Research¶

Complete evidence archives for every research instance. Every source scored, every search logged. Nothing summarized away.

How This Is Organized¶

Research Instance → a specific investigation (e.g., R0005). Tagged by subject matter. Multiple runs over time.

Run → a specific execution of the research. Timestamped, versioned by prompt and model. Runs within an instance are re-executions of the same research, not additions of new items. Each run is independent.

Entity → the individual items within a run: claims or queries. Each has a full evidence archive.

The drill-down: Instance → Run → Entity → Evidence (sources, searches, scorecards). Every level has a summary. Every level links to the next level of detail.

ID Namespace¶

Within a single claim or query, all entity IDs are unique — no prefix needed. Fully-qualified IDs (e.g., C001-SRC01-E01) appear only in page titles to establish position in the hierarchy. Everywhere else, use the short form.

Entity	Short form	Scope	Notes
Hypothesis	`H1`	Entity	H1=affirmative, H2=negative, H3=nuanced
Search	`S01`	Entity	Unique within entity
Source	`SRC01`	Entity	Unique within entity
Search Result	`S01-R01`	Entity	Scoped to search
Evidence Extract	`SRC01-E01`	Entity	Scoped to source

Research Archive¶

R0055 — A0022 Sycophancy Article Fact-Check

Tags: ai, sycophancy · Runs: 2026-04-01 · 28 claims

3 corrections (synthetic data effectiveness, Microsoft Research claim removed, CaTE acronym), 5 softenings. 3 Certain, 5 Almost certain, 8 Very likely, 10 Likely, 2 Very unlikely.

A0022 — AI Sycophancy: The Yes-Man Problem

ai

R0054 — A0013 Prompt Article Claims v2

Tags: methodology · Runs: 2026-03-31

Second-pass claim verification for Part 2 methodology article.

A0013 — The Truth is Out There. Now Go Find It. (Part 2)

methodology

R0053 — A0013 Prompt Article Claim Verification

Tags: methodology · Runs: 2026-03-31 · 7 claims

Choe uniqueness claim softened (other frameworks exist). Enforcement language nuanced (contradicts general guidance but empirically effective for behavioral override). All structural claims confirmed.

A0013 — The Truth is Out There. Now Go Find It. (Part 2)

methodology

R0052 — A0012 Methodology Article Claim Verification

Tags: methodology · Runs: 2026-03-31 · 14 claims

10 Almost certain, 4 Very likely. Blanket negatives softened. All framework descriptions confirmed. Negative existence claims acknowledged as inherently limited.

A0012 — The Truth is Out There. But How Do You Find It? (Part 1)

methodology

R0051 — Fact-Checking Methodology Gap Analysis

Tags: methodology · Runs: 2026-03-31 · 3 queries

No formal epistemological framework in fact-checking comparable to GRADE, IPCC, or ICD 203. W3C Credibility Coalition dormant since 2020. Academic literature documents the gap but scattered across disciplines.

A0012 — The Truth is Out There. But How Do You Find It? (Part 1)

methodology

R0050 — Journalism and Other Truth-Seeking Disciplines

Tags: methodology · Runs: 2026-03-31 · 3 queries

Journalism is principles-based, not methodology-based. Five disciplines contribute novel concepts: law, auditing, FMEA, historical source criticism, SIFT. Wardle-Derakhshan taxonomy widely adopted as vocabulary but not operationalized.

A0012 — The Truth is Out There. But How Do You Find It? (Part 1)

methodology

R0049 — Published AI Research Methodology Prompts

Tags: methodology · Runs: 2026-03-31 · 3 queries

No complete AI prompt implementing a full analytical rigor framework. No unified IC + scientific methodology published. Rich AI tool ecosystem but none implements structured evidence evaluation.

A0012 — The Truth is Out There. But How Do You Find It? (Part 1)

methodology

A0013 — The Truth is Out There. Now Go Find It. (Part 2)

methodology

R0048 — Corporate AI Training and Sycophancy Awareness

Tags: ai, methodology · Runs: 2026-03-29 · 3 queries

No corporate or government training warns about sycophancy under any name. 82% have AI training but it's superficial. 40% zero-scrutiny rate. Hallucination taught as undifferentiated, missing the spectrum.

R0047 — Source-Back Verification Test

Tags: methodology · Runs: 2026-03-29 · 1 query

Validated the source-back verification methodology. Caught two framing discrepancies in R0045 that the process self-audit missed.

R0045 — 2001 OSCON Prediction Verification

Tags: technology, history · Runs: 2026-03-29 · 7 queries

Verifying predictions from the 2001 OSCON keynote: Linux vs Solaris market share, Apache trajectory, industry consensus on open source, Doc Searls article, SCO chilling effect, Wall Street open source adoption timeline.

R0044 — Expanded Vocabulary Sycophancy Research

Tags: ai, methodology · Runs: 2026-03-29 · 4 queries

Re-searched sycophancy topics using human-side vocabulary (automation bias, overtrust, complacency). Central finding: all regulations address system design but not system output. Design vs output gap confirmed.

R0043 — Sycophancy Vocabulary Mapping

Tags: ai, methodology · Runs: 2026-03-28 · 3 queries

System-side vs human-side vocabulary asymmetry. AI safety says "sycophancy"; regulated industries say "automation bias," "complacency," "overtrust." No shared vocabulary bridges them. Every bridging taxonomy omits sycophancy.

R0042 — Private AI Enterprise Motivations

Tags: ai, methodology · Runs: 2026-03-28 · 3 queries

Enterprise private AI motivated by data sovereignty, security, and compliance. Behavioral customization (sycophancy control) absent from the conversation. Anti-sycophancy treated as model provider's problem.

R0041 — Enterprise Sycophancy Reduction

Tags: ai, methodology · Runs: 2026-03-28 · 3 queries

No vendor offers enterprise anti-sycophancy products. No deployment has sycophancy reduction as a stated requirement. RLVR limited to verifiable domains.

R0040 — RLHF Alternatives and Sycophancy Root Cause

Tags: ai, methodology · Runs: 2026-03-28, 2026-03-29 rerun · 2 queries

Six RLHF alternatives mapped. Critical finding: sycophancy root cause is biased preference data, not the RL algorithm. Rerun found sycophancy is mildest form of reward hacking; covert sycophancy risk identified.

R0031 — A0021 Claim Verification (blind)

Tags: ai, methodology · Runs: 2026-03-27, 2026-03-29 rerun · 14 claims

Blind verification of A0021 claims. Two corrections applied (C002 misattribution, C012 wrong journal name). Rerun caught NeurIPS exception to AI authorship prohibition.

R0029 — AI Attribution Frameworks and Public Sentiment

Tags: ai, methodology · Runs: 2026-03-27 · 5 queries

AI attribution frameworks (IBM toolkit, AIA icons, CHI 2025), public sentiment (46% global trust, 57% hide AI use), journal AI authorship policies (universal prohibition), Kurosawa/Shakespeare filmography.

R0028 — A0019 Blind Claim Verification

Tags: ai, methodology · Runs: 2026-03-26 · 33 claims

Blind fact-check of A0019. 504 files. Five corrections applied. 100% source convergence with query research. All disagreements directional (blind checker more conservative).

A0019 — Prompt Engineering Is Not. Engineering, That Is.

Published ~2026-03-26 · technology, ai

R0027 — Multilingual Prompt Engineering Challenges

Tags: ai, methodology · Runs: 2026-03-26 · 3 queries

Cross-language performance gaps: 3-30pp depending on language. Vendor guides all English-only. No ISO/IEC standard for prompt engineering.

A0019 — Prompt Engineering Is Not. Engineering, That Is.

Published ~2026-03-26 · technology, ai

R0026 — Pretendgineer Prior Art

Tags: ai, methodology · Runs: 2026-03-25 · 1 query

Origin search for the portmanteau "pretendgineer." Earliest documented use: January 2009 (Urban Dictionary). Multiple independent coinages.

R0024 — Sycophancy, Addiction, and Vendor Incentives

Tags: ai, methodology · Runs: 2026-03-25 · 4 queries

Vendor disincentive to reduce sycophancy confirmed. AI chatbot ruled a "product" under liability framework. "Dark addiction pattern" at CHI 2025. 42-state AG coalition demanded commitments.

A0019 — Prompt Engineering Is Not. Engineering, That Is.

Published ~2026-03-26 · technology, ai

R0023 — Counterproductive Advice and Prompt Lifecycle

Tags: ai, methodology · Runs: 2026-03-25 · 4 queries

Expert personas degrade accuracy. Three-tier authorship telephone. GPT-4 dropped 84%→51% in 3 months. No lifecycle framework exists.

A0019 — Prompt Engineering Is Not. Engineering, That Is.

Published ~2026-03-26 · technology, ai

R0021 — Engineering Definitions and Standards

Tags: ai, methodology · Runs: 2026-03-25 · 8 queries

Formal engineering definitions, PE licensing, 84% subjective vendor guidance, RFC 2119 absence, 430:1 ambiguity gap.

A0019 — Prompt Engineering Is Not. Engineering, That Is.

Published ~2026-03-26 · technology, ai

R0020 — Prompt Engineering Gaps

Tags: ai, methodology · Runs: 2026-03-25 · 4 queries

Testing frameworks, sycophancy in vendor docs, imperative constraints, theory-practice gap.

A0019 — Prompt Engineering Is Not. Engineering, That Is.

Published ~2026-03-26 · technology, ai

R9990 — STAR Interview Neurodivergent Impact

Tags: current-events · Runs: 2026-03-20 · 1 claim

Does the STAR behavioral interview format disadvantage neurodivergent individuals (ADHD, dyslexia)?

R0007 — AI Made Everyone Faster

Tags: ai, technology · Runs: 2026-03-19 · 15 claims

Performance distributions, toxic worker effects, AI leveling studies, and the gap between individual and organizational productivity gains.

R0005 — AI Company Profitability Before 2030

Tags: ai, technology · Runs: 2026-03-17 · 1 query

Company-by-company analysis across pure-play labs, diversified tech, and infrastructure providers.

R0002 — Research Standards for AI-Assisted Writing

Tags: methodology · Runs: 2026-03-13 · 12 claims

Investigation of the nine frameworks underlying the unified research methodology: ICD 203, GRADE, IPCC, PRISMA, Cochrane, CONSORT, ROBIS, NAS.