Skip to content
DAY ZERO

About

Why DAY ZERO exists

Early-stage sourcing mostly ranks people by prestige and popularity, because those are the fields that are easy to query. DAY ZERO is an attempt to rank nothing and read everything — to surface builders from what they have verifiably constructed, then do the technical work an introduction deserves before anyone's time is spent.

It was built as an independent work sample for an AI Analyst application at Array Ventures — a role about sourcing technical founders at day zero and interrogating their architecture, not tracking announced rounds. The system is the argument: sourcing that starts from public artifacts, verification that reruns the claim, diligence that reads the code, and a methodology that publishes its own failures.

What it does not claim: that its tallies are investment performance; that a convergence pass predicts a good company; that any person analysed here is or should be raising; that absence of public evidence is evidence of absence; or that a system this young has validated its own usefulness. The methodology page is deliberately the load-bearing one.

Principles

Privacy

  • Professional artifacts only

    The project analyses repositories, papers, project domains and organisation records — what people build in public. It does not profile people.

  • No private identifiers

    No emails, no phone numbers and no precise personal locations reach the public interface or the public repository. The export pipeline refuses to emit those fields, and tests enforce it at both the export and the full-Git-history level.

  • No inferred employment changes

    Nobody is inferred to be leaving a job from silence, inactivity, a deleted post or a bio edit. Only explicit first-person public statements count.

  • No enrichment brokers

    No data brokers, people-search sites, facial recognition, WHOIS registrant lookups, or guessed username matching — excluded by written policy, not by omission.

  • A small public surface

    The full research universe (267 identities, 102 repositories, 1,586 evidence records) never reaches this site. Only the material needed for the analyses shown here is exported, and a build test verifies the rest stayed home.

  • Raw caches are not redistributed

    Reading a public profile once for research and republishing hundreds of them as a downloadable file are different acts. The raw GitHub profile caches were removed from the entire public Git history before first publication; the collection code, schemas and aggregate outputs remain. The rewrite is documented rather than quietly performed.

  • Descriptive, not judgmental

    Attention metrics appear only as description. No person on this site is scored, ranked or labelled a good or bad founder.

Disclosure

AI disclosure

This project was built with substantial AI assistance. Claude Code assisted with research organisation, software implementation, structured extraction from public sources, classification assistance, testing, debugging, experiment implementation, drafting and synthesis — across every phase, including this website.

The boundaries that keep that honest: public sources are the factual evidence, and AI output is never treated as primary evidence. Observed facts are structurally distinct from inference in the data model, and every material claim carries an evidence state. Final sourcing selections and the recommendation on the Sandlock page are analyst judgments. Not every builder in the universe received equal manual diligence. And the system contains no global founder score — the schema has no column for one, and a test fails if one appears.

Disclaimer

Independence

DAY ZERO is an independent research project and is not affiliated with, sponsored by, or endorsed by Array Ventures, Shruti Gandhi, Multikernel Technologies, Sandlock, Headroom Labs, or any person or company referenced in the analysis. It is based entirely on public information and is not investment advice.

Nobody referenced in this research has been contacted about it. No introduction has been made. The financing statements on the diligence pages describe what is public, never what is true of any company’s actual capitalisation.

System

Architecture & reproduction

Architecture

Python research pipeline (canonical)└─ frozen rules · evidence store · experimentsscripts/build_frontend_data.py└─ validated public export · forbidden-key guardNext.js static export (this site)└─ no runtime DB · no API · no auth · no AI calls

Every research value on this site originates from the canonical export — components hard-code no numbers, and drift tests compare the built HTML against the export.

Reproduction

Three different things get called “reproducible”, so they are separated here. Code reproducibility: the frozen rule manifests, the pre-registered protocol, the collection code and the test suite are all public, and the export regenerates deterministically. Public-data reproducibility: the sourcing validations rerun against the frozen rules, and 34 of the reproduction’s 35 inputs ship in the repository — one profile-derived sample is withheld for third-party privacy, with a 34-sample sensitivity check reported separately. Original-run traceability: every recorded measurement, hash and commit is published, so the original run can be audited even where it cannot be re-executed byte-for-byte. The commit hashes on the methodology page are the audit trail.

export: outputs/** + experiments/headroom/datasets/manifest.json + research/sandlock/source_audit.md