Declination Join the group
A proposal · Version 0.1 · October 2026

Declination

de·cli·na·tion /ˌdek-lə-ˈnā-shən/ · the angle between true north and magnetic north

A fully online research group measuring the gap between governance on paper and governance in practice: how boards in the Americas really oversee audit, risk and AI.

Prepared by
Jimmy, Principal Investigator
For
Co-authors, data partners, funders
Working languages
English · Español · Français
Read the proposal See open seats
Each section below is a bearing: read clockwise from N.
N 000°The problem NE 045°What we propose E 090°How a fully online group works SE 135°Research agenda S 180°Method and AI SW 225°90-day plan and deliverables W 270°Team and open seats NW 315°Budget and the ask
N 000°
The problem

Boards are asked to oversee more, with less evidence about what good oversight looks like.

AI outpaces the board

In PwC's 2026 Annual Corporate Directors Survey, 71% of directors named AI as the top skill their boards need to strengthen.

Rules diverge by country

US, Canadian and Mexican issuers answer to different audit committee and independence rules. They are rarely compared clause by clause.

Evidence is thin outside the US

Most empirical governance research uses US or European samples. Mexico and Central America are rarely in the same dataset.

NE 045°
What we propose

A small, online, AI-assisted research group that publishes evidence anyone can check.

AI agents collect and code public filings, laws and governance codes. Named authors validate every coded field, own every finding, and release the code, codebook, prompts and validation sample with each paper.

We are
  • A research group producing working papers, journal articles and open datasets
  • Fully online, with members across the Americas
  • Trilingual: papers in English, summaries in Spanish and French
  • Independent: funders never see drafts or veto findings
We are not
  • A think tank or advocacy body
  • A consultancy selling opinions
  • An AI content mill: AI is never an author
  • Tied to one university, at least in year 1
E 090°
How a fully online group works

No office, no campus. Every step runs online and leaves a record.

Virtual lab

One shared repository holds code, codebooks, prompts and the source ledger. Every change is versioned.

Async by default

Written threads and comments replace most meetings, so members in Mexico City, Toronto and San José work on their own hours.

Online seminars

A monthly works-in-progress session, recorded. One paper, one discussant, open to invited readers.

Digital sign-offs

Each of the four review gates is an e-signed checklist stored with the project. No gate, no release.

Open repository

Pre-registrations on OSF. Datasets and replication packages get a DOI. Working papers go to SSRN.

Distributed members

Co-authors join from any jurisdiction we study. Recruiting, onboarding and authorship agreements all happen online.

A month in the group
Cadence Format Output
Weekly45-minute video call + written updatePaper pipeline status, gate decisions
Every two weeksAsync review of agent runsError log, accuracy sample, prompt changes
MonthlyOnline works-in-progress seminarOne paper presented, discussant notes
Twice a yearAdvisory group callAgenda review, outlets, collaborations
SE 135°
Research agenda

Two research streams, each built on one reusable dataset.

A

AI, Cyber & Technology Oversight

How boards disclose oversight of AI and cyber risk, and whether disclosure quality relates to incidents, restatements or audit findings.

Dataset: Board Tech-Oversight Disclosures (US, Canada, Mexico)
First papers: WP1 board AI oversight disclosures · WP2 SEC cyber incident disclosures and board attributes
B

Comparative Audit Committee Governance

How audit committee rules and practice differ across the US, Canada, Mexico and Central America, and what changed after Mexico's December 2023 securities-law reform.

Dataset: Americas Audit Committee Rules & Practice
First papers: WP3 audit committee composition in three jurisdictions · WP4 effects of the Mexican reform
S 180°
Method and AI

AI does the legwork. Four human gates own every finding.

  1. 01 · Gate 1 · HumanCharterQuestion, authorship, pre-registration
  2. 02 · AgentsCollectFilings, laws, codes, literature
  3. 03 · AgentsCodeEvery field with its citation
  4. 04 · Gate 2 · HumanValidate10% sample, 100% on any error
  5. 05 · Agents + authorsAnalyzePre-specified tests, methods draft
  6. 06 · Gate 3 · HumanReviewInternal, external reader, replication
  7. 07 · Gate 4 · HumanReleaseSSRN, then journal; data DOI
  8. 08 · AgentsMonitorRule changes trigger errata
AI is never an author. Authors write the introduction, discussion and conclusions.
Every field is traceable. Unsourced output is rejected automatically.
Accuracy is reported. Target: at least 97% against a hand-coded benchmark, published in each paper.
SW 225°
90-day plan and deliverables

First working paper online at day 90, if it clears every gate.

Days 1–30
Foundation
  • Members recruited online
  • Four group policies e-signed
  • Academic co-author seat filled
  • Codebook v1, 30-issuer pilot
Gate A · policies + authorship signed
Days 31–60
Build
  • Site live in EN, ES and FR
  • Benchmark set hand-coded
  • WP1 data coded and validated
  • First online seminar held
Gate B · ≥97% pipeline accuracy
Days 61–90
Release
  • WP1 internal and external review
  • WP1 posted on SSRN
  • Dataset v0 + codebook with DOI
  • Practitioner note in EN and ES
Gate 4 · WP1 release sign-off
W 270°
Team and open seats

One principal investigator, five open seats.

PRINCIPAL INVESTIGATORJimmyAudit and governance practitioner (CIA, CISA, CFE, CPA). Leads the agenda, signs Gates 1 and 4.
OPEN · CO-AUTHORAcademic, law or accountingLeads methods for Stream B and journal strategy.
OPEN · CO-AUTHORSecurities lawyer or ex-regulatorOwns rule coding for Mexico and Central America.
OPEN · CO-AUTHORDirector or CAEGrounds questions in boardroom practice, leads practitioner notes.
OPEN · RESEARCH ASSISTANTGraduate student or analystRuns the pipeline and Gate 2 validation, remotely. Paid per project.
OPEN · ADVISORY GROUP2–3 senior advisersTwo online calls a year to challenge the agenda.
NW 315°
Budget and the ask

A lean online budget, and four ways to join.

Year-1 budget · Standard scenario (USD)
LineWhat it coversAmount
Online toolsClaude Pro + API for the coding pipeline, domain; hosting, forms and repositories are free860
Research assistantRemote graduate student, ~400 hours at about MXN 200/hour4,300
Translation review~20,000 words of Spanish review; French reviewed in-house1,080
PublishingSSRN (free) plus two virtual conference registrations540
IVA (16%)On invoiced fees for the assistant and translator860
Contingency (10%)Price changes, extra coding hours765
TotalLean scenario: about USD 3,300 · converted at MXN 18.58 per USD8,400
We are asking for
  1. Co-authors who bring a jurisdiction, data access or legal expertise we lack
  2. Data partners willing to share or co-build a dataset
  3. External readers for working papers before release
  4. Seed funding for the budget lines above, with no say over findings
Express interest

Tell us how you would like to take part: co-author, data partner, external reader or funder. It takes about two minutes, in English, Spanish or French.

Open the form

Or write to hello@declinationresearch.org (hello@declinationresearch.org)