Method, coverage and reproducibility
Useful hypotheses.
Visible limitations.
Public-source research and independent checks inform the report. A deterministic compiler validates and renders the frozen evidence; it does not prove the science or predict the future.
What was included
42 names from column A, rows 2–43 of the supplied workbook. Other columns were excluded. The names are a supplied cohort, not a complete university directory. Source-spelled English names are mapped to Hebrew roster entries with an explicit identity status.
The output covers professional biographies, representative work, CV links where found, patent evidence, an individual future direction, external candidates, proposed internal teams, funding routes and a five-year institutional plan. Representative publications remain selective. The expanded inventor search includes all affiliations and reports its per-profile coverage; it is not a certified exhaustive patent audit.
Evidence versus proposals
Official profiles establish what the institution reports. A publication record establishes attribution and title; an abstract does not justify full-text efficacy claims. Patent records establish named records and inventors only within the captured source. External partners are suggestions, not endorsements, commitments or proof of a previously unknown connection.
All future collaboration directions and institutional benefits are hypotheses with low forecast confidence. No claim is made that the proposed research is novel, that a team will succeed or that reputation will improve. Direct feedback, feasibility work and a prior-collaboration/novelty check are required.
Independent review
Three research agents investigated separate roster groups; cross-review records identify the actual different agent, sources inspected and verdict. The coordinator gathered funding evidence and synthesized proposals. Review coverage is reported per claim; neither agreement among agents nor a machine receipt authenticates a source.
426 reviewed, 13 partial, 1 pending, 0 insufficient packet claims. The gates compute conservative confidence ceilings. Source credibility and claim confidence are separate ordinal categories, not probabilities.
Known coverage gaps
- Researcher-confirmed interests, unpublished work, available equipment, current grants and team capacity were not available.
- No exhaustive coauthor-network, novelty or patent-family/legal-status audit was performed. A proposed connection may already exist.
- Public-source coverage favors English/Hebrew institution pages and accessible records. Missing records cannot be treated as absent achievements.
- University grant-directory login, subscription databases and private CVs were not accessed. No accounts, outreach, grant applications or paid data purchases were made.
- External candidate geography reflects the bounded source search, not a systematic global ranking.
- Funding pages must be checked again at submission. Future EU framework proposals and historical policy announcements do not establish the current legal position or 2031 eligibility.
CBG contract
CBG Enforcer was created by Meir Gabay, copyright © 2026 Meir Gabay. Here the LLM drafts the meaning and evidence; accepted code owns validation, candidate filtering, canonical order and rendering.
- Research guarantee
- Validity-only: retrieval, source interpretation and proposal drafting can vary.
- Render guarantee
- Exact-artifact for the frozen JSON, engine sources and Python environment. Compare SHA-256 of every generated artifact.
- Selection
- Only explicit proposals; verified roster identities and complementary roles are hard gates. Sort by declared pilot/develop/explore stage, then stable ID. Initiative order is unchanged. Profile matches use a separate, visible 1–10 proposal-fit rubric; these are not scores of people.
- Boundary
- No network, runtime model calls, clock, randomness or fallback generation in the compiler.
- Bundle identity
5a55584dfaf643e55215906ab54c59d621a6a5723586d87d3fad3ca652ca761c
Build provenance · Frozen report data · Five-part CBG contract · Validation receipt
Local handling
The report is a local static artifact outside the application’s public asset directory. It includes real names and speculative professional matches. Review it before distribution. Static files do not enforce access control when copied or hosted; the noindex tag is not a security mechanism.
How to read the missing-data labels
“Not verified” means this search did not establish the fact. “Reported only” distinguishes a profile’s patent statement from a verified individual patent record. “Identity provisional/unresolved” means the name-to-person mapping or evidence needed for research attribution remains incomplete. Such profiles are excluded from generated team selection.
How connection fit is rated
Each connection is a proposal. Topic overlap contributes 0–4 points, complementary expertise 0–3, and a concrete first experiment 0–3. The displayed score is their sum, with a minimum of 1. It is an ordinal discussion priority, not a probability, personal ranking or claim of an existing relationship. Forecast confidence remains low.
Matches sort by descending score, then original proposal first on ties, then stable identifier. Original candidates and initiative members are preserved; each match records its rationale, first test and ranking explanation. Provisional identities remain conditional on confirmation.
Expanded patent coverage
The original snapshot used representative records and a five-record schema limit. The additive inventor search has no Bar-Ilan applicant filter. Publication records retain identifiers and family keys; a family can include multiple jurisdictions or applications. Five records are initially shown, with the remaining records expandable. An inaccessible database or an empty bounded search does not prove there are no patents.
New evidence, per-profile search coverage, source confidence ceilings and independent review records are available in the frozen report data. Research coverage remains partial; unpublished applications, name variants and incomplete indexes can leave gaps.