Appendix E — What This Book Cannot Certify
A book that grades its evidence owes the same grading to itself. These are the weaknesses I know about. There will be others.
1 · The evidence is skewed toward the United States and the United Kingdom
Almost every independent evaluation cited in Chapters 3, 6 and 11 is American or British: NAS, PCAST, NIST, RAND, inspector-general audits, the Innocence Project, the National Registry of Exonerations, Byford, Macpherson, the College of Policing.
This is not because those countries do investigation better. It is because they publish evaluations, and they publish them because they have inspectors general, freedom-of-information regimes, civil liberties organisations with standing to sue, and a legal culture that generates public documents when things go wrong.
The consequence for the reader is specific and uncomfortable: this book is disproportionately a description of what has been evaluated, not of what has happened. Where a country's failures do not generate documents, they do not appear in these pages. Nothing in the Thai chapters should be read as evidence that Thai enforcement fails less or more than the systems described elsewhere. It is evidence that it is measured less.
The Dutch material in the closing chapters and Appendix A is included partly to break the pattern, and it is not enough.
2 · The Thai institutional data is thin because it is not released
The Thailand chapter and Appendix B rest on three declared absences: no published clearance rate for online fraud, no aggregate conviction count, no current investigator headcount. Those absences are honestly reported and they are also a real limitation on what this book can say.
Everything in the Thailand chapter about where the bottleneck sits is inference from volume, not measurement of throughput. It is a well-supported inference and it is not a measurement. If CCIB publishes staffing and case-outcome data tomorrow and it contradicts the argument, the argument loses.
Two enforcement figures in Appendix B were also captured only through search-engine synthesis rather than a primary document fetch, and are marked accordingly.
3 · The PDPA question is unresolved and it is the one you most need answered
The open-source chapter states plainly that no source found addresses how Thailand's Personal Data Protection Act applies to an individual, non-commercial investigator, as distinct from a corporate data controller.
That is not a decorative caveat. For a reader intending to do this work in Thailand it is the single most consequential open question in the book, and it requires an opinion from a Thai-qualified lawyer. Nothing in that chapter should be treated as legal advice, and the chapter's Move — writing the lawful basis in the file — is a discipline that helps you demonstrate good faith, not a substitute for knowing the law.
4 · Every [CLAIMED] figure is still uncorroborated
At press, the following remain agency or vendor self-report with no independent audit located:
- Thailand's entire enforcement results table (accounts closed, assets frozen, arrests, percentage
declines) — Appendix B §2
- Singapore's recovery figures, including the 25% recovery rate that is the most useful number in
the money chapter
- The Bedfordshire 1.4TB / 100,000-messages-in-one-day claim, and its counterfactual timeline
- i2 Analyst's Notebook adoption figures
- The tracing-methodology narratives in the crypto cases, as distinct from the DOJ-confirmed seizure
amounts
- Thailand's SHIELD database and IBOC, whose capabilities I could not verify against any primary
government source
If you cite any of these, cite them as claims. The book's own Move on claimants applies to the book.
5 · Specific things I could not verify
- The exact verbatim wording of the NAS 2009 report's central conclusion. Paraphrases are
well-sourced; the primary text should be pulled before anyone quotes it as a sentence.
- PCAST's precise finding on footwear-impression evidence.
- A "52% of Innocence Project cases involved forensic misapplication" figure that circulates in
secondary literature. It could not be traced to a primary source and is probably a conflation of the 43% (n = 375) and "more than 50%" (n = 225, pre-2009) figures. This book uses 43% with its denominator stated.
- A field-wide official error rate for bloodstain pattern analysis. It appears not to exist; the 2021
black-box study is the best available data point, not a field consensus.
- The PEACE-versus-Reid false-confession comparison of roughly 0% against 50%, which appears in
secondary summaries. Reported as a claim, not a finding.
- Verbatim clauses from the Berkeley Protocol on hashing, preservation and minimisation. The
document's structure and OHCHR's summary of its contents are confirmed; the clause text is not.
- Whether the Texas Forensic Science Commission's Willingham review constitutes a formal finding of
invalid science or a narrower administrative closure. Sources differ on how strongly to characterise it, and the forensic chapter separates the fire-science consensus from the Commission's own conclusion.
6 · Three structural bets the book makes
That the linking problem is the right problem. The book argues the bottleneck is between report and network. That is an argument from volume and from the absence of a detection deficit. It is not established by any published study of where Thai investigative time actually goes, because no such study was found. If the real constraint turns out to be prosecutorial capacity, or cross-border cooperation, or the simple fact that suspects sit in another country, then the machine described in Part II improves a step that was never the binding one.
That citation-gated machine assistance helps. The book's central rule is that a machine may produce only pointers. The rule structurally eliminates fabrication. Whether the resulting workflow actually makes investigators faster or more accurate has not been demonstrated — the one formal evaluation identified, the UK College of Policing's CoPA study, had not reported as of August 2026. That rule is written on a promise.
That the five-per-cent capture estimate for Thailand is the right order of magnitude. The Thailand chapter derives it from a survey percentage and an approximate adult population, and it names four assumptions that could each move it by a factor of two. It is shown with its working precisely because it is the weakest number in the book. [CONTESTED]
Appendix D is a fourth bet, and it declares itself: its diagnosis is proven mathematics and its remedy has no documented deployment anywhere in investigative practice.
7 · What this book deliberately does not contain
No inference tutorial. No Bayes' theorem derivation, no bestiary of cognitive biases, no extended argument about abduction. That work is in the companion volume in this library, The Science of Deduction / ศาสตร์แห่งการอนุมาน, which covers the reasoner as this book covers the job. Where the two touch — the opening, the linking chapter and the closing protocol — this book compresses and points rather than repeating.
No investigative techniques that require breaking a law. The open-source chapter draws that line explicitly at the Navalny case, and the book stays on the near side of it throughout. This is a real limitation on coverage and it is intentional.
No methods for investigating a private individual. Nothing here is a guide to building a file on a person in your own life. That chapter's account of when data-protection obligations begin — the moment you start building a pattern of life on someone — applies to the reader as much as to anyone the reader might investigate.
8 · Further reading
The primary documents worth owning rather than reading about:
- NAS (2009), Strengthening Forensic Science in the United States: A Path Forward —
- PCAST (2016), Forensic Science in Criminal Courts —
- OHCHR & UC Berkeley (2022), Berkeley Protocol on Digital Open Source Investigations —
- Byford (1981), report into the police handling of the Yorkshire Ripper case —
- National Academies (2003), The Polygraph and Lie Detection —
- NISTIR 8280 (2019), face recognition demographic effects —
- The Markup (2023), the Geolitica audit, with published methodology —
- Bellingcat (2020), Hunting the Hunters — the Navalny methodology, including its own account
of the legal boundary — bellingcat.com
- Shafer (1976), A Mathematical Theory of Evidence, and Belnap (1977), *A Useful
Four-Valued Logic* — for Appendix D.
Full source dossiers, with every URL and retrieval date, are in source/research/.
Retrieval date for all sources in this book: 29 August 2026. Live counters — the National Registry of Exonerations figure in Chapter 3, weekly Thai case statistics in Chapter 2 and Appendix B — have moved since. Where a number matters, check it.
จบเล่มAfter Sherlockกลับหน้าหนังสือ