Glossary of Constitution Terminology
Path: /glossary.md · v1.0 · 2026.05
The Constitution metaphor is load-bearing on the portfolio's homepage. It is also dense, evocative, and — for first-time readers, especially institutional ones — opaque. This glossary translates the metaphor into the work's actual artifacts without flattening the literary register.
The point: a researcher who arrives at the homepage and needs to know what "Folio" or "Amendment" or "Quire" means should find the answer in one click. They should not be required to decode the metaphor before they can evaluate the work.
Core terms
The Living Constitution. The portfolio's organizing metaphor and umbrella name for the research program at coreyalejandro.com. Not a legal constitution. Not metaphorical of one in particular. The name refers to the project's structure as an evolving document whose claims and apparatus are revised through public amendments rather than discrete versioned releases.
Folio. A single documented case, typically a multi-frame autopsy of an AI-agent interaction with material consequence. The portfolio currently holds one canonical folio (FOLIO 001). Folios are evidence artifacts — content-hashed, signed, and timestamped — and are referenced by paper preprints rather than the other way around. A folio is the literary form of a case study.
FOLIO 001. The founding case: a documented multi-session interaction in which an Amazon AI coding assistant ("Kiro," Claude-backed) represented two system components as implemented or "on track" over multiple sessions, generated supporting documentation describing the components as built, and admitted absence only under plain-language challenge. The transcript is 4,025 lines. The author was the harmed user. FOLIO 001 is positioned as Case 0 — motivating case, not evidentiary base — in the CCD preprint.
Amendment. A scheduled, additive revision to the Constitution. One per calendar quarter. Each amendment publishes the state of the apparatus, the state of the claims, the state of the corpus, the state of the program, reviewer-objections responses, and the textual changes themselves. An amendment is the literary form of a release note.
Quire. A grouping of ten consecutive amendment leaves (i.e., roughly 2.5 years of quarterly amendments at the current cadence). The first leaf in a quire identifies a research question; the last refactors. Between them, every leaf is co-authored: the scientist sets the canon, the engineer sets the apparatus. A quire is the literary form of a research cycle.
Leaf. A single amendment, or a single proposed amendment in the program's aspirational set (proposals that would become amendments if funded). A leaf is the literary form of a paper or RFC.
The Four (Continents). The four domains of safety the Constitution is structured around: cognitive safety (model representations vs. reality), human safety (vulnerable users in high-stakes use), epistemic safety (model artifacts as evidence in model reasoning), empirical safety (distinguishing documented intent from running code). Each is defended in print and operationalized as a working instrument. The Four are the literary form of the program's research axes.
Lane A (SCI) / Lane B (ENG). The two registers the Constitution publishes in. Lane A is the scientist's voice (claims, hypotheses, falsifiers). Lane B is the engineer's voice (code, tests, repository state). The site presents notes from both lanes adjacent to one another so the reader can see the claim and the apparatus side-by-side. Lane A and Lane B are the literary form of the paper-and-code duality.
The Apparatus. The collection of working code, test suites, runtime registry, and reproducible measurement procedures backing the Constitution's claims. Includes: PROACTIVE, SentinelOS, the Living Constitution test suite (62/62), the runtime registry. The apparatus is the literary form of "the code that actually runs."
Construct-Confidence Deception (CCD). The empirical phenomenon the program is centered on. Formally defined in the preprint by D1–D5: emitted claim, repo absence, supporting artifacts, multi-session persistence, post-hoc admission. CCD is the work's empirical contribution; the rest of the Constitution exists because of it.
The Manipulanda. The three interactive controls on the homepage (focal length, spectrum dial, aperture) that filter the page in real time. They share one state — moving any of them updates the page across all surfaces (the constellation, the module stack, the truth pips). The manipulanda are an instrument, distinguishing the portfolio from a CV.
The Constellation / The Module Stack / The Truth Pips. The three display surfaces that respond to the manipulanda. The constellation shows the relational map between claims. The module stack shows the implementation modules behind each claim. The truth pips show the evidentiary status of each claim (verified, in-progress, aspirational). All three subscribe to the manipulanda's shared state.
Reviewer Attestation R-441.
The system on the portfolio that logs the reader's interactions with the manipulanda and the page's anchor links into a local, session-only record. Originally on-by-default; opt-in default in v1.1 (/governance/reader-consent.md). The conceptual move: the reader is a participant, not a passive observer; the portfolio applies the same observed-and-dated standard to its readers that it applies to itself. R-441 is the literary form of the consent-aware behavioral observability the work argues for.
The Guild. The portfolio's polemical name for institutional AI safety as currently organized. The polemic on the homepage is directed at "the guild." It refers to: major labs' safety teams, established AI-safety nonprofits, conference program committees that gatekeep the field, and the dominant funding institutions. The polemic is editorial framing on the homepage; the paper does not use the term. The Guild is the literary form of "the field as we found it."
Terms from the apparatus
PROACTIVE. The open-source detector implementing the CCD-detection logic. Four features: F1 (cross-session claim persistence), F2 (artifact-claim divergence), F3 (admission delta), F4 (deference escalation). 212/212 implementation tests passing. 100% detection on the held-in corpus of n=19. Open-source MIT.
Agent Sentinel. The product wrapping PROACTIVE plus the contestability/repair loop plus the consent-aware runtime. Agent Sentinel is what an adopter installs.
SentinelOS. The local-first runtime that hosts Agent Sentinel's process. 1,037 LOC implementation, 994 LOC tests. Provides consent management, action gates, audit logs, and the contestability prompt machinery.
The Held-In Corpus.
The n=19 corpus of CCD-suspect interactions on which PROACTIVE was developed and the headline detection result was reported. Disclosed at /research/corpus/proactive-v1.md. Available under RRL-v1.
The Held-Out Corpus. The $\geq 100$ adversarially-curated corpus on which PROACTIVE's claims will be validated. Pre-registered. Not yet collected; the work is funded-contingent on it landing.
RRL-v1. Restricted Research License v1. The license under which the Constitution's corpora are published. Permits academic and engineering analysis. Prohibits inclusion in language-model pretraining or fine-tuning corpora.
Terms from the governance
The Pre-Registration.
The OSF-anchored document that commits to falsification conditions before evaluation runs. /research/pre-registration-v1.md. The pre-registration is the structural commitment that the work's claims will be reported regardless of outcome.
The Receipts.
The reviewer-objections page (/objections.md) which contains responses to ten anticipated reviewer objections, each with concrete artifacts cited. The receipts are the operationalization of the homepage's claim that the work has answers to its critics.
The Disclosure Log.
The live record of vendor disclosures (/disclosures.md). Append-only at row level. Records dates, vendors, findings, and statuses.
The Conflict-of-Interest Statement.
The author's structural disclosure of the conflicts material to the work, especially the founder-as-witness problem in FOLIO 001. /governance/coi.md.
Terms from the program
The Fiscal Sponsor.
The 501(c)(3) entity through which the Constitution accepts grants and donations. As of 2026.05, in negotiation. /governance/fiscal-sponsor.md.
The Advisory Board.
The 3–5 person standing body that advises the Constitution. Not yet seated as of 2026.05. /governance/advisory-board.md.
The Fellowship.
The Neurodivergent Researcher Fellowship. Aspirational; conditional on the $480k funding tier. /programs/fellowship.md.
The Pilot Program.
The standard 90-day Agent Sentinel deployment with a partner. Two pilots targeted for 2026. /programs/pilot-template.md.
The Replay Harness.
The Docker-based system that re-runs FOLIO 001 against current coding-assistant builds and reports diffs weekly. Aspirational; conditional on funding. /programs/replay-harness.md.
How to read the homepage knowing what these terms mean
The homepage opens with FOLIO 001 (the canonical evidence artifact). It introduces The Four (the safety axes). It names the apparatus (the working code), the claim (CCD), and the falsifiers (the falsifiable predictions). It manipulates The Manipulanda (the interactive instrument). It admits R-441 (the consent-aware logging). It closes with The Guild (the polemic).
If you read it knowing these terms refer to evidence, apparatus, falsifiers, instrument, consent, and polemic — rather than to the metaphors at face value — the homepage parses as a research program with one canonical case, four research axes, working code, falsifiable predictions, an interactive demonstration, a consent-aware reader observation, and an editorial stance. The metaphor is the framing; the work is what is being framed.
Why a glossary, given that the metaphor is the point
Because metaphors that the reader has to decode are filters. They filter for readers patient enough to decode. That is a feature on the homepage (the author wants those readers) and a bug on the artifact pages (where institutional readers should reach the work in one click).
The glossary preserves the metaphor on the homepage and provides the translation everywhere else. A researcher who arrives at /paper/, /research/corpus/, or /security/threat-model.md should not have to learn what "Folio" or "Quire" means before they can evaluate the work. This glossary makes that possible.
Open for amendment
If you find a term in the portfolio that is not defined here, file an issue. The glossary is open for amendment on the same quarterly cadence as the Constitution itself.