• Sketches and a larger preview (2026-10-24 → 27, Mark):
  • Sketches quick filter (694e1345eb): PSDs in any folder named for sketches, 120 files, 5.1 GB. It leaves out .sketch (Sketch app) files, the dynamic-sandbox PNG copies, and the 24 Procreate illustrations in design.3.0/illustration/originials/.
  • ↑/↓ preview (1ce48f3487): in the folder tree, the selection follows focus, so arrowing opens each file and updates ?f=. Tabbing in doesn't select. Built on the page (a focusin handler), not in UxFileTree; a selectionFollowsFocus prop is the layer candidate if another page wants it.
  • Preview sizes (5e12ec76a5): S / M / L / XL (360 / 480 / 680 / 900 px) and a drag handle between tree and preview (←/→, double-click for M), remembered per browser; the tree keeps 360 px.
  • Large sketch previews (6d44fa9c64): drive-previews --large writes 1,600 px WebPs to public/library/large/. The 117 sketches with a picture have one (5.3 MB; the 3 blank PSDs are skipped). /library offers both in srcset, so the large one loads at XL on a normal screen and at any size on retina.
  • Deployed (2026-10-27) as ba4ff7ce7a = design 6d44fa9c64 + 40 uncommitted from other threads. 17,877 files, 815 MB (Cloudflare's cap is 20,000). Verified on design-f39: arrows step 2023-sf-roundtable → early-deck-2.1 → early-deck-2.2 with the panel following, XL is 900 px and loads the large WebP (200), the handle shows, no page errors. The build's error lines are older ones (cycle-doc 404s, /cycles/weeks/1999-12-27).

• Discovery (2026-10-23): formats on disk, existing archive tooling, and a measured cycle-page audit (discovery.md). Mark's answers (answers-prompt.md): Procreate exports live in the shared Design drive; several Figma files from different times; audience is us first; detail page only; everything deploys behind Access. A first Drive survey (drive-survey.md): 1,141 files, 14.2 GB, generations 1.0–4.0, 105 sketch PSDs, permanent Drive IDs, nine cycle candidates.

• Catalog (2026-10-23), Phase 1 (catalog-prompt.md):
  • scripts/drive-catalog.mjs (npm run drive-catalog, ~3 s) walks the drive read-only. It lists folders, stats files, reads each item's Drive ID from the com.google.drivefs.item-id#S xattr, and reads the online-only (SF_DATALESS) flag in one stat pass. It never opens a file.
  • It writes server/data/drive-catalog.json: 1,137 files, 248 folders (every folder has a Drive ID), 14.1 GB, 78 temp/dot files skipped.
  • Each file records path, Drive ID, title, format, kind, generation, discipline, size, whether it's on this Mac, and a date with precision and source. The date sources: 203 from the file name, 277 from a folder name, 657 from the file's own times.
  • 52 files (0.7 GB) are on this Mac; the rest are online-only. 56 files have no Drive ID yet, all of them local, so Drive for desktop may not have uploaded them.
  • GET /api/library adds a startup phase inferred from the date against the archive's eras: 946 ideation, 87 pre-seed, 101 seed, 3 before Qpoint.
  • /library (off the main menu for now):
    • generation cards with file count, size and year span;
    • search, plus kind and phase filters;
    • the layer's UxFileTree over the drive's own folders;
    • a detail panel: the date and where it came from, the phase, size, on-Mac state, and links to the original and its folder on Drive.
    • The selected file lives in ?f=<drive path>, so a link points at one file; the tree scrolls to it on arrival.
  • Verified headless (Playwright, system Chrome, 1440 × 900): deep link in view; Design 1.0 + Raster + "early-deck" → 42 files; click → URL and panel update; no page errors.
  • Incident (Phase 1): while removing a debug line, an edit joined two lines in the page's script. The syntax error took down SSR for every route on :3000 ("IPC connection closed"). Mark's terminal-started dev server was stopped and restarted under the design launch config. Found by moving the new files aside and adding them back one at a time; fixed in the page.

• Previews (2026-10-23), Phase 2 (previews-prompt.md):
  • scripts/drive-previews.mjs (npm run drive-previews) writes one 640 px WebP per file to public/library/thumbs/<drive id>.webp (14 MB for all of them), plus server/data/drive-previews.json. Options: --only <prefix>, --list <file>, --retry, --force.
  • Downloads: QuickLook can't make Drive download an online-only file; it waits forever. So each online-only file is copied to a temp folder (the copy makes Drive download it), previewed from the copy, and the copy deleted. With a non-blocking copy and 4 workers, the whole drive (~14 GB) took under 10 minutes. Adobe's InDesign QuickLook extension hangs even on local files, so every tool runs under a timeout.
  • Methods tried in order, per format:
    • sharp for images and SVG;
    • sips → QuickLook → the XMP thumbnail for PSD;
    • QuickLook → XMP for PDF/AI;
    • XMP for InDesign/EPS;
    • previews/preview.png for .sketch.
    
    A blank result moves on to the next method. White-on-transparent art is retried on ink.
  • Results: 1,111 previews (354 sharp, 128 SVG, 334 QuickLook including the 25 blanks, 234 sips, 52 XMP, 9 Sketch); 6 rendered on ink.
    • 25 stay blank: PSD/AI files saved without a flattened image or a thumbnail, which only Photoshop/Illustrator can draw. The page says so and names the save option that would fix it.
    • 2 .indb InDesign books have no image of their own.
    • One pair of files shares a Drive ID (QPT.2025.Q3/Q4.overview.ai), so shared IDs take a path-hash suffix.
  • /library: the API merges the previews. Generation cards count previews. The detail panel shows the preview, or the reason there's none. A Folders ⇄ Grid switch (?view=grid) shows the filtered files as pictures, 240 at a time. The 2023 sketches: 110 of 111 under sketches/ have previews.
  • Verified headless at 1440 × 900: grid, selection, deep link, the blank-file note, no page errors. A random 40-preview sheet read cleanly.
  • For Phase 3: the same artwork often appears as @2x/@3x PNG + SVG + .ai, so grouping variants by stem will shorten the library a lot.

• Illustrator artboards (2026-10-23) (artboards-prompt.md → artboards-plan.md):
  • The probe covered 15 files and 173 artboards. A PDF-compatible .ai is one PDF page per artboard, matching in order and size. Its Illustrator private data (zstd/zlib native PostScript) holds each artboard's name, canvas position and UUID. Page order isn't canvas order.
  • Plan revised on Mark's note: sequences and subjects are tagged and stitched by hand, in post; extraction stays automatic and records facts only.
  • Steps 1–2 built. scripts/drive-artboards.mjs (npm run drive-artboards) has the same read-only, temp-copy, timeout and cache contract as the previews. It uses pdfinfo + mutool + zstd (streamed through tr/grep) + pdftoppm + sharp.
    • Rendering: 1,200 px long edge (400 px short edge beyond 3:1) plus a 320 px thumb, as WebP.
    • Results: all 255 .ai files, 3,627 artboards, 120 MB, no failures. 246 files have canvas positions; 396 artboards have custom names, mostly on the history boards, where they name the source file.
    • 9 files kept only their first artboard in the PDF copy (3–38 artboards each, ~140 in all, e.g. QPT.dash-1.0.ai 1 of 24, QPT.lab.dash-3.15.topnav-filter.ai 1 of 38). They show "1 of N readable" and wait for the Illustrator pass (plan step 5).
    • Two fixes on the way: one file's private data decompresses to 1.4 GB, more than a Node buffer can hold, so the records are now filtered in a shell pipeline. A zstd early exit no longer crashes the run.
  • /library:
    • "Inside " draws an .ai file's artboards at their real canvas positions and sizes. Clicking one opens UxLightbox at 1,200 px; arrows step in reading order, and ?a=<n> links one artboard.
    • The detail panel gets an Artboards row, and the generation cards count artboards (Design 1.0: 1,268).
    • The /api/library payload is now 1.2 MB.
  • Quick filters (Mark, same day):
    • Illustrator (.ai): 255 files.
    • Presentations: 381 files, matching any folder named for a presentation, deck, pitch or slides, or "deck" in the file name. Deck folders also hold their image assets.
    • The two combine with AND: 45 .ai presentation files.
  • Deployed (2026-10-23) as 3e5d09dc10 = design 7b2668823f + 43 uncommitted from other threads, layer 0.9.35. 17,175 files, 777 MB; Cloudflare Pages caps a deployment at 20,000 files, so the next large image set needs a plan (R2, or fewer derivatives).
    • /library had no inbound link, so the static crawl from / would have skipped it. 7b2668823f links it from the /archive intro and adds it to nitro.prerender.routes.
    • Verified on design-f39 (headless, system Chrome): /library, its data, previews and artboard images all return 200. The QPT.index-v20.ai deep link renders 7 artboards in ~0.35 s, a click opens the lightbox with ?a=3, and the quick filters give 45 .ai presentation files. No page errors.
    • The build's error lines are older ones (todos cycle-ID 404s, /components/data/data-timeline 500).
  • Verified headless at 1440 × 900: QPT.index-v20.ai (one row, canvas order 2, 1, 3…), the 46-artboard dashboard history board (clusters with their title artboards), the lightbox, arrow keys and Escape with the URL following, and a partial file's note. No page errors.

• Mark's lab pass, 2026-10-23 (7c804d990): the Policy Config submenu reads Config File · History · Proposed Changes. The publishing tray closes to a chip, and staging an edit reopens it. Lab links no longer carry ?dp, so every page opens at phase 4 with exceptions shown. The mock bar folds its knobs behind "controls ▸" and names any that are off their default. These shipped uncommitted in the 2026-10-27 deploy.

• kick-off (2026-10-21): cycle created. AI governance replaces data governance as the textbook case in the wicked-problems section, with the three properties restated for AI use. The morning's refocus paragraph becomes one line of history: the framing began with data governance in web-app back-end traffic, and the method didn't change when the focus did. Product and Brand in Forming the Problem name AI governance. /method matches. The foundation's §07 sentence and the CLAUDE.md summary follow. The appendix gets a one-line note. Uncommitted.

• callout (2026-10-21): "Qpoint's core value proposition is an act of design: giving a shapeless problem tangible form" read awkwardly, and so did a first rewrite ("Qpoint does what a designer does first…"). After six drafts, Mark wrote the line himself: the /method callout now reads "We use this method to understand and deconstruct the AI ecosystem's wicked problems into well-defined micro problems resolvable by product-shaped solutions." The bold lead and the slogans after it are gone. Foundation §07 is back to its original sentence for now. Uncommitted.

Outcomes (2026-10-21): closed briefs now archive with what they produced. outcome front matter (summary, postUrl, published, hero, listing, screenshot, screenshotFull, sketches, sketchesNote) in the schema; /briefs renames the done group Outcomes and shows the listing image on the card; /briefs/<slug> renders an Outcome section above the stepper (hero + listing, article screenshot opening to the full page, sketch grid, one UxLightbox across all of them). First archived: your-agents-need-a-dynamic-sandbox (live 2026-10-15), with the three sketch sheets from the Design drive, the live hero/listing, and scripts/capture.mts shots of the post in public/briefs/your-agents-need-a-dynamic-sandbox/. Playwright's browser wasn't installed; PW_CHROMIUM_PATH pointed at Chrome.

• build (2026-10-20): Mark re-exported five masters as logo-1…5.svg. All are clean paths: two in grape, three duotone (a grape mark with an unfilled wordmark, which renders black, and the two never overlap). They're saved as symbol, stacked-caps, stacked, lockup and lockup-caps; brand/logos/logos.json holds the names, labels and Drive source files. scripts/build-logos.mjs (npm run logos) builds 5 logos × 3 colors × 4 formats = 60 files, 2.1 MB, in public/brand/logos/, plus app/data/logo-downloads.json. A Download section on /brand/logo offers logo cards, a Format and a Color UxTabGroup, a preview of the actual file on its matte, and the filename, size and Download. Checked: contact sheet by eye, all 60 files serve 200 with the right MIME types, dark mode, 375 px. The site nav overflowing at phone width predates this cycle, as does a hydration warning also on /brand/colors. Uncommitted.

• tweaks (2026-10-20): the pickers are now labelled "Select a logo / format / size / color". PNG, GIF and JPG come in four sizes, 256, 512, 1024 and 2048 px on the long edge, each rendered straight from the vector. The size picker shows only for those three formats and defaults to 1024. Files are named qpoint-<id>-<color>-<size>.<ext>; SVG keeps qpoint-<id>-<color>.svg. There are now 195 files, 3.7 MB, and all serve 200. Uncommitted.

• Sources on the design site (2026-10-20):
  • Two open Deutsch papers (arXiv; author PDF) were added, along with a Deutsch and Popper source list.
  • Seven photos of Roesler, Rittel and Webber were added for internal use. Rittel and Webber exist online only at low resolution.
  • /sources renders the bibliographies. /sources/files/ serves the PDFs and photos (nitro publicAssets), and /method links there.

c118 · the Oct 19 model in the lab (2026-10-20): c118 (policy structure and publishing) now draws the 2026-10-19 product session's landings in this cycle's lab: RULED runs the Oct 19 world (policy mode Off / Observe / Enforce, mode and advanced exceptions by org / team / user context with the narrowest winning, a policy snooze, resolve with a reason, no stored groups or per-scope enforcement), and C114_RULED keeps c114 as ruled through 2026-10-18 for the Phase 3 record pages and as the model page's alternative. The live Policies copy's toggle became a three-state Mode column. The record and the change list are in c118's lab-plan.md (Log); this cycle's rulings D0–D9 are unchanged.

composed S1 · Finding detail (2026-10-19): built from Mark's Illustrator screenshot (illustrator/c114-s1-finding-detail-design.webp) and his notes: "the table should be our uxTable with the appropriate header", "details strip, let's try to re-use any components like the framework item", "upper right–actions section … here's where the actions go", "session - re-use the existing transcript component", "related findings - related means two things here: any findings from the same session, any findings related to the same policy this finding violated". (prompt · plan)

• product sessions + first decision brief (2026-10-20): Mark asked for the latest product meetings (Rob's MWF "Qpoint Product" series), the Oct 19 one parsed first into a summary of its decisions for design.
  • Getting the calls: the series was invisible to the API, because Avoma hides private meetings from the org key, even by uuid. Rob set the series to Organization, and Oct 12, 14 and 19 then came through by --uuid (Oct 16 was cancelled).
  • Triage: all three declared ciso-advisory × roadmap-planning, as on Sep 3. Oct 19 declares the shown mocks design/phase4/{changeset-summary,policy-detail,finding-detail}.
  • Extraction: all PASSED, with session records of 5/9/6 decisions and 3 candidates each. 25 candidates now await review across the corpus. The index has 10 calls.
  • New brief kind decision: an internal session's rulings, what each changes in the design records, and a list of the workshopped ideas that were superseded so they don't get revived. Added to INDEX conventions.
  • First brief of that kind: briefs/policy-model-and-publishing.md, covering where Oct 19 landed on policy structure, findings, publishing and the dev server. Mark's steer: later statements win over workshopping.
  • Design picked it up as c118.policy-structure-and-publishing (initial prompt; design commits d1b6744f8, 1cb125e18). npm run consumers finds c118 against four problems.
  • Commits: 5c2964e, 17a62d0.

• product-input thread (2026-10-20, plan only): Mark asked how far market framing muddies agent context now that the project mostly feeds product work, and how well it serves product development. I read the agent-facing prompts and skills, the profile and standing code, and the Oct 12/14/19 extractions, then wrote product-input-plan.md (prompt: product-input-prompt.md).
  • Findings: role lines and schema order are still market-first. Product sessions get interest, urgency and positioning, and an advisor's design preference was captured as a value echo. The 44 session decisions have no cross-session state; Oct 14 rulings overturned on Oct 19 both read made. The brief template lacks the decision kind.
  • Recomputed standing without the October sessions: no change. Moto's support comes from Aug 20/31 and Sep 3, so the co-designer fix is preventive.
  • Five changes planned: (A) role lines and ordering; (B) call-type rules that omit market blocks; (C) a design-session call type whose extraction-tagged items count as design input rather than standing, with Oct 12/14/19 re-triaged and re-extracted; (D) a curated decisions ledger (data/decisions.json, statuses ruled / direction / open / tabled / superseded / rejected, supersedes chains, extraction tags decisionId + reaffirms / revises / reverses, /decisions page, decision:<id> consumer tokens); (E) brief template and skill drift.
  • Six decisions queued for Mark (plan §5). Nothing built.