Skip to content
DAY ZERO

Inputs

What actually feeds DAY ZERO

The honest description is GitHub-led discovery with multi-modal evidence — not “multi-channel sourcing”. One channel discovers; the others corroborate. Overstating that would be exactly the kind of claim this project exists to test.

GitHub

● LIVE — 100% of discovery

The discovery channel. Every one of the 102 repositories in the universe entered through it.

1,580 evidence records · 102 artifacts

Research papers

● LIVE — enrichment only

Evidence enrichment: papers corroborate depth and collaboration on already-surfaced leads (e.g. the Sandlock arXiv paper). Not a discovery source in the current run.

6 evidence records · 6 artifacts

Domains / web

● LIVE — formation evidence

Formation evidence: project domains, company sites, organisation records attached to existing leads.

73 formation signals

X

○ DISABLED

Ingestion is off by default and no credentials were present. Nothing in the current dataset came from X.

Hackathon automation

○ NOT USED

Manual only, by robots policy. No automated adapter exists.

Manual event layer

○ DESIGNED, NOT POPULATED

Designed (schema and adapter interface exist) but not populated in the current run.

GitHub: 1,259 API requests, 102 repositories, 1,580 evidence records, at an API cost of $0. The scale numbers on the methodology page all trace back to these channels.

Audit · Phase 4

The identity audit — an assumption, revised

Phase 1 assumed identity resolution was the bottleneck of the whole system. Phase 4 measured it. The assumption was wrong in an instructive way.

267
Live identities
Phase 2 collected universe, unchanged
166 · 62.17%
Mergeable
under conservative v2 rules
1
Verified X-linkable
0.37% of the universe
28
Bio @handles
deliberately NOT counted as X identities

Three measurements of one population

Live identities collected267 · 100.00%

the Phase 2 universe, unchanged

Mergeable under conservative rules166 · 62.17%

resolvable without fuzzy matching — the assumption that failed

Verifiably X-linkable1 · 0.37%

the measurement that did not improve

These are identity-resolution states, not candidate quality. Nobody was “lost” between the bars — the same people are counted three times, under three different evidentiary standards.

Resolution states

VERIFIED_CROSS_LINK109

two-way verified link between platforms

STRONG_ARTIFACT_MATCH57

same artifact referenced from both sides

POSSIBLE_MATCH60

suggestive but unverified — never merged

UNRESOLVED41

no cross-platform evidence at all

The revised conclusion

General identity resolution is not the bottleneck any more: 62.17% of live identities are mergeable under rules that forbid fuzzy matching. What remains broken is X-specific identity linkage: exactly 1 identity in 267 could be verifiably linked to an X account.

28 GitHub bios contain an @handle — and none was counted. On GitHub a bare @handle in a bio is overwhelmingly a GitHub organisation handle, not an X account. Treating them as X accounts would be the fuzzy matching v2 forbids. The handles are not exposed here, or anywhere in this product.

Methods excluded by policy: data brokers, people search sites, private email databases, guessed username matching, arbitrary x username search, facial recognition, location inference.

Audit · Phase 4

Presence is not signal: the domain example

Most projects have a domain. That is precisely why having one proves little — the question is whether the domain adds a formation event the evidence graph did not already contain.

Repositories with a project domain

70.59%

Domain adds a distinct formation event

32.35%

70.59% presence → 32.35% independent signal, measured on 102 repositories

Why the gap

A project domain is present on most repositories (70.59%), but it adds a DISTINCT formation event on only 32.35% of them - because in the other cases the GitHub organisation has already supplied one and the two describe the same decision. Domains earn their own event where an organisation has no domain of its own, or where the project domain differs from the company domain, which is a second and separate decision.

Excluded by policy: whois enrichment, domain registrant lookup, reverse whois, dns history broker data.

The core separation

Attention and construction are different axes

Stars measure attention. Commits by the people who own the work measure construction. Across the universe the ratio between them spans four orders of magnitude — which is why attention is barred from every surfacing decision.

1101001,000101,000100,000CONSTRUCTION — commits by the owning contributor (log)ATTENTION — stars (log)ComposioHQ/awesome-claude-skills: 73,056 stars, 38 commits0xSero/turboquant: 1,735 stars, 2 commitsrtk-ai/rtk: 77,106 stars, 530 commitslasso-security/claude-hooks: 262 stars, 2 commitsNirDiamant/agents-towards-production: 21,323 stars, 167 commitsBrain0-ai/brain0: 367 stars, 3 commitsOpenIntegrationEngine/engine: 216 stars, 1,489 commitsdredozubov/hazmat: 160 stars, 1,268 commitsrustledger/rustledger: 374 stars, 2,733 commitsvivekchand/clawmetry: 399 stars, 3,112 commitsmultikernel/kerf: 30 stars, 156 commitsFootprintAI/Containarium: 273 stars, 989 commits

Hover or focus a repository

Two independent measurements. Neither axis is a judgment about a builder, and no combined score exists anywhere in the system.

High attention · low construction

  • ComposioHQ/awesome-claude-skills73,056★ · 38 commits

    1,922.5 stars per commit

  • 0xSero/turboquant1,735★ · 2 commits

    867.5 stars per commit

  • rtk-ai/rtk77,106★ · 530 commits

    145.5 stars per commit

  • lasso-security/claude-hooks262★ · 2 commits

    131 stars per commit

High construction · low attention

  • OpenIntegrationEngine/engine216★ · 1,489 commits

    0.1 stars per commit

  • dredozubov/hazmat160★ · 1,268 commits

    0.1 stars per commit

  • rustledger/rustledger374★ · 2,733 commits

    0.1 stars per commit

  • vivekchand/clawmetry399★ · 3,112 commits

    0.1 stars per commit

Descriptive divergence only. High construction is not a claim about investment quality, and low attention is not a claim about obscurity. A curated list at ~1,900 stars per commit and a working system at 0.1 stars per commit are not a ranking of their authors — they are proof that the two measurements capture different things, and that a system surfacing builders must read construction directly.

Output unit

What the funnel produces

INTRO_READY_AWU

One builder or team lead that survives evidence review and is genuinely worth spending relationship capital on. The current count is 3 — from 102 repositories, 1,586 evidence records and 273 registered sources.

Three leads is the honest number, not a vanity metric: no introduction has been made, and the count measures review survival, not conversions. The full records are on Current 3.