DDL · ATLAS · PROJECT PROPOSAL
Prompt Strategy System
Turn great moments into durable assets. A strategy library that is deterministic, replayable, searchable, and portable.
Metadata
Versionv0.1
StatusExploratory
OwnerDave + Dex
Updated2026-02-16
TemplateATLAS
Design Principles
Artifact-first
If it isn't an artifact, it doesn't exist
Strategy > Prompt
Prompts are implementation details
Portability > Polish
Survivability over aesthetics (early)
Tags must be consistent
Or the system collapses
Versioning prevents drift
And protects trust in reuse
Success Criteria (v0.1)
25Strategy Entries captured
< 2 minto answer 'best move here?'
10+entries validated on 2nd scenario
0reliance on platform memory
I
Problem StatementWe generate high-signal strategies in real time, but they decay if not externalized. They become unfindable when stored as raw chat. They drift when platforms and models change. They are hard to teach or reuse without structure.
This creates friction and lost leverage. The goal: turn "great moments" into durable assets.
💨
Idea Evaporation
Strategies generated in-session disappear after the thread closes
📦
Raw Chat Burial
Valuable patterns trapped in conversational noise
🌊
Model Drift
What worked on one model may not replay on the next
🧱
Teaching Friction
No structure means no reusability across contexts
II
ObjectiveCreate a strategy library with four properties:
A
Deterministic
Clear triggers, inputs, outputs, and boundaries
B
Replayable
Re-run against new project/model, get similar utility
C
Searchable
Indexed by outcome, skill level, domain, pattern type
D
Portable
Survives account loss, model drift, platform changes
III
ScopeIn Scope
Strategy Entries describing how to get outcomes with AI — intent, triggers, prerequisites, interaction choreography, guardrails, failure signals, examples, and provenance. Plus index, taxonomy, governance rules, and export formats (TXT/CSV first).
Out of Scope
Not a prompt marketplace. Not an investor pitch. Not a clinical tool. Not "the one true prompt." Not a mystical/anthropomorphic framing system. This is a practical craft library.
IV
Core Artifacts5.1
Strategy EntryDOSSIER
The canonical unit. One record capturing one reusable strategy.
5.2
Strategy IndexLEDGER
Master table — filterable by outcome, skill level, domain, pattern type, model compatibility, last validated date.
5.3
Strategy PacketsCODEX
Curated bundles of strategies by theme. Expandable, sequential, grouped.
5.4
Provenance LogLEDGER
Lightweight lineage log — origin context, author(s), version history, edits and rationale.
Planned Strategy Packets
Getting Unstuck
Strategies for when you're blocked, spinning, or overthinking
Specification to Build
Converting requirements into buildable artifacts
Data Cleanup & Structuring
Transforming messy inputs into governed data
Tone Calibration
Dialing voice, persona, and communication style
Council / Red Team Protocols
Multi-model coordination and adversarial review
Working with Constraints
Context limits, token budgets, platform boundaries
V
Data ModelStrategyEntry Fields (Core)
FIELDTYPEEXAMPLEREQ
StrategyIDstring✓
Titlestring✓
Outcomestring✓
WhenToUsestring✓
Inputsstring✓
Proceduretext✓
Guardrailstext✓
FailureSignalstext✓
Exampletext—
SkillLevelint 0-4✓
DomainTags#tags✓
PatternTags+tags✓
ToneTags%tags—
Provenancestring✓
Statusenum✓
LastValidateddate✓
Optional Fields (v0.2+)
ModelNotes
Claude/GPT/Local differences
Variants
A/B prompts or approaches
Dependencies
Requires other strategies
AntiPatterns
Common mistakes
VI
WorkflowFive-phase lifecycle from raw observation to published, reusable strategy entry.
📌
1.Capture
Flag a moment as 'Strategy Candidate' during work. Create a rough entry immediately as Draft.
🔬
2.Distill
Convert to canonical structure. Strip story, keep mechanics. Preserve one example.
🗂️
3.Index
Assign tags + skill level + status. Record provenance. File in the LEDGER.
✓
4.Validate
Re-run once on a new use case. Update LastValidated. If it fails, log FailureSignals and revise.
🚀
5.Publish
Mark ~Active when stable enough for reuse. Searchable, replayable, portable.
VII
GovernanceLightweight rules that prevent entropy without creating bureaucracy.
Append-firstDo not overwrite history without logging why
Attribution is factualHuman/model roles are recorded accurately
No canon warsMultiple strategies may exist for same outcome
Deprecated = searchableOld entries remain visible but clearly labeled
TIGHTEN is explicitConsolidation passes are invoked, never automatic
VIII
RoadmapOpen Questions (For Later)
Is "Capture → Distill → Validate" a CAE loop? Should strategies be decision records or playbooks? Do we need "Decision Lineage" fields for why a strategy exists?
Dropdown Logistics · Cottage — Humble surface. Cathedral underneath.
dropdownlogistics.com · 2026