Technology content — critical review

Source: technologies/*/README.md · technologies/_source/pack-wiki-audit.json · _catalog/gaps.json

Verdict

Mechanical DoD looked green (131/131 deep). Editorial truth was not. Roughly three-quarters of packs shared a mass-deepen template while wearing a deep badge. Trust early hand waves (~36); re-open deepen for the mass pass. Start from [`technologies`](/technologies) and the peer-cell stacks (e.g. `agentic-evaluations`, `agentic-frameworks`).

131

Packs marked deep

Share template body

Corrupted What-it-is

Hand-written Why

Quality split

Template Why/How/When/Limits
95
Hand-written Why
36
Corrupted What-it-is
41
Generic In→Comp→Out mermaid
95
How list-only (warning)
12
Suspicious Primary evidence
58

Counts from regex over FS READMEs on the review day (same day as the closed deepen ledger).

Findings by severity

False deep — new template passed the harness
P0
95/131 packs share the same Why/How/When/Limits and In→Comp→Out mermaid. `pack_wiki_audit` bans old landscape phrases only; LOOP_TECH_DEEPEN closed at 131/131 PASS while this paste is green.
What it is is often scraped garbage
P0
~41 packs glue duplicate blurbs, HTML badge scraps, or truncated sentences (e.g. `agentscope` alt= tags; `kubernetes` ends mid-phrase “across multiple.”; `langchain` restates itself three ways).
Evidence is theatre, not grounding
P0
386 MODERATE vs 5 STRONG. ~58 “Primary / product role” rows point at peer-contrast or unrelated files (`autogpt`→autogen-contrast; `kubernetes`→dapr-contrast; `langchain`→dspy-related-work with mislabeled quotes).
Broken primary links
P1
`kubernetes` URL and Docs both point at https://agentic-community.com — not k8s docs. Likely more soft link rot in Sources headers that still cite landscape only.
Coverage skew vs OS substrate
P1
Frameworks=26, Runtime=1 (`kubernetes` only), Skill Orchestration empty. No Commercial license tags. Landscape harvest was 95; +36 extras. Missing substrate the product actually needs (Postgres/SQLite, OTel, Docker, IMAP observation path, OpenAI Agents SDK, Pinecone, etc.).
Catalog gaps still open
P2
78 technologies missing document-management.md (G_DM_CROSSWALK). 12 packs still W_HOW_LIST_ONLY (evals + memory cohort, e.g. `ragas`, `promptfoo`, `mem0`) — warning only.

Cell distribution

Runtime is a single pack. Skill Orchestration wave closed with 0 packs.

Frameworks
26
Workflow Orchestration
9
Vector DBs
9
RAG
9
Evaluations
8
Observability
7
Agentic Gateways
6
LLM Gateway
6
Serving
6
Coding Agents
5
Search & Web Access
5
IAM
4
Guardrails
4
Structured Output
4
Memory
4
Voice Agents
4
Knowledge Graphs
3
Tool Integration
3
Fine-tuning
3
Code Sandbox
3
Agent Protocols
2
Runtime
1

Evidence strength stamp

MODERATE
386
STRONG
5
ANECDOTAL
2

pack_wiki_audit checked section presence, banned old phrases, and evidence file existence — not product-specificity, link correctness, or Primary↔slug match.

Strength	Rows
MODERATE	386
STRONG	5
ANECDOTAL	2

Concrete wrong examples

Slug	What is wrong
langchain	Triplicate What; generic Why/How; Primary evidence = dspy-related-work
kubernetes	Truncated What; URL/Docs = agentic-community.com; Primary = dapr-contrast
agentscope	HTML badge debris in What; full template body
autogpt	Primary evidence file is autogen-contrast.md

Packs that look actually deep

Sample of zero/low template-score packs (early waves + protocols / memory / guardrails). Not an exhaustive gold list — spot-check before citing.

What to do next (if reopened)

Extend pack_wiki_audit: fail on the new template phrases, generic mermaid, URL host ≠ product, Primary evidence filename mismatch, HTML debris in What.
Demote the 95 templated packs to stub (or a third status) until rewritten — do not leave Wiki-status: deep.
Rewrite waves by subcategory starting with Frameworks (largest + most templated) and Runtime/Links — begin with `agentic-frameworks` and `kubernetes`.
Decide substrate coverage deliberately (DB, OTel, observe-email) vs staying landscape-bound — absence is fine if labeled off-path.

Proof at review: python regex over technologies/*/README.md; pack-wiki-audit.json passing=131; gaps.json G_DM_CROSSWALK tech=78. Page is a frozen snapshot of that review, not a live dashboard.