Genius, but a few things to think about

Yes, it has genuine potential to be fun and skillful - but only if a long list of currently unresolved gaps are closed tightly. Right now the design is promising on paper and fragile in practice.

The core fantasy is strong and relatively rare: persistent physical objects (dice) that you own, customize at the probability level, risk, lose, capture, and watch accumulate history. That combination of building, reading, selecting, rolling under changing win conditions, and actual permanent loss/transfer is more interesting than most pure RNG dice games and cleaner than most card-battlers that bolt dice on as an afterthought. The design pillars are coherent. The “simple surface, deep probability” goal is the right target.

However, several systems are still at the level of attractive sketches rather than closed, testable rules. That is normal for a v0.3 GDD, but it means the fun and skill claims are still conditional. Below is a hard-eyed walkthrough of what works, where skill actually lives, where the design is thin or dangerous, and what has to be proven before you can claim the game is good.

Core Loop Strength and Skill Expression

The round structure is the best part of the document:

1. Objective revealed.
2. Secret 3-of-4 selection.
3. Simultaneous authoritative physics throw.
4. Qualification + comparison.
5. Point or no point.
6. Repeat until first to 3.

This creates repeated, meaningful decisions under incomplete information. Skill lives in several distinct layers:

• Roster construction: covering the three colors while still having a specialist profile, or deliberately going extreme and accepting holes.
• Pip distribution under a fixed 21-pip budget: reliable middles vs high-variance extremes vs weird multi-modal sets. This is real probability craftsmanship.
• Geometry choice (D4/D6/D8): fewer sides concentrate the same budget; more sides open face slots for specials and change the shape of the distribution. This is interesting if the costs and face-capacity rules make the choice non-obvious.
• Objective reading + selection: which three dice give the best chance against this filter and comparison rule. The fact that the objective is shown before selection is the single biggest skill lever in the document. Without it the game collapses into pure RNG.
• Material / settling order: probabilistic influence on left-to-right special resolution. This is subtle and high-ceiling if the physics differences are readable and the specials reward correct ordering.
• Risk management of the fourth die and of match stakes: every match you risk a permanent object.

These layers can support real skill growth. A strong player will develop intuition for “what does a good Red 5+ die look like versus a good Lowest Even die,” how much color coverage is worth, when to bench the high-variance piece, and how to price the opportunity cost of a special face. That is deeper than most casual dice games.

The catch is that skill is almost entirely probabilistic management. You never guarantee outcomes; you only improve your odds. That is fine and even desirable, but it means the game will feel frustrating to a large segment of players unless the feedback is excellent and the variance is bounded. Matches that are decided by a single unlucky face after correct decisions will feel bad. The design currently leans on “the die is the item” and ownership history to create emotional investment that softens pure RNG pain - that only works if players actually form attachments to specific dice.

Where the Design Is Still Soft or Dangerous

1. Objective generator and resolution edge cases
This is the highest-priority open risk. The grammar (COLOR + CONDITION + COMPARISON) is elegant, but the document correctly flags that trivial, impossible, or ultra-narrow objectives must be filtered. “Exact 4” on a color a player has abandoned, or objectives that almost never produce a qualifier, will produce empty rounds. Empty rounds are death for pacing.

The proposed multi-qualifier resolution (best, then second-best, then third) is sensible and rewards having more qualifying dice, but it needs exhaustive testing. Ties that survive all three dice, and the exact rule when nobody qualifies, must be deterministic and feel fair. Right now these are still open.

2. Special faces and positional resolution
Specials are correctly constrained to literal dice/result manipulations (Kick, Copy Lowest Enemy, Copy Left, Reroll, etc.). That is good discipline. However:

• Kick is extremely high impact (removes an enemy die from the throw). It will dominate balance discussions.
• Positional specials (Copy Left/Right) depend on settling order, which is only probabilistically controllable via material. If the physics variance is high, players will feel the special “failed” even when they made the correct material choice. That is a classic “skill-looking RNG” trap that feels worse than pure dice RNG.
• Interaction with the 21-pip budget when a face becomes a special is still undefined. Does the special face consume its share of the pip budget or sit outside it? This changes the relative power of specials versus pure number dice.

Specials must remain rare and expensive or the game stops reading as a dice game.

3. Physics and material strategy
Making material a real strategic lever is ambitious and cool. It is also the easiest place for the prototype to fail. Mass, restitution, and friction differences must create noticeable but not deterministic differences in settling order without one material being strictly best or the simulation becoming chaotic/unreadable. Physics bugs, non-determinism across clients, or latency in the authoritative simulation will destroy trust faster than any balance issue. The prototype order in the GDD correctly puts physics last; that is wise. Do not let it leak into early fun tests.

4. Economy and stakes
The free-base-die + salvage (invested CP + 1) + permanent transfer model is one of the stronger parts of the document. It solves the “I lost everything and rage-quit” problem while still creating real stakes and a progression sink. Destroying a captured die for CP is clean.

Remaining risks:

• Heavily customized dice become extremely painful to lose. Some players will love the drama; others will quit after one bad match. The 10-second surrender timer helps, but the emotional experience of choosing which die to give away needs careful UX.
• Matchmaking by approximate spend (±2) is necessary. The exact formula for “spend” must include every CP-relevant edit or players will game it.
• CP generation rate (salvage + future competitions) must be tuned so that customization feels rewarding without creating an arms race that free bases cannot keep up with for new or returning players.

5. Match length and pacing
First to 3 is short and good for 1v1. But if a non-trivial percentage of objectives produce no qualifier, effective match length becomes random and sometimes draggy. The design needs a soft pressure against pure stall or ultra-safe non-qualification strategies.

6. Prototype and validation order
The listed test order is excellent and should be followed ruthlessly:

1. Plain D6s + changing objectives + 3-of-4 selection.
2. Color coverage value.
3. Pip redistribution (hunt for degenerate distributions early).
4. Geometry differences.
5. Tiny special set with deterministic resolution.
6. Persistence / surrender / salvage.
7. Materials last.

If step 1 is not already fun, nothing else will save it. Many designs look deep on paper and die at the pure selection + roll stage.

Edge Cases and Hidden Complexity

• Extreme mono-color or dual-color builds versus the objective pool. The pool must punish over-specialization without making coverage mandatory to the point of boring.
• Zero or near-zero faces, and maximum face values. Legal ranges need hard caps or the 21-pip budget produces pathological dice.
• Special resolution order interacting with qualification (does a Kick happen before or after qualification checks? Does a Copy happen before comparison?).
• Timeout behavior on both selection and surrender. Random selection on timeout is necessary for online play but will occasionally produce absurd results that feel like the server robbed the player.
• What happens if a player somehow ends up with fewer than four dice (should be impossible if free bases are truly free and instant).
• Long-term collection and history features are delightful for dedicated players and irrelevant or invisible to everyone else. They should not be required for the core loop to feel good.

Overall Assessment

Fun potential: High if the objective system is tight, the 3-of-4 decision feels impactful every round, and variance is bounded enough that correct play is rewarded more often than not. The ownership and transfer layer adds a layer of emotional stakes that pure score-based dice games lack. It can feel like a hybrid of probability puzzle, light collection, and high-stakes gambling with objects that have identity.

Skill development: Real and multi-layered. Players can improve at building, reading, selecting, and risk management. The skill ceiling is probably medium-high rather than extremely high - there is no perfect information sequencing or complex combo chains, just better and worse probability profiles and decisions under uncertainty. That is fine for the intended game.

Biggest current risks (in rough priority):

1. Objective generation quality and empty-round frequency.
2. Special power level and positional reliability.
3. Physics determinism, readability, and balance of materials.
4. Emotional impact of permanent die loss versus the free-base safety net.
5. Whether the pure selection + roll loop is already addictive before any of the fancy systems are added.

The document is honest about its open questions. That is a strength. Many of the remaining problems are solvable with iteration, but they are not small. The design will live or die on whether the changing objectives + 3-of-4 choice produces repeated moments of tension and cleverness, and whether losing a die feels dramatic rather than purely punishing.

If the early prototypes (especially the plain-dice objective tests) feel good, this can become a distinctive and replayable 1v1 game with a strong identity. If those tests feel flat or overly swingy, no amount of materials, specials, or die history will rescue it. The foundation is the round structure and the objective language. Everything else is amplification.