Network Effect · Local Critical Mass

What this page is / is not
· This page is a “condition-preparation tool”: it tells you whether this user relationship is a real network effect, whether this one circle is dense enough, which condition is missing, how to fix it and what the first step is. It is not investment advice.
· The thresholds come from an internal methodology document 《经济盈利算法·09 网络效应算法 v1.0》 (domestic machine /opt/inventory/econ_algorithms/09-网络效应算法.md, 161 lines, finalised 2026-09-14, read-only), and have not been through any external empirical test. The document itself writes in §4 Step2: “these are the practical thresholds this algorithm gives, not an academic conclusion; industries differ enormously, and you must calibrate them against your own 2–4 weeks of real data”, and §9 (limits) repeats it.

Evidence chain (the md document has no line numbers, so sections are cited)
· The three real-or-not questions and “any ‘no’ means it is not a network effect” → 09-网络效应算法.md §4 Step1
· Penetration = users in the circle ÷ the circle's target headcount; the three practical reference lines 10% / 30% → §4 Step2
· Pairing success rate = successful pairings ÷ demands posted; the two lines 20% / 60% → §4 Step3
· The three negative-effect signals and “any one appears → expansion pauses” → §4 Step4; circles in parallel ≤ 2 → §4 Step5
· The six N1–N6 start criteria and their thresholds (≥ 30 active / ≥ 20 on each side / month-2 retention ≥ 20% (15% for content-community) / ≥ 2 times a week on average / free side ≤ 30% of the paid side) → §3
· The condition-preparation checklist and the “first step”, verbatim → §5.2 / §5.3

How this differs from the “network effect” slice already live (de-duplication note)
· This page comes from the md document 《09-网络效应算法.md》 and judges “penetration / real or not / pairing success rate / condition preparation inside one minimum unit” —— the inputs are the three real-or-not questions, the unit's headcounts, and retention and frequency.
· The live network-effect slice (/slices/network-effect.html) comes from /opt/algorithms/functional/network_effect.py and computes V = c×N^α / c×N^β and the Metcalfe value avg×N²÷2 —— the inputs are total users and connections, and the output is a scale-value comparison table.
· Conclusion: the inputs, the formulas and the outputs of the two do not overlap at all, so this page is a separate slice under the new id network-effect-econ, not a rebuild of that page. No formula has been carried over from that page, and that page needs no changes either.

Unaligned / simplified (stated plainly)
· Simplified ①: in the document's §4 Step4 the negative effects are checked as three separate signals, and this page merges the three into one dropdown “none / at least one” —— the deciding action (pause expansion as soon as any one appears) matches the document, but you cannot tell which one it is.
· Simplified ②: in N3 “cross-side: ≥ 20 on each side and able to pair”, the “able to pair” part is evidenced here by “demands posted > 0”; the document gives no quantitative standard for “able to pair”, so this step is the page's own placement.
· Unconfirmed: the N4 “content / community relaxed to 15%” is decided by you manually picking a category; the document gives no criteria for “which businesses count as content / community”.
· Not done: the document's §7 suggests “have an LLM make the minimum unit concrete (what the local circle is called, where to find people)” —— this page is pure front-end and has no model connected at all.
· Not done: the document's §6 “cross-check the minimum unit's target headcount / a competitor's share against external data” is not collected on this page — everything rests on you filling it in by hand; for numbers you cannot get (public statistics, for example), label them yourself as estimates.

[Evidence section · where to check every number]
① Source: internal methodology document 《09-网络效应算法.md》 v1.0 (domestic machine /opt/inventory/econ_algorithms/09-网络效应算法.md, 161 lines, read-only): start criteria N1–N6 = §3; real-or-not / penetration / pairing / negative effects / pace = §4 Step1–Step5; preparation checklist and first step = §5.2 / §5.3. References are in §1 (the Katz & Shapiro 1985 AER original was not matched on Crossref and is flagged as such; Rochet & Tirole 2003, DOI 10.1162/154247603322493212).
② Three categories
· [Verifiable] The three real-or-not questions, 渗透率 = 圈内用户数 ÷ 目标人数 and the three bands <10% / 10–30% / >30%, 配对成功率 = 成功配对 ÷ 需求发布 and the three bands <20% / 20–60% / >60%, the three negative-effect signals, circles in parallel ≤2, N3's ≥30 active and ≥20 on each side, N4's 20% retention (15% for community) and N6's free side ≤ 30% of the paid side —— all of it lines up with §3 / §4.
· [This page's display convention] The dropdowns for the three real-or-not questions, the “how many did you hit” summary and the three-in-one negative-effect tier are the page's paraphrase; this page does not compute any “users² = value” formula (the document forbids treating Metcalfe as a theorem).
· [No basis found] 10% / 30% / 20% (retention) / 60% / 20 people / 30 people / 30% (the subsidy line): the document's own §4 Step2 states “these are the practical thresholds this algorithm gives, not an academic conclusion”, and §9 repeats that you must calibrate them with your own 2–4 weeks of data → no external basis was found. Also: critical mass itself “has no single authoritative source”, and the document uses it as a general concept.
③ How to check it yourself: a unit with a target of 100 people and 35 already there → penetration 35% > 30% → it should be judged self-accelerating phase; change the existing number to 8 → 8% < 10% → it should be judged cold-start phase. On the pairing side: 10 demands posted, 5 successful pairings → 50% → it should land in the healthy band.