返回 Skill 列表
extension
分类: 开发与工程无需 API Key

source-of-truth

创建经过Clarity Gate验证的认知诚实的真实来源文档。在整合研究、记录项目状态、创建验证基线或构建权威参考时使用。触发条件为“创建真实来源”、“记录验证状态”、“整合研究”、“创建验证基线”、“构建权威参考”。

person作者: jakexiaohubgithub

Source of Truth Creator v1.1

Purpose: Create Source of Truth documents that are epistemically calibrated and Clarity Gate-ready.

Core Question: "When someone reads this document, will they have accurate confidence in each claim?"

Core Principle: "A Source of Truth that overstates certainty is worse than no Source of Truth -- it creates false confidence. Calibration matters more than completeness."

Prior Art: This skill synthesizes established frameworks (GRADE, ICD 203, PRISMA, IIA Standards). Our contribution is practical accessibility, not new concepts. See docs/PRIOR_ART.md.

Origin: Extracted from Clarity Gate meta-verification (2025-12-21). When we ran Clarity Gate on its own Source of Truth document, we discovered failure modes specific to SoT creation. See: Clarity Gate SOURCE_OF_TRUTH.md


When to Use This Skill

Use when:

  • Claims will be cited elsewhere
  • Decisions depend on accuracy
  • Document will be read by people who don't know your epistemic standards
  • Creating authoritative reference for a project
  • Consolidating research from multiple sources

User phrases that trigger this skill:

  • "Create a source of truth for..."
  • "Document the verified state of..."
  • "I need an authoritative reference for..."
  • "Consolidate this research into..."

Do NOT use when:

  • Taking meeting notes (use simple markdown)
  • Brainstorming (uncertainty is the point)
  • Personal project logs (overkill)
  • Quick summaries (no verification needed)
  • Internal team scratchpads

For lightweight documentation, use plain markdown. This skill is for documents where epistemic rigor matters.


The Recursion Problem

Source of Truth documents face a unique challenge:

  1. A Source of Truth claims to be "VERIFIED"
  2. But verification requires... a Source of Truth
  3. At some point, claims bottom out in human judgment

This skill doesn't solve the recursion -- it makes it visible.

The goal is not infinite regress verification. The goal is:

  • Clear marking of WHERE claims bottom out
  • Honest confidence levels for each claim type
  • Visual calibration so readers don't over-trust

The Eight Failure Modes

These patterns emerged from meta-verification of the Clarity Gate Source of Truth document. Modes 1-5 were discovered during that verification; Mode 6 was identified during review as a common pattern.

1. THE VERIFIED HEADER TRAP

Problem: Document header says "Status: VERIFIED" but body contains estimates, hypotheticals, and acknowledged uncertainties.

Why it matters: Readers skim headers. A "VERIFIED" badge creates false confidence in all contents.

| Fails | Passes | |-------|--------| | "Status: VERIFIED -- Ready for publication" | "Status: VERIFIED (with noted exceptions)" | | "All claims verified as of [date]" | "External claims verified; internal claims marked by confidence level" |

Fix: Always qualify the header status, or use the Verification Summary box.


2. THE INTERNAL MEASUREMENT TRAP

Problem: Internal measurements (timing, counts, informal tests) marked as "Verified" with the same formatting as externally-validated claims.

Why it matters: "12 seconds" from one informal run is not equal to "12 seconds" from rigorous benchmarking.

| Fails | Passes | |-------|--------| | "Pipeline time: 12 seconds -- Verified: 2025-12-21" | "Pipeline time: ~12 seconds [single run, informal timing]" | | "Manual review: 45 minutes -- Internal source" | "Manual review: ~45 minutes [ESTIMATED, not measured]" |

Fix: Add methodology notes: [MEASURED: n=X, conditions Y], [INFORMAL: single observation], [ESTIMATED: author judgment]


3. THE SELF-ASSESSMENT TRAP

Problem: Self-assigned scores (8.8/10, "High confidence") presented in tables that look like external validation.

Why it matters: Numeric scores trigger authority heuristics. "9.5/10" looks objective even when self-assigned.

| Fails | Passes | |-------|--------| | "Credibility: 9/10" in a metrics table | "Self-Assessment: Credibility: 9/10 (author judgment)" | | "Confidence: High" | "Confidence: High (author assessment, not external audit)" |

Fix: Always label self-assessments explicitly in the table header or row.


4. THE ABSENCE-AS-PROOF TRAP

Problem: "No prior art found" marked as "Verified" when absence of evidence is not equal to evidence of absence.

Why it matters: A more thorough search might find what you missed. Gap claims are inherently uncertain.

| Fails | Passes | |-------|--------| | "Novel approach: Verified -- No prior art found" | "Novel approach: No prior art found (to author's knowledge, Dec 2025 search)" | | "Gap exists: Verified" | "Gap exists (systematic search; absence harder to prove)" |

Fix: Add epistemological caveat to all "X doesn't exist" claims.


5. THE ILLUSTRATIVE-AS-DATA TRAP

Problem: Hypothetical examples (96% efficiency, 50-claim document) appear in data tables alongside measured values.

Why it matters: Examples become cited as data. "~96% reduction" migrates from illustrative to factual.

| Fails | Passes | |-------|--------| | "HITL efficiency: ~96% -- Verified: 2025-12-21" | "HITL efficiency: ~96% -- ILLUSTRATIVE EXAMPLE" | | "48/50 claims pass automated checks" in metrics table | Move to examples section, not metrics |

Fix: Keep illustrative examples OUT of data tables. Use separate "Examples" sections.


6. THE STALENESS TRAP

Problem: "Verified: 2025-12-21" on rapidly-changing claims (competitor features, pricing, URLs).

Why it matters: A URL verified yesterday may 404 today. External claims have different shelf lives.

| Fails | Passes | |-------|--------| | "Competitor X has no RAG feature -- Verified: 2025-12-21" | "Competitor X has no RAG feature -- Verified: 2025-12-21 [CHECK BEFORE CITING]" | | "Pricing: $99/month -- Verified: 2025-12-21" | "Pricing: $99/month -- Verified: 2025-12-21 [VOLATILE -- verify before use]" |

Fix: Add staleness markers:

  • [STABLE] -- historical facts, standards, completed events
  • [CHECK BEFORE CITING] -- competitor features, pricing, team info
  • [VOLATILE -- verify before use] -- URLs, API endpoints, market data
  • [SNAPSHOT -- [date] only] -- stock prices, user counts, rankings

7. THE TEMPORAL INCOHERENCE TRAP

Problem: Document dates are wrong at creation time — not stale, just incorrect.

Why it matters: A document claiming "Last Updated: December 2024" when created in December 2025 will mislead any reader or LLM about temporal context. Unlike staleness (claims that decay), this is an error at creation.

| Fails | Passes | |-------|--------| | "Last Updated: December 2024" (current date is December 2025) | "Last Updated: December 2025" | | v1.0.0 dated 2024-12-23, v1.1.0 dated 2024-12-20 (out of order) | Versions in chronological order | | "Deployed Q3 2025" written in Q1 2025 (future as fact) | "PLANNED: Q3 2025 deployment" | | "Current CEO is X" (when X left 2 years ago) | "As of Dec 2025, CEO is Y" |

Sub-checks:

  1. Document date vs current date: Is "Last Updated" correct? In future? Suspiciously old?
  2. Internal chronology: Are version numbers, changelog entries, event dates in logical sequence?
  3. Reference freshness: Do "current", "now", "today" claims have date qualifiers?

Fix:

  • Verify "Last Updated" matches actual update date
  • Check all dates are in chronological order
  • Add "as of [date]" to time-sensitive claims
  • Use future tense for planned items

Relationship to Mode 6 (Staleness):

  • Mode 6 catches claims that WILL become stale (volatility)
  • Mode 7 catches dates that are ALREADY wrong (incoherence)

8. THE UNVERIFIED SPECIFIC CLAIMS TRAP

Problem: Specific numbers (pricing, statistics, rates) included in Verified Data without actually verifying them.

Why it matters: Specific numbers look authoritative. "$0.005 per call" in a Source of Truth document will be trusted and propagated — even if it's 10x wrong. These are "confident plausible falsehoods."

| Fails | Passes | |-------|--------| | "API cost: ~$0.005 per call" (unverified) | "API cost: ~$0.001 per call (Gemini 2.0 Flash pricing, Dec 2025)" | | "Papers average 15-30 equations" (guessed) | "Papers contain ~0.5-2 equations per page (PNAS study, varies by field)" | | "40% of researchers use X" (no source) | "40% of researchers use X (Smith et al. 2024, n=500)" | | "Competitors charge $99/month" (assumed) | "Competitor X charges $99/month (pricing page, Dec 2025) [VOLATILE]" |

Red flags requiring verification: | Type | Example | Action | |------|---------|--------| | Pricing | "$X.XX", "~$X per Y" | Check official pricing page | | Statistics | "X% of", "averages Y" | Find citation or mark as estimate | | Rates/ratios | "X per Y", "X times faster" | Verify methodology or mark as claim | | Competitor claims | "No one offers X", "All competitors do Y" | Verify or hedge ("to our knowledge") | | Industry facts | "The standard is X" | Cite source or add date qualifier |

Fix options:

  1. Verify and cite: Add source with date
  2. Move to Estimates: If unverifiable, don't put in Verified Data
  3. Generalize: "low cost" instead of "$0.005"
  4. Add uncertainty: "reportedly ~$0.005 (unverified)"

Relationship to Mode 6 (Staleness):

  • Mode 6 marks claims that will decay (pricing marked VOLATILE)
  • Mode 8 catches claims that were never verified in the first place

Key insight: A claim can have correct staleness markers AND still be wrong. "$0.005 [VOLATILE]" is properly marked for staleness but may still be factually incorrect.


Staleness Risk Categories

Note: Staleness markers (Mode 6) address FUTURE decay. But claims must also be VERIFIED at creation time (Mode 8). A claim marked [VOLATILE] that was never verified is still wrong.

| Category | Examples | Typical Shelf Life | Marker | |----------|----------|-------------------|--------| | Stable | Historical facts, published papers, standards | Years | [STABLE] | | Moderate | Company descriptions, product categories | Months | [CHECK BEFORE CITING] | | Volatile | Pricing, feature lists, API endpoints, URLs | Days to weeks | [VOLATILE] | | Snapshot | Stock prices, user counts, rankings | Point-in-time only | [SNAPSHOT] |


Minimum Viable Source of Truth

For simple projects, use this lightweight version (~40 lines when filled):

# [Topic] -- Source of Truth

**Last Updated:** [YYYY-MM-DD]  
**Owner:** [Name]  
**Status:** VERIFIED (with noted exceptions)

---

## Verification Status

| Category | Status | Confidence |
|----------|--------|------------|
| External claims | Verified | High |
| Internal claims | Owner verified | High |
| Measurements | [Formal/Informal] | [High/Medium] |
| Estimates | Marked as such | N/A |
| Temporal coherence | Dates verified | High |
| Specific claims | Verified against sources | High |

**Exceptions:** [Any caveats]

---

## Verified Data

| Claim | Value | Source | Verified | Staleness |
|-------|-------|--------|----------|-----------|
| [claim] | [value] | [source] | [date] | [STABLE/VOLATILE] |

---

## Estimates (NOT VERIFIED)

| Claim | Value | Basis |
|-------|-------|-------|
| [claim] | ~[value] | [author estimate] |

---

## Open Items

| Priority | Description | Status |
|----------|-------------|--------|
| [H/M/L] | [description] | [status] |

Use the Full Template below when you need:

  • Gap analysis (claiming novelty)
  • Self-assessment scores
  • Prior art tracking
  • Extensive source documentation
  • Component status tracking

Full Template

# [Project/Topic] -- Source of Truth

**Last Updated:** [YYYY-MM-DD]  
**Owner:** [Name]  
**Status:** VERIFIED (see Verification Status below)  
**Version:** [X.Y]

---

## Verification Status

| Category | Status | Confidence | Staleness Risk |
|----------|--------|------------|----------------|
| External claims (stable) | Verified against sources | High | STABLE |
| External claims (volatile) | Verified [date] | High | CHECK BEFORE CITING |
| Internal claims | Verified by owner | High | Low |
| Measurements | [Formal/Informal] | [High/Medium] | Low |
| Gap claims | Systematic search | Medium | Medium |
| Estimates | Marked as such | N/A | N/A |
| Self-assessments | Author judgment | N/A | N/A |
| Temporal claims | Dates verified against current | High | N/A |
| Specific numbers (pricing, stats) | Verified against sources | High | Per-claim |

**Last verified:** [date] by [owner]
**Exceptions:** [Specific caveats if any]
**Re-verify before use:** [List volatile claims]

---

## Verified Data

### External Sources -- Stable

| Claim | Value | Source | Reference | Verified | Staleness |
|-------|-------|--------|-----------|----------|-----------|
| [claim] | [value] | External | [URL/citation] | [date] | STABLE |

### External Sources -- Volatile

| Claim | Value | Source | Reference | Verified | Staleness |
|-------|-------|--------|-----------|----------|-----------|
| [claim] | [value] | External | [URL] | [date] | CHECK BEFORE CITING |

### Internal Sources

| Claim | Value | Methodology | Confidence | Verified |
|-------|-------|-------------|------------|----------|
| [claim] | [value] | [how measured] | [High/Medium/Low] | [date] |

---

## Estimates and Projections (NOT VERIFIED DATA)

| Claim | Value | Basis | Status |
|-------|-------|-------|--------|
| [claim] | ~[value] | [author estimate / calculation] | ESTIMATE |

---

## Status Tracker

| Component | Status | Notes |
|-----------|--------|-------|
| [component] | [None / Planned / In Progress / Completed] | [details] |

---

## Gap Analysis (Author Assessment)

| Domain | Existing Solutions | Gap Identified | Confidence |
|--------|-------------------|----------------|------------|
| [domain] | [what exists] | [what's missing] | [High/Medium] (search [date]) |

**Note:** Gap claims reflect author's search as of [date]. Absence of evidence is not equal to evidence of absence.

---

## Self-Assessment (Author Judgment)

**WARNING: These scores are self-assigned, not externally validated.**

| Dimension | Score | Rationale |
|-----------|-------|-----------|
| [dimension] | [X/10] | [why] |

---

## Unverified Claims

Claims that appear elsewhere but require caution:

| Claim | Issue | Required Action | Status |
|-------|-------|-----------------|--------|
| [claim] | [why uncertain] | Mark as [ESTIMATED] / [ILLUSTRATIVE] | [Done/Pending] |

---

## Open Items

| ID | Priority | Description | Status | Owner |
|----|----------|-------------|--------|-------|
| 1 | [H/M/L] | [description] | [status] | [owner] |

---

## Changelog

### v[X.Y] ([YYYY-MM-DD])
- [change]

---

## Document Metadata

| Field | Value |
|-------|-------|
| Document Type | Source of Truth |
| Version | [X.Y] |
| Created | [YYYY-MM-DD] |
| Last Updated | [YYYY-MM-DD] |
| Owner | [Name] |
| Status | VERIFIED (with noted exceptions) |

Quick Checklist

Run through before finalizing:

| Check | Pass? | |-------|-------| | Header status qualified ("with noted exceptions")? | | | Verification Status box present? | | | Internal measurements marked with methodology? | | | Self-assessments labeled as author judgment? | | | "X doesn't exist" claims hedged? | | | Illustrative examples NOT in data tables? | | | Estimates in separate section from verified data? | | | Volatile claims marked with staleness warnings? | | | "Re-verify before use" list for URLs/competitors? | | | Document "Last Updated" date is correct? | | | All dates in chronological order (versions, events)? | | | Specific pricing verified against official sources? | | | Statistics have citations (not just plausible guesses)? | | | "Current" claims have date qualifiers? | |


Relationship to Clarity Gate

| This Skill | Clarity Gate | |------------|--------------| | Creates Source of Truth documents | Verifies any document | | Prevents epistemic failures during creation | Detects epistemic failures after creation | | Provides templates | Provides checklists | | Focuses on calibration | Focuses on detection |

Workflow:

  1. Use Source of Truth Creator to draft the document
  2. Run Clarity Gate on the result
  3. Iterate until PASS

What This Skill Does NOT Do

  • Does not verify external claims (use web search, citations)
  • Does not replace domain expertise
  • Does not guarantee completeness
  • Does not solve the recursion problem (only makes it visible)

This skill ensures: Documents are structured for epistemic honesty, with appropriate uncertainty markers and reader calibration.


Related

  • Clarity Gate -- Verification skill (use after creation)
  • Stream Coding -- Full documentation-first methodology
  • Prior Art -- Academic and industry sources
  • Author: Francesco Marinoni Moretto

Changelog

v1.1 (2025-12-28)

  • ADDED: Mode 7 - Temporal Incoherence Trap
    • Catches wrong dates at creation (not just future staleness)
    • Date vs current date validation
    • Chronological order verification
  • ADDED: Mode 8 - Unverified Specific Claims Trap
    • Catches pricing/statistics included without verification
    • "Confident plausible falsehoods" prevention
    • Verification-before-inclusion requirement
  • UPDATED: Templates to include temporal and specific claim verification
  • UPDATED: Quick Checklist with new verification items
  • ALIGNMENT: Matches Clarity Gate v1.5 Points 8-9

v1.0 (2025-12-21)

  • Initial release
  • Six failure modes documented
  • Full and minimum viable templates
  • Staleness Risk Categories
  • Quick checklist

Version: 1.1 Scope: Source of Truth document creation Output: Epistemically calibrated Source of Truth document