 5 boards (max d1972 Sekigahara); verdict not measured
Value Strategic Terrain not measured: Weighs bridges and chokepoints, abandons unreachable goalsMeasured8-23 knobbite: bridge 7/d1719, costfield 6; choke synthetic only; *abandon inert on all 8 boards
Counterattack From Defence better on: Lets defending units strike back when the chance appearsMeasuredN=100 -8 to -10pt if off
Allocate Activation Slots better on: Spends the activation slot budget on the moves with the best payoff, mission, or contact priorityMeasuredN=30 +3.3 to +6.7pt at parity (actauto)
Build The Foundation better on: Reads a cut separating the enemy from VPs and the rear, then pulls units to posts along itMeasured8-23 knobbite: bites on all 8 boards (max d759 Shiloh); verdict not measured
Withdraw Before Contact not measured: Once a side has decided to withdraw, units that are neither in contact nor assigned a post fall back in bounds from the first turn. Active only during a withdrawal.
Steer Reserves To Their Sector better on: Draws reserves toward the highest-threat sector instead of spreading them along the whole lineMeasuredN=30 ratio 0.4 best, decisive back to 100%
Grab CP Objectives better on: Sends idle units to CP objectives that raise or lower the activation budgetMeasuredN=30 x3 seeds +3.3 to +5.0pt; 8-23 Gettysburg AB score -42 = noise
Intent to Wake Standby Forces not measured: Decides at the start of each turn whether to wake an enemy group whose waking would also wake a linked friendly standby group; if not, no unit moves next to it, fires on it or advances beside itMeasuredNew on 10-04 (Takeda and the Hojo river line at Kato Ichiran). Not yet measured
Standby Forces by Remaining Sleep not measured: Treats an enemy standby group that wakes on a timer according to the turns of sleep it has left: a group about to wake is regarded as present and is not avoided; a group that will sleep for some time is not sought outMeasuredNew on 10-11. Games waiting to the last at Bowang 19%→0%; games waking Lü Bu early at Yangren 59%→12% (N=402 each)
Offensive Doctrine Marches To The Decisive Sector not measured: Under an offensive doctrine, units without a post and freshly arrived reinforcements aim at the main battle area rather than at a single scoring hex. Units holding a post, meaning a garrison on ground already taken, are left in place
Pipeline (Unified Intent) not measured: Orders from strategy to operations to tactics are issued from a single ledger, and results are reported back up so that intent stays unified. Units holding a line post keep their usual movement, and a side whose plan is to hold does not use it.MeasuredWith one side enabled across 22 boards, win rate rose 6.6 points (attackers +3.2, defenders +1.8) and units lost fell by 1.0 per game. The Gettysburg fishhook formed in 48% of games (55% without it; within chance). At Nancy, 76% fell to 70% over 100 games (within chance).
Bring Units Forward from the Rear (Combat Requires No Activation) not measured: Combat does not consume activations. Units already in contact with the enemy fight without one, so activations are spent on bringing rear units forward. The rule is to use the rearmost unit that can still arrive in time. Forward units move first only when an important hex that the enemy can reach next turn can be reached this move by forward units alone. In this way more units take part in combat than the limit alone would allowMeasuredPassed the board review at both An and Waterloo (at An, rear units flowed forward; at Waterloo, the British held the ridge and the French no longer bunched up). It has no effect on boards without an activation limit.
Forced March When Time Counts, Rotation Otherwise not measured: Activations go first only to moves where time itself counts: taking a victory or command point that no enemy occupies; closing on an enemy-held point in the final third of the game; stopping an enemy withdrawal off the board; reaching a river line before the enemy; and fixing an isolated enemy unit before it joins a line. All other units are brought forward from the rear in rotation as beforeMeasuredNew on 10-12. Share of qualifying units able to move (switch off → on): open objectives 28→100%, final third 66→97%, stopping a withdrawal 40→93%, river line 55→100%, fixing an isolated enemy 46→96%
Hold the Line, Strike the Weak not measured: A set of three, built from the designer's Sekigahara game record. (1) Nearby friendly victory points are linked by a path through hexes the friendly side reaches first, and a single line is drawn before this chain of positions. (2) In the line's reach race, enemy arrival is counted as never passing through friendly units and as stopping in friendly ZOC. (3) The movement plan is treated as a picture: for each unit, reachable hexes (next to an enemy or to an enemy's line of retreat) are tried in turn, and only changes that raise the score are kept over several passes. The score is the expected enemy losses from this turn's obligatory attacks, minus friendly losses, minus expected losses on the enemy's next turn, minus holes in the line.MeasuredSekigahara (Western side only, 32 games): mean margin -11 to -47; Eastern units destroyed per game 14.5 to 23.2, Western 14.4 to 14.1. Across 34 shipped scenarios (8 games each): the side using it improved on 27 boards as second player and 30 boards as first player. The Gettysburg fishhook still forms on turn 5.

Tactics 12

Avoid Encirclement not measured: Dislikes surrounded hexes, shifts to one with an escape routeMeasured8-23 knobbite: bites on all 8 boards (max d583 Shiloh); verdict not measured
Use Terrain Cover not measured: Values woods and hills when scoring a positionMeasured8-23 knobbite: bites on all 8 boards (max d750 Nancy); verdict not measured
Allow Holding Position not measured: Units stay put when the hex they already occupy is the best one, instead of being forced out into pointless shufflingMeasuredN=30 oscillation 61.2→37.7, win rate held
Discount Risk better on: When stronger, weighs dangerous positions more heavilyMeasuredN=30 +11.7 / +6.7 / 0
Rally Idle Units not measured: Brings units idling in the rear back to the main effortMeasured8-23 knobbite: Gettysburg (d1840) and Sekigahara only; verdict not measured
Review After Moving better on: Replaces suicidal or isolated placements with safer onesMeasuredN=100 +24 / +42pt
Avoid Losing Trades no difference: Stops pouring troops into a sector it is losing on exchangeMeasuredN=60 +1.6pp (neutral); 8-23 Gettysburg AB score +110 t=+2.09 (uncorrected, suggestive)
Take Firing Positions not measured: Artillery actively moves to hexes where it can shootMeasured8-23 knobbite: bites on 5 boards (max d2644 Gettysburg); verdict not measured
Discourage Backtracking better on: Penalises moving straight back to the hex just left, curbing oscillationMeasuredN=30 oscillation 19.2→7.8/game
Handle Standby Forces no difference: Avoids needlessly waking dormant enemy units and keeps enough strength watching themMeasured8-23 knobbite: only boards with standby units (sbguard Nancy/Nancy2/Sekigahara d626, sbsleep 2 Nancy boards)
Offensive Doctrine Attacks At Even Odds not measured: Under an offensive doctrine the attack planner commits at even odds (1:1) instead of requiring 2:1. It still aims for 2:1, piling spare units onto the target to raise the ratio. Where adjacency forces combat, an attacker cannot wait for the odds to arrive
Re-order stuck units not measured: A unit boxed in by friends is reconsidered on the board after they pull away

Combat 5

Order The Combats no difference: Resolves the best odds first and rebalances assignmentsMeasuredN=30 even
Suppress Overkill better on: Stops piling extra strength on a target already past 3:1 and sends the surplus to reserveMeasuredN=30 max ratio 6.33 (6:1 cap)
Pour Spares Into Best Odds better on: Sends spare attackers in auto-assign to the combat with the best oddsMeasured8-23 knobbite: bites on all 8 boards (max d922 Gettysburg); verdict not measured
Fortress Units Attack By Choice not measured: Units in a fortress have no duty to attack: they are never used for soak-off attacks, join an attack only when it improves the odds, and do not leave the fortress to advance after combat unless the hex is a fortress or a VP hexMeasured2026-10-07 rule application (no duty to attack in a fortress); not measured
Account for Replacements not measured: On boards with replacements, an enemy destroyed with a line of retreat returns at half strength. The value of destroying it is discounted accordingly when choosing targets. An enemy with no line of retreat does not return and keeps its full value, so encircling and destroying is rated higher. Boards without replacements are unaffected.

Rulings

Behaviours the author decided on, because measurements alone cannot settle them: the effect varies with the situation. On by default; turning one off removes the intended behaviour.

Strategy 17

A Foe That Never Wakes Is No Foe: Standby forces wake in three ways. A timed group with a joining chance, and a group that wakes when the screen in front of it thins, both wake on their own and are counted at full strength. A group with neither can only be woken by stepping next to it, so it is treated as absent: it projects no threat range, and it is dropped from the effective force ratio and from the activation-rate denominator. Forced attack also stops aiming at it, except for a unit named as an objective. A group that wakes when its main force falls to a set share of its starting strength is counted by how close the main force is to that share.MeasuredA/B seed42, both sides on/off: Okehazama (N=200) 52.5 to 51.5, 14,315 aim skips and 7,008 threat-range skips; Nancy 2 (N=60) 36.7 to 38.3; Sekigahara unchanged (Nanguzan is screen-gated, zero firings). The three boards with no standby force are bit-identical
Cash In for the Instant Win: When holding everything it already controls still falls short of its threshold, and an instant win (a sudden-death hex or the enemy commander) is reachable within the remaining turns, a side drops its own held objectives from the line and sends those troops toward what it has yet to take and toward the commander. Defensive dispatch to its own points stops, and the cap on hunters is lifted. The judgement comes purely from the arithmetic that a point which cannot carry you to the threshold no longer contributes to winning. The odds gate for attacking the commander is also lowered from two-to-one to even, since there is no reason to wait for favourable odds against a target whose death ends the game.MeasuredN=200/60 seed42: Okehazama 50.0:50.0 (fires on 278 side-turns, sudden-death reach 1.0%); Guandu 2 91.7:8.3 (340 turns, reach 41.7%); Shiloh 46.7:53.3 (294 turns); Gettysburg has no instant-win path and never fires, so it is bit-identical. A crude two-flag version fired indiscriminately and swung Guandu 2 by 8.3 points; weighing the two routes against each other settled it
The Rear Is What Wins: What must be shielded is the cluster of scoring locations, not the board edge. Ground worth nothing may be given up.Measuredgettysburg N=3 score +44.7 to +68.3
Hold Only Your Own Ground: The line shields only what this side currently controls plus the ground it has decided to take. It no longer stretches into enemy territory trying to enclose locations the enemy already holds. Retaking them is the attack plan's job, not the line's.MeasuredGettysburg N=1: the cut did not move (the north-west wall was not there to enclose enemy-held ground). Kept on for consistency of the rule
Defenders Wait On Strong Ground: A side on the defensive stops pushing as far forward as it can still reach in time, and settles its line on defensible ground: hills, ridges, built-up areas. A side on the offensive still advances to the enemy's face as before.MeasuredWorks only paired with linefit. Alone, 9 maps N=100: Gettysburg blue seal 53.0 to 43.2 (the line moved onto the ridges but grew too long)
Contest Key Ground: Send troops to the locations the victory calculation names, richest first, and keep defending ground already taken while the enemy can still reach it (2026-08-21). Applies to the defending side too.Measuredgettysburg N=3 score +68.3 to +107.0
Order Troops To Key Ground: Stops picking whoever happens to be free and instead looks across the whole army for whoever arrives soonest, then orders them there. The ordered unit drops its current job, including a post on the line.MeasuredN=100 across 3 maps: captures 519 to 1766 / 79 to 1083 / 627 to 1462, Gettysburg blue seal 46.2 to 53.3, unnatural stance flips 58.3 to 44.7 per battle, fewer losses
Guard Objectives Only Under Threat: Sends a garrison to a friendly-held objective only when the enemy can actually reach it, so that no troops are spent on rear points the enemy will not approach.
Exit Off The Map: On boards where leaving the map scores points, it chooses which units exit and by which exit, sends them, and blocks and pursues enemy units attempting the same.
Read the Defectors: Reads the odds a unit turns coat over the rest of the battle and credits that share of its strength to the other side. Does not strike an enemy likely to join you, and does not post a likely defector on the line or as a sudden-death escort. It also reads the block on defection: while the receiving side has fallen to or below the set share of its starting strength, the unit is treated as staying put.MeasuredSekigahara N=30 blue 86.7 to 73.3% (1.8 sigma)
Time the Surprise: Decides afresh each turn when to spend the one surprise it has, by three tests taken in order. A short or light scenario leaves no time to wait, so it strikes from the outset. When few turns remain it strikes regardless, because a surprise never spent is the worst outcome of all. Failing both, it weighs the enemy line by comparing the fighting strength still standing against the strength at the outset: thinned, the moment has ripened and it strikes; still thick, it waits. Waiting does not mean holding back from a fight, it means staying out of reach, for contact compels combat and combat is itself the trigger, so a waiting elite shuns the hex beside an enemy, is exempt from the drive to close, and will not fire at a target merely within range.
Reinforcements on the Clock: Counts off-board reinforcements as runners by when they arrive, so the victory calculation sees them. Force ratio and the foundation already counted them; only the victory calculation did not.MeasuredGettysburg N=30 only +4 objective calls, rest identical
Besiege Enemy-Held VPs: Puts victory hexes the enemy already holds on the list of objectives. Taking one moves both your score and theirs, so it weighs more than a neutral hex. The old code treated them as a race the enemy had already won, so they never appeared as objectives at all.Measuredfires on 3 maps; win rate within noise
Strategic Read (3 Axes): The choice of posture is made on three axes instead of force ratio alone. First the clock (does holding win, or must the line be broken?), then an effective force ratio that folds in activation limits and off-board reinforcements, and finally each side's share of units it can move per turn, compared against the opponent's. Turning this off returns to the old force-ratio-only judgement.MeasuredNot adopted on win rate (postures are not chosen by win rate). Verified by the clock/posture breakdowns, the quality-of-defeat line, and the board itself.
Main Battle Check: Picks one place where the battle will be decided, from the enemy's mass, the influence map, strategic value, the heat on the line, and the win plan. This is the area the sector drive uses.MeasuredN=30 -5.0/-1.7/+6.7pt
Count the Enemy We Know Is Coming: Enemy reinforcements still off the board count as a threat, their arrival reckoned from the turn they are due and the hex they enter by. Until now off-board reinforcements did not exist until they stepped onto the map, so a force whose arrival is written on the scenario went unanswered until it was already there. Even where the detachment would appear was known, and the danger went unseen until that very turn.Measurednot yet measured (checked on the board)
Strike at the Commander Before His Army Gathers: Acts only when twice the strength around the enemy commander can be brought to him before his reserves are free to move. Until then the force waits beyond the distance that would rouse the troops around him; once it is assembled, it closes in a single turn, drives off his escort, and cuts his line of retreat before the final attackMeasuredOkehazama (Yoshimoto MP 4, Oda under AI), 20 games: Yoshimoto slain 40% to 80%; Oda victories 75% to 90%. It acts only in scenarios with a commander unit (Okehazama, Hastings).

Operations 43

Defend Outside The Pressed Zone: Does not try to hold inside the area the enemy is pushing into; the defensive line is set outside it.
Take Every Depot, Leave One Guard per Sector: Where a friendly depot (CP) can be reached by an enemy, including off-board reinforcements, within four turns, one guard per sector stays at its post; if no enemy can reach it, the post may be abandoned as before, and the other guards go to the front. After combat, units advance onto an enemy-held depot just as they would onto a victory point. A single unit is sent to take any vacant depot with no enemy on or adjacent to it; if it can arrive this turn, one unit per sector puts this ahead of its post, the forward drive and the march from the rear, while all others continue to the front.
Abandon Unreachable Objectives and Advance: A unit standing on a friendly-held victory point or depot leaves it and moves toward the enemy if no enemy, including off-board reinforcements, can reach that hex before the final turn. Reach is estimated from distance and movement allowance alone; zones of control and unit positions are disregarded, since both will change. Control is retained after the unit leaves. This takes precedence over posts, garrisons and guard orders, and such units receive activation slots first.
Mass the Surprise Force Before Striking: Once the surprise attack is released, the elite units do not engage one at a time. They first form up in front of the enemy, two hexes from the nearest enemy unit and just outside its zone of control, and strike together on the turn they are assembled. Units that cannot reach the line before the game ends are not waited for. They strike at once if the surprise has already been triggered, if an elite unit is already in contact, on the final turn, or after three turns spent forming up.
Objective Over Contact: Forced attack narrows a unit's options to squares that make contact, so a march toward a distant, far more valuable objective is dropped before it is ever scored. When the contact at hand is a fight over a small victory point and the named objective is worth at least twice as much, squares that close on that objective are added to the choice. No contact square is ever removed, and a plain field fight with no victory point at stake is left untouched.MeasuredA/B N=60 seed42, both sides on/off (blue delta): Okehazama -6.7, Gettysburg +6.7, Shiloh +3.4, Sekigahara +1.7, Guandu2 -1.7. Four boards move toward parity, none worse. Fires on 0.8-4.5% of forced contacts
Hold the Far Bank: Score rivers at the doubled combat value and count only the bank the enemy is on, so the line forms on our own side of the water.Measuredshiloh score +58.7 to -47.8, spread up
Lock the Line: On the defending side, a unit standing on its post does not move at all. Standing there is the job; a gap is an invitation.Measuredgettysburg N=3 score +107 to +84
No Line You Cannot Man: The length of the line is capped not only by how many troops exist but by how many can actually reach their posts before the enemy arrives. A line that can be afforded but never finished is not drawn.Measured9 maps N=100 (with holdground on): Gettysburg blue line-complete 67.6 to 93.1%, holes 2.06 to 0.18 per turn, seal flat at 52.5; Wuzhangyuan red 34.9 to 56.0, blue 25.8 to 34.2; Okehazama blue 74.6 to 93.0
Do Not Split The Army Over Scattered VPs: When friendly VP hexes are scattered rather than clustered, the army keeps its line instead of dispersing to sit on each of them.
No Falling Back To Rear Posts In Contact: While friendly units are engaged, a unit does not take a post farther from the enemy than its current position. Retreats and elastic withdrawals are exempt.
Leave Places We Can Reach in Time: Whether to send troops to guard one of our locations, or to keep a unit standing on it, is decided by a race between the enemy and us. If we can get there more than 2 turns before the enemy, we do not guard it now and the troops go to the front instead. A unit standing on the location stays only when the enemy can arrive within 2 turns.
No Returning To Unreachable Rear VPs: No sitting post is set on a friendly VP the enemy cannot reach in time; a line is needed only where the enemy will pass.
Attackers Hold No Line: Where an activation limit applies, the attacking side keeps only posts on scoring hexes and stops pushing its line forward. A limited budget is not spent on a line. On such boards this takes precedence over All-In Front (2026-08-21).Measuredgettysburg N=30: score +129.8 to +101.1 (1.7 sigma), stalls -27%
Attackers Garrison Only What They Hold: Where an activation limit applies, the attacking side keeps posts only on scoring hexes it already controls. A hex not yet taken is a target, not a seat: taking it, and holding it afterwards, is the objective routine's job. Control belongs to whoever stood there last, so an attacker scores by stepping on it and need not sit. Extends the 2026-08-17 rule, which failed to bite on boards with many scoring hexes (Gettysburg has 13).
Attackers Hold a Line from Mid-Game: Where an activation limit applies, the attacking side no longer reduces its posts to captured scoring hexes once the first third of the scenario has elapsed and two turns have passed since first contact, provided it has enough units for the line and a striking reserve. Units that reach the line may remain in place. The attacker's line serves to fix the enemy frontally, which is the precondition for passing through gaps and striking the flank and rear. Before contact and immediately after it, all units continue to attack as before.
Break Through in a Fan: The unit facing a gap in the enemy line leads the breakthrough, with up to two nearby reserve units on either side in a fan formation, and advances to a position that closes enemy lines of retreat. If no reserve is available, the leading unit proceeds alone. Having arrived, it remains behind the enemy line for several turnsMeasuredN=30 x 4 seeds. Gettysburg: 433 breakthroughs (155 in a fan formation), 57 reached their objective without being stopped (0.48 per game), of which 1 was counter-encircled and 47% made an effective attack. Austerlitz: 80 (23 in a fan formation), 9. No leading unit was stopped by the rear-area hunt
Attackers Ignore Terrain: The difficulty of rivers and heights is worth something to the side denying passage, not to the side attacking.Measuredgettysburg, 4 measured together: +102 to +175
Reinforcements Press On: A unit that enters spends the rest of its movement heading for the enemy, clearing the entry hex so the next may follow the same turn.Measuredgettysburg, 4 measured together: +102 to +175
Reinforcements Join the Line: Entering reinforcements and replacements head for an unmanned post in the line or, failing that, the hex directly behind a friendly unit in contact, and keep that destination until they have joined. They still spend all their movement moving forward.
Hunt the Sudden Death: Names any hex that wins outright on entry as the heaviest objective, puts enemy stacks holding a sudden-death unit first in the attack plan, and counts the enemy's own instant win as a route to reversal.Measuredshiloh N=30 enemy SD wins 11 to 7, outcome +10pt
Beware the Surprise: Knows which units on both sides are elite, and for an enemy elite force that has not yet spent its surprise, reckons its threat as reaching two turns deep, because such a force can move and strike twice in a single turn. The widened reach only extends around those pieces, so only friendly units nearby ever notice it. No fleeing and no counterpunch are added: the only counter is to endure, so this goes no further than nudging encirclement checks and the choice of post away from standing alone where the surprise can reach.
Hunt down infiltrators: Spots enemies that have slipped through a gap in the line and are running for a rear objective (victory point, supply point, sudden-death hex), plants a unit in front of that objective to block the road, and at the same time gathers overwhelming strength to destroy the runner. Pulls troops off the line if no spare ones are within reach. Active only against a human opponent.
Posture Sets Slot Order: Which roles get first call on the units that may move each turn now depends on the posture. An offensive army serves its strike plan and any unit that can reach the enemy first, and its defensive line last; a delaying army serves its posts first and gives nothing to a decisive-battle plan; a manoeuvre army puts the strike plan on top but gives no priority to a frontal contact. Turning this off returns to a plain offensive/not-offensive split that is applied only to whichever side has the larger force.MeasuredNot adopted on win rate. Measured at Gettysburg: under the old behaviour the Confederate side spent 1019 slots, 795 of them (78%) on the defensive line against 111 (11%) on the strike plan, and left another 883 unused.
All-In Front: Attack and defence are split by the win plan. On boards with an activation limit the attacking side follows Attackers Hold No Line instead (2026-08-21). The side that must take ground pushes its line right up to the enemy front; the side that wins by holding falls back as far as it can without surrendering a victory point it needs.MeasuredN=30 ±0.0/-3.3/+1.7pt
Front Line Check: Post spacing is no longer fixed. It takes the tightest the force can afford: shoulder to shoulder, then every second hex, then every third. Impassable ground, and holes a reserve reaches first, still count as a complete line.MeasuredN=30 -6.7/-1.7/+8.3pt
Drive Forward, Whole Board: Every unit on the board spends its movement to close with the enemy, including those holding a place in the line. If any destination makes contact, one of those is taken.MeasuredN=30 +11.7/+13.3/-5.0pt
Aim at the Line: When drive-forward and arriving reinforcements head for the enemy, they aim at their own post in the line of battle rather than the nearest enemy counter. A post sits between the enemy and our own objectives, so marching to it is marching at the enemy. Units with no post, and units already standing on theirs, head for the nearest enemy as before.Measurednot yet measured (checked on the board)
Measure by the Road: Distance to a post or objective is measured by the terrain cost of actually walking there, not by straight hex distance. In a straight line the far bank of a river reads as near, and a unit heading for it stalls on the near bank. Terrain does not move, so a distance field once built is reused for the rest of the battle.Measurednot yet measured (checked on the board)
Drive Forward, Sector: Only units inside the main battle area spend their movement to close with the enemy. Units holding a place in the line are exempt.MeasuredN=30 -5.0/-5.0/+10.0pt
Keep the Main Effort: A headquarters that jumps its target every turn is wrong. The main effort shifts only when the case is clear.Measuredwin rate neutral (+5.0pt did not repeat)
Budget by Sector First: Activations are shared out to sectors before units. Without the sector as a unit, 'hold still precisely because this is the fighting sector' cannot be expressed.Measuredunmeasured; adopted on military grounds
Shift a Neighbour: When a post falls empty, slide the neighbouring unit across before calling a distant reserve; the space it leaves becomes the next gap, and so on.Measuredgettysburg N=30 score down, spread 33 to 62
Hold What Arrived: A unit that has reached its post is shielded from other routines while a threat is near, and released once it is safe.Measuredgettysburg N=30 score up, spread 33 to 50
Form a Second Line: When defending while ahead, or holding many key locations, place a second rank behind the line.Measuredwas dormant; figures not on record
Price the Depot: A move that steps onto, or holds, a hex that raises our activation allowance or cuts the enemy's now scores in its own right. Until now a depot's worth appeared only in the table deciding which units get an activation slot, never in the merit of a destination, so a unit standing beside a depot had no reason to step onto it. The worth is measured on the same scale as victory points and discounted by the turns remaining, because victory points are counted at the end while a depot pays for every turn it is held.Measurednot yet measured (checked on the board)
Guard the Granary Too: A depot hex we hold counts as something to defend, exactly as a town worth victory points does. Until now only scoring locations were defended, so a depot worth no points appeared in neither the threat check, nor the test for whether a garrison is present, nor the bonus for staying put. The generals posted to guard a store of grain had no reason at all to remain there.Measurednot yet measured (checked on the board)
March to the Depot You Were Sent For: A unit sent to seize a supply depot walks toward that depot when it cannot reach an enemy this turn. Until now the order to close with the enemy aimed a unit at its line post, then the main battle area, then a key location, and a depot appeared nowhere on that list. The dispatch itself was written as a pull toward the target, but the order to close with the enemy is a shortcut that skips the scoring altogether, so that pull never once reached the unit it was meant for: at Guandu, fifty-one dispatches and no arrivals. Where the unit can reach an enemy, closing still comes first; what changes is only which way a unit walks when no enemy is within reach this turn.MeasuredN=30 seed 42 - Guandu: granaries held at the end 3 blue to 5 blue, depot arrivals 2 to 26, outcome split unchanged at 70.0/30.0. Gettysburg: depot arrivals 1 to 16, outcome split 26.7% blue to 53.3%
Nothing Left to Guard, Move On: A lone sentry on a location leaves its post only when two conditions hold at once. First, the nearest enemy cannot reach that location even by moving. Second, a location not yet ours lies within two turns' march. When both hold, the sentry always goes and takes it. If either fails it stays put as before, so a sentry with nowhere to go never abandons its post: it stretches toward the next location only while what it guards is genuinely safe. Scoring locations and supply depots are treated as different things. A scoring location is counted at the end, so its worth does not depend on when it is taken; a depot pays for each turn it is held, so it is not chosen as a destination when no turns remain to collect on it. A sentry on a location that would lose the game outright is left out of this rule.MeasuredN=30 seed 42 (blue outcome share, off to on, final version): Guandu 70.0 to 63.3, Guandu 2 90.0 to 96.7, Gettysburg 53.3 to 40.0, Okehazama 50.0 to 60.0, Sekigahara 30.0 to 26.7, Shiloh 50.0 to 40.0. Depot turns held at Guandu: blue 1913 to 2059, red 1982 to 1831. The ladder now outranks the line post (post overrides: 44 to 0). Cost: line-post occupancy falls 1.5 to 2.4 points on every board
The Sentry Does Not Leave His Post: Where a single unit is all that guards an objective, the bonus for staying put now opens as wide as the window in which threats are detected. The old window assumed a unit could leave and be back within two turns, but a unit drawn into the line does not come back. The window opened as the enemy drew near, and by the time it opened the sentry was already gone, missing by exactly one turn every time. Nothing is paid for an objective in the rear with no threat in sight.Measurednot yet measured (checked on the board)
Never Leave the Point Empty: A friendly objective the enemy can reach next turn is never left empty. The unit standing on that hex is ordered to hold it instead of leaving to seek a favourable fight. Until now the hold order was issued only when exactly one unit covered the point; with two or more, nobody was held. Since the number of defenders grows with the number of onrushing enemies, the more an objective was threatened the less likely anyone was to stay. Capture means stopping on the hex, so however many friends wait alongside it, an empty hex is one the enemy takes by walking on. Emptying a contested objective to go and punch elsewhere is abandoning the ground.MeasuredAn 589 BC, N=400: turns with an empty objective 3016 to 1671 (weighted by value, 73925 to 39415). Decisiveness moved from 54.2% to 60.5% blue and was brought back to 47.8% by +1 CS on the three Qi commanders.
Drive-Forward: Sentry: A third pattern of the drive-forward order. Where a single unit is all that guards a threatened objective, the order it receives is not "spend your full movement and close with the enemy" but "hold this place." It is not an exemption from the drive but a different order under it. Send the sentry forward and the objective changes hands without a shot. Only the destination is constrained to the objective itself, so a fight still happens whenever an enemy stands adjacent; what disappears is leaving the post to go and strike. The pattern lapses the moment a second unit arrives, and the unit returns to the general order. Turn this pattern off and the drive is absolute again.Measurednot yet measured (checked on the board)
Hunt down what breaks through: An enemy that slips past the line and runs for the rear is chased down and destroyed, whether the opponent is a person or the machine. Until now this only ran against human opponents; in machine-against-machine games not a line of it moved. The reason given was that the machine never breaks through in the first place - but measurement showed otherwise. Of the nine hundred and eighty-seven times the Union had its rear crossed at Gettysburg, every single one was a run at a victory location it held, and a quarter of them were the town itself. The last remaining exit of a unit about to be cut off is now guarded too: a unit with nowhere to retreat is destroyed outright the moment that square is taken. A side that has decided to hurry does not hunt, however. An army whose work is to press forward cannot spare troops for its own rear without wrecking the plan it is following.MeasuredGettysburg N=30 seed 42, one side switched off - both on: blue 50.0 / blue off: blue 50.0 / red off: blue 60.0. The side with it on does not lose in either direction (at N=30 three battles are worth 10 points, so the margin is not conclusive)
Hunt Enemies Behind the Line: An enemy unit standing in the rear area behind our front line is treated as having broken through, and becomes a target of the rear-area hunt even if it is in contact with our units or is not closing on an objective. If no objective is near, its own hex becomes the aim: blocking units surround it and others are gathered to destroy it.MeasuredN=30 x 4 seeds. Infiltrators destroyed by the rear-area hunt: Gettysburg 10 to 90, Sekigahara 5 to 260. Austerlitz: detections 12663 to 6954. Line fill within noise on all three maps

Tactics 9

The commander falls back: A unit that loses the game the moment it dies is defended only by standing where it cannot be hit. Penalising adjacent enemies and enemy reach alone merely stops it at the first hex that is not hit right now, so distance from the nearest enemy is itself rewarded. Given two safe hexes it takes the farther one, and stepping forward is never chosen. The reward is capped so it does not run to the board edge and abandon friendly support and defensive terrain.
Guard Only If the Enemy Can Arrive: A unit whose loss ends the game is given an escort only when an enemy can reach a hex adjacent to it before a friendly escort can, that is, only when the race to reach it would otherwise be lost. The enemy's arrival is measured along routes that pass through neither hexes occupied by friendly units nor hexes in friendly zones of control. While the line holds, no enemy can arrive; and while friendly units could still arrive first on the next turn, no units are withdrawn from the line. Enemy units already in contact cannot move and are not counted.MeasuredNot yet measured (implemented 2026-10-01)
Hold Objectives Only to Win the Race: Units are kept on, or sent to, a friendly-held victory point only when the enemy could reach it before friendly units can (a lost race; a tie is also guarded). Enemy arrival is measured along routes that avoid hexes occupied by friendly units and friendly ZOC hexes. A unit standing on the victory point is counted as needing at least one turn to return once it leaves. While the line holds, the enemy cannot reach it, and the unit is released to the front.MeasuredNot yet measured (implemented 2026-10-01)
Intercept Objective Raiders: When the enemy can reach a friendly-held victory point no later than friendly units, and the unit heading for it is isolated (no other enemy within two hexes), one or two uncommitted units that can move adjacent to it this turn with odds of 2:1 or better attack it instead of falling back to the objective. No more than one or two units are sent per objective.Measured10-07 6 boards x16: never triggered against the normal AI, games identical (no harm); against an all-out grabber 0.1 per game, -0.5±0.7
Attackers Never Halt: The attacking side never chooses to stand still. Standing still costs no activation, so the whole budget goes unused.Measuredgettysburg, 4 measured together: +102 to +175
Don't Stand on a Friend's Line of Retreat: Units avoid moving into the hex directly behind a friend who is in contact with the enemy. A retreat goes to one of the six adjacent hexes, but it cannot end in a hex already full of friends. Stand behind a friend and he has one fewer way out; fill them all and a retreat result becomes elimination. A move that would fill the hex is avoided strongly, one that still leaves room only lightly, and the unit is steered to the hex beside it. Hexes adjacent to the enemy are never penalised, so attacks are not held back. If there is nowhere else to go, the unit goes anyway.MeasuredGettysburg / Waterloo, N=20: no-retreat eliminations 804 to 791 / 551 to 538. The penalty changed the chosen move only 2 / 13 times in 20 games; it rarely gets the chance.
Resolve Contested Hexes: When units planned in the same turn aim for the same hex and the stacking limit sends some of them back, they queue for hexes that are being vacated, and any still left are reassigned by Monte Carlo trials so that the force spreads out. Always active for a side conducting a strike at the enemy commanderMeasuredGettysburg and Sekigahara, 40 games each: over 3,000 units sent back in total (about 80 per game). At Sekigahara the point margin moved from -7.0 to +17.5, so the victory line must be recalibrated.
ZOC Puzzle (Move the Front as One): Decides the destinations of units near the enemy together, region by region, rather than one at a time. Monte Carlo trials search for an arrangement that closes the enemy lines of retreat, keeps friendly units from being surrounded, and leaves no break in the chain of zones of controlMeasuredGettysburg, 40 games each, active for one side only: units of that side lost with no line of retreat fell by 15% and enemy losses rose by 15%, with the same sign for either side. No change at Sekigahara.
Don't Step in Front of a Full Stack: When moving to a hex the enemy can attack, a unit avoids it if a neighbouring hex outside enemy ZOC is full of friends. A retreat cannot end in a full friendly hex, so a unit pushed back there dies because of its own side, not because it was surrounded. The fewer free ways out remain, the heavier the penalty. Ways out closed by enemy ZOC are not counted; that is the normal cost of battle. If there is no better move, the unit goes anyway.MeasuredN=20: eliminations caused by friends, Gettysburg 185 to 151, Waterloo 196 to 153; of those, 'stepped in front of a full stack later' 84 to 48 / 95 to 65.

Candidates

Ideas that were measured and changed nothing either way. They are sound, so they are kept apart as candidates that could become the default at any time. Off by default.

Strategy 3

Judge Holding On Actual Score no difference: Decides whether holding wins the game from the score on the board, not from the projected scoreMeasuredN=100 across 6 maps: outcomes 45.0->44.0 / 61.0->59.0 / 50.0->56.0, all within noise
Pick Objectives By Our Own Arrival no difference: Lists a place as an objective based on whether our own troops can reach it within N turns, rather than at any point in the remaining gameMeasuredN=100 across 3 maps: drops 9.5-17.3% of objectives at 3 turns, seal rate 41.2->41.5 / 77.2->77.2 / 77.0->77.2, captures and outcomes unchanged
Send Whoever Arrives First no difference: Picks troops for a key location by how soon they arrive rather than by raw distance, so faster troops go even from further awayMeasuredN=100 across 3 maps: the set of troops sent changes in only 1.1/0.3/3.1% of cases, seal rate 46.2->46.4 / 76.2->76.2 / 80.9->79.8, Shiloh identical outcome

Operations 2

Refuse a Weak Flank not measured: When our strength at an end of the line is under 70% of the enemy's and the enemy can reach it within 2 turns, bend that flank back one step. Straighten it again once we reach 90%MeasuredGettysburg N=30 x 4 seeds: blue seal rate 62.3->61.4, line fill 14.9->14.95. Bending the flank does not shorten the line, so the line fill does not move
Keep Post Spacing In Phase not measured: Aligns the spacing of line posts with the previous turn so that posts do not shift from turn to turn.

Dormant

Ideas that were tried and made things worse. Off by default. Turning one on affects this side's PC only.

Strategy 9

Theater Plan not measured: Assigns each sector an attacking or holding role and follows itMeasured8-23 Gettysburg AB win -5.0pt, score -123 t=-1.84 = noise; still no verdict
Read Checkmate On Objectives worse on: Accepts defeat once the objective count is out of reachMeasuredN=100x2 -16pp
The Clock Comes From the Reinforcements not measured: Ends the scenario one turn after the last reinforcement arrives. Giving the attacker a long clock is neither kind nor fair: with enough turns anyone can push through, so the clock stops measuring skill.MeasuredGettysburg blue 96.7 to 83.3%
Decide When to Withdraw not measured: Withdraws the whole line when holding the present score wins, or when clearly outmatched; never withdraws when points must still be taken to win
Temporary Withdrawal not measured: A small, outnumbered force falls back for a few turns and then halts.
Time-Based Objective Defence worse on: Boosts an objective's defence based on whether the enemy can arrive within three turns, rather than at any point in the remaining gameMeasuredN=100 across 4 maps: boosts 44.5->27.3%, seal rate / captures / outcomes all unchanged
Count Reinforcements By Arrival Time worse on: Measures off-board reinforcements by when they can actually reach a place, factoring in their entry turn and entry hexMeasuredN=100 counter 0 -> 184234 firings. ON vs OFF gives 6660 vs 6658 boosts, near-identical
Only Race For What We Can Win no difference: Chooses places we do not hold yet by whether we can arrive before the enemy. A place the enemy garrisons is a siege, not a race, and is left inMeasured0 to 6 firings across the nine shipping maps. Almost no neutral locations exist (1 at Sekigahara, 3 at Wuzhangyuan), so there is nothing to act on
Only Send Troops Who Arrive Soon worse on: Sends only troops who reach the key location within N turns, falling back to the single fastest one when nobody makes itMeasuredN=100 across 3 maps: at 3 turns it defers 13959 troops, but Guandu red seal rate collapses 70.8->48.3 on both stances and captures fall 627->550

Operations 47

Monte Carlo Playout not measured: Advances both sides several turns for each candidate move and keeps only the better ones. Four thinking times: Light, Normal, Deliberate, Profound.MeasuredUsed to choose single-unit moves, all 15 variants across 22 boards stayed within the margin of error (-3.6 to +1.7 points); held pending. It takes about 2.2 times as long and overrides the standby-unit rules. Lookahead is to be applied at the operational level (ruling of 2026-09-22).
Concentration: Gather Forces in Advance not measured: Units without a post that stand where the enemy will prevail within a few turns are sent early to posts in the area the lookahead judges defensible.MeasuredNot yet measured.
Arrival Cut-off: Do Not Assign Orders That Cannot Be Reached not measured: A unit that cannot reach an order's destination within the remaining turns is not given that order; the next order in priority is tried instead. A unit that can reach none of them is held in reserve.MeasuredNot yet measured.
Lookahead: Take the VP Despite the Cost not measured: Even for a move judged dangerous, if the destination is a VP not yet held, the unit goes for it when the points outweigh the unit, when the game is near its end, or when the three-turn arrival race shows the position can be held.MeasuredFive boards, two seeds: win rate 48.8%, within the margin of error.
Monte Carlo: Read by Region not measured: The units moved in each trial are the allies of the same region (front) and the enemies in contact with them, rather than those within a fixed window around the destination.MeasuredNot yet measured on its own (asymmetric A/B pending).
Monte Carlo: Read Deeper with Spare Trials not measured: If trials remain for the turn, only the best few candidates are read again one turn deeper and re-ranked.MeasuredNot yet measured on its own (asymmetric A/B pending).
Enemy Intent: Measured Prior not measured: Adds weight to victory points that the enemy historically pressed hard, based on several hundred recorded battles on that map.Measured22 maps, N=30: diluted the effect of lookahead (not adopted).
Enemy Intent: Turn-by-Turn Prior not measured: Adds the recorded combat strength present on each hex at each turn as part of the perceived enemy threat.Measured22 maps, N=30: diluted the effect of lookahead (not adopted).
Cannae Reserve no difference: Thins the centre where strong and peels the surplus out to both wingsMeasuredN=30 +1.7 / -1.7 / 0 (neutral)
Race Against The Clock no difference: From midgame on (after 30% of the turns), sends idle troops to open objectives they can reach firstMeasured10-07 6 boards x16 one side: start at 30% +1.0±1.0 (within noise); starting on T1 negative on 2 boards (-4.0/-3.9); 0-1.3 chances per game
Midgame Objective Guard not measured: From midgame on only, sends one or two units from the line to guard an objective the enemy could take unopposedMeasured10-07 6 boards x16 vs all-out grabber: no difference by start (T1 -0.4, 30% -0.2, 50% +0.1, 70% +0.1, ±1.4)
All-Out Objective Grab not measured: Sends every unit not in contact to any open objective it can reach no later than the enemy, from the first turn, even from the line or an attack. Used as a test opponentMeasured10-07 48 boards one side: better on 22/48 (blue) and 21/48 (red), mean -0.2 (a coin toss); Gettysburg -30 and Shiloh -25 worse for both sides
Objective Grab by Value not measured: Sends a unit to an open objective only when the enemy cannot reach it before the battle ends, and only if its value exceeds that of any friendly objective left unguarded by the departure. Units in an attack plan or under holding orders are not sentMeasured10-07 48 boards one side: 113 dispatches, +2.6 points each against the baseline, 35% lost (all-out grab: -1.4 points, 58%). Overall +0.5±0.6; no change in 75 of 96 cases
Active Feint not measured: Draws the enemy reserve with a decoy, then strikes the thin spotMeasured8-23 knobbite: Gettysburg/Guandu/Shiloh (max d451); verdict not measured
Elastic Defense no difference: Falls back to the next line where it is losing the exchangeMeasuredN=120 neutral to slightly negative
Seal The Back Door worse on: Plugs the gap behind the line before the enemy can turn itMeasuredN=100 lost on opportunity cost
Concentrate When Thin no difference: Treats a poorly manned frontage as the few-troops case and concentratesMeasuredN=30 0 (seeds disagree)
Fill Gaps In The Line worse on: Reads empty hexes as holes rather than spacing between unitsMeasuredasym -11pt
Payoff Of The Outer Arc worse on: Values a position more highly the further the encircling arc reachesMeasured*8-23 Gettysburg AB (120 pairs): win -15.0pt z=-3.29, score -409 t=-3.26 = harmful even after multiple-comparison correction; dormant is right
Avoid Exposed Posts not measured: Will not occupy line posts the enemy can reach (only when holding a line)Measuredbundle part (not measurable alone); 8-23 knobbite: bites on all 8 boards (max d1506 Shiloh)
Curl The Wing no difference: Bends the end of the line forward to wrap around (only when holding a line)MeasuredN=30 0/0/+1.7 (redundant); 8-23 Gettysburg AB score +4.7 = noise (this is the only board it bites on)
Weight The Enemy Main Effort worse on: Prioritises objectives in the sector the enemy is committing toMeasuredN=60 -15pt
No Doomed Plugs worse on: Will not spend a unit plugging a hole where it dies at onceMeasuredN=100 better off: +3 / +9
Choose The Main Effort By Value no difference: Picks the main effort by strategic worth of terrain and objectivesMeasuredneutral on all metrics; 8-23 knobbite: synthetic/Gettysburg/Guandu/Sekigahara (max d493)
Press The Encirclement no difference: Doubles the bonus for encircling moves while on the offensiveMeasuredN=100 neutral (signs disagree); 8-23 Gettysburg AB still d0.00 at 120 pairs = noise
Absolute Front Line worse on: Plugs front-line gaps ahead of every other routine, whatever the force ratioMeasuredN=30 -5.0/+5.0/+3.3 (no gain); 8-23 Gettysburg AB win -5.0pt, score +31.7 = noise
Arc Contract no difference: Assigns only truly spare units, on a deadline, to close an encirclement on a pinned enemy stackMeasuredsolo +3.3pt (kept off by decision); *8-23 AB N=10 inert on synthetic board
Pre-prune Post Seed worse on: Seeds next turn's post spacing from the line as it stood before the attacker's posts were prunedMeasuredN=100 foundation fill: Gettysburg 21.5->21.5 / Guandu 18.5->17.3 (-1.2pt)
Additive Post Stickiness worse on: Keeps a post's previous holder if it arrives within one turn of the best candidate, instead of within 1.5 timesMeasuredN=30 foundation fill: Gettysburg 23.1->21.5(slack1) 20.0(slack2) 19.2(slack3), all down
Freeze Post Assignments worse on: Never moves a unit to another post once assigned, until its post disappearsMeasuredN=30 foundation fill down on 4 of 5 maps (Gettysburg 23.1->19.6) though churn fell 20-40%
Garrison Own Objectives Even If Late worse on: Posts a garrison on an objective the side already holds even when the enemy is expected to arrive firstMeasuredN=100 Gettysburg town breaches grew with the grace: 259-279 -> 290 (grace 1) -> 305 (2) -> 310 (unbounded)
Orderly Withdrawal not measured: Redraws the front line one step rearward each turn and moves disengaged units back to it; units held in contact remain as the rearguardMeasuredNot yet measured (V1 is a forced switch with no judgement; the decision to withdraw is planned for the win-condition planner)
VP Warden (One Unit per Objective, Held Forward) not measured: Assigns a single unit to each friendly victory point and holds it in reserve as far forward as possible, yet close enough to reach the objective before any enemy. Its duties are to close gaps in the line and to return to the objective before an enemy can take it unopposed. Nearby objectives may share one warden. No warden is assigned to an objective worth less than one tenth of all victory points, or to one that units on the line can already reach first. If the warden becomes engaged, the next nearest unit takes over the dutyMeasuredNot yet measured (implemented 2026-09-24)
No Pull toward Quiet Rear Objectives not measured: For friendly victory points that no enemy can reach within two turns, the attraction that draws units toward the objective is reduced almost to nothing. Units without a task therefore do not linger at quiet rear objectives and move toward the front instead. The normal attraction returns as soon as an enemy draws nearMeasuredNot yet measured (implemented 2026-09-25)
Initial Deployment for the Weaker Side Only not measured: Restricts the initial deployment, in which every unit is fixed to a line post before contact, to the weaker side defending a river and to a side forming a contracted front. A side of equal or greater strength advances instead of holding. This prevents a standoff in which neither side closes and no combat ever occurs on maps where the armies begin far apart.MeasuredStandard map, stronger side: win rate +8.3 pt (one seed, earlier code). The outcome reverses on Brandywine.
Bring Up Units from the Rear (Combat Is Free) not measured: Activation slots are assigned in order of distance from the enemy, the most distant units first. Missions and attack plans only decide the order among units at the same distance. Units already in contact can fight without a slot and are therefore served last. The only exceptions, which come first, are units clearing a reinforcement entry hex and units that must reach a key hex this turn before the enemy does.MeasuredNot yet measured (implemented 2026-09-26)
Sector Posture (Offensive and Defensive Sectors) not measured: Compares strength sector by sector along the front. If any sector holds a ratio of 1.5 or better, the strongest becomes the offensive sector and the rest are defensive. Units are not drawn off to other sectors one or two at a time; each fights within its own sector. If no sector is offensive, three or more units are transferred together to the weakest sector, but only when they bring it to a ratio of 1.5 on arrival; otherwise none are sent. Defensive sectors receive only the activation slots left over.MeasuredMeasured (2026-10-07), not adopted: piecemeal commitment at Blenheim was not reduced (losses just after arrival 0.78 to 0.68 per game)
Surround and Destroy not measured: For each enemy unit in contact or within reach, units moved with activation slots are sent to the hexes around it. Together with the units already in contact, they establish odds of at least 2:1 and close every hex into which the enemy could retreat before the attack is made. Within the available slots, the enemy units offering the greatest result per slot are chosen first, and the units forming the ring remain in place.MeasuredNot yet measured (implemented 2026-09-27)
Form a Line Before Enemy Reinforcements Arrive not measured: When enemy reinforcements can reach a friendly-held location whose loss reduces the activation limit, lying away from the main enemy force, within three turns, units far from the main enemy force are sent to form a line on the side of the objective facing the enemy. Units are sent until their combat strength reaches three-tenths of the arriving enemy strength; distant units approach over two turns.MeasuredNot yet measured (implemented 2026-09-27)
Wide Envelopment not measured: While their surprise attack is not being held back, elite units ride around the enemy flank, avoiding enemy units and their zones of control, and make for the hexes behind the enemy: empty hexes adjacent to an enemy unit and lying beyond it as seen from their own army. On arrival they attack from the rear.MeasuredNot yet measured (implemented 2026-10-01)
Drawing the Pocket not measured: On the side that has a sudden-death unit, infantry within two hexes of the enemy (excluding elite units and standby groups) do not move forward. They may still move sideways or back, stay in place, fight and advance after combat. The sudden-death unit withdraws straight back along the axis running from its own army towards the enemy. Because the centre does not return to its original line after being pushed back, the enemy is drawn deep into the pocket.MeasuredNot yet measured (implemented 2026-10-02)
Yield to Strength, Strike Weakness not measured: Having formed a front, the army pulls the line back where the enemy is strong and pushes forward to strike where he is weak (the advantage of exterior lines). For each post on the front, nearby enemy and friendly combat strength are compared: where the enemy is twice as strong or more, the post moves back to a line four hexes from the enemy and is held by stacks of two; where the enemy is half as strong or less, the post moves up to one step short of contact. In addition, any enemy that can be attacked at 4:1 or better is struck at once, even if its line of retreat is open. Active only when the opponent has an activation limit, so that following up costs him activations.MeasuredNot yet measured (implemented 2026-10-02)
The Line Runs from Edge to Edge not measured: The line (the posts of the foundation) is drawn to cut the enemy off from the friendly side's map edge, rather than as separate rings around each friendly victory point. A friendly victory point is added to what the line must protect only when the enemy could reach it before friendly units without passing through hexes occupied by friendly units or friendly ZOC hexes, that is, when a free capture is in sight. Enemy victory points that the army has decided to take remain objectives of the line as before.MeasuredDormant (author's ruling 2026-10-02, off by default): the line need not run from edge to edge. At Gettysburg the fishhook did not form by turn 5 (with this off it does form, same game). At Sekigahara the foundation fill rate was 0% (8–10% after fixing a defect that kept posts off occupied hexes).
Close the Holes in the Line not measured: Friendly units are divided into lateral slots two hexes wide, perpendicular to the direction from which the enemy comes. Any slot with no unit in it (a hole in the line) is filled with a spare unit from a slot that holds two or more, or with a reserve behind the front. The depth matches the front edges on either side, and the unit stays at least two hexes from the enemy.MeasuredNot yet measured (implemented 2026-10-02, based on the designer's Sekigahara game record)
Form the Line as a Whole not measured: Instead of moving units one by one, the placement of every unit is drawn at once. For each lateral slot two hexes wide, perpendicular to the direction from which the enemy comes, the enemy's front edge and the strength of the enemy units able to reach it are measured. Strong slots receive two units (strongest first) three to five hexes from the enemy; weak slots receive one unit two to three hexes away; only hexes with a line of retreat behind them are used. A unit that cannot reach its target stops at the reachable hex nearest to it. Several pictures with different depths are drawn and scored by the expected gains and losses of the next turn and by enemy breakthroughs behind each slot's front edge; the best is adopted if it beats the original plan.MeasuredNot yet measured (implemented 2026-10-02, based on the designer's Sekigahara game record)
Release Quiet Line Posts For Objectives worse on: Frees units posted on stretches of the line the enemy cannot reach, and sends them to objectives worth takingMeasuredN=100 across 4 maps, objectives captured 463->456 / 164->143 / 1008->1008 / 281->261 (never up); Nancy red seal rate 70.4->62.5
Require Two Turns Before Cutting the Tail worse on: Skips posts on the tail of the line only after the tail verdict has held for two consecutive turnsMeasuredN=100 phase improves on 3 maps but balance breaks: Nancy2 blue 45->12%, Shiloh 52->67%, Sekigahara 44->32%

Tactics 13

Lookahead: Avoid Hazardous Moves not measured: Discards moves that would leave the unit without a line of retreat once the enemy closes in (two-ply lookahead).MeasuredNot yet measured (only symmetric runs; asymmetric A/B pending).
Lookahead: Seize a Certain Encirclement not measured: When a move seals the enemy's retreat even if it withdraws, other moves are discarded. Units holding the line are exempt.MeasuredNot yet measured (asymmetric A/B pending).
Lookahead: Build Alternatives After Avoiding not measured: When lookahead finds that the best move is dangerous, safer alternatives are built from the adjacent hexes and from staying in place, and added to the candidates. When every move is dangerous, all reachable hexes are searched.MeasuredFive boards: +12.0 points with seed 42 but -4.5 points with seed 43. The direction reversed; the result is on hold.
Lookahead: Calculated Attack not measured: Even for a move judged dangerous, if it cuts off an enemy's retreat, three ways of making it safe are examined: the danger disappears once the enemy is destroyed; a trapped enemy is eliminated before it can counterattack; or the sealed enemy is stronger than this unit. If any holds, the attack is made.MeasuredFive boards, two seeds: win rate 49%, within the margin of error. Own losses decreased with both seeds (-0.75 and -1.24 units).
Safety Gate: Avoid Hexes with No Retreat not measured: Within the pipeline, hexes that would leave no line of retreat on the current board are removed from the candidates. Hexes saved by the calculated attack or by taking a VP despite the cost are kept, and nothing is removed when every hex is dangerous.MeasuredNot yet measured (judged by units destroyed with no line of retreat).
Translate the Victory Condition into Evaluation not measured: The victory condition is the difference in victory points plus the difference in strength destroyed, yet only the latter reached the position evaluation. This adds the former.Measured22 maps, N=30: win rate fell by 3.9 points (not adopted).
Artillery Comes Home no difference: Guns cover friendly objectives from within their own rangeMeasuredN=30 win rate 0 (no payoff)
Slip The Noose not measured: Shifts a nearly surrounded unit to a neighbour that still has an exitMeasured8-23 knobbite: Gettysburg (d938) and Guandu only; same AB score +26 t=+2.31 (uncorrected, suggestive - revival candidate)
Hold the Post (Defensive Deadline) not measured: For a side whose plan is to hold, a unit assigned to a line post that the enemy can reach next turn enters that post if it can do so this move, rather than advancing to attack. It advances as before only when it can attack at 2 to 1 or better. This takes precedence over the order to engage the enemy and over the routine that rearranges zones of control. A unit on its post that is adjacent to the enemy still attacks as usual.MeasuredNot yet measured (implemented 2026-09-26)
Hold the Post (Also When Currently Winning) not measured: Applies "Hold the Post (Defensive Deadline)" not only to a side whose plan is to hold, but also to a side that would win if the battle ended now. It takes effect only when "Hold the Post" is also enabled.MeasuredNot yet measured (implemented 2026-09-26)
Hold the Fortress not measured: Units inside a fortress move only between fortress hexes and do not sally out. They attack from within, and only when it pays. A scenario may assign this to a side (rules.ai_fort_hold).MeasuredNot yet measured (implemented 2026-10-11 for Yorktown)
Avoid Contacts That Cannot Be Won not measured: After all destinations have been decided, each move that creates a new contact is examined. Because every adjacent enemy must be attacked, a unit that would newly come into contact with an enemy it can attack only at 1:2 or worse stops instead at the hex closest to its original destination that is not adjacent to the enemy. Moves that join an enemy already engaged by friendly units, and moves against an enemy holding a victory point location, are not restricted.MeasuredNot yet measured (implemented 2026-09-27)
Do Not Advance into a Pocket not measured: After combat, a unit does not advance into a hex that would leave it with one or fewer hexes to retreat into while enemy units of at least twice its strength are within three hexes. Advances that take a victory point location are made as before.MeasuredNot yet measured (implemented 2026-09-27)