Griffin Hale
← All projects

And More Will Follow

Playable prototype

A tile-board game that's secretly a programming game: you rig bounded state machines to play a 2048-style board, then read the trace of what they did.

GameIncrementalPuzzleGodot

At a glance

  • Outcome: A deterministic tile game of doubling patterns, proofs and curious machines. Underneath the merge board is a programming game: you rig bounded state machines to play the board, then read exactly what they did.
  • Status: Playable, worked on across 29 days between January and August 2026. The full suite passes 178/178. The August release close-out was overtaken by a redesign, and the gates behind that are still open.
  • Role: Solo.
  • Stack & libraries: Godot 4.6 and GDScript with data-driven schemas, targeting iOS.
  • Validation: 178/178, pinned to an evidence packet with no test or source file changed since it was recorded, plus captured runtime evidence packets for board behavior.
  • Limitations: Human and device review of the evidence packet is still open, and target-device QA, motion-comfort review, signing and content freeze all remain before release.

What it’s actually about

And More Will Follow replaces the usual click-to-progress loop with a 2048-style sliding board. But the goal is narrower and more interesting than “incremental game with automation.” An engaged player should be able to express increasingly sophisticated board strategies without writing code, and a player coming back after a break should be able to work out exactly what happened while they were away.

That second requirement is what forces the automation to be readable instead of clever.

Automata are bounded machines, not move optimizers

This is the whole case study.

You arrange directional steps into sequences and guard them with rules that read only what’s visible on the board and what has already happened. Rules run top to bottom and the first true one wins.

There’s no arithmetic, no variables, no recursion and no call stack. A rule can read what’s on the board. It can never predict, and it can never pick a best move. A step that turns out to be invalid still costs an attempt and moves the sequence along, and a pass that produces no moves at all is a stall, which stops the automaton and opens the trace at the exact point it failed.

Those limits are the teaching design, not missing features. An automaton you can’t read is worth nothing to a player trying to work out why their strategy lost.

The strongest proof that the routes are real: deploying an automaton flips the precision scorecard to ineligible, and it doesn’t flip back. You can’t claim the manual credit and the automation credit for the same run.

Three routes, one destination

Precision play, authored automata and independent machines all feed the same threshold. Route choice earns no bonus.

A deterministic simulation across 28 scenarios measured how differently they get there. In cluttered boards manual play completes every valid move while automata miss some and expose their recoveries. The automata route runs many more attempts but front-loads the setup time. Pure machines out-earn pure precision several times over with no board input at all.

First proof lands within a day or two whichever way you go, and no strategy beats every other on timing, effort, failure recovery and earnings at once.

The QA packet states its own waiver next to those numbers, and it’s better copy than anything a portfolio would write for itself: no routes were played, and the durations, scores and observations are inferred not watched.

The reset that preserves your work

Proof is the repeatable economy reset. Theorem is the structural one, and there are 25 of them written.

A theorem clears currencies, machine levels and discoveries, but preserves everything you wrote. Your routines, automata, build orders and blueprints survive, held inactive until their systems reopen. Each grants one fixed axiom meant to change a rule instead of adding a percentage, plus constants that fund a permanent map with no exclusive locks and no respec.

A routine or blueprint can never confirm a theorem on your behalf. A test enforces it.

What got retired, and how

The redesign replaced the old progression, harmonic and attunement systems with the three routes above.

Those were removed outright, and three tests fail if any of their files, managers or save data come back. An older automation system was superseded instead of deleted. Its scripts and panel are still in the tree as unreachable compatibility code, fenced by an audit that rejects any new use of its vocabulary.

Leaving the dead code visible and fenced turned out to be more useful than deleting it. The obsolete action rail stays in the history the same way, as chronology instead of something quietly erased.

What’s there

  • The core board loop, the machine catalog, and exact independent production with optional synergies.
  • The automata language and the workbench that authors it.
  • Machines, precision scoring, and the run to proof to theorem to axiom spine.
  • Board modes, achievements, onboarding and in-game toasts.
  • Data-driven schemas throughout, plus QA support systems and a 178-script test suite.

Screenshots

And More Will Follow: the Classic 4x4 merge board part-way through a run, tiles from 1 to 128, with the precision scorecard above it.
The merge board that drives progression in place of clicking. Above it, the precision scorecard for the run — a grade and a credit multiplier that deploying an automaton makes ineligible.
And More Will Follow: the Automata Rigging Workbench, with sequence paths of directional steps on the left, ordered rigging links in the middle, and a rule rack of sensor plates on the right.
The workbench where an automaton gets rigged: bounded sequence paths of directional steps, ordered links instead of crossing wires, and a rule rack read top to bottom where the first match wins. It edits a draft — the deployed automaton does not change until Deploy.
And More Will Follow: the Workshop upgrade screen, showing tiered upgrades and a Board Path branch whose two tier-0 options bias tile spawns in opposite directions.
The Workshop, spending proof marks and credits across tiers. The Board Path branch is the shape of the whole economy: Spawn Sharpener and Spawn Softener are explicit counterweights, so taking one costs you the other.
And More Will Follow: the Proof Preview dialog, listing what resets now, what persists to theorem, and what is permanent.
The Proof reset, stated before you take it: what clears now, what survives to the next theorem, and what is permanent. Routines, automata, build orders and blueprints are on the survive side — the reset never eats what you authored.
And More Will Follow: the same Automata workbench in a narrow portrait layout, broken into three stages with rig diagnostics beneath.
The same workbench on a phone. The three columns become three stages you page through, which is what makes the authoring surface usable on the iOS target.