KneeSchool/About/Build status
Build status
What is done, what is waiting, and what is blocked
97 open items. A page that tells you a feature is not available yet links here, so this list is the answer to where that work has got to.
Blocked by something outside this repository
Work that cannot proceed from here, whatever effort is applied. Each one names what would unblock it.
| Item | State | What is in the way | What would move it | Pages |
|---|---|---|---|---|
| EV-1 Evidence verification has never run | Blocked | The build environment has no access to PubMed, Cochrane, NICE, any journal or any textbook. Agent 2 of the three agent pipeline is the fact checker and it cannot retrieve a single source. | Source access, or a human verifier. Until then no page on this site carries a reference, every verified handoff is marked NOT_RUN_TOOLS_UNAVAILABLE, and no citation has been invented to fill the gap. | All 245 published article pages |
| EV-2 Consultant review has reviewers and a return path, and nothing signed | Mechanism built, nothing signed | The client confirmed on 10 October 2026 that consultants are lined up, so the people exist. The mechanism existed only as prose packs in docs/review/ with no way for an answer to come back. tools/review_pack.py now generates one pack per chapter from the work and reads a filled pack back into the registers, so the route exists. What is missing is a reviewer's time. | A consultant to work through a chapter. 3,249 clinical recommendations across 635 pages and 526 measurements across 205 are outstanding, none signed, and chapter 3.17 adds 68 attributions to named authors, none checked against a paper. python3 tools/review_pack.py --status prints the breakdown by chapter; 73 chapters carry recommendations, the largest is 4.2 with 191 and the median is 33. | Every page on the site is an unreviewed editorial draft |
| EV-3 The sgit.ai deployment workflow was never read | Blocked | https://sgit.ai/llms.txt returns 403 at the network egress gateway. The host is not on the allow list for this environment. | Either the document contents pasted into the chat, or the host allowed through the environment's network policy. | |
| EV-4 The Bedrock pipeline has never been deployed; the static site has | Blocked | No AWS account, region or budget has been named, so the three agent pipeline has never run anywhere. That part is unchanged. What was wrong until 10 October 2026 is the claim that nothing is deployed: GitHub Pages has been building and publishing this repository on every push since late September, 61 successful runs, and a CNAME file in the repository root points kneeschool.com at it. See EV-21. | For the pipeline: an AWS account to deploy into and a monthly budget ceiling for Bedrock. For the static site: nothing, because it is already live. | |
| EV-5 Architecture received for sections 1 to 15, titles only | Partly resolved | The compiled Sections 1 to 15 architecture arrived on 9 October 2026 and is parsed into outline_1_to_15.json and section_conventions.json. It gives titles, chapters, parts and per section tiers and subsections. It does not give per page scope, must_not_cover, deeper_pages, page_type or priority, so briefs still cannot be generated for sections 4 to 15. The must_not_cover list is what stops the same fact being written on three thousand pages. | Per page scope and must_not_cover for sections 4 to 15, in the shape Part 1 used for sections 0 to 3. Also the pages of the 775 chapters that stop at chapter level: tools/outline.py --gaps lists them. See docs/findings/001. | 1,233 pages are outline only; 283 are briefable |
| EV-6 526 measurements on the site are unchecked | Live and flagged, and the count rose on 12 October 2026 because the scan was wrong | They are written from memory because no paper could be retrieved. Each page's references note says so and each figure is registered. The total rose from 381 to 442 on 12 October 2026, not because measurements were added but because the scanner could not see a measurement written in words. rule_fig_001's own docstring asserted that a quantity cannot be written without a digit. Chapter 4.9 disproved it by quoting six flexion angles as "ninety degrees" and similar, every one of which escaped the register decision 006 says lists every measurement on the site. 61 had accumulated across the site. The scan now reads a number word carrying a unit, and a bare enumeration such as three structures is still allowed. | Someone with source access to work the register: 526 figures across 205 pages, all UNVERIFIED_FROM_MEMORY. tools/figures.py --report --unchecked is the list. | 205 pages carry at least one measurement |
| EV-7 3,249 clinical recommendations on the site are unsigned | Live and registered, and the count rose on 10 October 2026 | They are written by a process with no clinical qualification, on pages that state they have not been reviewed. The client has chosen to keep them, and decision 007 makes the set finite and listed. The total rose from 976 to 1,295 on 10 October 2026, and chapter 3.12 added 93 more, not because advice was added but because the detector was found to be missing it. POS-001 read clinical direction off a list of imperative openers built from directive prose, and the directions opening with a weighing or a consultation verb had never reached the register. Two extensions of the verb list, decisions 018 and 019, added 319 between them. | A consultant to work the register: 3,249 positions across 635 pages, all UNSIGNED. tools/positions.py --report --unsigned is the list. Section 4 is where this count will grow fastest, because an examination page tells a clinician what to do by definition and 185 of its pages are still unwritten. | 635 pages carry at least one recommendation |
| EV-8 The Master Handbook has been supplied and read | Resolved | Supplied 9 October 2026 in Knee_school_2.zip and now in docs/KneeSchool_Master_Operations_Handbook.pdf with a text extraction beside it. Its text is stored as hex glyph ids from seven subset fonts, so it took a ToUnicode character map parser to read. 21 chapters, 10 appendices, 25,242 characters. | Nothing further. What it answers, what it adds and the three places it conflicts with the later Operations Handbook are in finding 002. | Affects the pipeline and the review route, not any page |
| EV-9 Which numbering names the academies | Resolved | In the Master Handbook the Anatomy Academy is chapter 6 and the Conditions Library is chapter 9. In the publishing architecture they are Sections 2 and 6. | Ruled on 9 October 2026: the architecture numbering is canonical. Nothing in the build changes, because the repository never used the other one. See decision 008. | None |
| EV-10 Section 0 is absent from the new architecture | Flagged | The compiled architecture covers Sections 1 to 15. The Junior Academy is not in it. The Master Operations Handbook of 13 June 2026 omits it too: its curriculum matrix has six levels, Patient to Consultant. The Master Playbook omits both Junior and Consultant and lists five. The two documents that do specify Section 0, the Junior Academy architecture of 28 August and the Operations Handbook of 22 September, are the later ones, so the tier appears to have been added after the June planning round. | Confirmation that Section 0 is still in the plan. | 38 published Junior Academy pages |
| EV-11 Publication QA, the fifth gate, is built | Resolved | Chapter 13 of the Master Operations Handbook names five QA tiers. The pipeline had four. The fifth, publication review, checks formatting, SEO, links, images, alt text, disclaimers, metadata and page status. | Nothing. Built on 9 October 2026 as tools/publication_qa.py, nine rules, 116 pages, 0 findings. The fields it checks did not have to be written by hand: 97 of 100 article summaries already fitted the meta description bounds, no title exceeded 60 characters, and the 20 pages that shared a title now carry their chapter in the title tag. See decision 010. | All 116 pages pass |
| EV-12 Which handbook has authority | Resolved | Three documents now set editorial rules. The Operations Handbook is version 1.0 of 22 September 2026 and claims authority over all other documents including prior playbook drafts. The Master Operations Handbook and both playbooks were written on 13 June 2026 and carry no version. On the dates, the June documents are the prior drafts that rule names. | Confirmed on 9 October 2026. The Operations Handbook of 22 September 2026 is senior on any disagreement; the Master Operations Handbook governs what it does not cover; both playbooks are historical. FAQs stay on the patient tier only and the spiral keeps seven tiers, so nothing is reworked. See decision 009. | None; the ruling confirms the existing state |
| EV-13 The prose reference report, and two faults it has found | Fixed twice, and tested | A bare reference in prose can name a real chapter and still be the wrong one. Seven pages across chapters 2.8, 2.9 and 2.10 sent the reader to Meniscal Examination, ACL Examination and PCL Examination for the posterior drawer, the medial side and the dial test. Every id was valid, so nothing caught it. | Nothing. 1,558 bare references across 202 pages, none pointing at an id the architecture lacks. The detector has itself been corrected twice: chapter 3.4's bare decimals were read as chapter references, one of which resolved to a real page, and fixing that uncovered a pre-existing miss where any reference ending a sentence was never checked. There were 56 of those. Six tests hold both fixes. | 7 pages corrected; 88 pages carry bare references |
| EV-14 Fifty-two study attributions on the site are unchecked | Writing complete, checking outstanding | This entry used to hold eleven pages. The client ruled on 11 October 2026, "Write it flagged unverified", and all eleven are now written: chapter 3.17's ten and page 3.1.8. Decision 022 sets the terms. The finding and the surname a study is taught under may be written, because they are recalled together and either checks the other. The year, the journal, the volume and the page may not, because they are recalled alone and a wrong one cannot be told from a right one. ATTR-001 refuses all six fabricable shapes at the gate and ATTR-002 requires every named attribution to be in a register. Nothing is held for source access any more. What is outstanding is the checking: 68 attributions across 14 pages, every one from memory, and some of them will be wrong. | Someone with a library to work the attribution register: 52 entries, all UNVERIFIED_FROM_MEMORY. tools/attributions.py --report --unchecked is the list, and the chapter 3.17 review pack carries the same rows with a Citation column, which is the one field this process could not fill. Separately, page 3.17.9 needs a ruling from the Imperial group itself: the architecture asked for their research programme summarised and the page delivers their published work instead, which is less, and whether a page bearing an institution's name should exist in that form is their call rather than this process's. | 11 pages published, 9 of them carrying unchecked attributions |
| EV-15 Fourteen pages fetched the wrong two typefaces | Fixed, and a check added | The MyKneeScore port changed the stylesheet to Cormorant Garamond and Hanken Grotesk. The 151 rendered article pages took the new font block from tools/site_chrome.py. The 14 hand written pages, the homepage and all seven level pages among them, kept the previous Newsreader and Source Serif block, so they asked for fonts the stylesheet never names and rendered in the Georgia and Helvetica fallbacks. A page that cannot load a font falls back silently, so the site verified clean for three days while its most visited pages looked wrong. | Nothing. The font block has one definition in tools/site_chrome.py, tools/apply_chrome.py rewrites it on every page, and tools/check_site.py fails any page that fetches a family the stylesheet does not name. Four tests hold it. | 14 pages corrected, all 165 now fetch the same two |
| EV-16 Two pages could have rendered to the same file in silence | Guard added, and tested | A render target is a page id mapped to an output path. Nothing checked that two ids did not map to the same path. One would have published and the other would have been lost, with every gate still passing: the file exists, it is valid, the links resolve, and nothing knows a second page was meant to be there. Chapter 3.6 came within one slug of it, because its Load Sharing, Shock Absorption and Proprioception pages share their titles with chapters 3.4 and 2.7. | Nothing. tools/render_site.py refuses the whole run and names both pages when two targets collide. Two tests hold it, one of which introduces a collision and checks the renderer refuses it. The titles stay as the architecture gives them and the three slugs carry a meniscal prefix. | No page was lost; the collision was caught before it was written |
| EV-17 163 clinical directions were never registered | Fixed, re-extracted and tested | POS-001 detects clinical direction from a list of imperative openers. The list was built from directive prose, so Assess, Avoid and Prefer were in it and Weigh, Stage, Correct, Offer and Confirm were not. A dry run across every professional page found 163 directions on 101 pages that had never reached the position register, among them 'Stage the corner before any cruciate reconstruction in a knee with lateral findings' and 'Correct varus alignment where it would load a corner reconstruction'. Both are clinical direction and both had been live for days. | Nothing in the tool. The verb list was extended, the register re-extracted and 7 tests added. With decision 019's consultation verbs the unsigned total rose from 976 to 1,295, which is a correction to the count rather than new exposure: the advice was always on the page and simply not on the list anybody is asked to sign. Found only because the consultant tier rewrite under decision 018 moved into weighing language the detector could not see. | 101 pages carried at least one unregistered direction |
| EV-18 POSITION_VERBS is a verb list, and it was extended twice in one day | Flagged, not fixed | POS-001 detects clinical direction from a list of imperative openers. Decision 018 extended it for weighing verbs and decision 019 extended it again for consultation verbs, both times because prose in a register the list had never seen went undetected. A list that needs extending whenever the writing changes register is keeping pace rather than leading, and the next new register will find the same gap. | No replacement is proposed, because the obvious alternatives are worse. A general imperative parser over-collects far more than the list does, and a model-based detector is not deterministic and cannot sit in a gate that has to give the same answer twice. Flagged so the next person does not assume it went unnoticed. | Affects every professional page, 190 of them |
| EV-19 Two QA rules could contradict each other, and chapter 3.12 made them | Fixed, and tested | Chapter 14 requires a unique SEO title per page, so render_site.py appends a colliding page's chapter name to its title tag. Chapter 13 bounds the title at 60 characters. The suffix had no length budget, so the fix for one rule could break the other, and it did as soon as a long chapter name collided: 'Effects of Malalignment, Tibiofemoral Contact Mechanics | KneeSchool' is 68 characters. Three existing pages in chapter 3.5 failed PUB-001 the moment chapter 3.12 was rendered, having passed for days. | Nothing. The suffix now has a budget read from site.json and trailing words are dropped from the chapter name until it fits, which keeps the distinguishing word and loses the generic one. A chapter name whose first word alone will not fit is left whole, so the finding is reported rather than the renderer shipping a mangled title. Seven tests, two of which check every published page's title for length and for uniqueness rather than checking the helper. | 6 pages carry a disambiguated title; every published page is now checked |
| EV-20 Measurements on seventeen pages were in neither gate and no register | Fixed, registered and tested | FIG-001 forbids figures where a brief says figures_allowed is false, and FIG-002 requires every figure to be registered where it says true. Seventeen briefs said neither, having been generated before the generator set the key. The condition evaluator read a missing key as false, so those pages silently acquired a rule written for Section 0's fees, and 1.1.3 sat published with eight style failures. The deeper consequence was the opposite rule: FIG-002 never applied, so the measurements on all seventeen were registered nowhere and a consultant checking the register would never have seen them. | Nothing. A conditional rule now applies only when the brief states the condition, the seventeen briefs carry the key, and 1.1.3's five flexion angles are registered as UNVERIFIED_FROM_MEMORY like every other measurement on the site. Found by running the style gate over the whole site for the first time, which is decision 021's main change. | 17 briefs completed, 5 measurements newly registered, total 352 to 357 |
| EV-21 The site is live, and it now publishes from main | Resolved 10 October 2026 | GitHub Pages deploys this repository and has done since late September. A CNAME in the root holds kneeschool.com. Until 20:26 on 10 October every one of 62 runs was on claude/great-meitner-kofr7g, the working branch, so each push went live immediately and unreviewed. That was the opposite of the README's arrangement and it was found by reading the Actions history rather than by anything in the build. | Nothing. The client changed the Pages source to main, and run 63 built 15a6ee5, the merge of pull request 2, in 47 seconds. The live site is now main's tip, so it is the copy that passed a green verify and went through a pull request. Decision 021's first rule is complete rather than aspirational. The safeguard held throughout: discoverable is false, so every page carried a noindex directive and robots.txt disallowed every crawler for the whole period. | 293 pages live at kneeschool.com, served from main, all noindexed |
| EV-22 Section 4's page map is provisional and its page count was wrong | Provisional map in place, replacement needed | Part 1 of the Master Publishing Architecture carries the full page map for Sections 0 to 3 and says so on its own first page. Section 4's map arrives in Part 2, which is not in this repository. What existed instead was outline_1_to_15.json, a bare title list whose own note says a brief generated from it would have no tiers and no must_not_cover list and would be worse than no brief. That note was right. Decision 024 built pages_4.json from Part 1's own tier sentence and authority boundaries rather than from invention, and the build status should say that the map is provisional until Part 2 lands. The page count was also wrong: the outline recorded chapter 4.2's five navigational sub section headings as pages and dropped the 34 four level leaves beneath them, so Section 4 is 227 pages rather than 198. The same extraction produced Sections 1 to 15, so any other section whose source uses four level ids is undercounted the same way and none has been checked. Chapter 4.2 has now been written at its correct size of 34 pages, which confirms the renumbering was right and leaves the eleven other sections unchecked. | Part 2 of the Master Publishing Architecture, for Section 4's real per page scope, tiers, deeper spiral links and priorities. Separately, a pass over Sections 5 to 15 in the source to find any other chapter organised with four level ids, because the whole site's page count is built from that outline. | 227 pages in Section 4, 8 published; 11 other sections unverified |
| EV-23 Section 4 was published and unreachable, and no gate noticed | Fixed, with two tests added | The Clinical Examination Academy's 65 pages passed every gate on this site and could not be reached by browsing. Each page had an inbound link, so an orphan check found nothing, but nothing outside the examination directory linked into it and the section was never added to the ribbon in tools/site_chrome.py. The whole section was an island with no path into it from anywhere a visitor starts, reachable only by typing a URL. Verified against main before the fix: no file outside examination/ contained a link to any page inside it. Six gates and 275 tests did not catch it because the property that matters is reachability from the homepage, not the existence of an inbound link. The client found it by asking to see the pages. Fixed by one line in the ribbon, and by EveryPageIsReachableByClicking, which walks the link graph from index.html and was checked by removing the ribbon line again to confirm it fails. | Nothing outstanding. Recorded because the class of bug is more interesting than the instance: a whole section can be built, published, indexed, gated and invisible, and the gates were all measuring something adjacent to the thing that mattered. | 65 pages were unreachable; all 384 published pages are now reachable from the homepage |
| EV-24 The build notice accumulated one copy per render | Fixed, with a test added | The work in progress notice added under decision 025 sits above the header and is returned by site_chrome.header(). apply_chrome's pattern matched the header alone, so each render prepended another copy rather than replacing the existing one. Three had accumulated on the homepage before it was noticed, and apply_chrome --check reported all 402 pages as drifted because the substitution was not idempotent. This is the third idempotency bug in this repository, after the figure register and the position register, which is why it has a test rather than only a fix. | Nothing outstanding. ChromeIsIdempotent asserts one notice per page and that applying the chrome again changes nothing. | 402 pages were affected and are now correct |
| EV-25 robots.txt named a sitemap that never existed | Fixed, with a test added | robots.txt has named sitemap.xml since the day it was written and the file did not exist. Harmless while discoverable was false, because every crawler was disallowed and none read robots.txt past the Disallow line. It became a 404 for every crawler the moment the client set discoverable true on 12 October 2026, and the site would have been indexed without the map it advertised. The same shape as the other latent bugs here: correct under one setting, wrong under the other, with nothing checking the setting it was wrong under. apply_chrome.py now writes sitemap.xml beside the robots.txt that points at it, so they cannot disagree, and removes it if the site ever returns to not discoverable. | Nothing outstanding. sitemap.xml carries 760 urls and test_the_sitemap_agrees_with_the_switch would have caught the absence. | 760 urls in the sitemap; every published page is listed |
| EV-26 The site is now indexable and nothing on it has been reviewed | Live and indexable, by the client's decision | The client set discoverable true on 12 October 2026, taking responsibility and asking for a disclaimer, which decision 026 records in their words. Search engines may now index 3,249 unsigned clinical recommendations and 526 unchecked measurements, and a reader can arrive from a search rather than from a link given by somebody who could warn them. Three notices stand between them and the content: the work in progress band on every page, the per page status line, and the reference note. None of them is review. Indexing is not fully reversible, because a page can stay cached after it is removed from the site. | The same thing as before, and now more urgently: a consultant through the review packs. 81 chapters are current and nothing has been signed. This entry exists so the cost of the decision is visible next to the work that would retire it. | Every page on the site is indexable and unreviewed |
| EV-27 The build status page quoted register totals that had fallen behind | Fixed, with a test added | This page tells a reader how much unchecked work the site carries, and until chapter 4.11 those numbers were typed into docs/build_status.json by whoever last edited it. They were right when written and wrong by the next chapter. At the point it was found the page said 448 unchecked measurements against a register holding 462, 2,139 unsigned recommendations against 2,149, that 32 chapters carried recommendations when 37 did, and that the largest was 3.7 with 152 when it was 4.2 with 191. The understatement was the problem: the page exists so the scale of unreviewed content is visible, and it was understating it. Same failure as the hand compiled figure list that decision 006 replaced, so it has the same answer. Every one of those numbers is now a placeholder filled from the registers when the page is built. | Nothing outstanding. TheBuildStatusPageAgreesWithTheRegisters fills every tokenised sentence from the live registers and looks for it whole on the page, so a register that moves without the page being rebuilt fails the suite. The first version of that test matched on the bare number and passed against a stale page, which is why it matches sentences. | about/status.html, generated from docs/build_status.json |
| EV-28 The renderer gave two pages the same title tag | Fixed, with a test added | PUB-009 requires a unique title tag per page, and the renderer gets one for a colliding page by appending the chapter name, trimmed to fit PUB-001's length bound. It trimmed from the back of the name, on the reasoning that the first word distinguishes. Chapter 4.15 disproved it. Patellofemoral Arthritis exists in chapter 3.12, Patellofemoral Biomechanics, and in chapter 4.15, Patellofemoral Examination, and both trimmed to "Patellofemoral Arthritis, Patellofemoral": two identical title tags for the two pages the disambiguation existed to separate. Trimming from the front instead breaks the opposite case, where "ACL Examination" loses the ACL that separates it from "Rotational Instability Examination". The renderer now ranks a chapter name's words by how many of the other colliding chapters contain them and drops the most shared word first, so the distinguishing word survives at either end. | Nothing outstanding. DisambiguatedTitlesSeparateThePages holds both cases and the length bound, the renderer now refuses to write two identical title tags rather than leaving it to the gate, and PUB-009 remains the backstop that found it. | One existing title changed: biomechanics/patellofemoral-arthritis.html now carries its chapter name |
| EV-29 The renderer's --only flag could publish a non-unique title | Fixed, with a test added | The same finding had a second cause. The renderer computed its collision set after --only had narrowed the targets, so a page rendered on its own could not see a page it collided with and was written without the chapter suffix. Chapter 4.15 was rendered with --only and published its title tag as though nothing shared it. This is the shape of several bugs in this repository: correct when the whole site is rendered, wrong on the narrower path, and nothing checking the narrow path. The collision set is now computed from every render target and only the rendering is narrowed. | Nothing outstanding. The gate caught it, and the renderer no longer depends on the whole site being rendered to get a title right. | tools/render_site.py |
| EV-30 The build status page quoted a stale sitemap count | Fixed, with a test added | This page quotes the number of urls in sitemap.xml, and the two tools disagreed by construction: tools/apply_chrome.py writes the sitemap and tools/build_status.py read the written file to get the count. Running build_status before apply_chrome, which is the order a page batch naturally produces, made this page quote the previous count. Chapter 4.15 failed the suite on it. The path list now lives in tools/site_chrome.py and both tools derive it, so the order they run in does not matter. Same answer as EV-27 one step further back: the number is computed rather than read from whatever happened to be on disk. | Nothing outstanding. TheSitemapCountAgreesWithTheSitemap holds the derived list against the file, and TheBuildStatusPageAgreesWithTheRegisters holds the page against the derived list. | sitemap.xml and about/status.html |
| EV-31 The style gate banned a page title the architecture had chosen | Fixed, with three tests added | GOV-001 bans a list of commercial terms, Sportshealing among them, on every page whose brief sets commercial_content_allowed to false. tools/architecture.py sets that flag to false for every page it generates a brief for. The architecture names page 4.22.7 Sportshealing Knee Score. So the gate banned that page's own title on its own page, and banned the correct cross link label on every page pointing at it, which is how it was found: chapter 4.18.6's link to 4.22.7 failed the lint and the link was replaced with a worse one rather than the gate being questioned. PHRASE-001 already carried the mechanism for exactly this class of problem, added when 0.5.5 Meet the Researchers tripped the meet the ban in its own H1. GOV-001 now sets the same excludes_architecture_title flag, so the H1 is out of scope when it matches the brief's title and the cross link labels are out of scope entirely. The boolean stays false, so buy, purchase, discount and our clinic remain banned everywhere, including in the body of 4.22.7 itself. | Nothing outstanding. Three tests in pipeline/tests/test_linter.py hold the behaviour: the architecture's own title and the cross link label pass, the term in prose on the same page still fails, and a title the brief does not name is not exempted. | pipeline/config/lint_rules.json and every page linking to 4.22.7 |
| EV-32 A cross link to a page that does not exist passed every gate | Fixed, with four tests added | tools/crossrefs.py --check exempted every unknown page id as a forward link to a section the architecture had not mapped to pages yet. That was true when the tool was written and only sections 0 to 3 were mapped. All sixteen sections now carry page ids, so the exemption protected nothing and hid a real typo: chapter 4.19.4 linked 4.15.11 as The Apprehension Test when chapter 4.15 ends at 4.15.10. The renderer counted it as one unresolved link of six and printed that count, the label check skipped it because it could not resolve it, and the prose check does not look at bracketed links, so the page published with a dead link and make verify passed. An unknown id with a valid shape is now a fault. A section placeholder stays exempt, because two future pages in one unpublished section share an id by design. | Nothing outstanding. Four tests in pipeline/tests/test_architecture.py hold it, including one that runs the check against the whole site. | examination/paediatric-patellar-instability.html, now corrected |
| EV-33 The attribution register could not see a name with a capital in the middle | Fixed, with six tests added and six attributions recovered | The surname shape the attribution detector used was [A-Z][a-z]{2,}, which cannot match McMurray, MacIntosh, Barber-Westin or O'Connor. Chapter 4.24.3 attributed the meniscal provocation test to McMurray in the first sentence of its What Was Done section and the register recorded nothing. The same shape had been losing names from chapter 3.17 since that chapter was published: re-extracting after the fix recovered LaPrade from 3.17.4 and 3.17.5, Terry and LaPrade, Griffith and LaPrade, and Goodfellow and O'Connor from 3.17.8, where the register had been holding the mangled entry Connor. A false positive in this register is visible and the next extraction withdraws it. A miss is invisible, and the register's whole value under decision 022 is that it comes from the pages rather than from someone remembering to add a name, so a silent miss is the worse failure. The shape now allows an internal capital, a hyphenated second part and an apostrophe. The same extraction run also read three capitalised common nouns as authors, Patient and Translation and Symmetry, each opening a sentence followed by a reporting verb; those words joined the stop list that already exists for that over-read. | Nothing outstanding. Six tests in pipeline/tests/test_linter.py hold the name shapes and the stop list. The six recovered attributions are now in the review packs for chapters 3.17 and 4.24, which means they are six more claims waiting on a consultant rather than six fewer. | pipeline/config/attributions, chapters 3.17 and 4.24 |
| EV-34 A banned phrase that straddled a line break passed the style gate | Fixed, with four tests added | PHRASE-001 built its pattern by escaping the banned phrase whole, so the space between words matched a literal space. Drafts are hard wrapped at about 79 columns, which means any multi word phrase that happened to break across lines was invisible to the gate. 82 of the 108 banned phrases are multi word. Page 4.22.7 carried "it is worth noting" with the wrap falling after "it": lint_local passed the markdown and tools/check_site.py caught it on the rendered page, where the break had become a space. The two gates read the same text through different whitespace and disagreed, which is how it surfaced. A space in a banned phrase now matches any run of whitespace. A scan of all 554 drafts for phrases hidden this way found exactly one, so the hole was real and its blast radius was a single page. | Nothing outstanding. Four tests in pipeline/tests/test_linter.py hold it: the phrase on one line, across a line break, across several spaces, and an ordinary pair of sentences that must stay quiet. | examination/sportshealing-knee-score.html, now corrected |
| EV-35 Flattening Section 5's chapters repointed links from published pages | Fixed, and the gate that caught it was added yesterday | Decision 027 renumbered chapters 5.2 to 5.5 flat, because the outline had recorded their navigational headings as pages and dropped the 78 real pages beneath them. That reassigned ids which two already published Section 2 pages were linking to. 2.1.1 and 2.1.8 carried [[5.3.2 | Distal femoral imaging]] and 2.1.8 also carried [[5.5.2 | Normal MRI Anatomy]]; after flattening, 5.3.2 is Lateral View and 5.5.2 is Sports Knee Protocol. The links now point at pages that carry the material: 5.3.1 AP View and 5.5.6 Bones. Worth recording for two reasons. First, renumbering a chapter is not a local change, and the same hazard will apply to any later section whose outline mixes navigational groupings with leaves. Second, this was caught only because EV-32 had been fixed the previous day: before that fix an unknown page id was skipped as a forward link, and these ids were unknown precisely because navigational headings are not pages. The gate earned its keep inside a day. | Nothing outstanding. tools/crossrefs.py --check is a gate and now fails on any link to a page the architecture does not have, and tools/extract_section5.py translates the spiral map's own navigational references so the osteoarthritis chain lands on the osteoarthritis pages rather than on Tunnel View. | anatomy/femur-gross-anatomy.html and anatomy/femur-imaging-anatomy.html |
Pages that describe a feature the site cannot yet do
Each of these pages is written and has passed the style gate. Where a page is published it says on itself that the feature is not available. Where the feature was the page's whole subject, the page is held and the syllabus shows it as coming soon.
| Item | State | What is in the way | What would move it | Pages |
|---|---|---|---|---|
| PG-1 Ask a Surgeon has no questions and no submission route | Page published, feature absent | No submission form, no moderation workflow, no answered questions. | A moderated submission route, and a consultant willing to answer. | 0.3.6 |
| PG-2 The guided read of a real paper is not attached | Page published, feature absent | The architecture asks for a guided first read of a named paper. No paper could be retrieved or checked. See EV-1. | Source access, then a chosen open access paper. | 0.5.3 |
| PG-3 Researcher profiles are not written | Page published, feature absent | A profile of a named person needs that person's consent and their own check of the text. Neither has been sought. | Named researchers willing to be profiled. | 0.5.5 |
| PG-4 The Future Surgeon Challenge is not playable, and its page is held | Written and held unpublished | The quiz ladder needs the adaptive question bank, which belongs to Section 14 of the publishing plan. Section 14 is not written. | The Section 14 adaptive question bank, or a decision to build a smaller one for the junior rungs. Held unpublished at the client's instruction. The draft and its handoffs are in pipeline/runs, render_targets.json records why, and the Junior Academy syllabus shows the page as coming soon. Publishing it is one entry moved back into render_targets.json plus tools/render_site.py. Chapter 0.6 is closed in the site's navigation under decision 004, so the page is not shown as coming soon either; it is simply not there. | 0.6.2 written and held |
| PG-5 Leaderboards and badges need accounts and a lawful basis | Both pages written and held unpublished | Both need progress tracking, which needs accounts. Because the users are children, that needs a lawful basis under UK GDPR, a completed data protection impact assessment, compliance with the ICO Age Appropriate Design Code, a named safeguarding lead and a pre moderation process. The site currently has no accounts and no server side anything. | A decision on whether to take this on at all. It is the largest single piece of compliance work in the plan. Held unpublished at the client's instruction. The draft and its handoffs are in pipeline/runs, render_targets.json records why, and the Junior Academy syllabus shows the page as coming soon. Publishing it is one entry moved back into render_targets.json plus tools/render_site.py. Chapter 0.6 is closed in the site's navigation under decision 004, so the page is not shown as coming soon either; it is simply not there. | 0.6.3 and 0.6.4 written and held |
| PG-6 No classroom files exist to download | Page published, feature absent | No slide decks, worksheets, diagram sheets or printable certificate have been produced. The lesson plans are therefore described on the page rather than attached. | A decision on format, then the files. Everything they would carry is already written on the pages. | 0.7.1, 0.7.2; also the printable certificate on 0.6.1 |
| PG-7 The outreach programme cannot be booked | Page published, feature absent | There is no booking route and no contact form anywhere on the site. | A contact route. This also unblocks the three dead links in the About menu. | 0.7.5 |
| PG-8 Three About menu links go nowhere | Placeholder | Authors and reviewers, References and sources, and Contact all point at '#'. Authors needs real names, references needs EV-1, contact needs PG-7. | Names of the editorial team, and a contact address. |
Decisions taken
Each of these was a gap in the handbook or the architecture, or a call the client made during the build. 001 to 004 and 006 to 007 are confirmed. 005 is partly reversed and partly withdrawn.
| Item | State | What is in the way | What would move it | Pages |
|---|---|---|---|---|
| D-001 Where the Section 1 depth boundary sits | Confirmed | The architecture excludes region level anatomy detail from Fundamentals pages without saying where the line falls. | Nothing. Adopted when the tier split ruling was chosen, and reconfirmed on 8 October 2026. | |
| D-002 Word counts for a page that carries one tier | Confirmed | The handbook's word ranges assume a page carrying several tiers. Section 0 pages carry one. The amendment adds room for the patient tier's FAQ block. | Nothing. The bands and the FAQ allowance were confirmed on 8 October 2026. Decision 005's attempt to add a third band was wrong on all six pages and has been withdrawn. | |
| D-003 Templates for Section 0 chapters 0.3 to 0.7 | Confirmed | The handbook defines three article templates and 27 Section 0 pages have a page type matching none of them. Four templates were defined locally. | Nothing. The four section orders were revised and confirmed on 7 October 2026 and the no figures rule is closed. Seventeen pages lost a heading in the revision and no prose was lost with it. | |
| D-004 Chapter 0.6 closed on the site | Confirmed and applied | Three of the chapter's four pages describe features that do not exist. A chapter heading with one live page and three greyed out rows advertises what the site cannot do, on the page where a new reader is choosing what to read. | Nothing. Reopening it needs the features in PG-4 and PG-5, then four steps listed in docs/decisions/004. | 0.6.1 moved into chapter 0.1; 0.6.2 to 0.6.4 held |
| D-005 Section 2 depth and tier pattern | Part 1 reversed by D-006; part 2 failed and was withdrawn; part 3 applied | Anatomy at MRCS and FRCS depth normally carries measurements, and none could be verified. A plausible wrong dimension on an FRCS page is worse than no dimension. The handbook's anatomy word band also does not fit a page about a single bony landmark. | Confirmation of the tier pattern, which is the part still standing. The no figures rule was reversed by the client. The narrow word band was wrong on all six pages it was applied to and has been withdrawn entirely. | All 79 Section 2 pages |
| D-006 Keep the measurements, and register every one | Applied | The client's instruction was to keep measurements provided they are based on current papers. That condition cannot be met here: there is no source access, so every figure is written from standard teaching and none is attributable to a paper. | Somebody with sources to work through the register. 12 figures are outstanding, all marked UNVERIFIED_FROM_MEMORY. The gate refuses an unregistered figure, so the list cannot fall behind the pages. | 6 of the 15 written Section 2 pages carry figures |
| D-007 A clinical recommendation needs a name against it | Applied | The review pack flagged one sentence that recommended when to operate rather than describing anatomy. The client said keep it. Acting on that exposed 76 such sentences across 27 pages, including Section 1 and the anterior cruciate page. | A consultant to work the register: 76 positions, all UNSIGNED. tools/positions.py --report --unsigned is the list, and POS-001 warns when a page carries one that is not on it. | 27 pages carrying a tier from medical student upwards |
| D-008 The architecture numbering is canonical | Applied | Two documents numbered the same academies differently. In the Master Operations Handbook the Anatomy Academy is chapter 6 and the Conditions Library is chapter 9. In the publishing architecture they are Sections 2 and 6, and the repository only ever used the architecture. | Nothing. Ruled on 9 October 2026 in favour of the architecture, so there is no migration: Section 2 means Anatomy and Section 6 means Conditions, which is what every config, brief, cross link and file path already assumed. A chapter number inside the handbook now means that document's own chapter, not a page id. | None; the ruling confirms the existing state |
| D-009 The authority order between the handbooks | Applied | Three documents set editorial rules and two of them disagree about FAQs and about how many tiers the spiral has. | Nothing. The Operations Handbook of 22 September 2026 is senior, the Master Operations Handbook governs what it does not cover, and both playbooks are historical. FAQs stay on the patient tier only; the spiral keeps seven tiers. Roughly 60 pages avoided rework. | None; the ruling confirms the existing state |
| D-010 Publication QA is built as the fifth gate | Applied | The handbook names five QA tiers and the pipeline had four. Nothing checked the metadata a search engine or a screen reader reads. | Nothing. Nine rules in tools/publication_qa.py: SEO title, meta description, slug and canonical, heading ladder, image alt text, the disclaimer, internal linking, metadata and site wide uniqueness. It also fixed a real defect: the renderer had been truncating descriptions to 160 characters mid word. | All 116 pages, 0 findings |
| D-011 The palette stays green, ivory and gold | Declined; palette unchanged | Chapter 6 of the handbook asks for a separate colour for learning highlights, muted teal. The stylesheet had none and was using the gold meant for emphasis. | Nothing. Ruled on 9 October 2026: three colours, no fourth. The muted teal is declined and the learning highlight keeps the gold. A token added earlier that day to carry a partner brand colour has been removed, because the brand reference was withdrawn and a token that always equals the gold is a hop with nothing at the end of it. See decision 011. | None; the palette is unchanged |
| D-012 The design system is MyKneeScore's | Applied | The site had its own palette and its own typography, built before the sister site's design system was available to read. | Nothing. Ported from SportsHealing/mykneescore at commit 4ce92e0 on 9 October 2026: its twelve tokens at their exact values, Cormorant Garamond and Hanken Grotesk, pill controls, card shapes and deep green chrome. One of its house rules came too, that no colour literal appears outside the stylesheet's :root block, and the site check now enforces it. One colour differs: small gold labels on a light background use a darker gold, because theirs is below the contrast minimum for text that size and this site carries those labels on pages for school age readers. See decision 012. | All 122 pages |
| D-013 The peroneal nerve page carries a fellowship tier | Applied | The architecture gives page 2.10.6 three tiers, and the client asked for a section on the nerve at risk from lateral surgery. That is approach anatomy, which is what the fellowship tier carries everywhere else in Section 2. | Nothing. A tier override mechanism was added, mirroring the word count overrides from decision 005: declared, additive only, each entry carrying a reason and a decision behind it. The page also has a declared 800 to 2400 band, because its fellowship tier is a single section of ten paragraphs. It came in at 2,227 words. See decision 013. | Page 2.10.6, Common Peroneal Nerve |
| D-014 The biomechanics and landmark papers templates | Applied, open to revision | Section 3's 188 pages had no template. Both its page types borrowed the anatomy one, which gave a page called What Is Biomechanics? a heading for Blood Supply and Innervation. | Client confirmation of the section order, as decision 003 received for Section 0. The biomechanics template is The Principle, At the Knee, What Changes It, How It Is Measured, Clinical Relevance. Two open questions in decision 014: whether biomechanics pages should be exempt from the junior closing message, and whether the landmark papers pages are worth attempting from memory. | Chapter 3.1 built on it, 7 pages published |
| D-015 Published, not publicised | Applied | The site can go live and will not be promoted until it is verified. Nothing on the site distinguished those two states, so a crawler finding it would have publicised it, and indexing is not cleanly reversible because pages stay cached after removal. | Nothing now. One value in site.json drives a noindex directive on all 262 pages and a robots.txt that disallows every crawler. Flipping it is one line and one command when verification is done. PUB-010 refuses the half flip, where the config says one thing and the pages say another. See decision 015. | All 262 pages carry the directive |
| D-016 The review pack is the return path | Applied | Eight hand written review packs covered batches that made sense on the day, went stale silently, and could only be answered in prose. A reviewer's decision had nowhere to go. | Nothing. tools/review_pack.py generates one pack per chapter from the chapter's own files and ingests a filled pack back into the figure and position registers. A blank row stays outstanding, one bad row writes nothing at all, and an amendment carries the reviewer's wording into the register. Twelve tests hold both directions, including that a signature survives the next extraction. See decision 016. | 25 generated packs in docs/review/packs/ |
| D-017 The review pack in three formats | Applied | A consultant filling eighty five cells wants a column to tab down, not a markdown table, and a printed pack for a meeting wants to be a document. | Nothing. Every pack generates as markdown, a spreadsheet and a Word file from the same data, all carrying the same row references, and --ingest reads any of them. Two faults were found while verifying: python-docx writes a schema-invalid settings.xml, now corrected, and LibreOffice does not work in this container, so the workbook sets fullCalcOnLoad and the Word file is verified structurally rather than rendered. See decision 017. | docs/review/packs/, markdown for all 81 chapters and both binaries for 3.2 to 3.10 |
| D-018 What the consultant tier is for | Applied | 3.11.9 set out two positions on lateral augmentation and called both defensible. The client's correction on 10 October 2026 is that the consultant tier is about how a decision gets made when the evidence does not make it, not about arbitrating between camps. The rule is now in article_template.json under tier_intent.consultant: name the decision, set out both sides, say what moves the weight, say what the measurements cannot settle, and leave the reader able to say why a colleague decided otherwise. | Nothing. All three consultant blocks on the site were rewritten: 3.11.9, 3.1.7 and 6.1.1. Whether the fellowship tier needs the same treatment is left open deliberately. | 3 pages rewritten, 1 template changed, 101 position registers re-extracted |
| D-019 The five patient factors | Applied | Client instruction, 10 October 2026, given straight after decision 018: decision-making should always take into account the patient's age and physiology, medical comorbidities, demands on their knee, their psychological state, and their own ideas, concerns and expectations. Decision 018 said the consultant tier is a balance argument. This says what goes on the scale. | Nothing outstanding. The five factors and their keywords are in article_template.json, the new gate DEC-001 reads them from there, and 3.11.9 and 6.1.1 now carry all five. 3.1.7 is correctly out of scope because its consultant block weighs a teaching decision with no patient in it. Open: whether the five factors should reach the FRCS and fellowship blocks, which also frame choices. | 2 pages extended, 1 new lint rule, 10 tests |
| D-020 Controversies skip the fellowship tier | Applied | 6.1.1 carried a Controversies and Evidence section at FRCS, fellowship and consultant, and the three repeated each other rather than pitching the same material three ways. The client cut fellowship on 10 October 2026. FRCS appraises the evidence, which is what the examination asks for, and the consultant block weighs the decision under decision 018. Fellowship is advanced practice, so its controversies belong inside its management and complications sections where the operation is. | Nothing. Section tier limits moved from a hardcoded dict in linter.py into article_template.json under section_tier_rules, which takes an explicit allow list as well as a floor, because a floor cannot express 'FRCS and consultant but not fellowship'. STRUCT-001 reads it, so the limit is checked rather than described: the condition template said frcs_and_above_only in prose and nothing read that string. The cut paragraph's one substantive detail, the name of the operation, moved into the fellowship complications section rather than going. | 1 page edited, 1 template changed, 1 linter dict demoted to a fallback, 5 tests. Applies to the 100 or so Section 6 condition pages not yet written. |
| D-021 One verify command, and how main moves | Applied | The gates were five separate commands held in one person's head, so a contributor could not run them, there was no CI to run them, and pull request #1 was merged carrying 26 commits and 276 pages with nothing checking it. The style gate had never run over the whole site: it ran per chapter at the time each chapter was written. | Nothing outstanding. make verify is the test suite plus six gates, the CI workflow runs that command rather than restating it, and main now moves only through a pull request that is green. Branch protection would enforce that rather than state it, and is left to the client because it needs admin rights and locks out a direct push. | 2 new tools, 1 new flag, 1 workflow, 17 briefs completed, 12 tests |
| D-022 How a landmark papers page is written without source access | Applied | A landmark papers page's whole subject is the literature, and no source could be retrieved. A year, a journal, a volume or a citation written from memory is the most fabricable detail on the site and the hardest for a reader to check. | Nothing outstanding. ATTR-001 refuses a year, a journal, a volume, a DOI, a PMID and an et al form on these pages, ATTR-002 holds every named attribution against a generated register, and the pages carry surnames and described findings with decades at most. | chapters 3.17 and 4.24, 62 attributions across 15 pages |
| D-023 The examination template, and the accuracy figures it refuses | Applied | Section 4 is 227 pages and the type map pointed examination at the anatomy template, whose headings are Structure and Location, Relations, Blood Supply and Innervation. A page called The Lachman Test has no blood supply. | Nothing outstanding. The examination template has five sections and ACC-001 refuses a sensitivity, specificity, predictive value or likelihood ratio written as a number. The template is unvalidated: no editorial authority has confirmed its five sections. | 227 pages, 1 template, 1 new rule, 5 new headings |
| D-024 How Section 4 is built before Part 2 of the architecture arrives | Applied | Part 1 governs tier coverage and page scope and gives Section 4 one sentence. Part 2, which carries the real page map, is not in this repository. | Part 2. Until it arrives the tier sets, the boundary lists and the scope sentences for 227 pages come from one sentence in Part 1. The script that generated the map lived in a scratchpad and was lost when the container was reclaimed, so Section 4's map can no longer be re-derived or checked against the source. Decision 027 does not repeat that mistake. | 227 pages, pages_4.json |
| D-025 Every produced page is visible, and the site says it is unfinished | Applied | Pages were being produced faster than they were being linked, so work existed in the repository and could not be reached from the site. A reader had no way to know which pages were reviewed and which were not. | Nothing outstanding. Every produced page is reachable, every page carries a status line, and the build notice says the site is being built in the open and that no page has been reviewed. | the whole site |
| D-026 The site is discoverable, and who decided that | Applied, on the client's instruction | The site was not indexable. Making an unreviewed site discoverable is the client's call rather than this process's, and it was taken in their name on their instruction. | Nothing outstanding. robots.txt allows all, sitemap.xml is generated from the same function that writes the pages, and no page carries a noindex. The disclaimer and the status page carry the honesty burden. | robots.txt, sitemap.xml, 574 pages |
| D-027 Section 5's page map, built from the source until Part 2 exists | Applied | The same gap as decision 024, one section later, and worse: Part 1 gives Section 5 one sentence and states no tier set for it at all, where Section 4 at least had the words Medical Student to Fellowship. The outline file recorded 161 pages because it took four chapters' navigational headings as pages and dropped the 78 real pages beneath them. | Part 2. Until then the four professional tiers are carried across from the section that defers its imaging here, which is a weaker derivation than Section 4's and is labelled as such on the Imaging Academy index. tools/extract_section5.py is committed so the 224 page map can be rebuilt and checked against the source, which Section 4's cannot. | 224 pages designed, pages_5.json, 1 committed tool, 6 tests |
| D-028 The imaging template, and the accuracy hole it found | Applied | Section 5 had no page type at all, so a brief could not be generated. Widening ACC-001 to cover imaging then exposed a hole in it: the term list held the nouns and not the adjectives, so a claim that a modality is 95 per cent sensitive had passed the gate, and would have passed on all 227 Section 4 pages. | Nothing outstanding on the rule. The template itself is unvalidated, as the examination template is, and chapter 5.20 Imaging Reporting is the loosest fit for its six sections and is the first thing to revisit. Chapter 5.19, the Magic Angle MRI Academy, was held on the client's instruction of 15 October 2026 and published on their instruction of 19 October 2026. Its ten pages were written for that second instruction, so the chapter went from unwritten to published in one step and has had no separate drafting review. Its eighth page, Imperial Magic Angle Programme, is published as a declared gap: its subject is a named research programme and this environment has no source access, so the page names no investigator, date or result and says so in its first paragraph. | 1 template, 4 new headings, ACC-001 widened, 9 tests |
Written and awaiting review
Finished work with nothing blocking it except review.
| Item | State | What is in the way | What would move it | Pages |
|---|---|---|---|---|
| W-1 Section 0 Junior Academy, 35 published of 38 written | 35 published and awaiting review, 3 held unpublished | None. Review packs are in docs/review/. | Consultant review of the 35 published pages. 0.2.4 and 0.2.5 are singled out in the chapter 0.1 and 0.2 pack; 0.7.4, 0.7.3 and 0.3.5 in the chapter 0.3 to 0.7 pack. The three held pages, 0.6.2 to 0.6.4, need no review until a decision is taken to publish them, and chapter 0.6 is closed on the site under decision 004. | |
| W-2 Section 1 Knee Fundamentals, all 16 pages | Written, style gate passed, awaiting review | None. The review pack is docs/review/section-1-batch.md. | Consultant review. | |
| W-3 Anterior cruciate ligament injury, the condition pilot | With a consultant for review | None. | A ruling on the pilot, which was left for the consultant to decide. | |
| W-4 Section 2 Anatomy Academy, complete | All 79 pages written, style gate passed, awaiting review | None on the written pages. The remaining 31 are a matter of time. | Consultant review of the whole section. It is the largest body of professional content on the site and none of it has been verified or reviewed. The fellowship sections on 2.10.6 and 2.12.6 are approach anatomy written without a source and should be read first. | 79 published pages, 111,493 body words as the style gate counts them |
| W-5 28 wrong forward link labels corrected across 36 pages | Fixed and pushed | None. | Nothing. The ids came from Part 1's deeper_pages lists and were right; the labels were invented because the target sections had no architecture. Three were worse: the anterior cruciate page linked to PCL, MCL and posterolateral corner injury using page ids inside the ACL chapter. crossrefs now checks every forward link against the outline, including chapter level targets. | 36 pages across Sections 0, 1, 2 and 6 |
| W-6 Section 3 Biomechanics Academy, chapter 3.1 | All 8 written, style gate passed, awaiting review | The first chapter of the largest section. It carries all five tier shapes Section 3 uses, from junior to consultant, which is why it was built first: whatever template serves it serves the section. Page 3.1.8 was written last, on 11 October 2026, and completes the chapter and the section. Its subject is the classic studies in general, and the architecture's must_not_cover list gives the per topic detail to chapter 3.17, so the page teaches how to read a landmark study and names no study at all. It sets out the five designs almost everything in Section 3 came from, what each one can and cannot support, and the three traps that recur across every subject in the section. It is the only landmark papers page on the site with no attribution register, which is the honest result of that boundary rather than a gap in it. | Consultant review. The claim worth arguing with is the page's own framing: that a study is landmark because it changed practice rather than because it was cited often, and that teaching the five designs serves a trainee better than teaching ten named studies. Both are editorial judgements presented as orientation. | 8 published pages, 12,382 body words as the style gate counts them |
| W-7 Section 3 Biomechanics Academy, chapter 3.2 | All 10 pages written, style gate passed, awaiting review | Functional Anatomy and Motion. Ten pages on the six degrees of freedom, the six motions taken one at a time, how they couple, and stability through the arc. Three tiers on every page, medical student to FRCS, and no junior or patient tier: the architecture does not ask for one here. The chapter orients rather than apportions, because per ligament restraint belongs to chapters 3.7 to 3.11 and surface motion to 3.3. | Consultant review. Nothing in the chapter is held: the chapter's ten pages are its whole plan. | 10 published pages, 12,889 body words as the style gate counts them |
| W-8 Section 3 Biomechanics Academy, chapter 3.3 | All 14 pages written, style gate passed, awaiting review | Knee Kinematics, the largest chapter in Section 3 so far. Eight pages on the mechanics: how motion is described, roll and glide, the two chains, the screw home mechanism and its reverse, the instant centre of rotation, and how all of it changes through flexion. Six more on the activities: walking, running, squatting, stairs, landing and cutting. Three tiers on every page, medical student to FRCS. | Consultant review. Kinetics is handed to chapter 3.4, gait pathology to 3.14 and 4.4, and sport specific application to 3.15, all by the architecture's own must_not_cover list rather than by choice. | 14 published pages, 17,774 body words as the style gate counts them |
| W-9 Section 3 Biomechanics Academy, chapter 3.4 | All 10 pages written, style gate passed, awaiting review | Knee Kinetics, the other half of chapter 3.3. Ten pages on forces and moments: what kinetics is, the ground reaction force, joint reaction forces, lever arms, moments and their external and internal halves, energy absorption, force transmission and shock absorption. Three tiers on every page, medical student to FRCS. | Consultant review. The chapter carries more measurements than any other in Section 3, because a kinetic claim without a magnitude says nothing, and every one of them is written from standard teaching. Contact area and stress are handed to 3.5. | 10 published pages, 12,196 body words as the style gate counts them |
| W-10 Section 3 Biomechanics Academy, chapter 3.5 | All 10 pages written, style gate passed, awaiting review | Tibiofemoral Contact Mechanics, where chapters 3.3 and 3.4 arrive. Four pages on the normal state: contact areas, contact pressures, load distribution and the effect of flexion. Three on what changes it: meniscal loss, malalignment and body mass. Three on the knee after an event: cruciate injury, meniscectomy and osteotomy. Three tiers on every page. | Consultant review, and this chapter needs it most. Every page turns on the distinction between force and stress, and almost every figure in it is a cadaveric contact measurement quoted from memory. If the area and pressure numbers are wrong, the mechanical argument of three chapters is wrong with them. | 10 published pages, 11,859 body words as the style gate counts them |
| W-11 Section 3 Biomechanics Academy, chapter 3.6 | All 14 pages written, style gate passed, awaiting review | Meniscal Biomechanics, where four chapters arrive. Eight pages on the structure and its functions: the five roles, hoop stress, the two fibre systems, load sharing, damping, lubrication and proprioception. Three on the roots and what fails: root biomechanics, extrusion, root tears. Three on the rest: radial tears, repair constructs and transplantation. The last two carry a fellowship tier, which the architecture asks for and no other page in Section 3 has. | Consultant review by someone who repairs menisci. Every mechanical statement in the chapter rests on hoop tension, and the chapter says so on its first substantive page. If that framing is overstated anywhere, fourteen pages are wrong together. Tear classification is handed to chapters 6.28 to 6.43 and MRI appearances to 5.6. | 14 published pages, 16,295 body words as the style gate counts them |
| W-12 Section 3 Biomechanics Academy, chapter 3.7 | All 20 pages written, style gate passed, awaiting review | ACL Biomechanics, the largest chapter in Section 3 and the most referenced from elsewhere in it. Eight pages on what the ligament does, two on how the knee behaves without it, four on what loads it, one on injury mechanism, one on sex differences, and five on grafts and the reconstructed construct. The last five carry a fellowship tier. | Consultant review by a knee surgeon who reconstructs. The chapter carries the graft property figures, the strain figures and the apportionment percentages, all from memory, and they are the numbers trainees will quote. 3.7.15 on sex differences needs reading for tone as well as accuracy: it states a rate difference, separates group level findings from individual prediction, and says the modifiable factors are the trainable ones. | 20 published pages, 23,062 body words as the style gate counts them |
| W-13 Section 3 Biomechanics Academy, chapter 3.8 | All 10 pages written, style gate passed, awaiting review | PCL Biomechanics. Six pages on function and behaviour, one on gait, three on reconstruction, the last three with a fellowship tier. The chapter is shaped differently from the anterior cruciate's because the ligament is: stronger, injured less often, partly healing, and reconstructed less successfully, and each of those four facts is mechanical. | Consultant review. Two things in it need checking by someone who does these operations: the bundle arrangement is the reverse of the anterior cruciate's and the chapter says so repeatedly, and the step off figures and grading thresholds are quoted from memory. The tensioning angle and the instruction to reduce the sag before fixation are the practical claims most worth confirming. | 10 published pages, 11,423 body words as the style gate counts them |
| W-14 Section 3 Biomechanics Academy, chapter 3.9 | All 10 pages written, style gate passed, awaiting review | Medial Knee Biomechanics. Two pages on what the side does, four on the individual structures, one on the corner as a unit, one on the rotatory instability pattern, one on the reconstruction and one on the combined injury. 3.9.8 carries a fellowship tier. | Consultant review. The chapter's organising claim is that the medial side divides its work by flexion angle, with the superficial ligament primary at 20 to 30 degrees and the posteromedial corner taking over near extension. Two practical claims most worth confirming: the two tibial attachments of the superficial ligament behaving differently, and the tensioning angles given for the two limbs of a medial reconstruction. | 10 published pages, 10,966 body words as the style gate counts them |
| W-15 Section 3 Biomechanics Academy, chapter 3.10 | All 9 pages written, style gate passed, awaiting review | Lateral Knee Biomechanics, the mirror of 3.9 and not a symmetrical one. Two pages on what the side does, three on the structures, one on the corner as a unit, one on the mechanics behind the dial test, one on reconstruction and one on the combined injury. 3.10.8 carries a fellowship tier. | Consultant review by someone who reconstructs corners. The three component reconstruction, the tensioning position, and the claim that a missed corner injury is a leading cause of graft failure on both cruciates are the statements that carry the most weight. The dial test page states the mechanics and hands the technique to 4.13, which is worth checking for the boundary as well as the content. | 9 published pages, 9,865 body words as the style gate counts them |
| W-16 Section 3 Biomechanics Academy, chapter 3.11 | All 9 pages written, style gate passed, awaiting review | Anterolateral Complex Biomechanics, the chapter where the anatomy is still argued over. Three pages on the structures, two on what they control, one on the Segond lesion as evidence of an attachment, one on the tenodesis, one on the combined reconstruction and one on the controversies. Four pages carry a fellowship tier and 3.11.9 is the first page in Section 3 to carry a consultant tier. | Consultant review by an ACL surgeon who uses lateral augmentation. The pages that carry the most weight are 3.11.3, which separates the naming question from the functional one, 3.11.7, where the clinical evidence is strongest and the anatomical rationale weakest, and 3.11.9, which sets out two defensible consultant positions rather than one answer. The indication threshold for augmentation is the single statement most likely to be contested. | 9 published pages, 10,546 body words as the style gate counts them |
| W-17 Section 3 Biomechanics Academy, chapter 3.12 | All 12 pages written, style gate passed, awaiting review | Patellofemoral Biomechanics. The organising idea is that the patella is a sesamoid held in a groove by three things, bone shape, soft tissue and the line of pull, and almost every clinical problem here is one of the three failing. 3.12.6 exists to say why the Q angle, the historical attempt to reduce all three to one number, predicts poorly. Chapter 3.5 is the tibiofemoral equivalent and already owns Contact Areas, Contact Pressures and Effects of Malalignment, so this chapter's three namesakes each say which joint they are about in the first line and link the other. Twelve pages, three tiers each, no fellowship or consultant tier. | Consultant review by a surgeon who does patellofemoral work. The statements carrying most weight are 3.12.9's, that patellofemoral pain has mechanical contributors and is not a mechanical diagnosis, and 3.12.11's, that femoral version is under measured and no soft tissue or tubercle procedure addresses it. 3.12.8's four factor account and the claim that two mild factors are better addressed than one severe one is the other statement likely to be contested. | 12 published pages, 16,455 body words as the style gate counts them |
| W-18 Section 3 Biomechanics Academy, chapter 3.13 | All 10 pages written, style gate passed, awaiting review | Alignment and Biomechanics. The chapter is about one number nobody can measure in the state that matters: the static angle is taken standing still, the load it governs is applied during movement, and corrections are planned to a target no trial has tested. 3.13.5 and 3.13.7 are where that lands, with constitutional varus saying neutral is not normal for everyone and dynamic alignment saying the adduction moment predicts where the angle does not. Ten pages. 3.13.9 and 3.13.10 carry a fellowship tier. | Consultant review by a surgeon who does osteotomy. The statements carrying most weight are 3.13.10's, that the classical overcorrection target came from a single 1980s series and has never been compared with an alternative, and 3.13.7's, that the variable which predicts is neither the one measured nor the one surgery corrects. 3.13.6's claim that the accepted joint line obliquity limit is a convention with no outcome study behind it is the third. | 10 published pages, 14,532 body words as the style gate counts them |
| W-19 Section 3 Biomechanics Academy, chapter 3.14 | All 12 pages written, style gate passed, awaiting review | Gait Analysis. The organising idea is that gait is the one place the whole knee is measured doing its actual job, and what it measures best is adaptation: every disease page here is about a compensation the patient makes without being told to and usually without knowing. The recurring honest point is that this is the most mechanically complete measurement available and it is almost never available in a clinic. Two overlaps are handled on the page rather than left to the reader: chapter 3.3 owns kinematics during walking, running, landing and cutting, and 3.8.7 owns the PCL deficient knee from the ligament's side. 3.14.1 carries a patient tier, which brings the FAQ block and the when to seek help message. | Consultant review by someone who reads gait literature. The statements carrying most weight are 3.14.9's, that quadriceps avoidance is the best known finding and not the commonest, and that altered kinematics persist after reconstruction; 3.14.11's, that the adduction moment predicts progression better than the static angle surgery is planned from; and 3.14.12's, that a successful knee replacement does not restore normal knee mechanics. 3.14.8's claim that the landing screen has underperformed while the training has held up is the fourth. | 12 published pages, 15,507 body words as the style gate counts them |
| W-20 Section 3 Biomechanics Academy, chapter 3.15 | All 10 pages written, style gate passed, awaiting review | Sports-Specific Knee Biomechanics, ten sports. The organising idea is that each sport asks a different question of the same joint and its characteristic injury is the answer: football asks for a direction change on a planted foot, skiing asks the knee to be the weak link in a long lever, cycling asks for a million low load repetitions, golf asks a fixed foot to absorb a rotating body, dance asks for control at the end of range. Every page carries patient, medical student and MRCS and none goes to FRCS, which the architecture sets. That makes these the most public facing pages in Section 3 and the shallowest, so each carries a real when to seek help message rather than a formality. | Consultant review by someone with sports medicine exposure, and ideally by more than one person, since no single reviewer will know ten sports. One discipline runs through the chapter and should be checked: no injury rates are quoted anywhere, because epidemiology needs sources and none has been retrieved. Each page says so where a reader would expect a percentage. The statements carrying most weight are 3.15.6's, that modern ski boots moved the injury from the tibia to the knee, and 3.15.7's, that recreational running is not established as a cause of knee osteoarthritis. | 10 published pages, 13,105 body words as the style gate counts them |
| W-21 Section 3 Biomechanics Academy, chapter 3.16 | All 10 pages written, style gate passed, awaiting review | Advanced Biomechanics, ten measurement methods. The organising idea is that every number in Section 3 came from one of these methods, and each buys its precision by giving something up. Cadavers give control and lose life. Robots give repeatability and lose physiology. Models give a continuous field and inherit their assumptions. Motion capture gives the living knee and loses precision. Wearables give volume and lose accuracy. This is the chapter that audits the other sixteen. It is also the first chapter where every page carries a consultant tier, so decision 018 and the DEC-001 gate apply on all ten: each consultant block is a balance argument that names the five patient factors rather than a verdict. 3.16.1 is where that was tested, because the obvious consultant line on cadaveric work is a preference between two study designs, and the gate was answered rather than suppressed. 3.16.10 turns the chapter's argument on this site: every measurement here was written from standard teaching, is registered as unverified, and an unverified number spoken aloud to a patient becomes a fact to them. | Consultant review, ideally split, since no one reviewer will know cadaveric rigs, finite element practice, machine learning method and wearable validation. Two things to check hardest. First, the accuracy figures: the chapter quotes marker-based and model-based error bands from standard teaching and they are the kind of number that dates, so all ten pages are in the figure register as unverified. Second, 3.16.6 on artificial intelligence and 3.16.9 on digital twins describe what the methods can do today rather than what is claimed for them, which is a judgement a reviewer should either back or correct. | 10 published pages, 12,530 body words as the style gate counts them |
| W-22 Section 3 Biomechanics Academy, chapter 3.17 | All 10 pages written, style gate passed, awaiting review | Landmark Biomechanics Papers, the chapter that was held from 9 to 11 October 2026 and is the only one on the site written under a rule invented for it. Each page takes one subject and asks what the studies asked, what they did, what they showed, what they changed, and what to read now. Three disciplines run through it. No page carries a citation, a year, a journal or a page number, which is decision 022 and is enforced by ATTR-001 rather than intended. No page carries a measurement either, which was not planned and follows from the same discipline: a chapter that cannot cite a source has no business quoting a figure, and the figure register gained nothing from these ten pages. And every consultant block is a balance argument under decision 018, which matters more here than elsewhere because a chapter about landmark papers invites a verdict on each one. 3.17.9 is the exception and says so on the page: the architecture asked for the Imperial group's research programme summarised and the page delivers their published work instead, because describing a named group's current work from memory is a different act from writing standard teaching. 3.17.10 collects the seven gaps the other nine pages named and ranks them by what answering them would change. | Consultant review, and this is the one chapter where the review needs a library rather than only expertise. 68 attributions are listed in the pack with a Citation column, and confirming or correcting them is the work. Three things to check hardest. First, whether the author named beside each finding is the right one, since the whole chapter rests on that pairing. Second, 3.17.9, where the question is not whether the content is right but whether the page should exist in this form, which is the Imperial group's call. Third, 3.17.10's ranking, which is a judgement about clinical consequence rather than a reading of the literature and should be argued with. | 10 published pages, 13,689 body words as the style gate counts them |
| W-23 Section 4 Clinical Examination Academy, chapter 4.1 | 8 of 227 written, style gate passed, awaiting review | The first chapter of Section 4, and the first page on the site built on the examination template that decision 023 invented for it. The type map had pointed examination pages at the anatomy template, whose headings are Structure and Location, Relations, Blood Supply and Innervation, Function and Clinical Relevance, and a page called The Lachman Test has no blood supply. The organising idea of the chapter is that examination is not a cheaper scan: it answers a different question, and the commonest error in knee practice is treating an image as the answer to a question only the examination asked. 4.1.5 is the page a clinician should read before the individual tests rather than after. 4.1.8 argues that the pre-operative record is part of the operation, because a procedure that cannot be audited cannot be improved. No page in the chapter carries a diagnostic accuracy figure, which ACC-001 refuses, so every claim about how good a test is, is written as a comparison. | Consultant review, and two structural confirmations before the rest of the section is written. First, the five section examination template: it is invented and no editorial authority has signed it, and 219 pages will be built on it. Second, the tier set: decision 024 took medical student to fellowship from one sentence in Part 1 of the architecture, because Part 2, which carries Section 4's real page map, is not in this build. Beyond those, the claims worth arguing with are the comparative accuracy statements, since they are the ones a reader will act on and the figures behind them were refused rather than retrieved. | 8 published pages, 9,927 body words as the style gate counts them |
| W-24 Section 4 Clinical Examination Academy, chapter 4.2 | All 34 pages written, style gate passed, awaiting review | History Taking, the largest chapter in Section 4 and the chapter that only exists at this size because EV-22 found the outline had collapsed it to five pages. Each page is one symptom or one question, in five groups: the presenting complaint, pain assessment, injury history, sports-specific history and previous treatment. The examination template fits history taking better than expected, and the reason is worth recording: a history question is a diagnostic test. What It Tests is what asking tells you, How It Is Performed is how to ask, What a Positive Means is what the answer licenses, How Good It Is is how discriminating the question is, and Where It Misleads is where the answer deceives. The organising idea of the first seven pages is that a patient presents with a word and the clinical meaning of that word is almost never what they mean by it, so the presenting complaint is a translation exercise. 4.2.5 on locking carries that argument furthest: almost every patient reporting locking means something else, and the follow-up question about how the knee was freed is the most valuable single addition in the chapter. The treatment history pages at the end carry its conclusion, that the history determines the plan more often than the imaging does. | Consultant review. Two things to check hardest. First, every comparative claim about which question discriminates: those are the claims a reader will act on, they are written from standard teaching, and ACC-001 refused the figures that would support them rather than inventing any. Second, the repeated argument across 4.2.32 and 4.2.34 that an incomplete rehabilitation trial is the commonest route to an unnecessary operation, which is an editorial judgement stated as a clinical one and should either be backed or removed. The chapter also makes no claim about frequency anywhere, which is deliberate and worth confirming as acceptable. | 34 published pages, 40,193 body words as the style gate counts them |
| W-25 Section 4 Clinical Examination Academy, chapter 4.9 | All 11 pages written, style gate passed, awaiting review | Meniscal Examination, the first chapter in Section 4 whose pages are individual named tests, so the examination template applies in full. It is also the chapter ACC-001 was written for: the meniscal tests are the ones whose sensitivities are quoted most often and recalled least reliably, and no accuracy figure appears on any of the eleven pages. What appears instead is the ranking, which is standard teaching: the tests are specific rather than sensitive, so a positive is informative and a negative excludes very little, and joint line tenderness is the most sensitive finding and the least specific. The chapter's argument is that no single test performs well enough to rely on, the combination performs better than any of them, and one finding acts alone. That finding is a confirmed mechanical block, which 4.9.8 treats as the chapter's most important page and the one exception to the meniscectomy trial evidence, because those trials recruited mechanical symptoms rather than locked knees. 4.9.9 states plainly that the examination performs worse for root tears than for any other lesion here and that its job is to raise the question rather than answer it. 4.9.10 and 4.9.11 carry the chapter's conclusion: deficiency is the predictable end state of the resection these tests lead to, and in a degenerate knee the examination exists to find the exception rather than confirm the rule. | Consultant review, and this chapter needs it more than most because the comparative accuracy claims are the whole content and the figures behind them were refused rather than retrieved. Three things to check hardest. First, the ranking itself: that joint line tenderness is the most sensitive finding and the provocative tests the most specific. Second, 4.9.4's statement that the Thessaly test performs comparably to the older tests rather than better, which contradicts its original reports and is the kind of claim a reviewer should either back or correct. Third, 4.9.11's position on the meniscectomy trials, which the chapter treats as settling the degenerate tear with pain and not settling a true block. | 11 published pages, 13,886 body words as the style gate counts them |
| W-26 Section 4 Clinical Examination Academy, chapter 4.10 | All 12 pages written, style gate passed, awaiting review | ACL Examination. The best performing examination in the knee, and the chapter where the gap between what the tests measure and what the patient complains about is widest. 4.10.9 states that gap as the chapter's conclusion: the translation tests are accurate, translation is not the complaint, and the test that measures the complaint is abolished by the patient being awake. Almost every dissatisfied reconstruction in referred practice sits in it. ACC-001 held on all twelve pages, so the comparative ranking is written and no percentage appears: the Lachman outperforms the anterior drawer substantially, the pivot shift is specific and insensitive awake and its sensitivity rises substantially under anaesthesia. 4.10.5 is the clearest case on the site of a position making a test worse, and the page says so rather than defending the anterior drawer. 4.10.7 holds the lever sign provisionally and names the pattern it shares with the Thessaly test, strong initial reports followed by less favourable independent ones. 4.10.11 separates four causes of a disappointing reconstruction and notes that the largest group is confidence and rehabilitation rather than structure. 4.10.12 closes the chapter by arguing that the revision clinic is where the cost of an incomplete pre-operative examination is paid. | Consultant review, and the comparative claims are again the whole content with the figures refused rather than retrieved. Three things to check hardest. First, 4.10.9's central claim that rotational instability is the commonest thing missed before reconstruction and the commonest residual complaint after it, which is an editorial judgement stated as a clinical fact. Second, 4.10.7's provisional handling of the lever sign, which a reviewer may consider either too cautious or not cautious enough. Third, 4.10.11's assertion that a small increase in translation with a firm endpoint is a satisfactory graft rather than a failure, which is the kind of claim that changes how a trainee reads their own findings. | 12 published pages, 15,616 body words as the style gate counts them |
How to read this. Nothing on this list is hidden from the pages themselves. Where a
page describes something the site cannot yet do, the page says so in its own words. This
list exists so that the reason, and what would change it, is recorded in one place.