An AWS-native attestation service, built as a public demonstration of principled engineering with AI as collaborator rather than substitute.
The primary objective is not the product alone. It is evidence that a senior engineer can close a real knowledge gap quickly—using AI for mechanics while owning architecture, judgment, and every decision that must be defended cold. The secondary objective is commercial viability as attestation-as-a-service; that upside is measured separately and is not allowed to expand the demonstration’s scope.
Credibility line (non-negotiable): AI here contributes mechanics, pushback, and rigor — drafting, modeling, adversarial review, even recommendations. What it never contributes is ownership: every load-bearing decision is read cold by the author, ratified or overridden, and defensible by the author without the AI in the room. Anything less proves the deskilling failure mode this project exists to refuse.
Tony Mason
Initiated 10 June 2026
This README reflects the repository as of July 2026 (Band 0 — the formal foundation, H0).
Tessera provides a neutral third party’s cryptographic binding to a package of information at a fixed point in time. Attestation here means: this is what was presented in a bundle, tied to a temporal anchor that cannot be trivially regenerated. It does not by itself vouch for the authenticity or truth of the contents; those guarantees, if any, ride in the package.
The service is a cousin of Willay: shared cryptographic and modeling ethos, forked edges for cloud-native operation (KMS, S3 Object Lock, queues, observability, IaC).
The method is as load-bearing as the artifact:
- Pre-registration — objectives, hypotheses, properties, and success criteria are written, signed, and OpenTimestamps-anchored before outcomes are known. The journal is measured against that record, including honest divergence.
- Formal foundation first (Band 0 / H0) — machine-checked models of the core guarantees gate later code. No Phase 1 until the foundation holds or is amended on the record.
- Public, falsifiable process — multi-round adversarial review, broken companions that must go red, named residuals, and a living property tracker. Theater is a failure mode, not a feature.
| Layer | Status |
|---|---|
| Phase 0 pre-registration | Signed and amended (Amendments 1 and 2 in force; A2 adopted 2026-07-21 after three non-author review rounds and the clock-precedence ruling) |
| Band 0 formal models | Active — TLA+ (P4, P5/P6, P5c issuance, bridge), ProVerif spike; see formal/ |
| Cloud-independent crypto core | Not yet frozen (gated on Band 0) |
| LocalStack / AWS vertical slice | Not started |
| Application code | Minimal scaffolding only |
We are deliberately early in product code and late in protocol precision. That ordering is intentional.
Recommended order for a new reader (human or collaborator instance):
- This file — orientation and credibility line.
docs/phase-0-prereg.md— the signed intent and ranked objectives.docs/phase-0-prereg-amendment-1.md— property list, adversary model, temporal-anchor semantics, agreement gate as falsification task.docs/phase-0-prereg-amendment-2.md— refinement of confirmation, refusal, anchor standing, and the clock-roles ruling (adopted 2026-07-21, three review rounds).docs/tessera-cloud-demo-spec.md— demonstration architecture and capability map.docs/travelog.md— contemporaneous journal of reality against the pre-registration.formal/PROPERTIES.mdandformal/tla/— discharge status and the models themselves.docs/reviews/— adversarial review artifacts and dispositions.
If the tracker and an amendment disagree, the amendment wins.
- GPG-signed commits (author identity — who).
- OpenTimestamps anchors on commits (existence in time — when). These are two mechanisms; both are required for the public record.
- Amend, do not silently rewrite historical pre-registration documents.
- AI as collaborator for mechanics and pushback; author owns all decisions (division of labor is explicit in the pre-registration).
- Falsification-style review before load-bearing text is signed (see Amendment 1 §A1.7 and the review archive).
Setup for local hooks and OTS tooling lives under scripts/ (see
scripts/install-hooks.sh).
- Not a highlight reel. Pre-registration exists so success and failure are both legible.
- Not formal-methods theater. Models state what they prove, what they assume, and what remains unclaimed (especially liveness vs. safety).
- Not an unbounded product fantasy. Commercial upside is secondary and deliberately constrained.
See repository metadata and commit history for authorship. The public record of the work is the git history: signed, timestamped, and open to scrutiny.
For the living narrative of how the work actually felt and where it diverged from plan, start with the travelog.