Evidence rules

These rules apply at every stage of the pipeline. They exist because a platform that claims to be evidence-based is only worth its weakest claim — and the characteristic failure mode of AI-assisted research is a confident number with no source behind it.

• No invented numbers. A statistic without a source is marked [UNVERIFIED — verify before publication] and blocks stage advancement.

• Preferred sources, in order:

  1. The municipality’s own published record — approved budget, council and committee protocols, bylaws (חוקי עזר), tenders, plans, and the municipal auditor (מבקר העירייה).
  2. CBS (הלשכה המרכזית לסטטיסטיקה) — the local authorities file and neighbourhood-level statistical areas.
  3. Jerusalem Institute for Policy Research (מכון ירושלים למחקרי מדיניות) — the Statistical Yearbook of Jerusalem and Jerusalem: Facts and Trends.
  4. State Comptroller (מבקר המדינה), including the local-government audit.
  5. Ministry of the Interior (משרד הפנים).
  6. Knesset Research and Information Center (מרכז המחקר והמידע של הכנסת).
  7. Bank of Israel, OECD, peer-reviewed research.
  8. Quality journalism — last resort, always flagged. A council row that made the local press is reported, not proven; find the protocol.

• Instruments are cited precisely, and municipal work runs mostly on subordinate instruments. A bylaw, ministerial circular, licensing condition, tender clause or budget line gets the same treatment as a statute: official name in Hebrew and English, the issuing authority, the date, and the specific clause. Verify against the actual text; do not cite provisions from memory.

• No power without a citation. “The municipality is responsible for X” is a claim like any other and needs the provision that says so. See the powers doctrine; this is the rule the whole project turns on.

• No promise without a mechanism, and no mechanism without a funding source. Every costed recommendation names the budget line and says what gets less. “Efficiency savings” is not a source. Anything relying on arnona (ארנונה, municipal property tax) income must establish the current rule on raising it and address the approval risk.

• Testimony is evidence of experience, not prevalence. Prevalence claims require data. At city scale this bites harder than nationally: one neighbourhood’s problem can be entirely real and entirely unrepresentative. Say which neighbourhoods.

• Disaggregate. City-wide averages hide the thing municipal policy exists to address. Report by neighbourhood or statistical area wherever the source allows, and where a source does not cover part of the city, record that — missing data about a neighbourhood is a finding, and often the most important one.

• Adopted is not implemented, and implemented is not funded. State which, with the date checked.

• Trade-offs mandatory. Every proposal names its losers — including the neighbourhood that loses and the department that absorbs the work — and answers the strongest objection, not the weakest.