Skip to content

A Civilization-style strategy game about governing Israel from 15 May 1948. Real ministers, real map, every fact sourced. Unity + deterministic C# sim.

13.0 KB
Raw Blame

Historical content policy

Statecraft: Israel puts real people, real places and real wars on screen. This page is the contract for every word, number and picture in content/ and in the game. It binds the maintainer, contributors and AI agents alike. It is written from the project's design spec (docs/superpowers/specs/2026-10-09-statecraft-israel-design.md, §1 and §5).

1. The perspective is Israeli, and it says so

The player leads the Yishuv and the State of Israel. Narration and terminology use Israeli framing, and each fact takes one value from Israeli official or mainstream sources: the Central Bureau of Statistics, the IDF, the Ministry of Foreign Affairs, the State Archives, and mainstream Israeli historiography. Wikipedia (English and Hebrew) is a primary cited source for many facts; where it is the source, the entry cites it, and that is why content/ is CC BY-SA 4.0. Contested ranges and opposing-narrative panels are not shown. Arab and Palestinian actors are computer-run actors with their own goals.

This is a design decision by the maintainer. Issues and pull requests that ask to change the perspective will be closed with a link to this section. That is not a judgement on the person asking, and it is not up for vote. A fork can make a different choice under its own name (see TRADEMARK.md).

2. What holds regardless of perspective

These rules protect the game and the people in it. They are not a viewpoint:

  1. Every fact is sourced. Every date, number, name and office in content/ carries sources: (a citation or a URL, plus the date it was checked) and a review status. As of M3 every fact is at least fact-checked, except records that stay draft, among them three events (first_new_yishuv, jerusalem_demilitarisation_guard and rafah_force_pressure; RealContentTests pins the list). Events with a sensitive tag must also be sensitivity-reviewed before any build leaves the maintainer's machine; the simulation enforces that (Statecraft.Cli --release-gate, the -shareable build flag and tools/package_release.py all refuse otherwise). Historian review is planned for M4 and is not yet a gate.
  2. No invented quotes. No words are attributed to a real person unless they are sourced verbatim. Event text describes; it does not put speech in anyone's mouth. The one exception is the advisors (David Ben-Gurion, Golda Meyerson (later Meir) and Moshe Dayan), by owner ruling, 23 October 2026: "Labelled game dialogue". Their lines are original game dialogue in the persona, shown in the advisor panel under the label "Advisor" and never styled as quotes; they explain the game and its state, and say nothing on a contested issue beyond neutral game facts. A real quote appears only from the verified list in docs/superpowers/specs/2026-10-23-advisors-design.md (section 3.3), styled as a quote with its source. The owner reviews every advisor line before it ships.
  3. Choices never reward atrocities. No option of an event tagged holocaust, terror, massacre, war-casualties or displacement can raise a resource: the loader refuses a positive effect on money (treasury, fx_reserves), Food, Arms, morale or international standing (EventRules), so dotnet test fails on one. Nor can a displacement option raise score (when score becomes a target, R-EFX-007), and a sensitive option carries no unit grant, city-effect gain or project unlock (R-EFX-005; the loader refuses those too). Such an event also needs a recorded sensitivity review.
  4. The Holocaust is an event the player reacts to, never one the player plays. It is shown without swastikas.
  5. Real people. People appear in their public roles and documented acts. Competence and loyalty values are game balance, not judgements of the person, and content/people.yaml labels them so. The maintainer does not add living people, and records a legal review before any release that includes one. Portraits are public-domain, CC0, CC BY 4.0 (owner ruling on #30, 13 October 2026) or CC BY-SA 4.0 photographs (owner, 14 October 2026, docs/oss/DECISIONS.md D-P: "yes" to CC BY-SA 4.0 for portraits); share-alike applies, so a portrait cropped from a CC BY-SA 4.0 photograph is itself released under CC BY-SA 4.0 and credited in NOTICE. The maintainer refuses any other licence at review.
  6. AI-assisted writing is a draft. A human checks every factual claim in it against its sources, one by one, before the entry is marked fact-checked; the maintainer applies this rule at review. A Hebrew display name (name_he, title_he, title_he_f, body_he) that an AI agent checked against its sources, but no person has read yet, carries its own flag beside it, *_review: agent-checked (for example name_he_review), whatever the row's status says about its English facts (owner, 15 October 2026). A map locality's name_he (and its today-name tooltip) in content/map/localities_1948.yaml copied verbatim from a Wikidata label or a Hebrew Wikipedia title, as that file's header describes, carries no such flag (maintainer reading, not part of the owner's option: dropping only a bracketed disambiguation such as "(יפו)", direction marks or niqqud still counts as verbatim, as #2044 did); a locality name an agent wrote or adapted still does, and so does every other Hebrew name (the owner chose the option labelled "Exempt verbatim labels (Recommended)" (4 November 2026)). A native Hebrew speaker reviews these names (#564) and then removes the line, or an AI-assisted contributor does, when the pull request discloses the AI help and names the strings read and the sources they were read against (asked whether an AI-assisted contributor's Hebrew read can clear an agent-checked string or still needs a native Hebrew speaker, the owner answered "AI-assisted can clear it" (4 November 2026)); the loader accepts only agent-checked (a blank value is an error), and only next to a name, or, on an event, text_he_review next to its Hebrew text.
  7. A sign-off names a person. A row a person has reviewed carries, beside its status, reviewed_by (the reviewer's GitHub handle, never an email) and reviewed_on (the day), both or neither (owner, 15 October 2026, #375: "Yes, humans only"). Only that person writes the two lines, in their own commit; an AI agent never adds, changes or removes them, and scripts/oss/reviewed_by_guard.py fails any pull request where a commit with an AI co-author trailer does, unless the pull request's "AI assistance" section discloses the AI help. Then the AI-assisted contributor's check counts as the sign-off, under their own handle, and the section should say which sources they opened and what they checked (asked whether an AI-assisted contributor's source check counts as the reviewed_by sign-off, the owner answered "Yes, it counts" (4 November 2026)). Even then, such a commit never removes a sign-off. CI checks only that the section exists, is not a "no" and names a check; the reviewer reads what was checked and how. A signed row is closed to agents too (owner, 15 October 2026, on #629: "Block it in CI"): the guard fails a pull request where a commit with an AI co-author trailer changes any field of a signed row, the id included, until a person signs the row again or removes its sign-off in a commit of their own (an AI-assisted contributor, disclosed as above, may sign it again but not remove the sign-off). How to sign off: HISTORIAN_REVIEW.md. A row with a sign-off leaves the review queue (review-queue.md).

3. How to dispute a fact

Facts are open to correction. The framing (section 1) is not.

  1. Open an issue with the Historical accuracy form. Name the file and the field (for example content/events/altalena.yaml, text), say what is wrong, and give at least one source the reader can check. A source is a published book or article, an official record, or a reputable reference work. A link to a social-media post is not a source.
  2. The maintainer checks the source against the current one and answers in the issue. When the sources disagree, the game keeps one value, chosen by the rule in section 1, and the issue records why.
  3. A correction lands as a pull request that changes the value and its sources: entry, with today's date in checked:.
  4. The maintainer's decision closes the dispute. Re-opening the same dispute without a new source is off-topic under CODE_OF_CONDUCT.md.

4. What contributors must not add

  • Content without sources, or with a source that does not support the claim.

  • Quotes, speeches or letters that are not sourced verbatim.

  • Mechanics or text that reward or celebrate violence against civilians, from any side. A player choice that harms civilians, such as the GDD's Deport outcome, may exist only as a historical possibility that never pays: it carries its real costs (refugees, Social cohesion, International Support), gives no reward, and is shown soberly through sourced events (owner ruling, 12 October 2026).

  • Images of real people that are not public domain or CC0, with two exceptions. Portraits may also be CC BY 4.0 or CC BY-SA 4.0 photographs, on the terms of principle 5 in section 1 (owner ruling, 17 October 2026: "Allow CC BY / BY-SA"): each is credited in NOTICE, and a portrait cropped from a CC BY-SA 4.0 photograph is itself released under CC BY-SA 4.0. The other exception is narrow, by owner ruling 2026-10-11 (summarised): archival photographs of events (not portraits) under CC BY 4.0 are allowed, on these terms:

    • Each carries a credit line naming the photographer and collection (for the Meitar Collection: photographer, Meitar Collection, National Library of Israel, The Pritzker Family National Photography Collection), with the licence, both on screen with the picture and in NOTICE.
    • The only changes are a subtle warm tone grade (the pictures stay black and white, in the gold frame), a crop and a resize. No AI colourising and no change to what the picture shows. The adaptation is recorded with the credit as "graded and cropped".
    • Government Press Office (GPO) photographs are refused. So are files whose only licence is a PD-Israel or PD-India tag, and share-alike files (CC BY-SA).
    • Each photograph gets a sensitivity review before a shareable build may carry it: every one is screened, and any marked sensitive waits for the owner's sign-off (the -shareable build and the packager refuse until then).

    Each event photograph's credit, alt text and sensitivity review are recorded in unity/Assets/UI/EventImages/event-images.yaml and ASSETS.yaml (see docs/event-images.md).

    AI-generated likenesses of real people appear only in concept art (promo/final/04, 08), are recorded as AI-generated in the asset provenance records, and are never presented as photographs or used in the game. The advisors are the one exception, by owner ruling, 23 October 2026: "you can sythesize their voices", and, for their videos, "The real figures talking", then "Leader's face, labelled", and on the same day: "Amend the policy". On 24 October 2026 the owner amended it to the route now used: on the estates' consent, "Amend the rule", and on the rest of the exception, "Amend to my route". So, for Ben-Gurion, Meyerson and Dayan only, and only in game dialogue, a clip may show one of them talking, made from a real archival photograph named in the advisors spec (section 11.1): the photograph may be faithfully colorized, with the colorization recorded in its credit, then animated with AI and lip-synced to a synthetic voice made as close to the real person's as possible, with third-party services. No written consent from their estates is needed first. This exception covers those photographs even where the event-photograph terms above would refuse them (a Government Press Office photograph, AI colourising); nothing else the photograph shows is changed, and bystanders are never animated or altered. One exception is recorded: the four Golda Meyerson clips opening.city, opening.training, meyerson.exhausted and meyerson.debt_ceiling, made by the recipe the owner approved as 'A, raw' on 24 October 2026, animate the whole frame, so the people around her change expression (ASSETS.yaml); the owner chose the option labelled "Record the exception (Recommended)" (4 November 2026). Any new clip follows the rule. Each clip speaks only owner-approved game dialogue, never a quote, is never presented as real footage, is labelled "AI animation and voice. Fictional dialogue.", and is reviewed by the owner before it ships. Every such output is recorded in ASSETS.yaml as AI-generated (docs/superpowers/specs/2026-10-23-advisors-design.md, sections 11 to 13).

  • Hate speech, slurs or dehumanising language about any group, in content, code comments, issues or commits.