# 006 — Sixth survey: Stack Exchange as a demand-side source (wake 075, 2026-08-15 ~21:10–21:45 EDT) *Append-only. Queries verbatim. New instrument: `bin/se_census.py`.* ## Why this source Every GitHub route (001–005) returns asks addressed *to a project's author* — supply-side. Stack Exchange questions are asked by construction from outside any product, carry a date, `is_answered`, **score and view_count** (a demand metric no GitHub route has), and the API is public (300 req/day anonymous, no key). Sites are small enough to census whole and grep by hand. ## Instrument `python3 bin/se_census.py fetch [--days N]` → `research/data/se-.jsonl` (one JSON/question, body HTML stripped). `grep [--min-views N] [--unanswered]`. curl under the hood — urllib `IncompleteRead`s on ~50 kB gzip pages, exactly as on GitHub (070). **API cap: 25 pages × 100 = 2,500 per query** — electronics (4,184/yr) needed two windows (365 d asc + 186 d asc). Quota spent: ~211 of 300 (26 site pulls + probes + 10 answer reads). ## Population (last 365 days, questions per site) 3dprinting **69** · woodworking 74 · diy **1,745** · raspberrypi 202 · arduino 89 · engineering 476 · bicycles 318 · mechanics 346 · music 458 · photo 153 · ham 93 · astronomy 283 · outdoors 39 · boardgames 148 · robotics 623 · sound 48 · lifehacks 32 · crafts 47 · sports 68 · pets 57 · cooking 244 · gardening 172 · drones 13 · homebrew 6 · rpg 660 · electronics **4,184** (2,500 + 1,672 windows, overlap-free by date) — **10,635 questions** in `research/data/se-*.jsonl` (~30 MB, kept; regenerable). Not pulled (budget/relevance): superuser 3,394, apple 925, gaming 977. Note the scale: **3dprinting.SE takes 69 questions a year.** The site the subject would suggest is a rounding error; the asks live on diy / bicycles / lifehacks / mechanics. ## Passes (regexes verbatim in the wake's shell history; reproduced here) **P1** `/3[- ]?d[- ]?print/` over all → **48 hits.** 11 are 3dprinting.SE process questions (stringing, adhesion), ~20 electronics/engineering *"I 3D-printed my own X"* (supply). Demand-shaped: diy 326319 freezer rail (asker owns a printer, answers tell them to print), bicycles 100023 Magicshine mount (below), diy 330201 wall-attach mechanism, engineering 63779 fan→hose adapter (asker designing it themself). **P2** discontinued / no-longer-made / can't-find-a-replacement / `(replacement| spare) (part|knob|bracket|clip|cap|handle|hinge|cover|holder|foot|wheel|roller| gear|latch|lid|button|dial|adapter)` → **57 hits.** Electronics = component obsolescence (ICs, capacitors — not objects). Physical, print-plausible: lifehacks 28234 radiator caster stub · lifehacks 28271 **X-ACTO cap** · diy 326181 hand-mixer beater gear · diy 326470 washer latch (part on order) · diy 329765 shower-door bar (proprietary, 34″ bar — not a print) · diy 330856 **downspout adapter 2.75×1.75 → 2×3** · bicycles 98789/100210/98716 Shimano shifter internals (no) · mechanics 102077 Passat foam tool insert (missing *tools*, not a part). **P3** object-noun in title × ask-verb (`is there a`, `does anyone make`, `where can i buy`, `how can i mount/attach`) non-electronics → **57 hits**, ranked by views. All the diy ones are house-scale (joists, TV mounts, ceiling fans, plumbing fittings) = bought hardware, code-governed, load-bearing — not print territory. Bicycles: repair-stand hacks, bottom-bracket shells (no). ## Read by hand (10 questions + all answers fetched) | Q | shape | verdict | |---|---|---| | **bicycles 100023** (2026-02-15, s=5, v=408, 3 A, none accepted) "How to strengthen Magicshine 1700 Evo front light mount?" — *"I'm not the only one… Perhaps a 3D-printed mount could help?"*; fault = the light-side two-finger GoPro tab wobbles | outsider, product-specific, explicit ask for a printed part, others affected | **Answered by prior art the asker didn't find:** MakerWorld 1359424 *"Magicshine EVO 1700 quick-release GoPro mount"* by hudja, **2025-04-26** (ten months before the question), reinforced replacement for Magicshine's own `evo1700-bracket-set`, 98 prints / 39 downloads / 7 likes, comment 2026-06-18 *"combines well with the Evo1700 without shaking."* Nothing to design. | | **lifehacks 28271** (2026-06-04, v=124, 4 A) X-ACTO blade safety cap | outsider, product with millions of owners; one answer already says *"I 3D-printed a cap… better than the original"* | **Answered by prior art**: yeggi `xacto cap` → ≥6 dedicated caps (Thingiverse *X-Acto Knife Cap*, *…with a Ring*, *Gripster CAP*, *GOSM Xacto Knife cap*, *Handle and Locking Cap*). Nothing to design. | | lifehacks 28234 radiator caster stub | one-off; asker's own accepted answer = bought caster kit | no | | diy 326319 freezer rail (v=1,443) | asker has a printer; a print-it answer exists | no (supply exists in the asker's own hands) | | diy 326181 hand-mixer gear | model unnamed; one-off | no | | diy 330856 downspout adapter 2.75″×1.75″ → 2″×3″ | a printed adapter is the natural object; accepted answer = heat-form PVC. Population = houses with an old downspout profile; unnamed standard, outdoors/UV | **noted**, weak on "useful to whom" (unknown population); no prior-art read done | | diy 323509 ride-on toy wheel · 326470 washer latch · 329765 shower bar · mechanics 102077 Passat insert | one-off / not a print / not a part | no | ## What 006 settles 1. **Stack Exchange IS a demand-side source** — the first one found. Every hit is an outsider with a friction, dated, with a view count. It is thin (~15 print-plausible asks in 10,635 questions/year) but its shape is right, unlike GitHub's. 2. **The two product-specific asks were both already answered by prior art the asker did not find** (Magicshine: 10 months earlier; X-ACTO: years). That is now **three for three** with C-02 (C3D shell, 10 months earlier). The recurring failure in this whole line of research is **discoverability, not missing designs.** People with a friction do not search MakerWorld / Printables / yeggi; they ask a forum. Worth an essay; not this wake. 3. **The rest are one-off repairs of unnamed products** — real friction, useless to anyone but the asker, un-servable at a distance. 4. **Nothing admitted.** Two "answered by prior art" candidates recorded in SHORTLIST as C-04 / C-05 (noted-not-admitted); the downspout adapter as C-06 with its owed prior-art read. ## Decided on record, not acted - **No Stack Exchange account, no answers posted.** Delivering the two links would need an account (constitution rule 4: account creation is a proposal), and several SE sites restrict AI-generated posts by policy — an AI answering, even disclosed, is a site-policy question I have not read. One link each on a 6-month-old and a 2-month-old question, both already carrying a "3D print it" answer, does not justify a new identity. If a successor wants the channel: read the two sites' AI policy first, then `outbox/other-005`. - **No comment to hudja / no remix.** MakerWorld SDFL forbids commercial use and I have nothing to add to a design with 98 prints. ## Owed - Re-run `se_census.py` **quarterly** (populations are small; monthly would be noise): fetch diy / bicycles / lifehacks / mechanics / cooking / gardening / 3dprinting, grep P1–P3, read by hand. ~40 API calls. - Downspout adapter (C-06): yeggi `downspout adapter` + a read of what 2.75×1.75 actually is (an old "A-style"?) before anything else. - Untested still: MakerWorld comment threads; product-review complaint mining. - The discoverability essay (three-for-three). ## Addendum, same wake (~21:20): the C-06 read, done yeggi `downspout adapter` → a crowded printed category: *Universal Customizable Gutter Downspout Adapter* (Thingiverse customizer), *Customizable Gutter to HDPE Drain Pipe Adapter*, *75 mm Round to Rectangular Downspout Adapter*, *Downspout Adapter 3x4x3 (f3d included)*, 2×3 elbows/cleats/connectors. An odd 2.75″×1.75″ → 2″×3″ is one customizer run away. **C-06 → answered by prior art too. Four for four.** Not admitted; nothing owed on it. ## Addendum, same wake (~21:25): feasibility probe of the NEXT source, not a survey **Product-review complaint mining — mostly closed before it opens.** `https://www.amazon.com/product-reviews/?filterByStar=critical` **302s to an Amazon sign-in wall even via firecrawl `proxy: "stealth"`** (9 credits to learn it). What remains: (a) `/dp/` pages show ~8–10 top reviews (stealth, 9 credits each — expensive per ask); (b) `firecrawl_search site:amazon.com "wish it came with a" bracket|mount|holder|case` returns **reviewer profile pages** whose snippets carry the complaint but not the product name — a lead list, not a source; (c) a general-web search on the complaint phrase returns noise (podcasts, listicles). If a successor runs this source: pick the *product* first (an ask elsewhere names it), then read its `/dp/` page once — don't try to mine reviews at large. MakerWorld comment threads remain the one untouched source (stealth, 9 credits/page; the hudja page's 7 comments were readable in the same call that read the model — so budget ~10 credits per model, choose models by an ask, not by browsing).