USR369 wanted maintenance items added for keeping the trackers themselves current. Rather than register a new duty (would contradict his T2538 direction to narrow down, not multiply), broadened tools-index-maintenance's scope and updated its owning SOP (SOP-TOOLS-MAINTENANCE.md) to explicitly include: grepping for live pages missing from page-inventory.html entirely, and keeping backend/index.html's links box in sync. Immediately ran this check as a real first pass rather than just describing it -- found maintenance-log.html (built earlier this session) was never added to Page Inventory. Added it as row #176, updated the page count 170->171 and date stamp.

Two different pages share similar names -- backend/index.html (linked from panel.html as 'Backend') and backend/info/index.html (linked as 'Info', where I'd built the Tracking & Reference section + Actively Being Built box). USR369 was likely looking at the former the whole time and finding nothing there. Added a simple, high-contrast bordered box directly under the page title with 4 plain clickable links (Tools Index, Maintenance Schedule, Maintenance Log, Page Inventory) -- no expand/collapse, no JS behavior, just links, per the explicit ask.

USR369 said he had no visibility into the schedule and couldn't drill down. Made each duty's name clickable -- expands to show the SOP link (what the duty actually does) and the 3 most recent MAINTENANCE-LOG.md entries for that owner system, reusing the same parsing logic as maintenance-log.html, cached once per page load so opening multiple duties doesn't re-fetch. For Tools Index, added a read-only 'View Code' link per tool pointing to file-reader.php?path=<the tool's own real path> -- confirmed file-reader.php's auth (cai_check_token) already checks the PHP session cookie before falling back to a URL token, so this works for a real logged-in browser without needing to embed a token in the page (my earlier curl-based 403 test was a false negative from curl having no session cookie, not a real limitation).

USR369 reported JANUS 'has nothing' when clicked and asked where the rest of the tools were -- both stemmed from the same root cause: PEEK/JANUS/Login/FIX-LOG/API Index had all been dropped into a 'Data & DB' section that didn't fit any of them, and Tier-1 systems/commands/ tools (OPEN/CLOSE/TASKGATE/BACKUP/etc) were entirely absent from this page (deliberately, to avoid duplicating commands.html, but that left them looking missing). Fixed both: JANUS now sits in its own 'Action Tools' section with an ACTION tag and an explanation of why a plain click shows an auth error (needs POST params, used by a session mid-work, not opened by a person) -- its Open Tool button now points to SOP-JANUS-PEEK.md instead of the dead-end endpoint. Added a Platform Commands section at the top linking to Command Reference so Tier-1 tools are visible without duplicating that list.

USR369 clarified the tools page should work as: click a tool -> see what it does (explanation) -> then a separate action to go to the actual tool, not click=immediate-navigate. Converted the anchor-tag cards to div-based cards with a JS toggle (only one open at a time) and an inner Open Tool button that stops click propagation so it navigates independently of the expand/collapse.

USR369 flagged (with a screenshot) that maintenance.html's log preview was an unreadable wall of raw text -- it was a literal unparsed tail of MAINTENANCE-LOG.md's dense per-line format, several lines from Server[40]'s verbose SOP-HOUSEKEEPING-SCHEDULE runs. Built a new page that parses each log line into date/time/system/action/outcome, color-codes outcome (ok/partial/alert), puts the dense internal detail behind a click-to-expand, and adds a per-system filter. Reduced the inline preview on maintenance.html itself to 3 short one-liners (date, system, short action) with a click-through to the full page.

Placed directly under the page title, above all category sections, styled distinctly (accent border/background) so it reads as a separate, temporary zone rather than part of the permanent index. Contains Tools Index, Maintenance Schedule, Page Inventory with one-line dated notes on what changed. Documented in the section's own HTML comment that this box should be kept current every session that builds/changes a page, and an entry should move out to its regular category once it's no longer new.

USR369 wanted a separate organized spot (not mixed into Administration) specifically for read-only tracking/reference pages, linked from the front panel, and asked for suggestions on what else belongs there. Reused the existing Info Index page (backend/info/index.html) rather than building a new hub -- it already existed as a categorized reference page, just didn't have this category or these 4 links yet, and panel.html didn't link to it at all.

Added includeFull flag (full=1 or include_content=1). Default response: sop_scope truncated to 240 chars with a note of the true total length, sop_content=null, content_length reports the real file's size, fetch_note tells the caller exactly how to get the rest (file-reader.php on sop_path, or re-call with full=1). When full=1 is set, behavior is byte-identical to the pre-patch version -- nothing removed, only made optional. Also: found and reconciled a duplicate-effort risk -- Master[10] independently created SOP-HTML-PAGE-MAINTENANCE.md + a 'html-page-check' duty in maintenance-schedule-api.php today, its KNOWN BLOCKER section claiming page-inventory.html was still 404 (true when written, false now after my earlier fix this session) -- corrected that note, and registered a second complementary duty 'tools-index-maintenance' (content-curation, narrower) rather than letting the two overlap silently.

Confirmed via TASKGATE.php source that a successful match injects the FULL matched SOP file (sop_content, file_get_contents of the whole file) plus the full index-entry description (sop_scope) into whichever session's conversation triggered the match -- every time, regardless of file size on disk. Rewrote SOP-TOOLS-MAINTENANCE.md from a narrative style (WHY/HISTORY/RELATED sections) to execution-only steps, moving rationale into decision/knowledge DB references only (D862/D863/K403/K430). Trimmed its SOP-INDEX.md entry from a restated paragraph to a bare trigger+pointer line.

Read TASKGATE.php's source directly: its regex (line 114) only recognizes SOP-INDEX.md entries in the ' SOP-NAME.md path' (2-space indent) + 4-space-indented description block format. The newer '## SOP-NAME.md (System, added date)' header-block style used for SOP-GROCERY-TRACKER.md (Finance, 08/24/26) and my own new SOP-TOOLS-MAINTENANCE.md entry is completely invisible to that regex -- confirmed by replicating the exact regex locally in Python against the live file before my fix: 0 entries in that style parsed. Reformatted both entries into the parseable indented style.

Server[40] drafted a patch removing the hardcoded admin token from inventory.html and staged it at backend/tmp/inventory_html_PATCHED_08-22-26.html, correctly deferring the HTML deploy to Builder per lane rule. Diffed the staged file against live before deploying -- confirmed the only changes were: removing the hardcoded var TOK line, and switching apiFetch()/error handling to the session-cookie + login.php redirect pattern already live on dashboard.html/panel.html/commands.html. Backed up live file first.

Root cause: dba_write_api.php's $paths registry keyed systems by OLD sequential single-digit codes (sys-01=Master...sys-10=Travel) while db-admin.html's DB_CONFIG (frontend, already correct) keys every system by its real decade code (sys-10,sys-20...sys-95). Only sys-00 (Admin) coincidentally overlapped, so every other system's Pick/edit action in db-admin.html silently 404'd with db not found. Fix: rekeyed all 10 non-Admin entries in $paths to decade codes matching DB_CONFIG exactly, no other logic touched (additive rename only, same structure). Files touched: backend/tools/critical/dba_write_api.php (backed up first to backups/20-builder/php/Builder_dba_write_api_CURRENT.php). Normally Server[40] lane (PHP/API) per D231 -- done directly on USR369's explicit go-ahead since Server was occupied elsewhere this session.

USR369 asked for 'final close' -- CLOSE.php's final_close=1 mode, which internally calls HANDOFF.php to build a structured handoff. It failed with 'unknown system code' on the very first try. Root cause: HANDOFF.php's $sys_dirs map still used the OLD legacy single-digit system codes (01, 02, 03, 05-09, and '10' for Travel) -- it was never migrated to decade codes when the rest of the platform was (roster_lib.php, inbox-api.php, transfers/index.php all did this migration already, per their own comments). CLOSE.php passes system=20 (Builder's real decade code); '20' was never a key in HANDOFF.php's map, only the old '02' -- hence the failure. Separately, and worse: the old map's '10' meant Travel, which directly collides with the CURRENT decade code for Master ('10') -- had anything ever hit this path with system=10 intending Master, it would have silently written Master's structured handoff into Travel's directory instead. Fixed by replacing the whole map with the canonical decade-code map from roster_lib.php's DIR_MAP, matching what every other migrated endpoint uses -- no legacy codes, no collision risk. Second, unrelated finding caught only by reading the actual handoff-20.md output after the fix (not just trusting the API response): HANDOFF.php has 'OPEN QUESTIONS FOR DAVID' hardcoded into its own template -- a direct, literal violation of the platform's own never-use-the-owner's-real-name rule, baked into a shared command file every system's final close runs through. Fixed to USR369. Also corrected the one already-written handoff-20.md that had the old text baked in before I caught it.

Built login.php, extended auth-lib.php additively for session-based auth, migrated file_write_web.php off a hardcoded admin token onto the shared source (D791). Full E2E verified live with a real cookie jar, not assumed.

Multi-system/Name-of-Task fields added (D785); corrected a false Finished status neither CC nor I wrote (K390, D788 correction); fixed \r\n insertion bug proven via Node test against real file content (D790).

USR369 correctly rejected the sessionStorage+prompt fix -- asking for the token on every visit isn't a fix, and per direction to 'search the web' / 'how do other systems do it', researched current practice (OWASP, PHP manual, Auth0/FusionAuth guidance): the standard answer is a server-side HttpOnly session cookie set by a real login page -- JavaScript can never read it, so it can't be exposed via page source or XSS, and the browser just sends it automatically after one login. Built: (1) auth-lib.php -- added cai_start_session()/extended cai_check_token() to ALSO accept a valid PHP session, purely additive, every existing token-param caller (API calls, this Claude session, Claude Code, etc.) works identically to before. (2) login.php -- one-time form, POSTs token, validates via cai_check_token(), sets ['cai_authed'] + session_regenerate_id(), redirects back. (3) DISCOVERED while wiring this up: file_write_web.php -- the endpoint every Save action actually hits -- had the raw admin token hardcoded directly in server-side PHP, completely bypassing auth-lib.php (a straggler that missed the 08/18/26 Phase 2 migration every other endpoint got). Same risk class as the prior K372/K373 token-in-files incidents, plus meant every future rotation required remembering to hand-edit this one file. Migrated it to use cai_check_token() too. (4) fix-log.html -- removed every token/TOKEN reference entirely, including the sessionStorage/prompt version from the previous fix; fetch calls now carry no auth of any kind, relying entirely on the browser's session cookie; added isAuthError()/redirectToLogin() so an unauthenticated visit bounces to login.php once instead of failing silently.

Second bug found while verifying the fix-log.html token fix was truly complete: the 'Add to Log' button's insertion logic checked for a bare ' --- ' separator, but FIX-LOG.md actually uses line endings throughout (confirmed directly: ' --- ' present, ' --- ' absent). The check silently always failed, so every entry added through the web UI landed appended at the absolute end of the file instead of after the header where new entries belong -- correct display order was preserved on-screen only because renderLog() re-sorts by parsed date/time client-side; the raw .md file itself would have drifted out of the newest-first convention that SOP-FIX-LOG.md documents and that other systems' manual/API writes rely on. Fixed: checks for ' --- ' first (the real format), falls back to bare ' --- ', falls back to plain append only if neither separator is found at all.

fix-log.html had the platform admin token hardcoded in plain text directly in its client-side <script> block -- visible to anyone who viewed the page's source, regardless of auth. This existed before my v3 rebuild (I noticed but didn't flag or fix it at the time -- should have) and I made it worse by writing the NEW rotated token into the same exposed spot when fixing the earlier break. USR369 called this out directly: the entire purpose of the rotation is to stop tokens sitting in plaintext 'everywhere'. Fixed properly this time: removed the hardcoded token from the file entirely, replaced with sessionStorage.getItem('cai_platform_token') with a window.prompt() fallback on first use per browser tab -- token now lives only in that browser's session storage, never in the file's static content on the server, cleared when the tab closes. This is very likely NOT an isolated issue -- any other browser-facing tool page built the same way (db-admin.html, panel.html, dashboard.html, cmd-popup.js/ml-popup.js, and any system-built app like Gym Logger/DoughCalc/Tech's inventory viewer that fetches file-reader.php/file_write_web.php directly from client JS) plausibly has the identical hardcoded-token pattern. Not audited this session -- flagging as a real, likely-widespread follow-up rather than claiming this one fix covers the actual exposure surface.

Landed PEEK.php (read-only cross-system status snapshot: COMMS open/closed, inbox/transfers pending counts, gov file freshness) that CC[100] had built locally but couldn't reach the server for. CC provided no attachable file content this session, so rather than block on file transfer or guess at its exact source, rebuilt it directly on the server matching CC's stated design (?system=all|decade, read-only-enforced) and platform conventions confirmed by reading CHECKPOINT.php and inbox-api.php source (auth-lib.php token check, roster_lib.php for the system list, direct SQLite3 prepared-statement queries against comms.db/community.db/transfers.db instead of calling inbox-api.php's pickup action, which has a side effect of marking items seen -- PEEK must never mutate state). Deployed first to systems/commands/PEEK.php per the intended path -- that path returns HTTP 403 for every request regardless of query string, confirmed to be path-specific (not a PHP error) since the same content at backend/tools/critical/PEEK.php returns 200 clean. Logged as K391 for Server[40] to investigate; left a DEPRECATED stub at the original path pointing to the working one.

Added multi-system support (comma-joined chip picker, + Add button) and a new NAME OF TASK field (separate from the existing TASK # code field) to the FIX-LOG web editor at backend/tools/fix-log.html. Log line format bumped systems/governance/FIX-LOG.md and its governing SOP-FIX-LOG.md from 5-field to 6-field: date/time | system(s) | T### | name-of-task | status | fix. Backward-compat parser (parseLineParts) auto-detects 4/5/6-field legacy vs current lines by field count and whether field 4 is a status keyword. Root cause of TASKGATE matched:false on the original request: SOP-FIX-LOG.md existed live at systems/governance/SOP-FIX-LOG.md since 08/12/26 but was never listed in SOP-INDEX.md, so TASKGATE's SOP-INDEX-based matcher could never have surfaced it. Fixed by adding it to SOP-INDEX.md (bumped to v2.8) instead of writing a duplicate SOP, per the NO PER-SYSTEM DUPLICATE SOPs rule. Also corrected a stale path claim in SOP-INDEX.md's existing FIX-LOG.md data-file entry (it said USR369/FIX-LOG.md; the real, live, populated file is at systems/governance/FIX-LOG.md). File backed up via BACKUP.php file-mode before editing (backups/20-builder/html/Builder_FixLog_CURRENT.html). All three files grew in size on write (no shrink-guard trigger).

id=427 assumed to be T2528 without verification, was actually unrelated. Reopened, flagged to Server[40] directly.

cmd-popup.js v5.7 and commands.html v1.4 both pull live version/date from Server endpoint, chose not to redirect to preserve real functionality

Consolidated duplicate Gym Logger rows, fixed 2 wrong system codes, corrected DoughCalc stale version and built_by field, fixed Toolbox dead file_path

Part 1: cmd-popup.js v5.6->v5.7. Added CMD_VERSIONS map populated via async fetch to records-api.php?action=get_commands_inventory, called after the existing synchronous render completes. Deliberately did NOT touch the list-row HTML/CSS at all given this file's real regression history (v5.5's row-level changes broke live rendering on both phone and desktop, reverted as v5.6, 08/10/26) -- injected version/date into the drill-down panel's existing single-line group subtitle instead, the lowest-risk element to extend. Fails silent on any fetch error. Part 2: backend/sys-com/commands.html v1.3->v1.4. Considered Server's suggested simpler option (redirect/iframe to backend/ed/reference/commands.html) but checked that page first -- it is a flat, non-interactive table (name/version/date/size columns only, no descriptions, no drill-down, no reference-page links), and this file's own v1.0 changelog says it was built specifically to replace that exact page when it was broken. Redirecting would have been a real functionality downgrade, not a simplification, so kept this page as the richer canonical one and wired its own drill-down panel to the same live endpoint instead, same safe async pattern as cmd-popup.js. Both hand-written editorial text (invoke instructions, full descriptions) left completely untouched per Server's own note that it's not derived data. Verified: node --check clean on both files' extracted script blocks (both before and after deploy, reading the actual live bytes back), backed up both files first, live endpoint confirmed returning real current data (CLOSE v2.9 08/10/26, SYNC v2.6 08/10/26, etc.) matching the exact field names both files read. Could not verify interactive rendering in an actual browser -- that requires USR369 confirmation.

Renamed to Gym-Logger-DIARY.md, pointer stub left at old path, ARTIFACTS-20.md updated, Build Diary DB backfilled (id 3)

Same blocking notice exists separately in inbox and transfers queues, each needs its own resolve call -- later confirmed by Master as a redirect+POST bug, not a DB issue

70-app.db and gym.db have incompatible schemas, not just staleness -- naive sync would corrupt cardio/pool data. Two options presented, not yet decided.

v5.44-v5.46, final color fix used border+background after text-color was confirmed not visible on device

Live diagnostic query timing (0.3-0.7ms) proved no server-side hang, retracted Server[40] escalation

Reviewed and deployed 3 fixes found already present in working file after flagging unexplained origin to USR369

Changed init restore gate from hasDraft() && savedHdr to hasDraft() || savedHdr, verified live

Created cai-platform-writer service account under cai-yttcom, hit and fixed an org-wide policy blocking JSON key downloads via a project-scoped override. Key downloaded successfully. Folder sharing and gdrive-api.php build remain open.

USR369 had accidentally created two Google Cloud billing accounts, both with $300 trial credit - one under ikre84i@gmail.com, one under yttcom.net. Closed the g account specifically, confirmed via closure banner and Gmail notification, verified y account remained active and untouched.

Enabled Drive+Sheets APIs, created cai-platform-writer service account under cai-yttcom. Google org-wide policy iam.managed.disableServiceAccountKeyCreation blocked JSON key download even after successful account creation - fixed via a project-scoped policy override (Organization Policies -> Override parent policy -> Enforcement Off for cai-yttcom specifically), requiring Organization Policy Administrator role first. JSON key downloaded successfully. Folder sharing and gdrive-api.php build (Server[40] Step 2) remain open - SOP-GDRIVE-SERVICE-ACCOUNT.md updated to v1.1 with full detail.

USR369 had accidentally created two Google Cloud billing accounts, both with 00 trial credit - one under ikre84i@gmail.com, one under yttcom.net. Walked through Cloud Console to close the g account specifically, confirmed via closure banner and Gmail notification. Verified y account billing remained active and untouched. Also shut down a leftover My First Project linked to the closed billing.

USR369 confirmed via screenshot that v5.45's orange text-color fix on the closed select box was not visually appearing on his actual Android device. Root cause: CSS color property on native <select> elements is commonly ignored by Android's rendering of the closed/collapsed state (separate issue from the already-known open-popup option-styling limitation noted in v5.45). Fixed by switching the unselected-state distinction to border-color (orange, 1.5px) plus a faint background tint (rgba(240,184,106,.08)) instead of relying on text color alone - box-model properties are reliably respected cross-browser where text color is not. Text color kept as a bonus. Deployed and verified live.

USR369 flagged via screenshot that the exercise dropdown empty-state text duplicated the row header. Changed placeholder from em-dash Exercise em-dash to Choose Exercise. Added CSS class ex-name-sel.unselected (orange) vs existing blue once chosen, toggled in onExSelChange and set on initial render based on prefill.name. Did not rely on per-option inline color since Android native select popups commonly ignore page CSS - targeted the closed state instead.

USR369 directed fixing the DLM timeout directly rather than waiting on cross-lane escalation. Wrote a temporary diagnostic PHP script (not in protected systems/commands/, no whitelist issue) that connected to the live gym.db directly and timed the real query for profile=DLM: connect 0.3ms, COUNT 0.7ms for 19 rows, full SELECT 0.3ms for 26,666 bytes, journal_mode=wal with 0-byte wal file (no lock contention). Compared against profile=david (16 rows, similar timing) - no meaningful difference. Conclusion: the database read itself is instant, there is no server-side data/query hang to fix. Retracted the earlier Server[40] escalation (inbox 870) with a correction (inbox 871) explaining the finding and that no action is needed on their end. Likely explanation for the original note: either the pre-K-GYM-03 double-? URL bug (fast error, not a slow timeout - imperfect fit) or an egress-proxy blip on a testing session (matches platform K215/K216 pattern). Diagnostic script neutralized to a 404 stub immediately after use (no delete API exists, T741).

USR369 confirmed via screenshot he does not use the Hide Done or Compact toggle buttons in the Exercises section header. Removed both <button> elements (id=hide-done-toggle, id=compact-toggle) from the header markup. Edit List button unaffected. toggleHideDone()/toggleCompact() JS functions left in place - both already null-guard on getElementById (if(!btn) return), confirmed safe to leave unreferenced rather than risk a wider edit for a cosmetic-only cleanup. node --check clean, deployed via file_write_web.php, read back live and confirmed both buttons absent from HTML.

Found K-GYM-02/03/04 already present in the working file at deploy time, not authored via a tool call I could account for this session. Flagged the discrepancy to USR369 explicitly rather than silently accept or discard. Reviewed each fix line-by-line before trusting: K-GYM-02 (toggleKeepSimple was localStorage-only, now writes session.keep + calls saveSessions/syncToServer - confirmed real gap, stale diary claim corrected), K-GYM-03 (LOAD_URL ends in ?key=..., 3 call sites were appending with ?profile= producing malformed double-? URL, confirmed real via source read, fixed to &profile= at all 3 sites), K-GYM-04 (sync disabled permanently on single failed request, now requires 3 consecutive failures matching platform EGRESS RETRY DISCIPLINE philosophy). All verified: node --check clean, 3/3 LOAD_URL sites confirmed fixed, 0 broken usages in executable code, deployed via file_write_web.php, read back live and confirmed. v5.42->v5.43, 223959->228141 bytes. NOT verified: none of K-GYM-01 through 04 tested live on real device yet.

Root cause: DOMContentLoaded restore gate used if(hasDraft() && savedHdr) - AND-gate requiring both draft AND header present or else-branch wipes both cai_gym_editing_id and cai_gym_hdr plus resets form. loadSessionIntoLog() writes draft (saveDraft()) and hdr/editing_id in separate try/catch blocks - if saveDraft() silently failed (empty catch, no trace), draft missing but hdr present would still trigger destructive fresh-start branch on refresh. Matches reported symptom exactly. Fix: changed gate to hasDraft()||savedHdr, header-restore now independently guarded by if(savedHdr). Also fixed saveDraft() empty catch to log+toast on failure instead of silent swallow. Verified: backed up original first (backups/20-builder/file/Builder_gym-logger_CURRENT.html), node --check passed on extracted JS before deploy, deployed via file_write_web.php, read back live and confirmed functional if(hasDraft() || savedHdr) present with old AND-gate no longer in executable code, file size 222117->223959 bytes. NOT verified: live on-device reproduction of the original bug being gone (needs USR369 to test on real device). KEEP checkbox server-save, minor session-info-lbl position, dead loaded-session-header div, and sync-offline root cause all remain open from original request.

654 rows audited. 21 of 30 decisions from the last 3 days have blank detail (70%), 18 of those 21 from Server 40s 08/05/26 session. Blank subject stopped occurring after 07/17/26. Travel 95 self-corrected this exact gap on 08/06/26. Recommended add_decision get the same min-content-length validation as the handoff gate fix.

Same failure class as documented K107 loopback bug (LiteSpeed shared hosting blocks server-to-self HTTP calls). Todays v1.4 patch narrowed but did not eliminate it - the hardcoded open default is still hit whenever the internal lookup fails. Recommended fix: replace with direct SQLite query against comms registry, same pattern used to fix K107.

Testing add_decision 500-error fix is still live and holding, per Builder[20] verification request

Confirmed live: OPEN.php's handoff_content/todo_content injection reads from handoff-20.md/todo-20.md (decade suffix), not the -02 (code suffix) paths -- verified by direct comparison, exact match. ORIENT-REG.md's WHERE DO I LOOK table claims CODE suffix is 'confirmed' standard for handoff/todo, which contradicts this live behavior -- a real doc-vs-code mismatch (SOP-GOVDIFF pattern), not just a Builder-specific inconsistency. Fixed for Builder: relabeled internal headers in handoff-20.md/todo-20.md to reflect decade as canonical. Left deprecation stubs (not hard deletes, D209) at handoff-02.md and todo-02.md pointing to the real files -- todo-02.md previously 404'd entirely, now resolves to a useful stub instead. Did NOT change ORIENT-REG.md itself -- that is Admin[00] jurisdiction and the code-vs-decade mapping differs per system, so a Builder-only test does not prove it holds platform-wide. Flagged to Admin/Master to verify via SOP-GOVDIFF and correct the doc if confirmed elsewhere.

Built backup.html covering BACKUP-REG.md v2.4 in full: 7 backup groups, 20+ file-registry categories (~100 individual file paths), all 11 systems' LEGACY files, platform + per-system databases, snapshot retention, registry-of-registries. Added a live search box that filters+auto-expands matching categories, given the scale of this registry vs CLEAN-REG.md. Cross-linked into dir.html, nav.html, clean.html (all bumped to v1.2/v1.1) and system/index.html (v1.2). THIS TIME (per USR369 catching the gap on T572): also registered both clean.html AND backup.html in page-inventory.html (was missing entirely before, confirmed live). Checked site-index.html -- deliberately did NOT add entries there, since dir.html/nav.html aren't linked there either (it only links the raw STANDARDS-REG.md registry) -- adding clean/backup only would put them ahead of the older pages in discoverability. Left for T-PAGE-AUDIT to decide whether site-index.html should be restructured into real categories. All pages backed up before edit, JS syntax-checked before every deploy.

Built clean.html following dir.html/nav.html template exactly (same top-bar, collapsible rule-row pattern adapted to cat-row for tiered groups, same color tokens/fonts). Converted CLEAN-REG.md v1.2's Tier 1/2/3 protection lists, TMP rules, and sweep patterns into a JS-driven data array rendered as collapsible sections. Cross-linked into dir.html, nav.html (bottom tabs) and system/index.html (card list) so it is not a dead end per nav standard. Backed up all three modified pages before editing (dir.html, nav.html, system/index.html) plus deployed clean.html fresh. Version-bumped dir.html and nav.html to v1.1, system index to v1.1. Syntax-checked JS before deploy.

Fixed 08/05/26: closeSystem() in backend/sys-com/kill-switch.html previously called CLOSE.php directly with no notification to Master[01], creating a silent gap. Added dropCloseReport() function, called on both the success path and the fallback path. Bumped file to v1.1, backed up original v1.0 to backups/platform/html-pages/ per SOP-HTML-BACKUP.md. Syntax-checked (node --check) before and after deploy.

btab font-size:7px padding:3px 7px gap:2px — locked for all hub pages

Bottom tab mobile size standard

readGov() with &raw=1 for all large shared text files. Panel shows text inline with Edit (purple, new tab), Download, Close buttons. Never link directly to file-editor.php.

Gov reader panel standard locked

Section order, labels (orange), Quick Access inline, gov reader Edit button, btab mobile sizing, sys-link dim color, Standards+Commands links — all 5 Core pages conform — standard at systems/20-builder/standards/HUB-PAGE-STANDARD-S.md

Hub Page Standard v1.1 locked

Every backend system hub page uses identical tab bar. Order: ← Hub, Admin, Master, Builder, Server, Tech, Backend (current, highlighted), Frontend (links to that systems frontend page).

Standard backend tab bar format locked

Tech [03] is part of the Core Five alongside Core Four. All backend system hub pages include Tech in the tab bar. Standard: ← Hub | Admin | Master | Builder | Server | Tech | Backend | Frontend.

Core Five — Tech [03] included in all backend tab bars

Moved from frontend/70-Health/Gym/ to frontend/70-Health/apps/Gym/. All links updated. apps/ is the standard subdir for user-facing apps within each system.

Gym Logger moved to frontend/70-Health/apps/Gym/

Each of 7 domain systems has --sys (primary color) and --text (lighter tint). Applied to page title, breadcrumb, section labels, card hover borders, active tab. Front Door cards show system name in its color.

System color identity — --sys and --text per domain system

All gov files named [type]-[decade].md — e.g. knowledge-20.md, todo-50.md. Applied to all 11 systems. Files live in systems/[dir]/gov/ only — no loose files in system root.

Gov file naming standard — type-decade.md

Final name: frontend/. Contains 7 domain system dirs + index.html (Front Door). Each system has 7 dirs: apps/ media/ web/ temp/ cache/ exports/ assets/.

frontend/ is the user-facing root — replaces 05-Front_end

Machine-readable pointer at backend/tools/TOOLS-INDEX.md. Lists file-editor, ml-editor, file-reader, db-admin, db-viewer with URLs, params, common paths, and standard link markup. Referenced in OPEN/REFRESH/build.php as tools_index field.

TOOLS-INDEX.md deployed — stop sign for backend/tools/

Removed path restriction whitelist — any path allowed, token still gates access. Fixed white-space:pre to pre-wrap + -webkit-user-modify:read-write for mobile textarea input.

file-editor.php v1.2 — open all paths, mobile fix

Lightweight write event log. Fields: file_path, event_type, system, bytes, note, created_at. API actions: log_file_event, list_file_events. Logged on every ML editor save.

records.db file_events table — audit log for file writes

New table in backend/records/db/records.db. Fields: record_id, label, detail, priority, status, list_file, added_by, added_at, completed_at. API actions: add_item, complete_item, update_item, get_item, list_items.

records.db list_items table — R001+ record IDs for all list items

ml-editor.html rebuilt with expandable detail per item, R001+ record IDs from DB, add_item logs to records.db, complete_item marks done with timestamp. Saves MD with Label — RXXX — Detail format.

Master List Editor v2.0 — detail field, record IDs, DB logging

Card added to backend/index.html (v6.2, orange). Tab added to backend/info/index.html bottom tabs.

Standards Hub linked from backend nav

All three command endpoints now return standards_index field with URL to STANDARDS-INDEX.md. Lightweight pointer only — no content injection.

standards_index pointer wired into OPEN/REFRESH/build.php

Machine-readable pointer at backend/standards/STANDARDS-INDEX.md — maps rule codes to URLs and build.php scope values. 12 sections, 4.7kb.

STANDARDS-INDEX.md deployed as stop sign

All 13 detail pages now have 5-10 rules each, up from skeleton 2-4. Categories: HTML/MD/System/PHP.

Standards Hub expanded to full rule set

Real branching pages at backend/standards/ (HTML/MD/System/PHP categories), standard card-button style reused from systems/hub/index.php, rollup format reserved for in-page rule lists only per David direction.

Standards Hub built — 17 pages

Two silent JS errors: field-name mismatch (timein/loc vs timeIn/location) between collectFormData and applyFormData, plus loadSessionIntoLog calling nonexistent switchTab() function. Both fixed, v5.25 live.

Gym Logger history-nav root cause found and fixed

Any system generating multiple files must: (1) check current series letter, (2) increment to next letter for ALL systems, (3) ZIP all files into one package, (4) drop ZIP in backend/369/ (David personal dropzone). Introduced in Series E EXEC_OPEN_Builder.20.E.txt 07/08/26.

D108 Multi-file output rule

Server TZ is UTC. Added date_default_timezone_set(America/Los_Angeles) to 10 system api.php files and dba_api.php. All DB timestamps now return PT (12-hour). Fixed 07/07/26.

Timezone fix — PT across all system APIs

systems/20-builder/knowledge.md created 07/07/26. Environment, Patterns, Lessons, Open Questions populated. Written at session close per D091 mandatory rule.

knowledge.md created — D091 compliance

Built perPage selector (25/50/100/All) and match highlighting into db-admin.html. Infinity-safe pagination. Resets to 50 on table switch. Deployed via bootstrap 07/07/26.

DB Admin v2.2 — perPage selector match highlighting

Added DB Admin v2.2 and D091 knowledge.md as done tasks. Changelog updated. Deployed 07/07/26.

Board v5.9 deployed