# Review: realizing any monotone identification pattern with a noiseless linear measurement system ## 1. Precise restatement **Setup.** - There is a finite set N of instruments. Each instrument may output several real-valued channels. - Each reading is a linear function, with no noise, of the scalar target θ ∈ ℝ and a vector of nuisance parameters z ∈ ℝ^m. The nuisance parameters are completely unrestricted: no bounds and no prior. - A subset A ⊆ N **identifies** θ if the readings of all instruments in A determine θ uniquely, over all (θ, z). **Claim.** - Let 𝓕 be any family of nonempty subsets of N that is closed under supersets. - Then there is such a system in which A identifies θ exactly when A ∈ 𝓕. - Moreover, every reading has a nonzero coefficient on θ. - The construction gives something stronger than non-identification: for A ∉ 𝓕, every value of θ is consistent with the data. **Implicit assumptions.** - Upward closure is necessary. More readings can never destroy identification, so upward-closed families are exactly the realizable patterns. - "Direct sensitivity" means a coefficient in one particular parametrization of the nuisance. - Instruments are allowed to return multiple channels, one per minimal group they belong to. ## 2. Checking the argument and re-deriving it **Sufficiency.** Take a minimal set S of size k. Its block sums to (k−1)θ/k + θ/k + Σz_j − Σz_j = θ. So any A ⊇ S recovers θ. When k = 1 the single reading is simply y = θ, which is fine. **Necessity.** Suppose A contains no minimal set. Then each block is missing at least one reading. Fix any proposed value θ′ and handle each block separately: - **Last reading missing.** Set z_j = y_j − θ′/k for each observed j. Choose the remaining offsets freely. - **Last reading observed, but some earlier reading j₀ missing.** Set the other observed offsets as above. Then solve the equation for y_k for z_{j₀}. This works because z_{j₀} enters y_k with coefficient −1. - **Blocks with no observed readings.** These impose no constraint. The offsets are disjoint across blocks, so all these choices can be made simultaneously. Hence every θ′ is consistent with the data. ✔ **Independent re-derivation.** In a noiseless linear model y = aθ + Bz, the target θ is identified from the rows in A exactly when (1, 0) lies in the row space of [a_A | B_A]. - Within one block, the k rows span (1, 0) only through their full sum. - Removing any row leaves k−1 rows with k−1 free offsets. That restricted map is surjective onto ℝ^{k−1} through z alone, so θ is invisible. - Disjoint offsets mean the blocks combine as a direct sum. θ is identified iff some block is complete. This confirms the result. ## 3. Attempts to break it **Small example: 2-of-3 threshold.** Instruments are {1, 2, 3} and 𝓕 is all sets of size at least 2. | Channel group | Readings | |---|---| | Block {1,2} | θ/2 + a, θ/2 − a | | Block {1,3} | θ/2 + b, θ/2 − b | | Block {2,3} | θ/2 + c, θ/2 − c | - Instrument 1 alone sees θ/2 + a and θ/2 + b. These are consistent with any θ. - Any pair contains a complete block and so identifies θ. **Small example: 3-of-3.** Here 𝓕 = {{1,2,3}}. No pair identifies θ, but all three together do. This illustrates the "abrupt gain" in the final paragraph. **Edge cases.** - **Instruments in no minimal set** get no readings, including when 𝓕 is empty. "Every reading is sensitive" is then vacuous for them. This is easily patched: give each such instrument an extra channel θ + w_i with a private free offset w_i. That channel never helps identification. - **Multi-channel instruments.** If each instrument were restricted to a single scalar reading, the claim would be a different and more delicate question. It would resemble ideal linear secret sharing, which is known *not* to exist for all access structures. The claim's "finite … system" quietly allows many channels per instrument. **The substantive caveat: "direct sensitivity" is not invariant.** - Reparametrize with z′_j = z_j + θ/k for j < k. Then y_j = z′_j and y_k = θ − Σz′_j. - In these coordinates, k−1 of the k readings have zero θ-coefficient. The family of distributions is unchanged. - In the original coordinates, each reading taken alone carries no information about θ, because its offset is free. - So the nonzero coefficient on θ is a parametrization artifact, not "genuine target sensitivity". The theorem is true, but the clause "even when every reading has nonzero direct sensitivity" adds almost no content. ## 4. Novelty This is the classical construction of perfect secret sharing for general monotone access structures: - Ito–Saito–Nishizeki (1987), the per-minimal-set additive sharing construction; - Benaloh–Leichter (1988). The construction is transplanted from finite fields to ℝ, with "the secret is uniquely determined" in place of "authorized set" and "every θ is consistent" in place of perfect privacy. Its realization by linear maps is the standard link between linear secret sharing and monotone span programs (Karchmer–Wigderson). The identification criterion used, that the target functional must lie in the row space, is standard linear-model identifiability. So the result is known, apart from the measurement-science framing. ## 5. Fairness of the summary The mathematics matches the claim. However, the summary oversells two points: - **"Genuine" sensitivity.** The sensitivity condition is coordinate-dependent and essentially vacuous, as shown above. - **Hidden structural freedom.** The construction relies on unrestricted nuisance offsets and on multi-channel instruments. Two broader caveats also apply to the closing interpretive paragraph: - It treats the claim as a general statement about measurement systems, yet it is an existence result for adversarially constructed systems. - It is purely noiseless. It says nothing about precision or diminishing returns in noisy settings. ```json {"verdict": "correct with caveats", "novelty": "known", "fair_summary": "partly", "confidence": 4, "one_line": "The construction is valid and realizes any monotone identification pattern (it is the Ito–Saito–Nishizeki/Benaloh–Leichter additive secret-sharing scheme over the reals), but it needs multi-channel instruments and unrestricted nuisance offsets, and its 'nonzero direct sensitivity' is a parametrization artifact rather than genuine sensitivity."} ```