Notes from an ongoing experiment in governing AI-assisted work
By Douglas P. Galullo
Founder, Dog House Ventures LLC
This repository is Dog House Ventures LLC’s public claim ledger: the permanent public home for (1) published research, (2) high-level engineering, product, validation, and governance milestones that materially support public claims, (3) program status with explicit non-claims, and (4) a claim→evidence map with explicit evidence classes.
Philosophy: Don’t ask people to trust us. Give them enough structure and evidence that they can evaluate our claims — including where verification stops at company attestation and where artifacts are public.
See the Public Claim Ledger Charter.
- Product source code or private development monorepos
- The internal constitution stack or business-record filesystem
- A marketing site
- A promise that proprietary tests are independently re-runnable unless a claim is classed
PUBLIC-ARTIFACTorINDEPENDENTLY-ATTESTED
| If you are… | Start with… |
|---|---|
| Reading the argument | Full article · PDF |
| Evaluating engineering / product claims | STATUS.md · history notes · Evidence Index |
| Checking how evidence is classed | Evidence classes · Corrections |
| Understanding programs / products | Programs |
| Citing the thesis | CITATION.cff · PUBLICATION.md |
Maintainer preflight files (AI_START_HERE.md, AGENTS.md, CLAUDE.md) are for authorized maintainers; they are not part of the public argument.
This article argues that artificial intelligence's central challenge is not capability, but responsibility. The argument emerged from an ongoing applied study of AI-assisted work in which multiple systems were used for real operational tasks and the surrounding process was repeatedly observed, tested, and refined.
Recurring failures - lost context, evidence detached from conclusions, ambiguous approvals, and authority without clear limits - revealed a common gap. Conventional activity logs can show that something happened, but not that it was properly authorized, supported, reviewed, or accepted.
The resulting hypothesis is that consequential AI-assisted work requires responsibility infrastructure: bounded authority, preserved evidence, meaningful review, adaptive procedures, institutional memory, and identifiable human accountability.
Authority enters through a mandate. Evidence and decisions persist through a governed record. Responsibility exits through explicit acceptance.
The work remains an ongoing experiment, but the hypothesis has held through each iteration so far: AI will not become governable simply by becoming more intelligent. Responsibility must be built into the process itself.
- Full article
- Abstract
- Publication PDF
- Publication record
- Author background
- Citation metadata
- Version history
- Thesis publication milestone
As artificial intelligence becomes more deeply involved in consequential work, increased capability alone is insufficient.
The surrounding process must be able to establish:
- who authorized the work,
- what authority an AI system received,
- what evidence supported its output,
- which limitations were known,
- where the work was reviewed or challenged,
- how procedures changed,
- who approved the resulting decision,
- and who accepted responsibility for the consequence.
The distinction at the center of this work is the difference between an activity record and a responsibility record.
An activity record may show that something happened.
A responsibility record must also show that the work was authorized, evidenced, reviewed, decided, and accepted.
This repository contains an original working thesis developed through an ongoing applied process study.
It does not claim:
- a completed scientific experiment,
- universal validation across industries,
- a final model of AI governance,
- or a substitute for law, regulation, professional standards, or institutional judgment.
The work is published so the argument, its development, and its revisions can be examined over time.
For multi-program status — Mission Control, external validation, Investment Vault, Proof Ready, Doghouse, and governance — see STATUS.md.
Highlights (all non-public-artifact metrics are company-attested unless noted):
| Program | Public posture |
|---|---|
| Research | Thesis v1.0 published 2026-07-18 |
| Mission Control | M1 closed 2026-07-20 (112/112 at close); M1-A+ trust maturity closed 2026-07-21 (201/201 + 13/13 adversarial); not a production control plane |
| External validation | Gold-standard multi-cycle campaign: 49% → 65% → 94% → 100% on 500-case cycles; final Strong, not Elite (ATTESTED-INTERNAL) |
| Investment Vault | In development; MVP audit remediation 2026-07-15; 2026-07-20 separate evaluations (App Store C / tests later 94/94 / product RC A / MVP readiness B — not one grade) |
| Proof Ready | In development; HOLD FOR MARKET VALIDATION — production build not authorized |
| Governance | Constitution activation 2026-07-20 (summary); Publication Safety Gate enrolled |
Mission Control conceptual layers (named only): ARCHITECTURE_OVERVIEW.md.
Quantitative claims and program status facts in STATUS.md and the engineering/product highlights below resolve to the Evidence Index. Thesis narrative is evidenced by the article/PDF themselves.
Ordered by engineering / event date (public documentation date may be later; see each note).
| Event date | Note |
|---|---|
| 2026-06-20 | External gold-standard validation campaign (public doc 2026-07-20) |
| 2026-07-15 | Investment Vault MVP audit remediation |
| 2026-07-17 | Proof Ready readiness gate HOLD |
| 2026-07-18 | Thesis v1.0 publication |
| 2026-07-20 | Mission Control M1 |
| 2026-07-20 | Company governance activation |
| 2026-07-20 | Investment Vault verification & readiness (separate evaluations; not one merged grade) |
| 2026-07-21 | Mission Control M1-A+ trust maturity |
Full index: docs/history/README.md.
articles/— canonical Markdown articlepdf/— publication PDFassets/— cover artworkresearch/— sourcing agenda and future-paper scopereleases/— fixed release notesdocs/PUBLIC_CLAIM_LEDGER_CHARTER.md— repository identity and admission standarddocs/EVIDENCE_CLASSES.md— public / attested / independent evidence classesdocs/EVIDENCE_INDEX.md— claim-to-source mappingdocs/PROGRAMS.md— program registerdocs/CORRECTIONS.md— corrections and supersessionsdocs/history/— public milestone notesdocs/PUBLICATION_SAFETY.md— Publication Safety Gate enrollmenttesting-dashboard/— aggregate test and validation counts (see field semantics in its README)STATUS.md— multi-program public statusARCHITECTURE_OVERVIEW.md— Mission Control conceptual layers onlySECURITY.md— security contact and repository scopeCITATION.cff— citation metadataPUBLICATION.md— authorship and publication recordCHANGELOG.md— version / ledger history
| Surface | Role |
|---|---|
| This repository | Authoritative public claim ledger (research + engineering/product/governance record + evidence map) |
doghouse-public |
Company front door / high-level product narrative |
doghouse-governed-ai-evaluations |
Evaluation protocols/results when published; plan-only material is not completed evidence |
| Internal business filesystem | Constitutions, operational standards, legal/financial records (not this repo) |
| Internal engineering OS | Implementation, audits, completion packages (not this repo) |
Internal Dog House Ventures governance (Constitution, AI Control Constitution, File Structure Constitution, operational standards) lives in the company business filesystem — separate from this public tree. As of 2026-07-20 that internal stack is treated as activated for operations; this repository remains the public claim ledger, not the operating control system. See governance activation note.
Products (Doghouse as internal platform, Investment Vault, Proof Ready, and others) inherit internal governance rather than replacing it.
This publication discloses thesis, process, results, and high-level milestones. It does not disclose proprietary implementation (source code, prompts, schemas, lock models, or security design).
- Thesis version: 1.0
- Initial public research record: July 18, 2026
- Public claim ledger framing (Phase 2): July 20, 2026
- Mission Control M1-A+ public admission: July 21, 2026
- Author: Douglas P. Galullo
- Affiliation: Founder, Dog House Ventures LLC
Galullo, Douglas P. "AI Does Not Have an Intelligence Problem. It Has a Responsibility Problem." Version 1.0, July 18, 2026.
Substantive criticism, corrections, counterarguments, and relevant sources are welcome through GitHub Issues or Discussions.
Feedback should address the argument and public claims rather than speculate about proprietary implementation details.
Copyright © 2026 Douglas P. Galullo. All rights reserved.
No open-source or Creative Commons license is granted for the article, abstract, cover image, or associated publication materials. See LICENSE.md and NOTICE.md.