Antifragile Option

This page is a conditions-preparation tool, not investment advice.It recommends nothing; it only lays out the doc's six start conditions (O1–O6) and the banding rules.
Every threshold (per-attempt outlay ≤ 5% of disposable funds, spare time ≤ 3 months, convexity ratio ≥5:1, best ≥ 10× the per-attempt cost, ≥3 times a year, floor covering 70%+ of your livelihood, portfolio ≤3 attempts / ≤15%, walk-away lines ≤8 weeks and a 50% top-up) comes from the internal methodology doc 《03-反脆弱期权算法.md v1.0》 (domestic machine /opt/inventory/econ_algorithms/, read-only, untouched this round) and has not been externally validated —— these are internal practice thresholds, not theorems.
Where it is not aligned / has been simplified (said plainly)
· ① The doc's §7 advises “force yourself to enter the most conservative version of the best outcome (discount it to 30%)” —— this page does not discount automatically; the discounting is yours to do while entering the figures, and the page only hints at it in the label instead of changing your numbers.
· ② “Barbell allocation, 80–90% ultra-safe + 10–20% ultra-volatile” (end of §3) is described by the doc itself as “a practice rule, not a theorem”; this page does not turn it into a hard gate and gives no allocation advice.
· ③ The expected value uses the success rate you enter (blank falls back to the doc's §6 most conservative assumption of 10%); if the rate you enter is higher than reality the expected value will read too rosy —— that is decided by the input, not a miscalculation by the page.
· ④ “Best outcome ÷ per-attempt cost ≥10×” is only checked numerically when you pick “Money (CNY)” as the unit; when you pick “Time (days)” it follows O3's choice alone.
Unconfirmed
· Unconfirmed ①: which band §4 Step3's table assigns at the 20 / 5 / 2 points (see above; this page takes ≥20 and <3 times → 🟢).
· Unconfirmed ②: §3 O5 says “an attempt below 5:1 is not worth committing resources to” but gives no remedy for “how to raise the convexity ratio” —— when O5 is missing this page only gives the qualitative hint “either squeeze the bad outcome or open up the ceiling” and invents no specific actions.
· Unconfirmed ③: §4 Step4's timeline only says “suggested ≤ 8 weeks”, which is advice and not a hard threshold, so this page repeats the word “suggested”.

[Evidence section · where to check each number]
① Source: internal methodology doc 《03-反脆弱期权算法.md》v1.0 (domestic machine /opt/inventory/econ_algorithms/03-反脆弱期权算法.md, read-only): start conditions O1–O6 = §3; decision rules = §4 Step1–Step5; preparation list and first step = §5.2 / §5.3; external data and hard rules = §6. The doc's §1 cites literature only for option pricing / real options (Black & Scholes 1973, DOI 10.1086/260062; Myers 1977, DOI 10.1016/0304-405x(77)90015-0).
② Three categories
· [Verifiable] The O6 hard gate (build the floor first → stop), the opportunity-card fields (worst / best / convexity ratio / repeat count), convexity ratio = best ÷ worst, §4 Step3's four band lines (≥20:1 / 5:1–20:1 / 2:1–5:1 / <2:1), the three walk-away lines, portfolio ≤3 and ≤15% —— all of these match §3 / §4.
· [This page's presentation choice] The “expected value” block is the page substituting your figures into §2's formula once and is not source-code output; the doc does not put it among the decision rules. “Best ≥ 10× the per-attempt cost” is only checked numerically when the unit is money.
· [No source found] §4 Step3's band table has overlapping ranges at the 20 / 5 / 2 points and the doc does not say how an overlap is assigned; §3 O5 gives no remedy for “how to raise the convexity ratio”; §4 Step4's “≤8 weeks” is worded as advice, not a hard threshold; the “barbell strategy” is labelled by the doc itself as Taleb's popular phrasing, not an independent academic theorem.
③ How to check it yourself: worst 1000, best 30000, repeatable 3 times → convexity ratio 30:1 and ≥3 times → should give 🔵 worth repeating; change the repeat count to 2 → should land in 🟢 worth a try (🔵 requires both conditions at once). Then set “there is a stable floor” to No → should output “build the floor first” directly, with not one number computed afterwards.