svemagie/Substrate
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: Add Alternative Estimates section to schema and summary

Browse Source
This commit is contained in:
Daniel Miessler
2025-12-10 14:26:55 -08:00
parent dede1eacc8
commit 64ad357dce
2 changed files with 42 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
Data/DATASET-TEMPLATE.md
View File

@@ -78,17 +78,25 @@ Use this template for all new Substrate datasets. The key principle: **put the a
---
## Commonly Confused Metrics
## Alternative Estimates & Why We Differ
[Optional section - use when the metric is often confused with similar-sounding metrics]
[Recommended section - use when other estimates exist that might seem to contradict yours]
| Metric | Value | What It Actually Measures | Source |
|--------|-------|---------------------------|--------|
| **This dataset** | [Value] | [Description] | This research |
| **Similar metric 1** | [Value] | [Description] | [Source] |
| **Similar metric 2** | [Value] | [Description] | [Source] |
| Estimate | Source | What It Actually Measures | Why It Differs |
|----------|--------|--------------------------|----------------|
| **[Alternative 1]** | [Source] | [What it measures] | [Why different from ours] |
| **[Alternative 2]** | [Source] | [What it measures] | [Why different from ours] |
| **[Our estimate]** | This research | [What we measure] | [Our approach] |
**Do not compare these - they measure different things.**
### Why Our Approach
[2-4 bullet points explaining why you chose this measurement approach over alternatives:
- What makes it more appropriate for the question being answered
- Why it's more directly measurable or verifiable
- What constraints or sanity checks it passes
- Why apparent contradictions aren't actually contradictions]
**Key insight:** [One sentence explaining that different estimates often measure different things, not that one is "wrong"]
---
@@ -117,7 +125,8 @@ Use this template for all new Substrate datasets. The key principle: **put the a
4. **Quick Context** - 2-3 sentences max
5. **Methodology Summary** - How was this derived
6. **Sources** - Where did data come from
7. **Changelog** - Track all revisions
7. **Alternative Estimates & Why We Differ** - Recommended when other estimates exist
8. **Changelog** - Track all revisions
### Mandatory Fields in BEST ESTIMATE Table

24
Data/Knowledge-Worker-Global-Salaries/SUMMARY.md
View File

@@ -95,6 +95,30 @@ The wide global range reflects genuine uncertainty in international data, not he
---
## Alternative Estimates & Why We Differ
Various estimates for knowledge work value exist in the literature, but they often measure different things:
| Estimate | Source | What It Actually Measures | Why It Differs |
|----------|--------|--------------------------|----------------|
| **$5-7 trillion** | McKinsey Global Institute | Economic value of *automatable* knowledge tasks | Measures AI productivity potential, not compensation |
| **$2-3 trillion** | Various tech industry | Professional services market revenue | Revenue ≠ compensation; excludes in-house knowledge workers |
| **$70+ trillion** | Some extrapolations | Knowledge worker share of all economic output | Confuses GDP contribution with compensation; exceeds labor share ceiling |
| **$35-50 trillion** | This research | Actual wages + benefits paid to knowledge workers | Direct compensation measurement |
### Why Our Approach
We chose to measure **actual compensation paid** rather than productivity value or market revenue because:
1. **It's directly measurable** - BLS, ILO, and OECD track wages and benefits; productivity value requires modeling assumptions
2. **It's the right denominator for AI impact** - If you want to know what's at stake in the AI transition, you need to know what we actually pay people today
3. **It passes the math check** - Any estimate must fit within total global labor compensation (~$58T); productivity-value estimates often don't face this constraint
4. **It's definition-transparent** - We show exactly which occupational codes we include at each confidence level
The key insight: estimates that seem wildly different often just measure different things. A $5T automation-value estimate and a $40T compensation estimate can both be correct—they're answering different questions.
---
## Sources
**Primary (High Weight):**