PROXY SEDUCTION FRAMEWORK

Research Hub

Central workspace for the PSF empirical phase and the wider paper pipeline. Cards below track research design instruments, fieldwork materials, coding rubrics, data, analyses, and papers in the pipeline (active, queued, and held). Each card links to a standalone file (HTML or Markdown) that can be opened independently. Placeholder cards mark tools that will be built as the research progresses.

Target: ~50 semi-structured interviews Population: 30-35 frontline, 15-20 boundary activity Primary domain: software development Phase: Pre-fieldwork (design and recruitment)
PSF Theory Hub →
01 Research Design 3 tools
DESIGN
Interview Protocol 2x2
Semi-structured interview questions organized by a consequence-exposure x engagement-depth matrix. Four cells (Cell 4 structurally empty). Each cell has a practitioner profile, tailored questions, and "listening for" annotations. Cross-cutting probe on friction episodes. Sequencing notes: Cell 2 first to establish pre-engagement vocabulary baseline.
4 cells Cross-cutting probe Sequencing rationale
Active
DESIGN
Dumbbell Coding Rubrics
Four paired-measure specifications for coding interview transcripts into dumbbell chart visualizations. Pairing 1: self-assessed improvement vs. criteria articulation depth. Pairing 2: proxy vocabulary ratio vs. consequence vocabulary ratio. Pairing 3: friction episode count vs. confidence rating. Pairing 4: reported team metric vs. boundary performer's actual monitor. Includes coding workflow, inter-rater protocol, and JSON output format.
4 pairings 1-5 rating scales Vocabulary taxonomy JSON spec
Active
DESIGN
Empirical Phase Checklist
Progressive Web App covering the 11 elements of the PSF empirical design, broken into 47 sub-questions. Tracks readiness across interview protocol, recruitment, ethics, analysis framework, and domain coverage. Originally on the Theory Hub, migrated here as an empirical-phase instrument.
11 elements 47 sub-questions PWA
Active
02 Fieldwork 4 tools
FIELDWORK
Candidate Tracker
Unified pipeline for ~50 frontline and 15-20 boundary activity interviews. Frontline grouped by cell (C1: senior + heavy AI, C2: senior + limited AI, C3: junior + heavy AI). Boundary grouped by category (engineering mgmt, DevEx/DevRel, UX/design, consultancy CTOs, PM, internal AI champions). Tracks status, source channel, recruit type (friend/cold), and alumni flag. Includes a cell-assignment helper for frontline candidates. Data persists locally in browser. Export/import JSON for backup.
~50 frontline + ~18 boundary Cell assignment helper PWA
Active
FIELDWORK
Discourse Traps of Proxy Seduction
Field diagnostic for identifying how practitioner discourse naturalizes evaluative erosion. Eight observed traps (inductively derived) and six hypothesized traps (deductively predicted from PSF). Each trap includes the discourse move, PSF counter-question, and observable indicators. Use for interview prep and real-time field note coding. Also on the Theory Hub.
8 observed + 6 hypothesized Interview prep Field note coding
Active
FIELDWORK
Notes from the Field (Apr 2026)
Running log of observations from informal conversations, coffee meetings, and chance encounters during San Francisco fieldwork, April and May 2026. These are not formal interviews. Each entry uses a three-column framework covering raw observations, PSF flags, and follow-ups. Direct quotes and outgoing correspondence are preserved verbatim for later compilation into the fieldwork report.
Informal conversations 3-column entries Apr–May 2026
Active
FIELDWORK
Interview Data Repository
Centralized store for transcripts, field memos, and post-interview coding sheets. Each interview produces a transcript, a 200-word first-pass memo, and coded scores for all four rubric pairings. Will link to individual transcript pages as interviews are completed.
Transcripts Memos Coded scores
Planned
03 Analysis 3 tools
ANALYSIS
Transcript Analysis Framework
Three-pass coding workflow. Pass 1: holistic memo within 48 hours. Pass 2: rubric scoring (Pairings 1 and 3). Pass 3: vocabulary extraction and classification (Pairing 2). Inter-rater reliability protocol after every 10 interviews. Codebook will evolve iteratively as patterns emerge.
3-pass coding Inter-rater protocol Iterative codebook
Planned
ANALYSIS
Cross-case Pattern Tables
Within-cell and cross-cell comparison tables following Miles, Huberman and Saldana (2020). Will organize patterns by 2x2 cell, testing whether PSF predictions hold within each population (Cell 1 widest gaps, Cell 2 narrowest, Cell 3 premature arrest signal). Outlier cases flagged for theoretical refinement.
Within-cell patterns Cross-cell comparison Outlier analysis
Planned
ANALYSIS
Boundary Activity Translation Map
Analysis of Pairing 4 data (reported team metric vs. boundary performer's actual monitor). Maps the translation gap across matched teams. Tests whether boundary activity performers hold criteria the frontline has dropped, and whether that holding is visible or invisible to the organization.
Pairing 4 specific Matched teams Translation gap
Planned
04 Outputs 3 tools
OUTPUT
Pairing 1 Dumbbell: SAI vs. CAD (Illustrative)
Interactive dumbbell chart with synthetic data for 40 practitioners demonstrating what the visualization would look like if PSF predictions hold. Sortable by gap magnitude, 2x2 cell, work style (O/P/S), or SAI score. Hover reveals score detail and key quote per practitioner. Summary stat cards by cell. Uses illustrative data only, not from real interviews.
40 synthetic practitioners 4 sort modes Work style dimension Standalone HTML
Active (illustrative)
OUTPUT
Dumbbell Visualization Gallery
Interactive dumbbell charts built from coded interview data. One chart per rubric pairing. Each row is a practitioner, two dots show paired scores, bar reveals the gap. Color-coded by 2x2 cell. Sortable by gap magnitude, cell, seniority. Hover detail shows practitioner ID, scores, and key quote. React component reused from the unicorn density visualization.
4 chart pairings Interactive React/JSX
Planned
OUTPUT
Empirical Findings Report
Synthesis document connecting interview evidence to PSF theoretical predictions. Organized around the five falsification conditions from the AMR paper. Each condition mapped to the relevant 2x2 cells, coding pairings, and pattern tables. Includes confirming evidence, disconfirming evidence, and cases requiring mechanism refinement. Feeds directly into the thesis empirical chapter.
5 falsification conditions Thesis chapter input Mechanism refinement
Planned
05 Paper Pipeline 8 papers
PAPER · ACTIVE
The AI Alibi
Co-authored with Neeti Gupta and Florian Urmetzer for MIT Sloan Management Review. Argues corporations deploy AI as rhetorical cover for workforce decisions driven by financial pressure (Block/Dorsey, Klarna, Baker McKenzie, Accenture, Commonwealth Bank, Amazon). Shares the PSF evidence base but makes a distinct argument about accountability displacement at the enterprise-vendor interface. v6 draft. Submission target end of May or early June 2026.
MIT Sloan target v6 draft Co-authored
Active · drafting
PAPER · CONCEPT
Proxy Traps of AI Engagement
Practitioner-oriented paper on how discourse naturalizes the erosion of evaluative capacity. Three-phase epistemological progression: 8 inductively observed traps, 6 deductively predicted (all confirmed), 7 emerged from re-analysis and cross-domain extension. Four contributions: epistemological progression as method, co-occurrence clustering as compositional grammar, falsifiability architecture, cross-domain portability classification.
Concept stub 21 traps Post-mechanism paper
Active · stub complete
PAPER · 1.5
The Context Coverage Gradient
Paper 1.5 in the PSF pathway. Specifies the boundary condition on the proxy seduction mechanism. The proportion of the accountable criterion that AI's inputs can capture varies along a continuum from high-coverage (structured tasks, AI genuinely effective) to low-coverage (clinical judgment, organizational strategy). Trust calibrated during high-coverage carries forward into low-coverage phase without recalibration. Reconciles the binary debate between "AI works" and "AI does not."
Paper 1.5 Boundary condition Concept
Active · stub complete
PAPER · ADJACENT DOMAIN
AI Diagnostics in India
Adjacent-domain research interest. PSF mechanism translated to AI diagnostics in Indian healthcare (radiology, pathology, clinical laboratory). Tests cross-domain portability of mechanism-level traps and surfaces domain-bound traps absent from the software evidence base. Workforce shortage as structural context that reshapes the proxy-criterion relationship. Cross-domain prediction generates four additional discourse traps beyond the software base.
Adjacent domain Cross-domain test Healthcare AI
Active · research interest
PAPER · QUEUED
Performativity in the Regulatory State
Field-level extension of PSF anchored in public-sector AI evaluation infrastructure (NIST AI RMF, UK and US AISI, EU AI Act conformity assessments). Cabantous and Gond's three mechanisms (conventionalising, engineering, commodifying) provide the analytic spine. Pollock and Williams 2026 "After Hype" supplies the production-side evidence, PSF supplies the consumption-side companion. Felicity conditions specify when regulatory proxies successfully produce the world they claim to describe.
Memo ready Cambridge SoG Felicity conditions
Draft · memo ready
PAPER · QUEUED
The Recursive Proxy
AI observability as the engineering-layer analog of the productivity J-curve. Observability instruments (LangSmith, Arize, RAGAS, DeepEval, Galileo) relocate the proxy ladder rather than eliminating it. LLM-as-judge is the recursive case where models score model outputs under criteria the model can read. Anthropic Constitutional AI as the bounded case. Four conditions for legible recursion as the contribution.
Memo ready Engineering register LLM-as-judge
Draft · memo ready
PAPER · HELD
Organizational Subduction
Tectonic theory of active burial. Surface-level processes do not just neglect accountable criteria, they actively force them beneath organizational awareness through their own operational energy. Distinct from passive sedimentation (Cooper et al.) and layering (Streeck and Thelen). Closest existing concepts: surrogation, normalization of deviance, means-ends decoupling, Strathern's visibility-concealment paradox. PSF as the engagement-shaped variant.
Held Tectonic metaphor Theory contribution
Held · idea captured
PAPER · HELD
Developer Archetypes, 20 Years On
Practitioner essay (with Steven Clarke) recovering 2004 to 2005 archetypes work from Microsoft (opportunistic, pragmatic, systematic, paranoid developer dispositions) as prior art for the developer-decomposition framing reappearing in vendor AI discourse. Asks whether dispositional categories still apply cleanly under AI engagement or moderate the erosion mechanism. HBR or California Management Review target.
Held HBR/CMR target Co-authored Prior-art recovery
Held · post-OSP/IJMR