Draft faster, ship copy your editors still recognise

A quest with four branches is easy. A quest with four branches that all have to respect the codex is where teams lose weeks.

Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a human editing pass and budgets time for it.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

Version the prompts alongside the content. When a line changes six months later, you want to know which prompt produced it.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

What we changed

• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a
• Branching content needs a validation step, not just generation. We run every branch against the

Writers on Unity keep hitting the same wall: the barks file grows faster than anyone can review it. Here is the workflow we settled on.

Branching content needs a validation step, not just generation. We run every branch against the codex and flag contradictions before a human ever reads it.

Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it is in the localisation pipeline.

Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it is in the localisation pipeline.

Branching content needs a validation step, not just generation. We run every branch against the codex and flag contradictions before a human ever reads it.

Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a human editing pass and budgets time for it.

What we changed

• The single biggest quality gain came from feeding the codex in as structured context rather tha
• Keep the tone guide short. Long style documents produce worse output than three sharp examples
• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• Version the prompts alongside the content. When a line changes six months later, you want to kn
• Version the prompts alongside the content. When a line changes six months later, you want to kn

A quest with four branches is easy. A quest with four branches that all have to respect the codex is where teams lose weeks.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

What we changed

• Branching content needs a validation step, not just generation. We run every branch against the
• Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a
• Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it i
• Keep the tone guide short. Long style documents produce worse output than three sharp examples

Writers on Bevy keep hitting the same wall: the barks file grows faster than anyone can review it. Here is the workflow we settled on.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

Keep the tone guide short. Long style documents produce worse output than three sharp examples of the voice you actually want.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

Version the prompts alongside the content. When a line changes six months later, you want to know which prompt produced it.

What we changed

• Version the prompts alongside the content. When a line changes six months later, you want to kn
• Branching content needs a validation step, not just generation. We run every branch against the
• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• Version the prompts alongside the content. When a line changes six months later, you want to kn

Quest Outliner started as an internal script. After six studios asked for it, we rewrote it properly and it changed how our own team drafts Unity content.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a human editing pass and budgets time for it.

Keep the tone guide short. Long style documents produce worse output than three sharp examples of the voice you actually want.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

What we changed

• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• The single biggest quality gain came from feeding the codex in as structured context rather tha
• Branching content needs a validation step, not just generation. We run every branch against the
• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• Version the prompts alongside the content. When a line changes six months later, you want to kn

Quest Outliner started as an internal script. After six studios asked for it, we rewrote it properly and it changed how our own team drafts GameMaker content.

Keep the tone guide short. Long style documents produce worse output than three sharp examples of the voice you actually want.

Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a human editing pass and budgets time for it.

Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it is in the localisation pipeline.

What we changed

• The single biggest quality gain came from feeding the codex in as structured context rather tha
• Branching content needs a validation step, not just generation. We run every branch against the
• Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it i

Name Foundry started as an internal script. After six studios asked for it, we rewrote it properly and it changed how our own team drafts Twine content.

Branching content needs a validation step, not just generation. We run every branch against the codex and flag contradictions before a human ever reads it.

Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it is in the localisation pipeline.

Version the prompts alongside the content. When a line changes six months later, you want to know which prompt produced it.

What we changed

• Version the prompts alongside the content. When a line changes six months later, you want to kn
• Keep the tone guide short. Long style documents produce worse output than three sharp examples
• Branching content needs a validation step, not just generation. We run every branch against the

Writers on Unreal Engine 5 keep hitting the same wall: the barks file grows faster than anyone can review it. Here is the workflow we settled on.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

Version the prompts alongside the content. When a line changes six months later, you want to know which prompt produced it.

What we changed

• Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a
• The single biggest quality gain came from feeding the codex in as structured context rather tha
• Barks are the best place to start: high volume, low narrative risk, and the review cost per lin
• Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a
• The single biggest quality gain came from feeding the codex in as structured context rather tha

Localisation is where generated text usually falls apart. This is what we changed in Lore Keeper after the first round of QA came back.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

Keep the tone guide short. Long style documents produce worse output than three sharp examples of the voice you actually want.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

Version the prompts alongside the content. When a line changes six months later, you want to know which prompt produced it.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

What we changed

• Keep the tone guide short. Long style documents produce worse output than three sharp examples
• Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a
• Keep the tone guide short. Long style documents produce worse output than three sharp examples
• The single biggest quality gain came from feeding the codex in as structured context rather tha

Writers on Godot 4 keep hitting the same wall: the barks file grows faster than anyone can review it. Here is the workflow we settled on.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

Branching content needs a validation step, not just generation. We run every branch against the codex and flag contradictions before a human ever reads it.

Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a human editing pass and budgets time for it.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a human editing pass and budgets time for it.

What we changed

• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• The single biggest quality gain came from feeding the codex in as structured context rather tha
• Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it i

Tone Matcher started as an internal script. After six studios asked for it, we rewrote it properly and it changed how our own team drafts Twine content.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

Keep the tone guide short. Long style documents produce worse output than three sharp examples of the voice you actually want.

Keep the tone guide short. Long style documents produce worse output than three sharp examples of the voice you actually want.

What we changed

• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• The single biggest quality gain came from feeding the codex in as structured context rather tha
• Branching content needs a validation step, not just generation. We run every branch against the
• Version the prompts alongside the content. When a line changes six months later, you want to kn
• Version the prompts alongside the content. When a line changes six months later, you want to kn

Branch Mapper started as an internal script. After six studios asked for it, we rewrote it properly and it changed how our own team drafts Twine content.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it is in the localisation pipeline.

Branching content needs a validation step, not just generation. We run every branch against the codex and flag contradictions before a human ever reads it.

What we changed

• Do not let the tool name characters. Naming is cheap for a human and expensive to fix once it i
• Treat generated lines as a first draft, never a shipping asset. Every workflow below assumes a
• Branching content needs a validation step, not just generation. We run every branch against the
• Barks are the best place to start: high volume, low narrative risk, and the review cost per lin
• Keep the tone guide short. Long style documents produce worse output than three sharp examples

Name Foundry started as an internal script. After six studios asked for it, we rewrote it properly and it changed how our own team drafts Godot 4 content.

The single biggest quality gain came from feeding the codex in as structured context rather than pasting it into the prompt. Consistency errors dropped by more than half.

Barks are the best place to start: high volume, low narrative risk, and the review cost per line is small enough to measure improvement.

Measure editing time, not output volume. A tool that writes twice as much but takes the same time to fix has not helped anyone.

What we changed

• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• Branching content needs a validation step, not just generation. We run every branch against the
• Measure editing time, not output volume. A tool that writes twice as much but takes the same ti
• Branching content needs a validation step, not just generation. We run every branch against the