The GoldenEye Arms Reference
Methodology


Site Navigation
Home Page - Introduction and credits.
Principles - General behavior of firearms and explosives.
Weapon Details - Detailed descriptions of each weapon.
Weapon Comparison - Ranking of weapons by performance metrics.
Multiplayer Modes - Analysis of multiplayer weapon combinations.
Realism - Authentic and inaccurate aspects of gameplay.
Methodology - Techniques used to measure several metrics.

Many of the facts presented in these web pages are easy to come by; figures like weapon clip size and the quantity of ammunition that can be carried are openly displayed by the game engine. The sharing of ammunition types may be determined by the symbols used to represent each weapon's bullets and confirmed through simple experiments. But some figures were won only with labor, with the techniques described below.

Firing Rate

Ten clips were fired from each weapon and the process recorded on videotape. The tape was then reviewed with equipment featuring an accurate counter, showing minutes, seconds, and NTSC frame number (the thirty frames a second were numbered zero to twenty nine). The times of the first and last muzzle flash in each clip were noted, the difference being the time required to fire all but the last of the bullets in the clip. The ten trials for each weapon were averaged, then a simple quotient produced the number of rounds per second.

In the data sets below each time stamp is abbreviated by an integer of the form `...mssff', ending with a two digit second and two digit frame number; the minute counter is omitted when both times fell in the same minute. They are arranged in pairs specifying, again, the times of the first and last muzzle flash of a full clip. For the Laser, which has no clip, measurements are given of the times of a number of consecutive shots.

US AR33: (2808, 3106) (3225, 3525) (3713, 4017) (4206, 4506) (4623, 4926) (5119, 5422) (5608, 5913) (103, 403) (523, 825) (1012, 1315) RC-P90: (1818, 2710) (2901, 3721) (3912, 4802) (4921, 5818) (10, 904) (1025, 1917) (2108, 3006) (3124, 4017) (4204, 5017) (75205, 80101) Deustche: (4520, 5004) (5121, 5609) (25729, 30214) (401, 817) (1004, 1422) (1609, 2028) (2218, 2705) (2820, 3306) (3422, 3912) (4027, 4512) Phantom: (5104, 5903) (22, 817) (1015, 1808) (2013, 2807) (2929, 3721) (3912, 4707) (4822, 5618) (45807, 50600) (719, 1516) (1706, 2502) ZMG: (5427, 5805) (15925, 20229) (416, 721) (910, 1215) (1400, 1709) (1825, 2204) (2410, 2719) (2907, 3215) (3405, 3715) (3905, 4214) Laser: (3401, 3409) (3409, 3418) (3418, 3427) (3427, 3505) (3505, 3514) (3514, 3523) (3523, 3602) (3602, 3611) (3611, 3620) (3620, 3629) Klobb: (328, 623) (816, 1113) (1303, 1600) (1716, 2014) (2200, 2428) (2613, 2912) (3024, 3324) (3505, 3804) (3920, 4217) (4405, 4702) KF7: (5112, 5529) (5714, 10202) (325, 813) (1001, 1418) (1607, 2026) (2214, 2629) (2816, 3303) (3421, 3910) (4025, 4510) (4701, 5119)

The nonautomatic weapons have smaller data sets. The author unloaded several clips from each weapon, trying in each case to optimize the firing rate, and selected the best times as representing the limits of the weapon's capability. These data sets are:

Shotgun: (507, 803) (1004, 1229) (1417, 1713) (1901, 2127) (2319, 2614) (2803, 3029) (3219, 3514) (3703, 3928) (4116, 4411) (4529, 4825) Sniper: (5711, 5808) (145926, 150023) (423, 522) (705, 804) (920, 1017) Magnum: (724, 811) (811, 829) (829, 918) (918, 1006) (1006, 1025) (1212, 1302) (1302, 1322) (1322, 1410) (1410, 1428) (1428, 1515) Auto Shot.: (5224, 5520) (85708, 90005) (123, 419) (607, 903) (1021, 1317) (1505, 1801) (1920, 2215) (2404, 2700) (2820, 3116) (3304, 3529) Dostovei: (2812, 2919) (3109, 3216) PP7: (3211, 3310) (3814, 3915)

Damage Rating

The damage reticle is a poor instrument for measuring partial damage - the illuminated fraction of the gauge ends gradually rather than abruptly, and the eightfold segmentation conceals portions of the scale. Accurate ratings must be based on the number of shots required to kill, not on visual guesswork based on the reticle.

Damage was first measured in multiplayer mode with no health bonus; the number of limb hits required to kill were measured. When the character was far from death they were racked up more quickly by substituting head hits for groups of four limb hits, and torso hits for pairs of limb hits. This revealed that while some weapons caused fractional damage, several seemed to cause exactly 4 damage points or 8 points. Prolonging the death by repeatedly getting body armor after the previous armor is completely eliminated (when is assured when a shot reduces the red health gauge) allowed these numbers to be confirmed to fairly high accuracy.

One of the 4 point weapons was chosen as a standard, and now a health bonus of +2 was selected - and the effectiveness of damage was reduced by a fraction close to 1.69. Since I expected the GoldenEye programmers to use an exponential health adjustment, it seemed significant that 1.69 it is the square of 1.3. Further experiments quickly confirmed my initial hypothesis, that a health bonus of n reduces effective damage by the fraction 1.3 to the nth power. The exception was the +10 bonus which instead simply reduced damage to one tenth its normal potency. (My guess is that the -10 health bonus also abandons the exponential rule, and increases damage by a factor of ten; but since either formula makes the least damage (limb hit from a Klobb) fatal, the cases are indistinguishable.)

The +10 bonus was then selected, and used to increase the precision of the ratings for the `fractional' weapons. These measurements converged on the values 2.4 for the Klobb, 5.6 for the US AR33, and 7.2 for the RC-P90; and since this gave weapon damage ratings that were all integral tenths of normal health (0.8 points times either 3, 5, 7, 9, 10, or 15), the results seemed reasonable. (This suggests that weapon damage might have been described more cleanly had full health been defined as ten points; but since the reticle interface suggests division into eighths, I have retained the eight point system.) The Phantom and a few of the novelty weapons could not be directly measured since they are not available in multiplayer mode, but by employing them against enemies of known strength during missions I derived provisional ratings for them. It was Hu Man Bing that suggested experiments with enemies whose health has been enhanced to 1000% by the 007 cheat, which when combined with the remarkable stamina of Jaws on the Aztec level revealed that the Golden weapons' damage was not actually infinite.

Medium Range Correction

Most mysterious and difficult to reproduce among my metrics is the Medium Range Correction or MRC. The idea is to assign each weapon a fraction that specifies how likely each round is to strike a human figure standing a moderate distance away.

The difficulty in designing this measurement is that one needs a stationary target a set distance away, and a mechanism for determining how many shots hit the target. It turns out that the game provides us with both! The procedure is as follows:

  1. Turn on the Invisibility cheat along with any cheats needed to select the weapon to be tested. All Weapons, Infinite Ammo, and Slow Animation are very useful for performing this test.
  2. Select the Caverns mission at the 007 Agent difficulty (The difficulty may not matter, but extraneous experimental variables should be controlled).
  3. Leave the elevator - since you are invisible Trevelyn will not react - and kill the two Janus marines standing by him with Unarmed, since this will not affect the accuracy statistics given at the end of the mission. This is most easily done by standing behind the nearest marine and delivering a fatal blow from behind, then turning to pummel the other marine who will run up after seeing his comrade killed. Stay out of view of the marines down the hall so they will not become alarmed and run up as well.
  4. Return to the elevator and stand facing Trevelyn with your back against the wall.
  5. Reopening the elevator door when necessary, fire entire clips from the weapon you are measuring right at Trevelyn's head; do not allow the sight to move off of the center of his face, since the only factors affecting each hit should be the inaccuracy of the weapon and its recoil. Time your activity so the doors do not close while you are firing.
  6. Quit the mission, and record the Total Shots, Head hits, Torso hits, and Limb hits from the second mission debriefing screen.
  7. The weapon's MRC is the number of head hits, plus half the number of torso hits, plus one fourth the number of limb hits, divided by the number of total shots.
So the MRC is the fraction of hits that strike an opponent when you aim at his head from medium range (which we define as exactly the distance from the back of the elevator to Trevelyn), where shots that struck his torso and limbs are counted off because they cause less damage than true head hits.

The above procedure will need to be repeated several times for each weapon, and a compromise MRC produced from the wide range of values you may get from each set of experiments. Higher power weapons tend to knock Trevelyn's head about like a punching bag; it is possible that the movement of his head degrades their apparent accuracy, but I know of no way to correct for this.

Different values may be produced by aiming elsewhere than the head. My experiments suggested, however, that for every weapon, aiming at the head produces damage that on average is at least as great as that produced by aiming elsewhere. So while inaccurate weapons certainly land more hits when aimed at the torso, this advantage is offset by the lower damage earned through each hit.


Rhodes MillCopyright © 1998 Brandon Craig Rhodes.