ZIM File Priority Analysis for OAK

Reference appendix. For a guide on using Kiwix and ZIM content, see Kiwix & Offline Content.

Purpose

This document identifies and prioritizes Kiwix ZIM files for inclusion in OAK releases. The goal is to maximize offline utility for people in disconnected or crisis settings — those who need factual information, medical guidance, practical survival knowledge, and educational resources — while respecting the tight space budgets of removable media.

Methodology

Evaluation criteria

Each ZIM file was scored across four dimensions:

1. Crisis relevance — Does this content help someone survive, stay healthy, or navigate an emergency? Medical references, water purification, food safety, and disaster preparedness score highest.

2. Knowledge breadth — How much general-purpose factual information does this provide? Wikipedia variants and encyclopedic references score high.

3. Practical skills — Does the content teach actionable skills (agriculture, energy, computing, navigation)? Hands-on content outranks purely academic material.

4. Size efficiency — What is the utility-per-gigabyte? A 67 MB medical encyclopedia that could save lives scores infinitely higher per-byte than a 168 GB video archive, however valuable the latter may be.

Variant selection strategy

Kiwix publishes many ZIM files in multiple variants:

Our strategy: prefer nopic for large reference works (Wikipedia, Wiktionary) since text carries most informational value in crisis settings. Use maxi for small, visually dependent content (medical atlases, PhET simulations, children's content). Use mini when nopic isn't available or when the full article set is too large for the target tier.

Data sources

• Kiwix OPDS catalog: https://library.kiwix.org/catalog/v2/entries
• Download mirrors: https://download.kiwix.org/zim/{category}/
• File sizes verified against directory listings as of March 2026

Key Tradeoffs

Wikipedia: which variant?

The single most important decision. English Wikipedia ranges from 3.1 GB (Simple English, maxi) to 115 GB (full, maxi):

Decision: Simple English for M, mini for L. Larger variants deferred until demand justifies a 64GB+ tier. Simple English is underrated — it's written at an accessible reading level, making it ideal for non-native speakers and young readers.

Breadth vs. depth

Including one massive resource (Wikipedia nopic at 48G) vs. many smaller ones is a constant tension. We favor breadth at lower tiers (many small, high-impact ZIMs) and depth at higher tiers (larger comprehensive references). An M-tier kit with medical guides, survival content, Simple Wikipedia, and TED-Ed videos is more useful than one with just Wikipedia mini and nothing else.

Language coverage

This analysis focuses on English-language content. Multilingual and non-English ZIMs are noted where they exist but are not included in tier allocations. Future work should create language-specific profiles.

Video content

Video ZIMs (TED, Khan Academy, Crash Course) are extremely large relative to text. TED-Ed at 5.3G is the most space-efficient video offering. Khan Academy (168G) and Crash Course (21G) are Max-tier only. Video is valuable for education but text content wins on information density per byte.

Space Budget

Existing source usage (estimated)

Available ZIM budget per tier

Note: S tier has no Kiwix tools/desktop or ZIM files — it is a pure privacy toolkit. ZIM content starts at the M tier, which includes kiwix-tools and kiwix-desktop.

Curated Priority List

Files are grouped by priority level. Within each level, they are ordered by utility-per-byte (highest first).

P0 — Critical (include in M and L tiers)

Tiny, high-impact survival and medical content. Total: ~1 GB. S tier has no ZIM budget — P0 starts at M.

P0 subtotal: ~1.04 GB

P1 — High priority (include starting at M tier)

Core reference and education content to fill the M-tier budget. Combined with P0, targets ~9.5 GB.

P0 + P1 subtotal: ~12.2 GB (exceeds M tier's ~9.5 GB ZIM budget — need to drop TED-Ed or Gutenberg agriculture from M, or accept a tighter fit by trimming P1 picks. See tier allocation summary below.)

P2 — Medium priority (include starting at L tier)

Broader reference, structured learning, and community Q&A. Combined with P0+P1, targets ~24.5 GB.

P0 + P1 + P2 subtotal: ~23.5 GB (adjustments below)

L tier note: At L, wikipedia_en_all_mini (11G) replaces wikipedia_en_simple_all_maxi (3.1G) from P1, so the net add is ~7.9G, not 11G. Adjusted total: ~20.4 GB within 24.5 GB budget. Remaining ~4 GB could accommodate additional P2+ picks from the expansion list below.

P3 — Extended (deferred — future 64 GB+ tier)

Comprehensive references and richer educational content. Not included in current S/M/L tiers. Revisit when demand justifies a larger tier.

Future 64 GB tier note: If published, wikipedia_en_all_nopic (48G) would replace wikipedia_en_all_mini (11G), net add ~37G. wikibooks_en_all_maxi (5.1G) replaces nopic (2.9G), net add ~2.2G. A 64 GB tier couldn't fit everything in P3 — prioritize Wikipedia nopic + Wiktionary + one TED collection + one Gutenberg collection. Estimated fit: items 24-27 + item 29 ≈ ~52 GB ZIM total.

P4 — Future Max tier only

Large archives that provide tremendous depth but require hundreds of gigabytes. Deferred until demand justifies publishing a Max tier.

Expansion List (ranks 43-100)

Additional ZIM files for consideration, especially at higher tiers or for language-specific builds. Ordered by estimated priority.

Medical & health (43-48)

Practical skills & sustainability (47-55)

Education & video (56-65)

Stack Exchange communities (66-75)

Reference & languages (76-90)

Specialized & niche (91-100)

Tier Allocation Summary

S tier (4 GB hard cap — no ZIM)

The S tier is a pure privacy toolkit. No Kiwix tools or ZIM files are included. Total content is ~1 GB (Tor Browser, Orbot, guides, keys), leaving comfortable margin within the 4 GB hard cap.

M tier (~9.5 GB ZIM budget)

Margin: ~3 GB. Can selectively add TED-Ed (5.3 GB, item 16) if it fits, or Gutenberg agriculture (4.2 GB, item 17) — but not both. Alternatively, keep the margin for quarterly growth.

L tier (~24.5 GB ZIM budget)

Margin: ~4.1 GB. Can add 1-2 items from expansion list (e.g., items 44+55 = 4.5 GB, or item 59 = 4.8 GB).

L tier note: At this tier, wikipedia_en_all_mini (11G) replaces wikipedia_en_simple_all_maxi (3.1G) from P1, so the net add is ~7.9G.

Future tiers (64 GB+, Max)

Not currently published. If demand warrants it, a 64 GB tier could add Wikipedia nopic (48G) and a Max tier could include the full P3/P4 lists. See P3 and P4 sections above for candidate content.

Integration Notes

Relationship to oak.yaml

ZIM files would be added as new sources in oak.yaml, likely as type: http sources downloading from download.kiwix.org. Each ZIM file (or group) would be a separate source entry with appropriate stage_path under a zim/ directory.

Kiwix tools and desktop are included in the M and L tiers. The S tier has no ZIM content and therefore does not need Kiwix software.

Suggested stage layout

zim/
  reference/
    wikipedia_en_*.zim
    wiktionary_en_*.zim
    wikivoyage_en_*.zim
  medical/
    zimgit-medicine_en_*.zim
    wikem_en_*.zim
    mdwiki_en_*.zim
    librepathology_en_*.zim
  survival/
    zimgit-post-disaster_en_*.zim
    zimgit-water_en_*.zim
    zimgit-knots_en_*.zim
    zimgit-food-preparation_en_*.zim
  education/
    freecodecamp_en_*.zim
    phet_en_*.zim
    ted_mul_*.zim
    wikibooks_en_*.zim
    wikiversity_en_*.zim
    vikidia_en_*.zim
  sustainability/
    appropedia_en_*.zim
    energypedia_en_*.zim
  literature/
    gutenberg_en_*.zim
  tech/
    superuser.com_en_*.zim
    stackoverflow.com_en_*.zim
    *.stackexchange.com_en_*.zim

Version management

ZIM filenames include dates (e.g., _2026-02). The download process should fetch the latest available version. This could be implemented via:

• OPDS catalog query to resolve current filename
• Directory listing scrape of download.kiwix.org/zim/{category}/
• Pattern matching similar to existing version_detect in oak.yaml

Verification

Kiwix provides SHA-256 checksums alongside ZIM files. The download source type should verify checksums after download.

Open Questions

1. Language profiles: Should OAK support language-variant builds (e.g., an Arabic-focused L tier)? This would change the ZIM selection significantly.

2. Update frequency: ZIM files are updated monthly to quarterly. Should OAK pin specific versions or always fetch latest?

3. Demand for larger tiers: If users request 64 GB+ builds, which Wikipedia variant should anchor it — nopic (48G, all articles, no images) or mini (11G) plus many more supplementary resources?