Topic leaf pages remain thin after the 01:28 guide+FAQ4 pass; hub IG must be cross-tool decision tables, workflows, and cluster examples—not more routing prose.

Date: 2026-09-02 09:48 Summary: Topic leaf pages remain thin after the 01:28 guide+FAQ4 pass; hub IG must be cross-tool decision tables, workflows, and cluster examples—not more routing prose. Visibility: people

[question] 目前的topic页面还是太薄，进行仔细分析，给出info gain方案

[try to solve]

Context

The site has 18 topic leaf URLs at /topics/{id} plus /topics hub. Catalog is 162 tools. A previous pass (2026-09-02 01:28) added a guide section (3–6 paragraphs mapping jobs to tools) and a fourth FAQ in ten locales. The user still judged the pages too thin.

Site strategy already required pillar pages to include a scenario comparison table (docs/2026-07-28-google-seo-strategy-implementation.md §4). That table was never built. The 01:28 pass added more prose in the same template instead.

Process

Inventoried leaf template in scripts/site/components/content-topics.mjs: intro → choose → limits → guide → primary card grid → optional secondary grid → FAQ×4.
Counted English unique editorial words (intro+choose+limits+guide+4 FAQs, excluding card blurbs that copy homepage homeDesc) and membership from tool-catalog.json.
Compared copy against tool-page IG (§3.1 nine dimensions), scaled-content policy, and the unused §4 pillar spec.
Listed every primary/secondary slug so hub modules can name real tools, not generic product types.

Audit (English unique editorial, 2026-09-02)

Topic	Primary	Secondary	Total tools	Editorial words	Guide paragraphs
health-body	8	0	8	439	4
ai-prompt-template-builder	9	0	9	436	6
personal-money	12	3	15	403	4
text-compare	5	0	5	402	5
business-finance	10	5	15	375	4
math-stats	14	2	16	354	4
time-calendar	4	1	5	329	4
image-media	27	5	32	314	5
pdf-docs	13	5	18	311	5
network-ip	11	0	11	307	5
geometry-measure	8	1	9	305	3
design-brand	4	2	6	301	4
seo-growth	11	7	18	299	4
dev-data	9	12	21	296	5
security-ids	4	6	10	292	5
science-physics	5	2	7	284	5
home-diy	4	3	7	284	3
files-archives	4	2	6	282	4

Worst mismatch: image-media (32 tools, 314 words) and pdf-docs / seo-growth / dev-data (18–21 tools, ~300 words). Tightest clusters (text-compare, prompt, health) are relatively less thin because scene→URL mapping already exists, but they still lack tables, worked cluster examples, and citations.

Root cause / analysis

Hub thinness is not “too few FAQ items.” It is a wrong IG unit.

Tool-page IG answers one job (formula, one Input→Output, that tool’s failure modes). Hub IG must answer questions that cannot live on a single tool page: which of N tools, in what order, with what inputs, and when to leave the topic.

What the current leaves actually do:

Template isomorphism. All 18 pages share the same seven sections. Combined with 10 locales, this is the scaled-content shape Google’s spam policy targets if unique facts are missing.
choose ≈ guide ≈ intro. Three sections restate “pick by job” at different lengths. No new verifiable fact.
Guide is a directory in prose. “Open BMI for screening; open TDEE for energy” restates tool titles. Guide body has no <a href="/tools/{slug}">.
Card grid is duplicate catalog. renderToolCardGrid reprints homeTitle + homeDesc already on / and on the tool page. Zero hub-unique sentences.
FAQ answers taxonomy, not jobs. Several answers explain why a slug is primary vs secondary (internal IA). Users searching the topic do not need that.
Missing the one artifact §4 already named: scenario comparison table.
Zero of tool IG dimensions 1, 4, 5, 8 at hub level: no formulas/rules unique to the cluster, no comparison table, no clickable WHO/NIST/WCAG/IETF citations, no worked numbers.
Secondary dumps. dev-data has 12 secondary slugs with only a heading “Also listed here.” Looks like stuffing unless each row states why.
Hub /topics is thinner still: label + blurb + count. Acceptable as a navigator if leaves are thick; not a ranking target.

Word-count padding would not fix this. A 900-word isomorphic essay is still thin if it does not add a table, a pipeline, or a number another SERP hub lacks.

Solution

Hub IG quota (distinct from tool nine-dim)

A topic leaf may ship only if it hits ≥4 of 6, and #1 is mandatory:

#	Dimension	Hub artifact
1	Decision table (mandatory)	Job × tool: inputs, default, when not to use, link to `/tools/{slug}`
2	Workflow	One ordered pipeline of ≥2 tools (e.g. merge → organize → compress → protect)
3	Cluster example	One worked number or file pair that spans 2+ tools
4	Cross-topic boundary	“This topic does not do X → `/topics/{other}`” with links
5	Trust block	Cluster privacy (local vs upload) and, for YMYL, ≥1 clickable authority URL
6	Secondary rationale	Each secondary slug has a one-line “why this job”, or do not list it

Do not copy How/Example/Formula from child tool pages onto the hub (duplicate + scaled). Hub text must be routing facts the child pages cannot own.

Page architecture

Keep: short intro, limits (strengthen citations), primary list, FAQ.

Merge/kill: choose into the table; guide paragraphs become table + workflow (stop the third restatement).

Add: decision <table>, workflow <ol> with tool links, cluster example, related-topics strip, secondary “why” column.

Cards: keep the grid for humans, but each card’s hub blurb should be a best-for line from the table, not homeDesc.

Large topics (image-media, pdf-docs, seo-growth, dev-data, math-stats) must subgroup the grid (platform specs / transform / compose / export) so 27 cards are not an undifferentiated dump.

FAQ: 4–6 items that the table does not already answer (legal, YMYL, privacy, “vs desktop suite”). Drop IA questions (“why is this listed here”).

Search intent: hubs support in-site routing and long-tail “which tool” queries. Do not retarget head terms (pdf tools, image editor online) as H1; that contradicts the long-tail gap strategy and doorway policy.

Per-topic modules (authoritative sketch)

health-body (YMYL): Table screening (BMI, ideal weight) vs energy (BMR/TDEE, macros) vs composition (body fat) vs training (HR, pace, 1RM) with required inputs. Example: 70 kg / 175 cm / 30y / moderately active → BMI then TDEE then macros—labeled educational. Citations: WHO BMI adult classification; ACSM-style HR zone caveat. Boundary: not diagnosis.

personal-money (YMYL): Checkout (tip, discount, VAT) vs pay (salary↔hourly) vs borrow (EMI, debt payoff) vs growth (compound, 72, PV, CAGR, inflation, % change). Example: 12% APR, 5y, $10k → EMI vs extra-payment payoff order. Citations: CFPB / local tax authority for VAT as illustration only.

business-finance: Unit economics (markup vs three margin flavors) vs break-even vs capital (ROI, payback, NPV, IRR). Example: cost $40, price $80 → 100% markup ≠ 50% margin; then break-even units. Boundary: not IFRS statements.

math-stats: Center/spread vs inference vs counting vs algebra/notation. Example: {2,3,100} mean vs median; then when z-score is the wrong next step. Boundary: not mixed models.

geometry-measure vs home-diy: Geometry = named shape formulas; DIY = coverage + waste + bags. Example: 12×10×0.33 ft slab → volume tool then concrete bags. Cross-link both hubs.

science-physics: Kinematics / Ohm / density / field helper. Example: 12 V, 3 A → 4 Ω; disclaimer not house wiring. Citations: NIST SI where relevant.

time-calendar: Age vs date-difference vs IANA zone vs Unix s vs ms. Example: 10-digit vs 13-digit epoch. Boundary: business-day calendars out of scope.

pdf-docs: Structure (merge/split/organize/rotate/page numbers) vs size (compress/crop) vs access (protect/unlock) vs export (MD/JPG/sizes) vs watermark. Workflow: merge → organize → compress → protect. Example: 12-page pack, email 8 MB limit. Privacy: local processing cluster statement.

image-media (highest priority): Four sub-tables—platform pixel specs (YouTube, IG, LinkedIn, OG, Amazon, App Store); transform (compress/optimizer/crop/flip/border/grayscale/format/SVG); compose (merge/overlay/GIF/watermark); export (PDF/PPT/Word, Base64, EXIF). Workflow: crop to spec → convert format → compress. Example: 4K PNG hero → WebP byte budget. Do not treat 27 cards as one list.

design-brand: Contrast gate → palette → tokens → style guide. Example: #1a73e8 on white vs AA/AAA. Citation: WCAG 2.2 contrast. Upstream: color-from-image is secondary.

seo-growth: Snippet vs technical files vs crawl notify vs CWV vs UTM. Workflow: visible copy → meta/OG → sitemap → IndexNow. Honest limit: no rank guarantee; FAQ/HowTo rich results retired. Secondary image compressors only as LCP helpers.

network-ip: Identity / DNS / headers / CIDR family (cheat sheet, v4 range, v6, private, Terraform, AWS VPC). Example: /24 → 256 addresses, 254 hosts (standard caveat). Authorized testing only. Citations: RFC 1918 / 4632 as appropriate.

dev-data: Encode vs structure (YAML/CSV/JSON Schema) vs HTTP/debug vs docs pipeline. Secondary compare/time tools listed only with “why.” Example: invalid JSON fails schema then diff is meaningless.

ai-prompt-template-builder: Artifact table (template vs writing vs MJ vs sketch vs film vs short drama vs product vs Android vs iOS) with primary field difference. Trust: local default vs Turnstile+Workers AI. Explicit: no per-model URL.

text-compare: Input-object table (paste / txt / docx / JSON / xlsx) already sketched in the topic brief—promote it to HTML <table> with loader rules and “when not.” Example: same two strings on paste vs file page.

security-ids: Generate (password, UUID) vs integrity (hash, metadata) vs inspect (JWT secondary). Example: UUID v4 is not a session authenticator. Never paste live cookies.

files-archives: Create vs unzip/extract vs magnet parse. Example: magnet URI does not start a download. Hash as secondary verify step.

Implementation notes (when building)

Source of truth: structured topic modules (table rows with slug ids) in build scripts, rendered to HTML—not a third isomorphic prose blob.
Lint: fail leaf if table missing, if a primary slug is omitted, if a named tool has no /tools/ link, if YMYL topic lacks a citation URL, if secondary has no why-line.
i18n: ten-locale retrieval rewrite, not EN template fill. Tables especially need local job language.
Schema: keep CollectionPage + ItemList + BreadcrumbList. Do not add FAQPage for rich-result KPI (retired 2026-05). Visible FAQ stays for humans.
Do not split 18 topics into more hub URLs. Do not noindex leaves unless a later GSC pass shows persistent crawl-not-index after thickening.

Notes / boundaries

Information Gain here is a content principle aligned with Helpful Content, not a confirmed ranking switch.
Hub copy must not explain SEO/doorway/thin-URL engineering to end users (localization visible-copy rule).
Indexing still depends on crawl + snippet-eligible people-first pages over time.
This log is the plan; code was not changed in this turn.

[actions]

Analysis only: catalog 162 tools, 18 leaves, editorial word counts, hub IG quota and per-topic modules (no site build)

Topic leaf pages remain thin after the 01:28 guide+FAQ4 pass; hub IG must be cross-tool decision tables, workflows, and cluster examples—not more routing prose. | Dev Logs