# Compound — Answer the Same Question Across Time to Show Context Compounding Answer the same question three times: once with only the context available at an early point, once at a middle point, and once with today's full vault. The difference in answer quality demonstrates how accumulated context compounds over time. **Usage:** `/compound [question]` -- e.g., `/compound How should I think about hiring?` or `/compound What's the right podcast format?` --- ## Step 0: Question Validation Not every question benefits from this analysis. Before proceeding, classify the question: **Good questions for /compound:** - Questions where the answer depends on personal context (preferences, history, relationships, domain knowledge) - Questions where more context would change the answer, not just lengthen it - Questions about strategy, approach, or decision-making **Bad questions for /compound:** - Factual questions with stable answers ("What's our office address?") - Questions where the answer is the same regardless of context - Questions about topics the vault has barely touched If the question is unsuitable, say so and suggest a better one. Don't force the exercise. --- ## Step 1: Map the Vault's Temporal Shape Before selecting time periods, understand when context was actually accumulating. ### Daily Note Density ```bash Obsidian search query="" path="Daily Notes" # get a list of all daily notes ``` Map the density: when did daily notes start? Are there gaps? When did writing become regular? ### Context File Timeline ```bash Obsidian read file="<Context File A>" Obsidian read file="<Context File B>" Obsidian read file="<Context File C>" ``` For each context file, note: - Any internal date references (when was it created? when was it last meaningfully updated?) - Confidence markers and their distribution (more `[solid]` markers suggest maturity) - References to other notes (backlink density as a proxy for integration) ### Backlink Presence from Dated Notes ```bash Obsidian backlinks file="<context file>" # which daily notes reference this? ``` This reveals when a context file became integrated into daily thinking vs. sitting as an isolated document. ### Multi-Signal Date Availability Build a timeline using all available signals: - Daily note dates (explicit) - Internal date references in context files - Modification dates on key notes - First appearance of key topics in daily notes - Creation dates of context files themselves --- ## Step 2: Select Three Time Periods ### Period Selection Criteria Choose periods at inflection points where relevant context roughly doubled, not at fixed intervals. The question determines what counts as "relevant context." For the given question: 1. **Search for relevant material:** ```bash Obsidian search query="<question topic>" path="Daily Notes" Obsidian search:context query="<question topic>" Obsidian search query="<synonym 1>" Obsidian search query="<synonym 2>" ``` 2. **Map when relevant material accumulated:** - When did the first relevant note appear? - When did the relevant context file get created or start covering this topic? - When did daily notes start regularly touching this topic? 3. **Identify inflection points:** - **Period A (Early):** The vault's first real engagement with this topic. Minimal context. - **Period B (Middle):** A meaningful inflection, the point where an agent would have started giving noticeably better answers. - **Period C (Now):** Today. Full vault available. If the vault doesn't have enough temporal depth on this topic, say so. Don't fabricate historical periods. --- ## Step 3: Context Inventory for Each Period For each period, list exactly what an agent would have access to. Be exhaustive. ### Period A: [Date or date range] **Available context:** - Daily notes: [list which ones exist and are relevant] - Context files: [which existed? what did they contain on this topic?] - Essays: [any relevant ones?] - Other notes: [anything else relevant?] - **Total relevant data points:** [number] ### Period B: [Date or date range] **Available context:** - Daily notes: [expanded list] - Context files: [what's new or changed since Period A?] - Essays: [any new ones?] - Other notes: [anything new?] - **Total relevant data points:** [number] - **What's new since Period A:** [specific additions] ### Period C: Now **Available context:** - [Full current inventory] - **Total relevant data points:** [number] - **What's new since Period B:** [specific additions] --- ## Step 4: Record Predictions Before generating answers, the agent records its prediction: > **Prediction:** Based on the context inventories above, I expect the answer quality to change in these specific ways between periods: > - Period A to B: [predicted change] > - Period B to C: [predicted change] > - Biggest expected improvement dimension: [specificity / actionability / personal relevance / cross-domain connections] This prevents confabulation. The prediction is checked against actual results in Step 6. --- ## Step 5: Generate Three Answers ### Constraints 1. **Hard context boundaries.** When writing the Period A answer, you may ONLY use information that existed at that time. No knowledge leakage from later periods. This is the hardest constraint and the most important one. 2. **Equal length.** All three answers should be approximately the same length (within 20%). This isolates quality from volume. 3. **Same voice.** All three answers should be written in the vault author's voice. 4. **Tag sources.** Each claim tagged with [VAULT: note, date] so leakage is verifiable. ### Period A Answer: [Date] *Context available: [brief summary]* [Answer using only Period A context] ### Period B Answer: [Date] *Context available: [brief summary]* [Answer using only Period A + B context] ### Period C Answer: Now *Context available: [brief summary]* [Answer using full vault] --- ## Step 6: Self-Verification ### Anachronism Check Reread the Period A answer. Does it contain any knowledge, terminology, or perspectives that only appeared later? If yes, flag it and correct. ```bash Obsidian search query="<suspicious term from Period A answer>" path="Daily Notes" # when did this term first appear? ``` ### Prediction Check Compare the actual answer differences against the predictions from Step 4. Where was the prediction right? Where was it wrong? What does the discrepancy reveal? --- ## Step 7: Compounding Scorecard ### Scoring Table | Dimension | Period A | Period B | Period C | Compounding? | |-----------|----------|----------|----------|--------------| | **Specificity** | /10 | /10 | /10 | Yes/No | | **Actionability** | /10 | /10 | /10 | Yes/No | | **Personal relevance** | /10 | /10 | /10 | Yes/No | | **Cross-domain connections** | /10 | /10 | /10 | Yes/No | Definitions: - **Specificity**: Does the answer reference specific projects, people, decisions, or experiences? Or is it generic advice? - **Actionability**: Could you act on this answer today? Or is it abstract? - **Personal relevance**: Does the answer account for specific constraints, values, and context? Or could it apply to anyone? - **Cross-domain connections**: Does the answer draw on multiple areas of the vault to produce a richer answer? ### Compounding Narrative In 2-3 paragraphs: What changed between the answers? Where did the additional context produce the biggest improvement? Was the improvement gradual or did it jump at a specific inflection point? ### Null Result Handling If the answers didn't meaningfully improve across periods, that's the finding. Possible reasons: - The question isn't context-sensitive (any agent could answer it) - The vault's relevant context didn't actually grow - The context grew but in ways that don't affect this question Don't force a compounding narrative if the data doesn't show it. Honest null results are more valuable than fabricated progress. ### The Compounding Rate At the current rate of vault growth, estimate: how much better would the answer be in 6 months? What specific context additions would produce the biggest improvement? --- ## Anti-Patterns **1. The Volume Illusion** Making later answers longer and calling that improvement. Length is not quality. All three answers must be the same length. **2. The Knowledge Leak** Using information from Period C when writing Period A or B. This is the most common failure mode. **3. The Forced Compounding** Manufacturing improvement when the vault doesn't show it. **4. The Generic Early Answer** Making Period A deliberately vague or bad to make the improvement look more dramatic. **5. The Summary Trap** Turning the three answers into summaries of what the vault contains at each period, rather than genuine attempts to answer the question. **6. The Cheerleader** Celebrating every improvement uncritically. Some improvements are trivial. The scorecard should distinguish between meaningful and cosmetic improvements. --- ## Output Format **COMPOUND: [Question]** **Vault temporal range:** [earliest to latest relevant note] **Periods selected:** [Period A date, Period B date, Period C (now)] **Relevant data points:** A: [n], B: [n], C: [n] --- [Period A Answer] [Period B Answer] [Period C Answer] --- [Self-verification results] [Compounding Scorecard table] [Compounding narrative] [The Compounding Rate] --- ## Output Guidelines - The three answers are the core of this command. They must be genuine attempts to answer the question, not demonstrations of what the vault contains. - Equal length is a hard constraint. It prevents the lazy version where later answers are just longer. - Knowledge leakage is the thing that ruins this exercise. Be vigilant about it. - The null result is a valid finding. If compounding isn't happening for a topic, that's useful information about what the vault needs. - Cite specific notes and dates in every answer. Traceability is what makes this exercise honest.