Leverage

The verdict comes from LLM inference, not from a deterministic algorithm (stated plainly)
· What is fixed: the source document's four core formulas/laws, the three states, the four live-tentacle tests, the four kinds of hanging, the four hanging-timeline bands, the five steps for handling a dead tentacle, the four quality tiers, the three expansion principles and the external-data requirement — all of these are rules copied out of 14-风雨方法论·八爪鱼算法-v1.md into the prompt and hard-coded in the proxy code; they do not change with the model.
· What the model generates: your list of tentacles, whether each tentacle has all three elements, its state, which element is missing, what kind of hanging it is, how to fix it, the possible space and the first step — all of that is produced by the model on the spot. Ask the same input again on another call and both the list and the verdicts can differ.
· So it can be wrong: the model does not know your real situation and may overstate your capability, overstate external demand, or call an unreachable path reachable. Treat this page as a questioning tool, not as a verdict tool.
· Which part deserves the least trust: when the model calls a tentacle “✅ live”, check it yourself against the source's four tests (does money really come in? do customers really come back? is it more than a one-off? can it be replicated at scale?) — “live” is something you compute, not something you declare.
· No numeric validation is done at all: this page involves no amounts or ratios, no number is computed by an engine — every number is something the model said.
· The Reverse Octopus section is a summary, not a full port: only Machine 9's five-star moat rating and three verification questions were carried over; moat decay, maintenance frequency and the four stages of the double helix were not — do not use it as Machine 9's full conclusion.
· The key is not in this page: the page only calls the same-origin /llm/leverage, and the key lives in the proxy process's environment variable LLM_API_KEY, so the front end cannot see it and it is not in the page source. When the proxy is not connected, the page says so plainly and never invents a verdict.
· Language note (known gap): the page copy, headings, labels and buttons on this English page are English, but the model's own reasoning text below is still generated in Chinese — the proxy does not yet accept a lang parameter, so the inference output has not been translated. That is a known gap, not an oversight.

[Evidence section · where every number can be checked]
(1) Source: the rules come from /opt/algorithms/康锦源药房/风雨方法论-v20260620/08-八爪鱼·八号机/14-风雨方法论·八爪鱼算法-v1.md (264 lines, read-only, unchanged in this round). The proxy code hard-codes those rules into the prompt, so they do not change with the model; the judging itself is done by the LLM.
(2) Three categories
· [Verifiable] The core formula (internal capability + external demand + connection path = live tentacle), the three states, the four live-tentacle tests, the four kinds of hanging, the four hanging-timeline bands, the five steps for a dead tentacle, the four quality tiers and the three expansion principles — each can be matched word for word against the corresponding section of the .md.
· [This page's display convention] Your tentacle list, whether each tentacle has all three elements, its state, which element is missing, how to fix it, the possible space and the first step — these are text the model generates on the spot, not source-code output; the same input on another call can produce different content. This page is a questioning tool, not a verdict tool.
· [No source found] The four hanging-timeline bands (0-1 / 1-3 / 3-6 / beyond 6 months) and the four quality tiers (1 / 3 / 5 / 10 tentacles) carry no stated origin and no cited source inside the .md → no source found (set by the methodology document itself).
(3) How to verify it yourself: the rules are checkable, the verdicts cannot be diffed. Take a tentacle that is obviously missing one element (for example: you have the internal capability and there is external demand, but there is no connection path at all) and run it — the model should call it a dead tentacle (the source's core law: “miss one of three = dead tentacle”). If it calls it “✅ live”, it contradicts the hard-coded rules and can be pointed out directly.