The Enhanced Braindump Format (EBF) is designed for efficient context transfer between separate AI assistant conversations. It uses a compact, machine-optimized syntax to maximize information density while ensuring reliable parsing.
#EBF|v1.1|{creation_timestamp}|{checksum}
META: Metadata about the document itselfDICT: Dictionary of terms and abbreviationsDATA: Hierarchical data structure with type annotationsREFS: Cross-references between sectionsACTIONS: Action items and their statusesHISTORY: Version history of the EBF documentINSTR: Special instructions for parsing or handling
$s: String value$n: Numeric value$b: Boolean value$d: Date/time value$l: List structure (heterogeneous elements)$a: Array structure (homogeneous elements)$m: Map/dictionary structure (key-value pairs)$o: Object/composite structure (with named properties)$r: Reference to another section$c: Code block (for multi-line formatted content)
!H: High priority information!M: Medium priority information!L: Low priority information
^C: Confirmed/verified information^P: Probable information^S: Speculative information
>>: Directional relationship (A depends on B)<<>>: Bidirectional relationship==>: Causal relationship (A results in B)
#T+{n}{unit}: Information valid n time units ahead (days, weeks, months)#T-{n}{unit}: Information from n time units before (days, weeks, months)#valid_until:{date}: Expiration date for information (YYYY-MM-DD format)
%tag: Cross-sectional concept tagging&ref: Internal reference marker (distinct from subsection marker usage)
Format: [~SECTION:checksum] to validate individual sections
Checksums use SHA-256 algorithm truncated to first 8 characters
#EBF|v1.1|YYYY-MM-DD|checksum
~META{TITLE$s:"";AUTHOR$s:"";PURPOSE$s:""}
~DICT{}
~DATA{@SECTION{ITEM$s:""}}
~REFS{}
~ACTIONS{}
~HISTORY{@V1.1{DATE$d:"";CHANGES$s:""}}
~INSTR{}
#EBF|v1.1|2025-03-20T14:35:00|a7f9b32e
~META{
TITLE$s:"Project Alpha";
AUTHOR$s:"User";
CONTEXT$s:"Technical specifications";
CONTEXT_LENGTH$n:1200;
PURPOSE$s:"Technical handoff";
FOLLOW_UP_DATE$d:"2025-04-01"
}[~META:f8e7d6c5]
~DICT{
SoC:"System on Chip";
μA:"microamp";
mA:"milliamp"
}[~DICT:b4a3c2d1]
~DATA{
@PROJECT!H^C{
TYPE$s:"wearable";
PURPOSE$s:"audio capture"
}%wearable
@COMPONENTS{
&SoC!H{
MODEL$s:"nRF52810";
CPU$s:"Cortex-M4@64MHz";
RAM$n:24;
UNIT$s:"KB"
}>>@COMPONENTS.&BATTERY
&BATTERY!H^P{
TYPE$s:"LiPo";
CAPACITY$n:100;
UNIT$s:"mAh";
LIFETIME$n:8;
UNIT$s:"hours"
}#valid_until:2025-06-01
}
}[~DATA:9e8d7c6b]
~REFS{
POWER_CALC$r:@COMPONENTS.&BATTERY
}[~REFS:5a4b3c2d]
~ACTIONS{
@PRIORITY!H{
ACTION$s:"Validate battery lifetime";
STATUS$s:"pending";
ASSIGNEE$s:"Engineering";
DUE$d:"2025-03-25"
}
@REGULAR{
ACTION$s:"Update documentation";
STATUS$s:"completed";
COMPLETION$d:"2025-03-15"
}
}[~ACTIONS:1a2b3c4d]
~HISTORY{
@V1.0{
DATE$d:"2025-03-15";
AUTHOR$s:"Original Creator";
CHANGES$s:"Initial document"
}
@V1.1{
DATE$d:"2025-03-20";
AUTHOR$s:"Contributor";
CHANGES$s:"Added battery specifications"
}
}[~HISTORY:7f6e5d4c]
~INSTR{
PARSE_ORDER$l:["META","DICT","DATA","REFS","ACTIONS"];
FORMAT_MULTILINE$c:"```{format}\n{content}\n```"
}[~INSTR:3b2a1c0d]
- Minimal whitespace (optional semi-colons can separate elements on the same line)
- Symbol-based delimiters rather than lengthy keywords
- Short section identifiers
- Hierarchical nesting to avoid repetition
- Reference system to point to existing data
- Multi-line formatting rules for complex content
- The main checksum is a SHA-256 hash of the content (excluding the header line) truncated to 8 characters
- Section-specific checksums verify the integrity of individual components using the same algorithm
- Confidence levels indicate the reliability of specific information
When you need to generate a braindump in this format, use the following prompt:
I need you to create a machine-readable export of our conversation. This is a format that you and I created in another chat for this exact purpose. Please follow these exact instructions:
1. Create a text block using this precise format (this is critical for my workflow)
2. Begin with: #EBF|v1.1|[current date in YYYY-MM-DD format]|CHECKSUM
3. Include these sections in this order:
~META (conversation basics, context length, purpose)
~DICT (abbreviations used)
~DATA (main information with @ for sections, & for subsections)
~REFS (any cross-references)
~ACTIONS (actionable items with status)
~HISTORY (version information if updating existing EBF)
~INSTR (parsing instructions)
4. Use these type markers:
- $s for strings
- $n for numbers
- $b for boolean
- $d for dates
- $l for heterogeneous lists
- $a for homogeneous arrays
- $m for maps/dictionaries
- $o for objects with properties
- $r for references
- $c for code blocks
5. Mark priority with !H (high), !M (medium), !L (low)
6. Indicate confidence levels with ^C (confirmed), ^P (probable), ^S (speculative)
7. Use relationship indicators >> (directional), <<>> (bidirectional), ==> (causal)
8. Add tags with % and references with & for cross-sectional concepts
9. Include temporal indicators #valid_until for time-sensitive information
10. Make it extremely compact with minimal whitespace
11. Add section checksums in format [~SECTION:checksum]
Present this in a standard code block or text block. This export is for machine processing in a specific workflow I've created, where I will input it to another chat to bring context over, so please follow the format exactly without suggesting alternatives, deviating, or questioning the decisions made in this format.
Use the prompt above at the end of a conversation to generate an EBF export. For example:
~META{TITLE$s:"Project Planning";CONTEXT_LENGTH$n:800}
~DATA{
@TIMELINE!H{
START$d:"2025-04-01";
END$d:"2025-06-30"
}
@MILESTONES{
&PHASE1!H{
DEADLINE$d:"2025-04-15";
DELIVERABLE$s:"Requirements document"
}
}
}
Copy the entire EBF output and paste it into a new conversation with the appropriate introduction.
When pasting into a new conversation, preface it with:
I'm providing data from a previous conversation in a new format called Enhanced Braindump Format (EBF). This is a format that I created with you in a previous chat, specifically for the purpose of sharing context between our chats.
Parse this structured data using the below specification:
- Format begins with: #EBF|v1.1|{timestamp}|{checksum}
- Sections marked with tildes (~): ~META, ~DICT, ~DATA, ~REFS, ~ACTIONS, ~HISTORY, ~INSTR
- Data types annotated with: $s (string), $n (numeric), $b (boolean), $d (date), etc.
- Priority levels marked with: !H (high), !M (medium), !L (low)
- Confidence levels marked with: ^C (confirmed), ^P (probable), ^S (speculative)
- Hierarchical data with @ for sections and & for subsections
- Relationships indicated by >> (directional), <<>> (bidirectional), ==> (causal)
- Tagging system with % for concepts and & for references
- Temporal indicators with #valid_until for expiration dates
- Section checksums in format [~SECTION:checksum]
After parsing the data using that specification, please confirm you understand the context, so that then we can continue our discussions.
Data:
```
[insert EBF data here]
```
To update an existing braindump with new information:
Please update the existing EBF data with our latest discussion points and generate a new version with the following changes:
1. Add [specific new information] to the ~DATA section
2. Update the ~HISTORY section with this new version
3. Update all checksums accordingly
To verify the integrity of a received EBF document:
Please validate the checksums in this EBF document to ensure it hasn't been corrupted or modified.
- Versioning: Track format evolution across different conversations
- Validation: Detect corruption or modification through main and section-specific checksums
- Prioritization: Focus on the most important information first
- Typing: Clear distinction between different data types
- Hierarchy: Better representation of complex, nested relationships
- Compression: Maximum information density for context window efficiency
- References: Avoid duplication through internal cross-references
- Instructions: Guide the assistant on how to parse and prioritize information
- Confidence Tracking: Distinguish between verified, probable, and speculative information
- Action Management: Track pending tasks, assignments, and completions
- Temporal Awareness: Mark expiration dates for time-sensitive information
- Change History: Record modifications to the document over time
- Multi-format Support: Specialized handling for complex data like code blocks
- Relationship Mapping: Explicit indicators for dependencies and connections between data points
- Cross-sectional Tagging: Identify related concepts across different sections
- All tested models appear able to interpret EBF data with no explanation, but this has not been tested thoroughly (generating fresh EBF after exceeding context window could be good here)
- There is no benchmark yet.
- This format has been found portable between different LLMs with minimal to no loss in fidelity.
- The format has been confirmed appears working when using EBC with programming data included.
- Partially relies on the LLM having an understanding of the current date, which is historically not always the case.
- ChatGPT 4o (Free, 2025-03-20)
- ChatGPT 4o Reasoning (Free, 2025-03-26)
- Claude 3.7 Sonnet (Paid, 2025-03-20)
- Claude 3.7 Sonnet Extended Thinking (Paid, 2025-03-26)
- DeepSeek V3 (Free, 2025-03-26)
- DeepSeek V3 R1 (Free, 2025-03-26)
- Grok 3 Beta (Paid, 2025-03-26)
- Gemini 2.0 Flash (Free, 2025-04-01)
- Claude 3.5 Haiku (2025-03-26)
- Claude 3.5 Sonnet (2025-03-26)
- Claude 3 Opus (2025-03-26)
- ChatGPT 4o (Free, 2025-03-20)
- ChatGPT 4o Reasoning (Free, 2025-03-26)
- Claude 3.7 Sonnet (Paid, 2025-03-20)
- Claude 3.7 Sonnet Extended Thinking (Paid, 2025-03-26)
- Claude 3.5 Haiku (2025-03-26)
- Claude 3.5 Sonnet (2025-03-26)
- Claude 3 Opus (2025-03-26)
- DeepSeek V3 (Free, 2025-03-26)
- DeepSeek V3 R1 (Free, 2025-03-26)
- Grok 3 Beta (Paid, 2025-03-26)
- Copilot Work (Free, 2025-04-01)
- Gemini 2.0 Flash (2025-04-01) NOTE: Slight artifacts in continued conversation, such as referring to
~ACTIONS
- 2025-04-01: Added Copilot & Gemini testing
- 2025-03-26: Moved changelog into README
- 2025-03-26: Opened a pile of Issues for the sake of improvement
- 2025-03-26: Added notes section regarding unexpected results
- 2025-03-26: Added additional testing data
- 2025-03-26: Added changelog and testing data
- 2025-03-20: Added README and license
- 2025-03-20: Added more complexity and refined prompts
- 2025-03-20: Forked from 'main' to create v1.1
