Open Design Questions

Unresolved decisions, ordered by how much they cost to discover late. Each carries a stable ID so the changelog can reference it. Nothing here is settled; entries are closed by writing a resolution and a justification into CHANGELOG.md.

Status values: Blocking (downstream work cannot proceed), Open (needs a decision before the first prototype), Backlog (worth resolving, not yet urgent).

Player Assistant reframe. C.A.T.S. assumes a digital assistant at every seat — see 03-player-assistant.md. Several entries below were written before that was established and are annotated where it changes the answer. The short version: bookkeeping and lookup objections mostly dissolve; game-feel, decision-cost, and distribution objections stand unchanged.

OQ-01 — Penetration is fully deterministic

Status: Open

Penetration Margin = Weapon Penetration + Attack Edges + Ammunition Modifier - Threshold contains no die roll. Against a given target, a weapon either always penetrates or never does. Matchups become binary — "my lasers cannot hurt that tank, ever" — which removes tension from the exact step where the design most needs it, and makes a single armor statistic a hard counter to whole weapon categories.

Candidate resolutions:

Interacts with OQ-02 and OQ-09.

OQ-02 — Edges are double-counted

Status: Open

Attack Procedure step 6 says Edges are spent on location control, penetration, extra damage, suppression, system targeting, or reduced Heat. The Penetration Margin formula adds them automatically. These cannot both be true.

Decide whether Edges are a spendable currency or an automatic bonus. If spendable, remove them from the formula. If automatic, remove penetration from the list of things Edges buy.

OQ-03 — No scale numerology exists

Status: Blocking

The cross-scale rule adjusts Penetration by ±2 per scale step. Armor Capacity across seven scales almost certainly grows geometrically, not linearly. Infantry-to-capital is six steps, so −12 Penetration against a Threshold that might plausibly be 60 — the modifier is meaningless without anchors.

Equally, the Penetration Margin result bands (−4 or lower / −3 to −1 / 0 to 3 / 4 to 7 / 8+) span only twelve points. If typical Threshold values differ by tens across scales, nearly every cross-scale exchange lands in the outermost band and the middle bands never fire.

Required before anything else: publish anchor values for typical Threshold, Armor Capacity, Structure, and weapon Penetration at each of scales 0 through 6. Then check whether ±2 per step survives contact with them, or whether the rule needs to be multiplicative.

This is the real first deliverable. The probability table was cheap; this is the expensive one, and the data schema, unit statlines, and worked combat example all depend on it.

OQ-04 — Component Checks are too lethal and roll the wrong direction

Status: Open

At Breach Severity 0, the published table yields:

So 36% of every single Structure box lost disables or destroys a component, before any severity bonus is applied. Units will reach the state where no meaningful choices remain far faster than the Playable Damage goal intends. Severity bonuses push this further.

Separately: this is a roll-high table inside a roll-under engine. Players will misread it constantly. Either invert it or flag the reversal loudly in the rules text.

PA note. The readability half is largely solved — the PA interprets the result, and Component Checks are designated a physical roll precisely because they are dramatic. The lethality half is untouched. A 36% disable-or-worse rate per Structure box is a game-feel problem, not a lookup problem, and the PA makes it arrive faster.

OQ-05 — Hit location centralizes hard

Status: Open

Applying a bell curve to hit location concentrates damage in one box, which works against the localized-degradation premise:

Mech arms carry weapons, and a 5% chance to hit a weapon-bearing limb means weapon loss almost never happens through normal fire. Cockpit and "attacker chooses" are both 1%.

The curve is doing useful work on the action check and the wrong work here. Consider a flat d20 or d100 for location specifically, decoupled from the resolution die.

PA note. This one gets easier, not harder. Hit location is a PA-side roll, so the distribution is now free to be anything at all — d100, per-facing tables, weighted by attack profile, different templates per chassis — with zero lookup cost and no requirement that it use a physical die at all. The constraint that made a single shared 2d10 table attractive has been removed. Design the distribution the game wants.

LIKELY RESOLUTION — depth tiers. See 07-hit-location-depth.md. Hit location rolls on an exploding d6, and a critical location cannot be reached without exploding. The die partitions itself into exact tiers of five separated by the unreachable multiples of 6: tier 1 at 83.3%, tier 2 at 13.9%, tier 3 at 2.3%.

This replaces a lateral distribution problem with a depth structure, which is what the game actually wanted. The complaint here was that a bell curve concentrated 44% of hits on one torso box while weapon-bearing arms took 5%. Depth tiers make the concentration meaningful instead of arbitrary: outer hits are common because outer surfaces are what gets shot, and reaching something vital requires punching through. The dice mechanic and the layered-armor premise become the same statement.

OQ-06 — Complication-triggered subsystems never fire

Status: Open

Complications (matching dice on a failure) occur at 1-3% for competent characters. Ammunition depletion in the Low state and reactor safety checks at Critical Heat both hang off Complications, so for anyone good at their job those subsystems effectively do not exist.

Note also that only even totals can be matching dice, so roughly half of all results can never produce a Complication or a Critical Edge at all. Total match frequency is a flat 10% at every Target Number — raising the Target Number converts Complications into Critical Edges rather than making matches more common.

Either give these subsystems their own trigger or accept that they are a low-skill-only rule.

OQ-07 — Edge frequency scales nonlinearly with skill

Status: Open

From the generated analysis:

A TN-18 shooter generates an Edge roughly seven times as often as a TN-10 shooter, against a success rate only about twice as high. Because Edges currently feed Penetration directly (see OQ-02), pilot skill scales lethality nonlinearly — a skilled gunner does not merely hit more often, they penetrate armor that an average gunner cannot scratch at all.

Decide whether that cliff is the intended fantasy. If not, cap Edges, raise the Edge step above 5, or decouple Edges from Penetration.

OQ-08 — Target Number has no ceiling

Status: Open

Skill + Attribute + Equipment + Situation - Difficulty is a four-term sum, but the brief states ratings normally range 4 to 18. Nothing constrains the total. A TN of 21+ is an automatic success that also generates 3 Edges on most rolls, and nothing in the engine prevents a build from reaching it.

Needs either a hard cap, a diminishing-returns curve, or a stated design ceiling on each term.

OQ-09 — Exploding dice — RESOLVED: adopted universally

Status: Resolved. Every roll explodes on a 6, recursively.

Raised during the initial review as a nice-to-have, promoted when the one-set constraint made multi-die damage impossible, briefly declared dead under one-roll resolution, and finally adopted as the engine's core.

Two corrections to earlier reasoning in this entry, both recorded rather than deleted:

• The claim that exploding dice conflict with one-roll resolution was wrong. An explosion continues the same roll toward one result rather than asking a second question. The situation is still resolved by one roll producing one Margin.
• The worry about redundancy with the matching-dice Critical Edge is moot. The explosion is the critical. It is self-announcing, needs no lookup, fires at a known 16.7% per die, and its magnitude scales with the chain length. One mechanic now covers both critical results and damage spikes.

Remaining sub-question: whether explosion chains are capped. Uncapped is dramatic but produces rare absurd results and occasionally stalls the table. Tracked in 04-dice.md.

Original entry follows for the record.

Should damage use exploding dice — roll the maximum face, roll again and add — or some other critical damage escalation mechanism?

Arguments to weigh:

Open sub-questions: does it explode on damage dice only, or on Penetration too? Once, or recursively? Does it stack with Critical Edge, or does Critical Edge become the explosion trigger? Does exploding damage bypass Threshold, or is it applied before the comparison?

The cheapest version worth testing first is probably folding this into the existing Critical Edge rather than adding a second mechanism — the matching-dice result already fires at a known 5-9% and would not add a roll to the common case.

Dice-constraint note — this question got significantly more important. One standard set per player means multi-die damage is impossible: there is one d6, one d8, one d12. 2d6 cannot be rolled. So damage variance cannot come from adding dice, and the realistic options reduce to a single die plus a flat modifier (flat, fixed variance) or a single exploding die (long tail, one physical die). Exploding dice went from a nice-to-have that risked redundancy with Critical Edge to the natural way of getting damage variance at all. See 04-dice.md. Priority raised.

OQ-10 — Missing subsystems that block prototyping

Status: Open

The brief never defines these, and a worked combat example cannot be produced without them:

OQ-11 — Heat and Power may be redundant

Status: Open

Heat and Power both exist to limit alpha strikes and both require per-round bookkeeping, and Power can be spent on cooling — which is Heat's own resource. Two overlapping economies doing similar work is a real cost in a system already targeting Standard mode as the default.

Consider scoping them: Heat for mechs and aerospace, Power for ships and shields, rather than both everywhere.

PA note — the objection changes shape rather than going away. The bookkeeping cost was half the argument and the PA absorbs it entirely. What remains is the sharper question: do Heat and Power generate two genuinely different decisions, or one decision wearing two costumes? Decision cost is not reduced by automation. If the honest answer is that players experience both as "can I afford to shoot everything," they should be merged regardless of how cheap they are to track. Status raised from Backlog to Open on that basis.

OQ-12 — The stated MVP is not minimal

Status: Open

The brief's Minimum Viable Prototype is 22 unit templates, 20 weapons, 20 components, 10 cybernetics, 4 species, and 6 scenarios. That is a small supplement, not a prototype. Separately, 14 damage properties against 10 armor properties is 140 interactions to balance.

A genuine MVP is closer to 2 mechs, 4 weapons, 2 armor profiles, and 1 scenario — enough to prove the core loop resolves cleanly, and cheap enough to throw away when it does not.

Related: the brief's own playtest target of 45 seconds per attack is worth checking early. The attack procedure is 12 steps with three separate table lookups (location, penetration band, Component Check). Sixty to ninety seconds is the realistic figure for a new player.

PA note. The 45-second target survives but stops measuring the right thing — lookup was the bottleneck it was aimed at, and the PA removes it. Replace it with taps per turn and time from physical roll to stated outcome. The 12-step procedure also stops being a player-facing cost; it becomes an implementation sequence, and only steps 1, 6, and the physical rolls are things a player actually performs.

The MVP scale objection is unchanged. 140 weapon-property against armor-property interactions still have to be balanced by humans, and the PA does not help with that at all. It makes them cheaper to use, not cheaper to design.

OQ-13 — Structural resemblance to existing games

Status: Backlog

Game mechanics are not copyrightable, so there is no legal exposure here. But head / right torso / core torso / left torso / arms / legs with inward damage transfer, combined with heat bands, ammunition explosions, and mass-and-slot construction budgets, will read as a BattleTech clone regardless of how originally it is worded.

If that perception matters commercially, deliberately diverge on at least one structural axis. Non-anthropomorphic location templates would be the cleanest place to do it.

Treat this as design-risk commentary, not legal advice.

OQ-15 — What happens without the PA

Status: Open — raised by the Player Assistant architecture

If the PA is assumed at every seat, the failure mode has to be designed rather than discovered. Dead batteries, a forgotten tablet, a player who refuses to use one, or a demo at a convention table with no devices are all ordinary occurrences.

Recommendation to evaluate first: graceful degradation. It gives the Narrative complexity mode a genuine purpose beyond "the simple version," and it means a table can always finish the session it started even if a device dies.

This needs deciding early. It determines whether every mechanic carries a hand-resolvable path as a design constraint, and retrofitting that later is expensive.

OQ-16 — Dice input and trust

Status: Open — raised by the Player Assistant architecture

Physical dice and digital state have to meet somewhere. How results get in is an interaction design decision with real consequences for pacing:

Related and worth deciding explicitly: the PA should trust player-entered rolls. This is a cooperative game at a shared table where everyone can see the dice. Building anti-cheat into a tabletop RPG is solving a problem that does not exist and signals distrust of the group.

Note the tension with the physical-roll list in the PA spec: every roll designated physical is an input step. That is the real budget constraint on how many physical rolls a turn should have, and it argues for keeping the count at two or three.

OQ-17 — Computability audit

Status: Open — raised by the Player Assistant architecture

Any rule the PA automates must be a deterministic function of game state. The current brief is not written to that standard, and several places quietly assume a human is reading:

• "the fiction clearly exposes a particular component" — genuine GM judgment, needs marking as an explicit handoff rather than sitting inside an automated path;
• "identify a vulnerable subsystem, exposed crew, opening, joint, sensor, weapon, or mission-specific weakness" in the cross-scale rule — same problem, and it gates whether damage happens at all;
• ground vehicle and aerospace location templates have no roll ranges (see OQ-10), so they are not implementable as written;
• no stacking rule is defined for modifiers;
• conditions lack explicit durations and end conditions.

Every rule needs classifying as automated, assisted (PA presents, human decides), or human (GM ruling, PA records only). Doing this as rules are written is cheap; doing it during implementation means rewriting them.

OQ-18 — Are complexity modes rulesets or presentation layers?

Status: Open — raised by the Player Assistant architecture

The brief defines four modes (Narrative, Standard, Tactical, Engineering) as progressively larger rulesets, targeting Standard as default. A PA permits a fundamentally different architecture:

The second is only possible because of the PA, and it is a genuinely attractive property: a group could run Narrative for the pilot who wants to fly and Engineering for the player who wants to manage the reactor, in the same fight, with consistent results.

It also interacts with OQ-15 — if modes are presentation layers over one simulation, "paper Narrative mode" is no longer a subset of the rules but a separate, simplified implementation.

Decide before writing rules text. The two architectures are not convertible after the fact.

OQ-26 — Suppression and volume of fire have no representation

Status: Blocking — raised by adversarial review, escalated by the coverage survey

A survey of the open corpus identifies suppression as the single most important military mechanic, and C.A.T.S. currently has none. Three structural approaches exist:

The third inverts the usual assumption and is the most military-feeling of the three.

The pattern to adopt is two-axis: a short-term suppression flag plus a persistent confidence tier stepping down through advance / hold / withdraw / break as a unit takes casualties or loses its leader. Better training absorbs more suppression before breaking, which makes asymmetric fights work without fudging lethality. Rally becomes a leader action tested against troop quality.

That single mechanism also answers squad-scale morale, which the survey identifies as its own gap — nothing in the open corpus sits between individual fear and army-scale mass battle.

Escalated to Blocking because it interacts with the LOCK_TN clamp: the clamp deletes exactly the low-probability shots that suppression is built on, so the two must be designed together.

Original entry follows.

The LOCK_TN clamp reports any shot past Target Number 21 as needing guidance, on the grounds that unguided fire has stopped working. That is defensible for a single aimed shot and indefensible as a general claim, because the engine has no burst or volume-of-fire mechanic.

Thirty rounds at 3% each is roughly a 60% chance of a hit somewhere. Suppression, massed fire, and area denial are real tactics built on exactly the low-probability shots the clamp deletes.

Options: a burst mechanic converting rounds fired into an effective hit chance; a separate suppression rule that does not need hits at all; or accepting the gap and documenting that C.A.T.S. models aimed fire only. The third is honest but loses a great deal of infantry tactics.

Related: the clamp is now correctly described as a playability decision rather than physics, and it applies to gunnery only — skill tests are deliberately unclamped.

OQ-25 — How many d6, and how does skill enter?

Status: Blocking — sets the entire numeric scale of the game

The d6-only exploding pivot settles the die but not the dice pool. These two questions must be answered together, because they interact.

How many dice?

How does skill enter?

Recommendation to evaluate first: fixed dice count with skill as a flat modifier. It keeps variance constant so the exploding tail means the same thing at every skill level, and it avoids compounding the convexity problem that already appears in OQ-07 and in the 9x expected-Margin spread noted in 06-margin-and-progress.md.

Real-world evidence says skill should barely touch accuracy at all. Under time pressure the measured expert-versus-average dispersion gap is about 1.7x at 100 yards, decaying to 1.0-1.25x beyond 200 yards — and expert performance declined under rapid fire against randomly presented targets while average performance did not. Across 200+ comparisons, range qualification score showed no relationship to combat hit rate. Real hit rates run 17-19%, and 18% when the target is shooting back against 34% when it is not.

The mechanism is not steadiness. Under threat, shooters make significantly shorter goal-directed eye fixations, leaving less time to refine the shot — a visual-attention failure, and a trainable one.

This argues for something better than a flat modifier: skill should buy tempo and attention, not steadiness. A large effect on time-to-first-shot, rate of aimed fire, and target discrimination; a small effect on dispersion. Given the exposure gate (below), tempo is lethality — a shooter who acquires a second faster gets shots that a slower one never gets at all. See 10-real-world-reference.md.

OQ-24 — Rebuild every bounded band for an unbounded roll

Status: Blocking

Exploding dice have no maximum. Several tables in the brief assumed a bounded 2-20 result and are now structurally invalid:

Bands must be rebuilt as open-ended tiers against the measured exploding distribution, with an explicit answer for what happens far out in the tail. Blocked on OQ-25, since the numbers depend on dice count.

OQ-23 — Roll-under or roll-over? — RESOLVED: roll-over

Status: Resolved by the d6-only exploding pivot

Exploding dice push totals upward, so under roll-under a 6 would drive a character toward failure — self-defeating, and it reads as punishment for a lucky roll. High must be good.

Resolution: succeed by meeting or beating the Target Number. Margin = Roll − Target Number.

Difficulty becomes a single clean number rather than a subtracted term, which suits the no-opposed-rolls lock and gives the PA one value to compute and display.

Original entry follows for the record.

The founding brief specifies roll-under: succeed on the Target Number or less, with Margin = TN − Roll.

The worked example given for progress accumulation describes roll-over: Target Number 11, rolled 15, earned 4 points. That is Margin = Roll − TN, and a 15 against a Target Number of 11 is a failure under the brief.

Both are coherent. They cannot both be the system.

The analysis cost of switching is zero. Roll-under and roll-over are mathematically symmetric — every distribution, Edge rate, expected Margin, and Margin shape measured so far mirrors exactly and transfers unchanged.

What actually differs:

Argument for roll-over. Difficulty becomes one clean number rather than a subtracted term, which suits the no-opposed-rolls lock and gives the PA a single value to compute and display. It is also what most players expect.

Argument for roll-under. Every existing number in the brief assumes it, and the brief valued the readability of "I need my number or less."

Decide alongside OQ-20. The two together fix the shape of every number in the system.

OQ-21 — What does a player roll when attacked?

Status: Blocking — the largest gap opened by one-roll, player-facing resolution

If the GM never rolls, an NPC attacking a player character has to become something the player rolls. This is unspecified and it affects every combat in the game.

Interacts directly with the action economy and with the Guard action already in the brief. Also determines whether "one roll per situation" means one roll per exchange or one roll per participant per exchange.

OQ-22 — Does "one roll" mean one result or one throw?

Status: Open

A simultaneous throw — d20, d10, and a damage die released together and read by size — is one physical action but produces three readings.

Under the strict reading (one result) it is disallowed, and everything downstream of the attack must derive from the single Margin. Under the permissive reading (one throw, no back-and-forth) it is the natural implementation, and the intent of the rule is really "no sequential rolls, no opposed rolls, no waiting."

The distinction decides whether hit location and damage can use dice at all, or must be computed by the PA from Margin. That in turn decides whether OQ-19 (die-size damage classes) is dead or merely constrained.

OQ-20 — Core resolution die — SUPERSEDED

Status: Superseded by the d6-only exploding pivot

The choice between 2d10, d20, and d100 roll-under is void. The engine is exploding d6.

The analysis retains value as a record of why a curve was wanted: the leading argument for 2d10 was that its decaying Margin distribution produced a typical result with rare spikes, recovering the feel of multi-die damage. Exploding d6 delivers that property more strongly, since the tail is unbounded rather than merely thin. The design goal survived; the die did not.

Preserved: analysis/resolution-comparison.md.

Original entry follows for the record.

OQ-20 (original) — Core resolution die: curve or flat

Status: ~~Blocking~~ Superseded — nothing about damage, Edges, or modifier design can be settled first

2d10 roll-under was inherited from the founding brief, written before the one-set dice constraint existed. It survives that constraint but has never been tested against alternatives. Measured comparison in analysis/resolution-comparison.md.

Because 2d10 is the only doubled-die roll a standard set can produce, this is a binary choice between a curve and a flat distribution. There is no third distribution shape available.

Reversal — one-roll resolution changes the answer. Under one-roll resolution, the Margin distribution is the damage distribution. Margin mass by quartile at ~75% success:

Scheme	Low	Mid-low	Mid-high	Top
2d10	38%	33%	21%	8%
d20	27%	20%	27%	27%
d100	25%	24%	25%	25%

2d10 gives a decaying damage curve — a typical result with rare spikes, exactly what 2d6 damage would have produced and which the one-set constraint otherwise forbids. The flat schemes make a maximum-damage result as likely as an average one, which reads as arbitrary rather than dramatic. The bell curve was a liability when it governed only hit-or-miss; it becomes an asset once it also governs damage. 2d10 is now the leading candidate.

The case for flat. C.A.T.S. stacks many situational modifiers, and a player evaluating a tactical tradeoff needs to know what one is worth. On 2d10 the honest answer is "it depends on your Target Number, and you cannot intuit it." The PA can show the delta but cannot make it feel legible, and automation never reduces decision cost. Flat distributions also partially defuse OQ-07 for free, since the convex Edge ramp is exactly that complaint.

This argument still stands on its own terms. It is now a cost knowingly paid rather than a reason to switch.

The case for the curve. Extreme results stay rare, so a rookie rarely embarrasses a veteran. Competence feels reliable rather than merely likely — the brief's stated reason for choosing it. Diminishing returns at the top end compress high skill without a hard cap, which bears on OQ-08.

Design headroom — the argument that is not about probability. 2d10 spends the entire d10 budget on the core check, permanently. Any later mechanic wanting a d10 or a percentile roll has nowhere to go.

This also constrains advantage and disadvantage. The cheap standard implementation — roll an extra die, keep the best — is impossible under 2d10, since a set has no third d10. Advantage would have to be a flat Target Number modifier, which on a bell curve is worth 4 to 10 points depending on skill and is therefore the single hardest thing to balance on that curve.

A fourth candidate exists but is not recommended. Step dice (skill selects the die, d4 through d20) uses the whole set and is highly tactile, but variance scales backwards — a d20 master is swingier than a d4 novice, which contradicts the goal that ordinary competence be reliable. Detailed in 04-dice.md.

A procedural option is orthogonal to all of this. A simultaneous throw — d20, d10, and a damage die thrown together and read by size — collapses an attack into one physical action with three readings. It composes with any scheme, but is easiest under d20 (where the d10 is free to be thrown alongside) and fights exploding dice, which are inherently sequential.

Consequences of switching. The Edge step, the Penetration Margin bands, the Component Check table, and every Target Number in the brief are all calibrated to a 2-20 range. d20 is nearly a drop-in (same range, different shape). d100 requires rescaling every number in the system — cheap now, expensive after scale numerology exists.

Decide before OQ-03. Scale anchors calibrated to one scheme do not transfer to another.

OQ-19 — Die-size damage classes — SUPERSEDED

Status: Superseded. There is only one die size.

Weapon identity must come from Penetration, how damage scales against Margin, range, arcs, ammunition, Heat cost, and properties — not from which die is picked up. The tactile "pick up a d12" appeal is gone; the design problem it addressed is not, and it now falls to OQ-24's band rebuild and to the weapon property set.

Original entry follows for the record.

A standard set leaves five dice unused once the core check claims the two d10s. Assigning them to damage classes gives weapon identity a tactile expression — picking up a d12 tells a player what kind of weapon they are firing before any number is read — and stops the game from demanding a full set while using two of it.

Provisional allocation in 04-dice.md: d4 light, d6 standard, d8 heavy, d12 very heavy, d20 either hit location or the heaviest damage class.

Open sub-questions:

• Does the d20 carry damage, or stay reserved for physical hit-location rolls? It cannot comfortably do both, and OQ-05 already established that the PA removes any need for location to use a physical die at all.
• Do damage classes map to weapon mass, scale, or role? Mass is most intuitive; scale interacts with the cross-scale rule and risks double-counting.
• Does armor Threshold interact with die size, or only with the flat Penetration value? If a d12 weapon and a d4 weapon can share a Penetration rating, die size expresses damage magnitude only — which may be the cleaner separation.
• Where does the flat modifier sit relative to the die, and does it scale with the die?

Blocked on OQ-03. Die sizes cannot be assigned to damage classes before the scale anchors exist, because "heavy" is meaningless until Armor Capacity values are known.

OQ-14 — Document errata

Status: Backlog

• The Layered Protection section says "four defensive layers" and then lists five (Evasion, Shield, Threshold, Armor Capacity, Structure).
• The Component Conditions table has columns for Impaired, Disabled, and Catastrophic but no Shock column, although Shock is a defined state.
• The facing-versus-section split for spacecraft (facing carries armor, section carries structure and components) is the most genuinely original idea in the brief and is under-specified. How a facing hit maps to a section is never stated.
• The brief never states the physical dice requirement. Now recorded in 04-dice.md: one standard seven-piece polyhedral set per player.