About & Credits

Acknowledgement of source

All documents presented on this site originate from the Independent Committee in relation to the Fire at Wang Fuk Court in Tai Po and are the work of the Committee, its counsel, experts, witnesses and the Government. Original PDFs are linked from every document page. This site reorganises, indexes and summarises those public materials; it claims no authorship of them.

Compiler

Compiled and maintained by Dr Simon Wang (simonwanghkteacher@gmail.com).

Disclaimer

Dr Simon Wang compiles this site as a private individual, not as a scholar of Hong Kong Baptist University (HKBU). HKBU is not involved in, and bears no responsibility for, this site. The site is provided for educational purposes only and is not for commercial use. Readers should consult the original documents for authoritative content.

Image count and filtering (8,785 extracted → ≈1,800 shown)

All 8,785 images embedded in the 243 PDFs were extracted programmatically. Most are not meaningful standalone items: automatic inspection classifies them into photographs (1,772), diagrams, charts and plans (1,036), full-page scans (834), blank (304) and tiny layout fragments (4,837 — single dots, sprites and formatting debris produced by PDF extraction). A vision model then reviewed the photographs and diagrams, rated each for importance (★1–5) and relevance; the ≈1,800 that met the bar (importance ★3 or above, or clearly substantive non-photographic material) are shown on the document pages — 'N shown of M' — so readers see evidence rather than extraction noise. The full extracted set remains available in the original PDFs.

How Terms-of-Reference counts are computed

Counts under each ToR reflect how many of the 243 documents contain substantive material related to that task, judged by a large language model that read the full text of every document (in ~2,000-character passages, semantic judgment rather than keyword matching). Documents may count under several ToR. Twenty-seven documents arrived as scans with no readable text; they have since been OCR'd and re-read by the model, so every document now has full-text tags. This full-text reading markedly raised the counts for ToR5 (collusion, bid-rigging) and ToR6 (laws) compared with the earlier summary-level classification: many documents that never use those keywords still contain directly relevant evidence — for example a registered fire-service contractor that never visited the estate, an oral subcontract with no written contract, and a contract attachment whose material-data folder was empty.

Scanned pages: 347 pages of evidence published as images

Of the 243 documents (6,377 pages) published by the Committee, 46 documents — 347 pages — carried their text only as scanned images with no machine-readable text layer: 27 documents (283 pages) were scans in full, and a further 19 documents contained 64 individual scanned pages mixed into otherwise text-based PDFs. None of these pages can be searched, quoted or read by a computer as-is; anyone who wishes to process the text — a journalist, a researcher, a member of the public, or the compiler of this site — must first run optical character recognition (OCR) page by page, then check and correct the output, because OCR of scanned pages is never perfectly accurate. This site performed that work for every affected page so the material could be searched and read. Publishing official evidence as images instead of text imposes a hidden, unnecessary burden on anyone who wants to work with the documents on a computer; this practice is to be strongly condemned.

Open source & press

This site is built from an open-source repository on GitHub ↗ (project under projects/FireReport): the crawler, PDF processor, OCR, LLM analysis and site generator are all public, so journalists, researchers and developers can reproduce or adapt the methodology. Our letter on the accessibility of the inquiry documents was published in the South China Morning Post letters column on 19 August 2026 ↗ — read it on the Our Letter page.