{"id":675,"date":"2026-09-05T17:33:55","date_gmt":"2026-09-05T17:33:55","guid":{"rendered":"https:\/\/xplain-r.com\/?page_id=675"},"modified":"2026-09-05T17:33:55","modified_gmt":"2026-09-05T17:33:55","slug":"working-maxims","status":"publish","type":"page","link":"https:\/\/xplain-r.com\/?page_id=675","title":{"rendered":"Working Maxims"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These maxims emerged from the Dossier Space pilot while testing AI-assisted evidential analysis, assurance, reproducibility, transportability, and human supervision against the Enron email corpus.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">They are not presented as formal rules. They are practical lessons distilled from repeated use over three months of configuration and analysis &#8211; failure, rerun, challenge, and review.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Evidence and relevance<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Relevance and usability beat exhaustive accumulation.<\/li>\n\n\n\n<li>Frequency is not relevance.<\/li>\n\n\n\n<li>A single message can be insignificant, or it can be the start of an investigation.<\/li>\n\n\n\n<li>Missing evidence can itself be analytically valuable.<\/li>\n\n\n\n<li>Absence from the corpus is not proof of absence from the real world.<\/li>\n\n\n\n<li>Interpolation must not be presented as observation, inference as fact, or coherence as proof.<\/li>\n\n\n\n<li>A connection is not a relationship; a relationship is not involvement; involvement is not culpability.<\/li>\n\n\n\n<li>The dot on the network is a person.<\/li>\n\n\n\n<li>AI is not analysing abstract records; it is making propositions about people.<\/li>\n\n\n\n<li>Reasoning and transparency<\/li>\n\n\n\n<li>Traceability should be proportional to materiality.<\/li>\n\n\n\n<li>Perfectly exposed reasoning may be worse than proportionately exposed reasoning.<\/li>\n\n\n\n<li>Every exposed reasoning step creates another object capable of challenge.<\/li>\n\n\n\n<li>The objective is sufficient transparency, not maximal transparency.<\/li>\n\n\n\n<li>We finally make reasoning visible enough to inspect, then discover that there may be far too much of it to inspect sensibly.<\/li>\n\n\n\n<li>A perfectly logged search can still search for the wrong thing.<\/li>\n\n\n\n<li>A reproducible analysis can reproducibly omit relevant evidence.<\/li>\n\n\n\n<li>A validated workflow can still answer a badly framed question.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Assurance<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Process assurance cannot substitute for professional judgement.<\/li>\n\n\n\n<li>The purpose of assurance is not to prove the AI right, but to make it possible to discover when it is wrong.<\/li>\n\n\n\n<li>Good assurance should make failure visible, not merely make success look convincing.<\/li>\n\n\n\n<li>The substantive conclusion may survive even when part of its supporting reasoning does not.<\/li>\n\n\n\n<li>Professional curiosity may be as important a safeguard as technical control.<\/li>\n\n\n\n<li>Sometimes the most valuable assurance step is simply asking the question again.<\/li>\n\n\n\n<li>The final test of a trustworthy system is not how well it works, but how safely and visibly it breaks.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Human-led AI<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Human in the loop is not enough if the human merely approves the answer.<\/li>\n\n\n\n<li>The stronger model is human-led work with AI in the loop.<\/li>\n\n\n\n<li>AI should sit inside the professional loop, not place the professional inside the AI loop.<\/li>\n\n\n\n<li>The professional should direct the enquiry, not merely approve its conclusion.<\/li>\n\n\n\n<li>The solicitor directs; the AI investigates; the solicitor challenges; the AI responds; the solicitor decides.<\/li>\n\n\n\n<li>AI is better treated as a supervised para-legal capability than as an oracle.<\/li>\n\n\n\n<li>Responsibility cannot be automated away by adding more controls.<\/li>\n\n\n\n<li>The more capable the AI becomes, the more important it is to know who is actually directing the work.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Methods and tools<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Govern the environment; keep the professional in charge; let the AI provide the capability.<\/li>\n\n\n\n<li>Primers should be foundations, not handcuffs.<\/li>\n\n\n\n<li>Structure the enquiry before the AI begins, then allow the evidence to change its direction.<\/li>\n\n\n\n<li>The method should follow the problem, not force the problem into the method.<\/li>\n\n\n\n<li>Governed enough to be defensible; flexible enough to investigate properly.<\/li>\n\n\n\n<li>Technical competence should be expected of professionals; technical proficiency should not be a condition of professional AI use.<\/li>\n\n\n\n<li>The durable asset is the professional operating method, not the particular AI platform.<\/li>\n\n\n\n<li>The technology will change faster than the professional principles should<strong>.<\/strong><\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Professional practice<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>The danger is not necessarily that AI replaces professionals, but that professional judgement is displaced incrementally without anyone deciding that it should be.<\/li>\n\n\n\n<li>If the profession does not choose its AI operating model, the market will choose one for it.<\/li>\n\n\n\n<li>Professional conservatism may protect against reckless adoption, but it may also allow others to define the operating model.<\/li>\n\n\n\n<li>The important question is not whether AI can find something, but whether the finding can be challenged, reproduced and properly relied upon.<\/li>\n\n\n\n<li>\u201cAI can search\u201d is no longer the interesting proposition.<\/li>\n\n\n\n<li>The unit of competition may become the governed workflow one professional can command, rather than the number of people an organisation can deploy.<\/li>\n\n\n\n<li>Large models increase capability; good method determines whether that capability is usable.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>The Dossier Space progression<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Raw evidence \u2192 assured handling \u2192 analysis \u2192 analytical finding.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The Dossier Space pilot suggests that increasingly capable, commercially available AI can support this progression on ordinary computing hardware. The continuing challenge is not simply to increase capability, but to preserve professional control, evidential discipline, proportional assurance, and the ability to recognise when the system has reached its limits.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>These maxims emerged from the Dossier Space pilot while testing AI-assisted evidential analysis, assurance, reproducibility, transportability, and human supervision against the Enron email corpus. They are not presented as formal rules. They are practical lessons distilled from repeated use over three months of configuration and analysis &#8211; failure, rerun, challenge, and review. Evidence and relevance [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-675","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/xplain-r.com\/index.php?rest_route=\/wp\/v2\/pages\/675","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/xplain-r.com\/index.php?rest_route=\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/xplain-r.com\/index.php?rest_route=\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/xplain-r.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/xplain-r.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=675"}],"version-history":[{"count":1,"href":"https:\/\/xplain-r.com\/index.php?rest_route=\/wp\/v2\/pages\/675\/revisions"}],"predecessor-version":[{"id":676,"href":"https:\/\/xplain-r.com\/index.php?rest_route=\/wp\/v2\/pages\/675\/revisions\/676"}],"wp:attachment":[{"href":"https:\/\/xplain-r.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=675"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}