Remove private evaluation details from task inputs

tasks/business_finance/bpmn_category_governance_restructuring_l3/base/input/starter_project/disruption_scenario_L3.md

@@ -136,7 +136,7 @@ Each of these is a forbidden anti-gaming pattern for the final design.
 
 ## External Context — Required Reading
 
-In addition to this scenario file, four stakeholder artifacts in `stakeholder_artifacts/` carry **binding, authoritative constraints** that feed business Rules 27-30 and several hidden evaluator checks (cross_rule_27 / cross_rule_28 / cross_rule_29 / cross_rule_30, c12 role-authorization). You must read them before designing the BPMN:
+In addition to this scenario file, four stakeholder artifacts in `stakeholder_artifacts/` carry **binding, authoritative constraints** that feed business Rules 27-30 and the cross-rule constraints (cross_rule_27 / cross_rule_28 / cross_rule_29 / cross_rule_30, c12 role-authorization). You must read them before designing the BPMN:
 
 - `stakeholder_artifacts/regulator_email_thread.md` — SAMR letter imposing: brand-risk precedence on compliance-crisis branch (Rule 27); `complianceHandled` re-entry latch (Rule 30); terminal escalation end-event `escalation_final_end` (reinforces Rule 4).
 - `stakeholder_artifacts/product_compliance_slack.md` — A/B guild ticket PLATFORM-AB-2211 mandating kill-switch variables `abTestEnabled` / `freeTrialEnabled` / `newMerchantAdmissionEnabled` as `in_*` formProperties on `team_drafts_initial_plan` (Rule 28).

tasks/computing_math/ising_post_measurement_1/n10_critical_u01_correlators/input/task_specification.md

@@ -138,7 +138,7 @@ Since `ancilla_mode` is `"critical"`, no ancilla file is provided; the agent use
 
 ## 7. Evaluation Criteria
 
-Hidden reference outputs for evaluation are generated using imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$.
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
 
 | Quantity | Metric | Tolerance |
 |---|---|---|

tasks/computing_math/ising_post_measurement_1/n10_paramagnetic_u01_correlators/input/task_specification.md

@@ -0,0 +1,158 @@
+# Post-Measurement State Preparation for a Critical 1D Quantum Ising Chain
+
+## 1. Task Goal
+
+Given a quantum system consisting of a critical 1D Ising chain and an ancilla chain (each with $N = 10$ qubits), execute the following protocol:
+
+1. Compute the critical ground state $|\psi_c\rangle$ of the transverse-field Ising model at criticality.
+2. Load the ancilla state $|\psi_a\rangle$ from the provided file `ancilla_state.npy` (paramagnetic ancilla).
+3. Apply the combined inter-chain unitary and projective measurement operator to produce post-measurement states and their probabilities for all $2^N = 1024$ measurement outcomes.
+4. Compute the one-site reduced density matrix (RDM) at site $j = 1$.
+5. Compute one-body correlation functions $\langle Z_j \rangle$ and $\langle X_j \rangle$ across all sites.
+
+---
+
+## 2. Conventions
+
+The following conventions are used throughout this specification and in all input/output files.
+
+**Pauli matrices.**
+
+$$Z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, \quad X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \quad I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$
+
+**Single-qubit states.**
+
+$$|0\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |1\rangle = \begin{pmatrix} 0 \\ 1 \end{pmatrix}, \quad |{+}\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$$
+
+Note: $Z|0\rangle = +|0\rangle$ and $Z|1\rangle = -|1\rangle$.
+
+**Computational basis ordering (big-endian).** An $N$-qubit computational basis state $|s_1 s_2 \ldots s_N\rangle$ (with $s_i \in \{0, 1\}$) is stored at array index
+
+$$\text{index} = s_1 \cdot 2^{N-1} + s_2 \cdot 2^{N-2} + \cdots + s_N \cdot 2^0$$
+
+Qubit 1 is the most significant bit. For example, with $N = 3$: $|000\rangle \to 0$, $|001\rangle \to 1$, $|010\rangle \to 2$, $|100\rangle \to 4$, $|111\rangle \to 7$.
+
+**Site indexing.** Sites are labeled $1, 2, \ldots, N$ (1-indexed). In the tensor-product Hilbert space $(\mathbb{C}^2)^{\otimes N}$, site $i$ corresponds to the $i$-th tensor factor (from the left).
+
+---
+
+## 3. Physics Protocol
+
+### 3.1 Hamiltonian
+
+The critical 1D transverse-field Ising Hamiltonian with periodic boundary conditions is:
+
+$$H = -\sum_{i=1}^{N} Z_i Z_{i+1} - \sum_{i=1}^{N} X_i, \qquad Z_{N+1} \equiv Z_1 \text{ (PBC)}$$
+
+with $N = 10$.
+
+### 3.2 Critical Ground State
+
+$|\psi_c\rangle$ is the ground state (lowest-energy eigenstate) of $H$. The agent may obtain it by any correct method (e.g., exact diagonalisation, imaginary time evolution, DMRG). For ITE, the recommended Trotterised scheme is
+
+$$|\psi_c\rangle \approx \prod_{k=1}^{K} \Bigl[e^{\Delta t \sum_i Z_i Z_{i+1}} \;\cdot\; \prod_i e^{\Delta t\, X_i}\Bigr]\;|{+}\rangle^{\otimes N}$$
+
+with $\Delta t = 0.0002$, $K = 50000$ (total imaginary time $T = K\Delta t = 10$), maintaining unit norm throughout. The Trotter ordering (whether the $Z$-part or the $X$-part is applied first) does not affect the converged result at the recommended parameters.
+
+### 3.3 Ancilla State
+
+This variant uses the **paramagnetic ancilla**. The file `ancilla_state.npy` is provided as input. The ancilla state $|\psi_a\rangle$ was generated by 5 ITE steps from $|{+}\rangle^{\otimes 10}$ using the same Trotterised scheme as in section 3.2. The agent must load this file directly and use it as $|\psi_a\rangle$.
+
+### 3.4 Inter-Chain Unitary and Measurement
+
+The site-$j$ unitary coupling the critical chain to the ancilla chain is
+
+$$U_j = \exp\!\bigl(i\,u\,(Z_j - \theta\,I)\otimes \tilde{X}_j\bigr)$$
+
+with coupling strength $u = 0.1$ and constant $\theta = -1$. The tilde denotes operators on the ancilla chain. After applying $\prod_j U_j$ to the initial state $|\psi_c\rangle \otimes |\psi_a\rangle$ and projectively measuring the ancilla in the $Z$-basis, each measurement outcome $|\tilde{s}\rangle = |\tilde{s}_1 \tilde{s}_2 \ldots \tilde{s}_N\rangle$ (with $\tilde{s}_j \in \{|0\rangle, |1\rangle\}$) produces a post-measurement state of the critical chain:
+
+$$|\psi_{\tilde{s}}\rangle = \frac{e^{iH'}\; e^{-H_m/2}\;|\psi_c\rangle}{\bigl\|e^{-H_m/2}\;|\psi_c\rangle\bigr\|}$$
+
+The effective Hamiltonians $H'$ and $H_m$ are both diagonal in the computational basis and **depend on the measurement outcome** $|\tilde{s}\rangle$: each of the $2^N = 1024$ outcomes yields a different pair $(H', H_m)$.
+
+**Ancilla coefficients.** For each outcome $|\tilde{s}\rangle$, define:
+
+$$a(j) = \frac{\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}, \qquad a(j,k) = \frac{\langle\tilde{s}|\tilde{X}_j\tilde{X}_k|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}$$
+
+Since $|\tilde{s}\rangle$ is a computational basis state, $\langle\tilde{s}|\psi_a\rangle = \psi_a[\text{index}(\tilde{s})]$, the amplitude of the ancilla state at array index $\text{index}(\tilde{s})$. Similarly, $\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle = \psi_a[\text{index}(\tilde{s} \text{ with bit } j \text{ flipped})]$, since $\tilde{X}_j$ flips qubit $j$ in each computational basis component.
+
+**Derived quantities:**
+
+$$V_{jk} = a(j,k) - a(j)\,a(k), \qquad m_j = (-2\theta)\Bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\Bigr)$$
+
+With $\theta = -1$, this simplifies to $m_j = 2\bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\bigr)$.
+
+**Effective Hamiltonians** (diagonal in the computational basis):
+
+$$H' = u\sum_j a(j)\,Z_j, \qquad H_m = u^2\sum_j m_j\,Z_j + u^2\sum_{j\neq k} V_{jk}\,Z_jZ_k$$
+
+**Measurement probability:**
+
+$$p(\tilde{s}) \propto \langle\psi_c|e^{-H_m}|\psi_c\rangle \cdot |\langle\tilde{s}|\psi_a\rangle|^2$$
+
+normalised so that $\sum_{\tilde{s}} p(\tilde{s}) = 1$.
+
+### 3.5 Reduced Density Matrix
+
+The one-site RDM at site $j = 1$ is $\rho_1 = \text{Tr}_{\overline{1}}(|\psi_{\tilde{s}}\rangle\langle\psi_{\tilde{s}}|)$, a $2\times 2$ Hermitian matrix, computed for every measurement outcome.
+
+### 3.6 One-Body Correlators
+
+$\langle Z_j\rangle = \text{Tr}(\rho_j Z)$ and $\langle X_j\rangle = \text{Tr}(\rho_j X)$ for each site $j \in \{1, \ldots, 10\}$ and each outcome $|\tilde{s}\rangle$.
+
+---
+
+## 4. Input Files
+
+All input files are in the `input/` subdirectory.
+
+| File | Description |
+|---|---|
+| `input/config.json` | Parameters: `N=10`, `u=0.1`, `theta=-1`, `dt=0.0002`, `K=50000`, `ancilla_mode="paramagnetic"`, `ancilla_steps=5`, `site_j=1`. |
+| `input/ancilla_state.npy` | Shape $(1024,)$, float64. Pre-computed paramagnetic ancilla state. |
+
+---
+
+## 5. Output Files (Agent Produces)
+
+| File | Shape, dtype | Description |
+|---|---|---|
+| `critical_state.npy` | $(1024,)$, float64 | Ground state $\psi_c$. |
+| `post_probs.npy` | $(1024,)$, float64 | Measurement-outcome probabilities. Entry $i$ is $p(\tilde{s})$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `rdm_site1.npy` | $(1024, 2, 2)$, complex128 | One-site RDM at site 1. Entry $[i, :, :]$ is $\rho_1$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `correlators.npz` | `Z_one_body` $(1024, 10)$ float64, `X_one_body` $(1024, 10)$ float64 | One-body correlators. Entry $[i, k]$ is $\langle Z_{k+1}\rangle$ (or $\langle X_{k+1}\rangle$) for outcome $i$; column $k$ corresponds to site $k+1$ (0-indexed columns, 1-indexed sites). |
+
+---
+
+## 6. Tools / Software
+
+- **OS**: Ubuntu 22.04 LTS
+- **Python**: 3.10.12
+- **NumPy**: 2.2.6
+- **SciPy**: 1.15.3
+- **QuTiP**: 5.2.3 (optional; may use for exact diagonalisation)
+- Single machine. Peak memory ≈ 40 MB. Hilbert-space dimension $2^{10} = 1024$.
+
+---
+
+## 7. Evaluation Criteria
+
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
+
+| Quantity | Metric | Tolerance |
+|---|---|---|
+| Critical state $\psi_c$ | State infidelity $1 - F$, where $F = \lvert\langle\psi_c^{\text{ref}} \mid \psi_c^{\text{agent}}\rangle\rvert^2$ | $10^{-4}$ |
+| Probabilities $p(\tilde{s})$ | Max absolute error over all 1024 outcomes | $10^{-2}$ |
+| RDM $\rho_1$ | Max element-wise absolute error over all 1024 outcomes | $10^{-2}$ |
+| Correlators $\langle Z_j\rangle$, $\langle X_j\rangle$ | Max absolute error over all outcomes and sites | $10^{-2}$ |
+
+**Pass condition:** all four criteria are satisfied simultaneously.
+
+---
+
+## 8. Notes on Reproducibility
+
+- The computation is fully deterministic; no random seed is needed.
+- All $2^{10} = 1024$ measurement outcomes are computed in parallel (no stochastic sampling).
+- The golden reference uses imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$. Agents using ITE with the recommended parameters or exact diagonalisation should meet the tolerances.
+- For this paramagnetic-ancilla variant, the agent receives `ancilla_state.npy` as input, so the ancilla state is pinned exactly.

tasks/computing_math/ising_post_measurement_1/n12_critical_u01_correlators/input/task_specification.md

@@ -0,0 +1,158 @@
+# Post-Measurement State Preparation for a Critical 1D Quantum Ising Chain
+
+## 1. Task Goal
+
+Given a quantum system consisting of a critical 1D Ising chain and an ancilla chain (each with $N = 12$ qubits), execute the following protocol:
+
+1. Compute the critical ground state $|\psi_c\rangle$ of the transverse-field Ising model at criticality.
+2. Set the ancilla state $|\psi_a\rangle = |\psi_c\rangle$ (critical ancilla).
+3. Apply the combined inter-chain unitary and projective measurement operator to produce post-measurement states and their probabilities for all $2^N = 4096$ measurement outcomes.
+4. Compute the one-site reduced density matrix (RDM) at site $j = 1$.
+5. Compute one-body correlation functions $\langle Z_j \rangle$ and $\langle X_j \rangle$ across all sites.
+
+---
+
+## 2. Conventions
+
+The following conventions are used throughout this specification and in all input/output files.
+
+**Pauli matrices.**
+
+$$Z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, \quad X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \quad I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$
+
+**Single-qubit states.**
+
+$$|0\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |1\rangle = \begin{pmatrix} 0 \\ 1 \end{pmatrix}, \quad |{+}\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$$
+
+Note: $Z|0\rangle = +|0\rangle$ and $Z|1\rangle = -|1\rangle$.
+
+**Computational basis ordering (big-endian).** An $N$-qubit computational basis state $|s_1 s_2 \ldots s_N\rangle$ (with $s_i \in \{0, 1\}$) is stored at array index
+
+$$\text{index} = s_1 \cdot 2^{N-1} + s_2 \cdot 2^{N-2} + \cdots + s_N \cdot 2^0$$
+
+Qubit 1 is the most significant bit. For example, with $N = 3$: $|000\rangle \to 0$, $|001\rangle \to 1$, $|010\rangle \to 2$, $|100\rangle \to 4$, $|111\rangle \to 7$.
+
+**Site indexing.** Sites are labeled $1, 2, \ldots, N$ (1-indexed). In the tensor-product Hilbert space $(\mathbb{C}^2)^{\otimes N}$, site $i$ corresponds to the $i$-th tensor factor (from the left).
+
+---
+
+## 3. Physics Protocol
+
+### 3.1 Hamiltonian
+
+The critical 1D transverse-field Ising Hamiltonian with periodic boundary conditions is:
+
+$$H = -\sum_{i=1}^{N} Z_i Z_{i+1} - \sum_{i=1}^{N} X_i, \qquad Z_{N+1} \equiv Z_1 \text{ (PBC)}$$
+
+with $N = 12$.
+
+### 3.2 Critical Ground State
+
+$|\psi_c\rangle$ is the ground state (lowest-energy eigenstate) of $H$. The agent may obtain it by any correct method (e.g., exact diagonalisation, imaginary time evolution, DMRG). For ITE, the recommended Trotterised scheme is
+
+$$|\psi_c\rangle \approx \prod_{k=1}^{K} \Bigl[e^{\Delta t \sum_i Z_i Z_{i+1}} \;\cdot\; \prod_i e^{\Delta t\, X_i}\Bigr]\;|{+}\rangle^{\otimes N}$$
+
+with $\Delta t = 0.0002$, $K = 50000$ (total imaginary time $T = K\Delta t = 10$), maintaining unit norm throughout. The Trotter ordering (whether the $Z$-part or the $X$-part is applied first) does not affect the converged result at the recommended parameters.
+
+### 3.3 Ancilla State
+
+This variant uses the **critical ancilla**: $|\psi_a\rangle = |\psi_c\rangle$ (an identical copy of the ground state). No ancilla file is provided.
+
+### 3.4 Inter-Chain Unitary and Measurement
+
+The site-$j$ unitary coupling the critical chain to the ancilla chain is
+
+$$U_j = \exp\!\bigl(i\,u\,(Z_j - \theta\,I)\otimes \tilde{X}_j\bigr)$$
+
+with coupling strength $u = 0.1$ and constant $\theta = -1$. The tilde denotes operators on the ancilla chain. After applying $\prod_j U_j$ to the initial state $|\psi_c\rangle \otimes |\psi_a\rangle$ and projectively measuring the ancilla in the $Z$-basis, each measurement outcome $|\tilde{s}\rangle = |\tilde{s}_1 \tilde{s}_2 \ldots \tilde{s}_N\rangle$ (with $\tilde{s}_j \in \{|0\rangle, |1\rangle\}$) produces a post-measurement state of the critical chain:
+
+$$|\psi_{\tilde{s}}\rangle = \frac{e^{iH'}\; e^{-H_m/2}\;|\psi_c\rangle}{\bigl\|e^{-H_m/2}\;|\psi_c\rangle\bigr\|}$$
+
+The effective Hamiltonians $H'$ and $H_m$ are both diagonal in the computational basis and **depend on the measurement outcome** $|\tilde{s}\rangle$: each of the $2^N = 4096$ outcomes yields a different pair $(H', H_m)$.
+
+**Ancilla coefficients.** For each outcome $|\tilde{s}\rangle$, define:
+
+$$a(j) = \frac{\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}, \qquad a(j,k) = \frac{\langle\tilde{s}|\tilde{X}_j\tilde{X}_k|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}$$
+
+Since $|\tilde{s}\rangle$ is a computational basis state, $\langle\tilde{s}|\psi_a\rangle = \psi_a[\text{index}(\tilde{s})]$, the amplitude of the ancilla state at array index $\text{index}(\tilde{s})$. Similarly, $\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle = \psi_a[\text{index}(\tilde{s} \text{ with bit } j \text{ flipped})]$, since $\tilde{X}_j$ flips qubit $j$ in each computational basis component.
+
+**Derived quantities:**
+
+$$V_{jk} = a(j,k) - a(j)\,a(k), \qquad m_j = (-2\theta)\Bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\Bigr)$$
+
+With $\theta = -1$, this simplifies to $m_j = 2\bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\bigr)$.
+
+**Effective Hamiltonians** (diagonal in the computational basis):
+
+$$H' = u\sum_j a(j)\,Z_j, \qquad H_m = u^2\sum_j m_j\,Z_j + u^2\sum_{j\neq k} V_{jk}\,Z_jZ_k$$
+
+**Measurement probability:**
+
+$$p(\tilde{s}) \propto \langle\psi_c|e^{-H_m}|\psi_c\rangle \cdot |\langle\tilde{s}|\psi_a\rangle|^2$$
+
+normalised so that $\sum_{\tilde{s}} p(\tilde{s}) = 1$.
+
+### 3.5 Reduced Density Matrix
+
+The one-site RDM at site $j = 1$ is $\rho_1 = \text{Tr}_{\overline{1}}(|\psi_{\tilde{s}}\rangle\langle\psi_{\tilde{s}}|)$, a $2\times 2$ Hermitian matrix, computed for every measurement outcome.
+
+### 3.6 One-Body Correlators
+
+$\langle Z_j\rangle = \text{Tr}(\rho_j Z)$ and $\langle X_j\rangle = \text{Tr}(\rho_j X)$ for each site $j \in \{1, \ldots, 12\}$ and each outcome $|\tilde{s}\rangle$.
+
+---
+
+## 4. Input Files
+
+All input files are in the `input/` subdirectory.
+
+| File | Description |
+|---|---|
+| `input/config.json` | Parameters: `N=12`, `u=0.1`, `theta=-1`, `dt=0.0002`, `K=50000`, `ancilla_mode="critical"`, `site_j=1`. |
+
+Since `ancilla_mode` is `"critical"`, no ancilla file is provided; the agent uses $|\psi_a\rangle = |\psi_c\rangle$.
+
+---
+
+## 5. Output Files (Agent Produces)
+
+| File | Shape, dtype | Description |
+|---|---|---|
+| `critical_state.npy` | $(4096,)$, float64 | Ground state $\psi_c$. |
+| `post_probs.npy` | $(4096,)$, float64 | Measurement-outcome probabilities. Entry $i$ is $p(\tilde{s})$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `rdm_site1.npy` | $(4096, 2, 2)$, complex128 | One-site RDM at site 1. Entry $[i, :, :]$ is $\rho_1$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `correlators.npz` | `Z_one_body` $(4096, 12)$ float64, `X_one_body` $(4096, 12)$ float64 | One-body correlators. Entry $[i, k]$ is $\langle Z_{k+1}\rangle$ (or $\langle X_{k+1}\rangle$) for outcome $i$; column $k$ corresponds to site $k+1$ (0-indexed columns, 1-indexed sites). |
+
+---
+
+## 6. Tools / Software
+
+- **OS**: Ubuntu 22.04 LTS
+- **Python**: 3.10.12
+- **NumPy**: 2.2.6
+- **SciPy**: 1.15.3
+- **QuTiP**: 5.2.3 (optional; may use for exact diagonalisation)
+- Single machine. Peak memory ≈ 600 MB. Hilbert-space dimension $2^{12} = 4096$.
+
+---
+
+## 7. Evaluation Criteria
+
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
+
+| Quantity | Metric | Tolerance |
+|---|---|---|
+| Critical state $\psi_c$ | State infidelity $1 - F$, where $F = \lvert\langle\psi_c^{\text{ref}} \mid \psi_c^{\text{agent}}\rangle\rvert^2$ | $10^{-4}$ |
+| Probabilities $p(\tilde{s})$ | Max absolute error over all 4096 outcomes | $10^{-2}$ |
+| RDM $\rho_1$ | Max element-wise absolute error over all 4096 outcomes | $10^{-2}$ |
+| Correlators $\langle Z_j\rangle$, $\langle X_j\rangle$ | Max absolute error over all outcomes and sites | $10^{-2}$ |
+
+**Pass condition:** all four criteria are satisfied simultaneously.
+
+---
+
+## 8. Notes on Reproducibility
+
+- The computation is fully deterministic; no random seed is needed.
+- All $2^{12} = 4096$ measurement outcomes are computed in parallel (no stochastic sampling).
+- The golden reference uses imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$. Agents using ITE with the recommended parameters or exact diagonalisation should meet the tolerances.

tasks/computing_math/ising_post_measurement_1/n12_critical_u02_no_correlators/input/task_specification.md

@@ -0,0 +1,151 @@
+# Post-Measurement State Preparation for a Critical 1D Quantum Ising Chain
+
+## 1. Task Goal
+
+Given a quantum system consisting of a critical 1D Ising chain and an ancilla chain (each with $N = 12$ qubits), execute the following protocol:
+
+1. Compute the critical ground state $|\psi_c\rangle$ of the transverse-field Ising model at criticality.
+2. Set the ancilla state $|\psi_a\rangle = |\psi_c\rangle$ (critical ancilla).
+3. Apply the combined inter-chain unitary and projective measurement operator to produce post-measurement states and their probabilities for all $2^N = 4096$ measurement outcomes.
+4. Compute the one-site reduced density matrix (RDM) at site $j = 1$.
+
+---
+
+## 2. Conventions
+
+The following conventions are used throughout this specification and in all input/output files.
+
+**Pauli matrices.**
+
+$$Z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, \quad X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \quad I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$
+
+**Single-qubit states.**
+
+$$|0\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |1\rangle = \begin{pmatrix} 0 \\ 1 \end{pmatrix}, \quad |{+}\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$$
+
+Note: $Z|0\rangle = +|0\rangle$ and $Z|1\rangle = -|1\rangle$.
+
+**Computational basis ordering (big-endian).** An $N$-qubit computational basis state $|s_1 s_2 \ldots s_N\rangle$ (with $s_i \in \{0, 1\}$) is stored at array index
+
+$$\text{index} = s_1 \cdot 2^{N-1} + s_2 \cdot 2^{N-2} + \cdots + s_N \cdot 2^0$$
+
+Qubit 1 is the most significant bit. For example, with $N = 3$: $|000\rangle \to 0$, $|001\rangle \to 1$, $|010\rangle \to 2$, $|100\rangle \to 4$, $|111\rangle \to 7$.
+
+**Site indexing.** Sites are labeled $1, 2, \ldots, N$ (1-indexed). In the tensor-product Hilbert space $(\mathbb{C}^2)^{\otimes N}$, site $i$ corresponds to the $i$-th tensor factor (from the left).
+
+---
+
+## 3. Physics Protocol
+
+### 3.1 Hamiltonian
+
+The critical 1D transverse-field Ising Hamiltonian with periodic boundary conditions is:
+
+$$H = -\sum_{i=1}^{N} Z_i Z_{i+1} - \sum_{i=1}^{N} X_i, \qquad Z_{N+1} \equiv Z_1 \text{ (PBC)}$$
+
+with $N = 12$.
+
+### 3.2 Critical Ground State
+
+$|\psi_c\rangle$ is the ground state (lowest-energy eigenstate) of $H$. The agent may obtain it by any correct method (e.g., exact diagonalisation, imaginary time evolution, DMRG). For ITE, the recommended Trotterised scheme is
+
+$$|\psi_c\rangle \approx \prod_{k=1}^{K} \Bigl[e^{\Delta t \sum_i Z_i Z_{i+1}} \;\cdot\; \prod_i e^{\Delta t\, X_i}\Bigr]\;|{+}\rangle^{\otimes N}$$
+
+with $\Delta t = 0.0002$, $K = 50000$ (total imaginary time $T = K\Delta t = 10$), maintaining unit norm throughout. The Trotter ordering (whether the $Z$-part or the $X$-part is applied first) does not affect the converged result at the recommended parameters.
+
+### 3.3 Ancilla State
+
+This variant uses the **critical ancilla**: $|\psi_a\rangle = |\psi_c\rangle$ (an identical copy of the ground state). No ancilla file is provided.
+
+### 3.4 Inter-Chain Unitary and Measurement
+
+The site-$j$ unitary coupling the critical chain to the ancilla chain is
+
+$$U_j = \exp\!\bigl(i\,u\,(Z_j - \theta\,I)\otimes \tilde{X}_j\bigr)$$
+
+with coupling strength $u = 0.2$ and constant $\theta = -1$. The tilde denotes operators on the ancilla chain. After applying $\prod_j U_j$ to the initial state $|\psi_c\rangle \otimes |\psi_a\rangle$ and projectively measuring the ancilla in the $Z$-basis, each measurement outcome $|\tilde{s}\rangle = |\tilde{s}_1 \tilde{s}_2 \ldots \tilde{s}_N\rangle$ (with $\tilde{s}_j \in \{|0\rangle, |1\rangle\}$) produces a post-measurement state of the critical chain:
+
+$$|\psi_{\tilde{s}}\rangle = \frac{e^{iH'}\; e^{-H_m/2}\;|\psi_c\rangle}{\bigl\|e^{-H_m/2}\;|\psi_c\rangle\bigr\|}$$
+
+The effective Hamiltonians $H'$ and $H_m$ are both diagonal in the computational basis and **depend on the measurement outcome** $|\tilde{s}\rangle$: each of the $2^N = 4096$ outcomes yields a different pair $(H', H_m)$.
+
+**Ancilla coefficients.** For each outcome $|\tilde{s}\rangle$, define:
+
+$$a(j) = \frac{\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}, \qquad a(j,k) = \frac{\langle\tilde{s}|\tilde{X}_j\tilde{X}_k|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}$$
+
+Since $|\tilde{s}\rangle$ is a computational basis state, $\langle\tilde{s}|\psi_a\rangle = \psi_a[\text{index}(\tilde{s})]$, the amplitude of the ancilla state at array index $\text{index}(\tilde{s})$. Similarly, $\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle = \psi_a[\text{index}(\tilde{s} \text{ with bit } j \text{ flipped})]$, since $\tilde{X}_j$ flips qubit $j$ in each computational basis component.
+
+**Derived quantities:**
+
+$$V_{jk} = a(j,k) - a(j)\,a(k), \qquad m_j = (-2\theta)\Bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\Bigr)$$
+
+With $\theta = -1$, this simplifies to $m_j = 2\bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\bigr)$.
+
+**Effective Hamiltonians** (diagonal in the computational basis):
+
+$$H' = u\sum_j a(j)\,Z_j, \qquad H_m = u^2\sum_j m_j\,Z_j + u^2\sum_{j\neq k} V_{jk}\,Z_jZ_k$$
+
+**Measurement probability:**
+
+$$p(\tilde{s}) \propto \langle\psi_c|e^{-H_m}|\psi_c\rangle \cdot |\langle\tilde{s}|\psi_a\rangle|^2$$
+
+normalised so that $\sum_{\tilde{s}} p(\tilde{s}) = 1$.
+
+### 3.5 Reduced Density Matrix
+
+The one-site RDM at site $j = 1$ is $\rho_1 = \text{Tr}_{\overline{1}}(|\psi_{\tilde{s}}\rangle\langle\psi_{\tilde{s}}|)$, a $2\times 2$ Hermitian matrix, computed for every measurement outcome.
+
+---
+
+## 4. Input Files
+
+All input files are in the `input/` subdirectory.
+
+| File | Description |
+|---|---|
+| `input/config.json` | Parameters: `N=12`, `u=0.2`, `theta=-1`, `dt=0.0002`, `K=50000`, `ancilla_mode="critical"`, `site_j=1`. |
+
+Since `ancilla_mode` is `"critical"`, no ancilla file is provided; the agent uses $|\psi_a\rangle = |\psi_c\rangle$.
+
+---
+
+## 5. Output Files (Agent Produces)
+
+| File | Shape, dtype | Description |
+|---|---|---|
+| `critical_state.npy` | $(4096,)$, float64 | Ground state $\psi_c$. |
+| `post_probs.npy` | $(4096,)$, float64 | Measurement-outcome probabilities. Entry $i$ is $p(\tilde{s})$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `rdm_site1.npy` | $(4096, 2, 2)$, complex128 | One-site RDM at site 1. Entry $[i, :, :]$ is $\rho_1$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+
+---
+
+## 6. Tools / Software
+
+- **OS**: Ubuntu 22.04 LTS
+- **Python**: 3.10.12
+- **NumPy**: 2.2.6
+- **SciPy**: 1.15.3
+- **QuTiP**: 5.2.3 (optional; may use for exact diagonalisation)
+- Single machine. Peak memory ≈ 600 MB. Hilbert-space dimension $2^{12} = 4096$.
+
+---
+
+## 7. Evaluation Criteria
+
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
+
+| Quantity | Metric | Tolerance |
+|---|---|---|
+| Critical state $\psi_c$ | State infidelity $1 - F$, where $F = \lvert\langle\psi_c^{\text{ref}} \mid \psi_c^{\text{agent}}\rangle\rvert^2$ | $10^{-4}$ |
+| Probabilities $p(\tilde{s})$ | Max absolute error over all 4096 outcomes | $10^{-2}$ |
+| RDM $\rho_1$ | Max element-wise absolute error over all 4096 outcomes | $10^{-2}$ |
+
+**Pass condition:** all three criteria are satisfied simultaneously.
+
+---
+
+## 8. Notes on Reproducibility
+
+- The computation is fully deterministic; no random seed is needed.
+- All $2^{12} = 4096$ measurement outcomes are computed in parallel (no stochastic sampling).
+- The golden reference uses imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$. Agents using ITE with the recommended parameters or exact diagonalisation should meet the tolerances.

tasks/computing_math/ising_post_measurement_1/n12_paramagnetic_u01_correlators/input/task_specification.md

@@ -0,0 +1,158 @@
+# Post-Measurement State Preparation for a Critical 1D Quantum Ising Chain
+
+## 1. Task Goal
+
+Given a quantum system consisting of a critical 1D Ising chain and an ancilla chain (each with $N = 12$ qubits), execute the following protocol:
+
+1. Compute the critical ground state $|\psi_c\rangle$ of the transverse-field Ising model at criticality.
+2. Load the ancilla state $|\psi_a\rangle$ from the provided file `ancilla_state.npy` (paramagnetic ancilla).
+3. Apply the combined inter-chain unitary and projective measurement operator to produce post-measurement states and their probabilities for all $2^N = 4096$ measurement outcomes.
+4. Compute the one-site reduced density matrix (RDM) at site $j = 1$.
+5. Compute one-body correlation functions $\langle Z_j \rangle$ and $\langle X_j \rangle$ across all sites.
+
+---
+
+## 2. Conventions
+
+The following conventions are used throughout this specification and in all input/output files.
+
+**Pauli matrices.**
+
+$$Z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, \quad X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \quad I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$
+
+**Single-qubit states.**
+
+$$|0\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |1\rangle = \begin{pmatrix} 0 \\ 1 \end{pmatrix}, \quad |{+}\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$$
+
+Note: $Z|0\rangle = +|0\rangle$ and $Z|1\rangle = -|1\rangle$.
+
+**Computational basis ordering (big-endian).** An $N$-qubit computational basis state $|s_1 s_2 \ldots s_N\rangle$ (with $s_i \in \{0, 1\}$) is stored at array index
+
+$$\text{index} = s_1 \cdot 2^{N-1} + s_2 \cdot 2^{N-2} + \cdots + s_N \cdot 2^0$$
+
+Qubit 1 is the most significant bit. For example, with $N = 3$: $|000\rangle \to 0$, $|001\rangle \to 1$, $|010\rangle \to 2$, $|100\rangle \to 4$, $|111\rangle \to 7$.
+
+**Site indexing.** Sites are labeled $1, 2, \ldots, N$ (1-indexed). In the tensor-product Hilbert space $(\mathbb{C}^2)^{\otimes N}$, site $i$ corresponds to the $i$-th tensor factor (from the left).
+
+---
+
+## 3. Physics Protocol
+
+### 3.1 Hamiltonian
+
+The critical 1D transverse-field Ising Hamiltonian with periodic boundary conditions is:
+
+$$H = -\sum_{i=1}^{N} Z_i Z_{i+1} - \sum_{i=1}^{N} X_i, \qquad Z_{N+1} \equiv Z_1 \text{ (PBC)}$$
+
+with $N = 12$.
+
+### 3.2 Critical Ground State
+
+$|\psi_c\rangle$ is the ground state (lowest-energy eigenstate) of $H$. The agent may obtain it by any correct method (e.g., exact diagonalisation, imaginary time evolution, DMRG). For ITE, the recommended Trotterised scheme is
+
+$$|\psi_c\rangle \approx \prod_{k=1}^{K} \Bigl[e^{\Delta t \sum_i Z_i Z_{i+1}} \;\cdot\; \prod_i e^{\Delta t\, X_i}\Bigr]\;|{+}\rangle^{\otimes N}$$
+
+with $\Delta t = 0.0002$, $K = 50000$ (total imaginary time $T = K\Delta t = 10$), maintaining unit norm throughout. The Trotter ordering (whether the $Z$-part or the $X$-part is applied first) does not affect the converged result at the recommended parameters.
+
+### 3.3 Ancilla State
+
+This variant uses the **paramagnetic ancilla**. The file `ancilla_state.npy` is provided as input. The ancilla state $|\psi_a\rangle$ was generated by 5 ITE steps from $|{+}\rangle^{\otimes 12}$ using the same Trotterised scheme as in section 3.2. The agent must load this file directly and use it as $|\psi_a\rangle$.
+
+### 3.4 Inter-Chain Unitary and Measurement
+
+The site-$j$ unitary coupling the critical chain to the ancilla chain is
+
+$$U_j = \exp\!\bigl(i\,u\,(Z_j - \theta\,I)\otimes \tilde{X}_j\bigr)$$
+
+with coupling strength $u = 0.1$ and constant $\theta = -1$. The tilde denotes operators on the ancilla chain. After applying $\prod_j U_j$ to the initial state $|\psi_c\rangle \otimes |\psi_a\rangle$ and projectively measuring the ancilla in the $Z$-basis, each measurement outcome $|\tilde{s}\rangle = |\tilde{s}_1 \tilde{s}_2 \ldots \tilde{s}_N\rangle$ (with $\tilde{s}_j \in \{|0\rangle, |1\rangle\}$) produces a post-measurement state of the critical chain:
+
+$$|\psi_{\tilde{s}}\rangle = \frac{e^{iH'}\; e^{-H_m/2}\;|\psi_c\rangle}{\bigl\|e^{-H_m/2}\;|\psi_c\rangle\bigr\|}$$
+
+The effective Hamiltonians $H'$ and $H_m$ are both diagonal in the computational basis and **depend on the measurement outcome** $|\tilde{s}\rangle$: each of the $2^N = 4096$ outcomes yields a different pair $(H', H_m)$.
+
+**Ancilla coefficients.** For each outcome $|\tilde{s}\rangle$, define:
+
+$$a(j) = \frac{\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}, \qquad a(j,k) = \frac{\langle\tilde{s}|\tilde{X}_j\tilde{X}_k|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}$$
+
+Since $|\tilde{s}\rangle$ is a computational basis state, $\langle\tilde{s}|\psi_a\rangle = \psi_a[\text{index}(\tilde{s})]$, the amplitude of the ancilla state at array index $\text{index}(\tilde{s})$. Similarly, $\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle = \psi_a[\text{index}(\tilde{s} \text{ with bit } j \text{ flipped})]$, since $\tilde{X}_j$ flips qubit $j$ in each computational basis component.
+
+**Derived quantities:**
+
+$$V_{jk} = a(j,k) - a(j)\,a(k), \qquad m_j = (-2\theta)\Bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\Bigr)$$
+
+With $\theta = -1$, this simplifies to $m_j = 2\bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\bigr)$.
+
+**Effective Hamiltonians** (diagonal in the computational basis):
+
+$$H' = u\sum_j a(j)\,Z_j, \qquad H_m = u^2\sum_j m_j\,Z_j + u^2\sum_{j\neq k} V_{jk}\,Z_jZ_k$$
+
+**Measurement probability:**
+
+$$p(\tilde{s}) \propto \langle\psi_c|e^{-H_m}|\psi_c\rangle \cdot |\langle\tilde{s}|\psi_a\rangle|^2$$
+
+normalised so that $\sum_{\tilde{s}} p(\tilde{s}) = 1$.
+
+### 3.5 Reduced Density Matrix
+
+The one-site RDM at site $j = 1$ is $\rho_1 = \text{Tr}_{\overline{1}}(|\psi_{\tilde{s}}\rangle\langle\psi_{\tilde{s}}|)$, a $2\times 2$ Hermitian matrix, computed for every measurement outcome.
+
+### 3.6 One-Body Correlators
+
+$\langle Z_j\rangle = \text{Tr}(\rho_j Z)$ and $\langle X_j\rangle = \text{Tr}(\rho_j X)$ for each site $j \in \{1, \ldots, 12\}$ and each outcome $|\tilde{s}\rangle$.
+
+---
+
+## 4. Input Files
+
+All input files are in the `input/` subdirectory.
+
+| File | Description |
+|---|---|
+| `input/config.json` | Parameters: `N=12`, `u=0.1`, `theta=-1`, `dt=0.0002`, `K=50000`, `ancilla_mode="paramagnetic"`, `ancilla_steps=5`, `site_j=1`. |
+| `input/ancilla_state.npy` | Shape $(4096,)$, float64. Pre-computed paramagnetic ancilla state. |
+
+---
+
+## 5. Output Files (Agent Produces)
+
+| File | Shape, dtype | Description |
+|---|---|---|
+| `critical_state.npy` | $(4096,)$, float64 | Ground state $\psi_c$. |
+| `post_probs.npy` | $(4096,)$, float64 | Measurement-outcome probabilities. Entry $i$ is $p(\tilde{s})$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `rdm_site1.npy` | $(4096, 2, 2)$, complex128 | One-site RDM at site 1. Entry $[i, :, :]$ is $\rho_1$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `correlators.npz` | `Z_one_body` $(4096, 12)$ float64, `X_one_body` $(4096, 12)$ float64 | One-body correlators. Entry $[i, k]$ is $\langle Z_{k+1}\rangle$ (or $\langle X_{k+1}\rangle$) for outcome $i$; column $k$ corresponds to site $k+1$ (0-indexed columns, 1-indexed sites). |
+
+---
+
+## 6. Tools / Software
+
+- **OS**: Ubuntu 22.04 LTS
+- **Python**: 3.10.12
+- **NumPy**: 2.2.6
+- **SciPy**: 1.15.3
+- **QuTiP**: 5.2.3 (optional; may use for exact diagonalisation)
+- Single machine. Peak memory ≈ 600 MB. Hilbert-space dimension $2^{12} = 4096$.
+
+---
+
+## 7. Evaluation Criteria
+
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
+
+| Quantity | Metric | Tolerance |
+|---|---|---|
+| Critical state $\psi_c$ | State infidelity $1 - F$, where $F = \lvert\langle\psi_c^{\text{ref}} \mid \psi_c^{\text{agent}}\rangle\rvert^2$ | $10^{-4}$ |
+| Probabilities $p(\tilde{s})$ | Max absolute error over all 4096 outcomes | $10^{-2}$ |
+| RDM $\rho_1$ | Max element-wise absolute error over all 4096 outcomes | $10^{-2}$ |
+| Correlators $\langle Z_j\rangle$, $\langle X_j\rangle$ | Max absolute error over all outcomes and sites | $10^{-2}$ |
+
+**Pass condition:** all four criteria are satisfied simultaneously.
+
+---
+
+## 8. Notes on Reproducibility
+
+- The computation is fully deterministic; no random seed is needed.
+- All $2^{12} = 4096$ measurement outcomes are computed in parallel (no stochastic sampling).
+- The golden reference uses imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$. Agents using ITE with the recommended parameters or exact diagonalisation should meet the tolerances.
+- For this paramagnetic-ancilla variant, the agent receives `ancilla_state.npy` as input, so the ancilla state is pinned exactly.

tasks/computing_math/ising_post_measurement_1/n14_critical_u01_no_correlators/input/task_specification.md

@@ -0,0 +1,151 @@
+# Post-Measurement State Preparation for a Critical 1D Quantum Ising Chain
+
+## 1. Task Goal
+
+Given a quantum system consisting of a critical 1D Ising chain and an ancilla chain (each with $N = 14$ qubits), execute the following protocol:
+
+1. Compute the critical ground state $|\psi_c\rangle$ of the transverse-field Ising model at criticality.
+2. Set the ancilla state $|\psi_a\rangle = |\psi_c\rangle$ (critical ancilla).
+3. Apply the combined inter-chain unitary and projective measurement operator to produce post-measurement states and their probabilities for all $2^N = 16384$ measurement outcomes.
+4. Compute the one-site reduced density matrix (RDM) at site $j = 1$.
+
+---
+
+## 2. Conventions
+
+The following conventions are used throughout this specification and in all input/output files.
+
+**Pauli matrices.**
+
+$$Z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, \quad X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \quad I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$
+
+**Single-qubit states.**
+
+$$|0\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |1\rangle = \begin{pmatrix} 0 \\ 1 \end{pmatrix}, \quad |{+}\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$$
+
+Note: $Z|0\rangle = +|0\rangle$ and $Z|1\rangle = -|1\rangle$.
+
+**Computational basis ordering (big-endian).** An $N$-qubit computational basis state $|s_1 s_2 \ldots s_N\rangle$ (with $s_i \in \{0, 1\}$) is stored at array index
+
+$$\text{index} = s_1 \cdot 2^{N-1} + s_2 \cdot 2^{N-2} + \cdots + s_N \cdot 2^0$$
+
+Qubit 1 is the most significant bit. For example, with $N = 3$: $|000\rangle \to 0$, $|001\rangle \to 1$, $|010\rangle \to 2$, $|100\rangle \to 4$, $|111\rangle \to 7$.
+
+**Site indexing.** Sites are labeled $1, 2, \ldots, N$ (1-indexed). In the tensor-product Hilbert space $(\mathbb{C}^2)^{\otimes N}$, site $i$ corresponds to the $i$-th tensor factor (from the left).
+
+---
+
+## 3. Physics Protocol
+
+### 3.1 Hamiltonian
+
+The critical 1D transverse-field Ising Hamiltonian with periodic boundary conditions is:
+
+$$H = -\sum_{i=1}^{N} Z_i Z_{i+1} - \sum_{i=1}^{N} X_i, \qquad Z_{N+1} \equiv Z_1 \text{ (PBC)}$$
+
+with $N = 14$.
+
+### 3.2 Critical Ground State
+
+$|\psi_c\rangle$ is the ground state (lowest-energy eigenstate) of $H$. The agent may obtain it by any correct method (e.g., exact diagonalisation, imaginary time evolution, DMRG). For ITE, the recommended Trotterised scheme is
+
+$$|\psi_c\rangle \approx \prod_{k=1}^{K} \Bigl[e^{\Delta t \sum_i Z_i Z_{i+1}} \;\cdot\; \prod_i e^{\Delta t\, X_i}\Bigr]\;|{+}\rangle^{\otimes N}$$
+
+with $\Delta t = 0.0002$, $K = 50000$ (total imaginary time $T = K\Delta t = 10$), maintaining unit norm throughout. The Trotter ordering (whether the $Z$-part or the $X$-part is applied first) does not affect the converged result at the recommended parameters.
+
+### 3.3 Ancilla State
+
+This variant uses the **critical ancilla**: $|\psi_a\rangle = |\psi_c\rangle$ (an identical copy of the ground state). No ancilla file is provided.
+
+### 3.4 Inter-Chain Unitary and Measurement
+
+The site-$j$ unitary coupling the critical chain to the ancilla chain is
+
+$$U_j = \exp\!\bigl(i\,u\,(Z_j - \theta\,I)\otimes \tilde{X}_j\bigr)$$
+
+with coupling strength $u = 0.1$ and constant $\theta = -1$. The tilde denotes operators on the ancilla chain. After applying $\prod_j U_j$ to the initial state $|\psi_c\rangle \otimes |\psi_a\rangle$ and projectively measuring the ancilla in the $Z$-basis, each measurement outcome $|\tilde{s}\rangle = |\tilde{s}_1 \tilde{s}_2 \ldots \tilde{s}_N\rangle$ (with $\tilde{s}_j \in \{|0\rangle, |1\rangle\}$) produces a post-measurement state of the critical chain:
+
+$$|\psi_{\tilde{s}}\rangle = \frac{e^{iH'}\; e^{-H_m/2}\;|\psi_c\rangle}{\bigl\|e^{-H_m/2}\;|\psi_c\rangle\bigr\|}$$
+
+The effective Hamiltonians $H'$ and $H_m$ are both diagonal in the computational basis and **depend on the measurement outcome** $|\tilde{s}\rangle$: each of the $2^N = 16384$ outcomes yields a different pair $(H', H_m)$.
+
+**Ancilla coefficients.** For each outcome $|\tilde{s}\rangle$, define:
+
+$$a(j) = \frac{\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}, \qquad a(j,k) = \frac{\langle\tilde{s}|\tilde{X}_j\tilde{X}_k|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}$$
+
+Since $|\tilde{s}\rangle$ is a computational basis state, $\langle\tilde{s}|\psi_a\rangle = \psi_a[\text{index}(\tilde{s})]$, the amplitude of the ancilla state at array index $\text{index}(\tilde{s})$. Similarly, $\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle = \psi_a[\text{index}(\tilde{s} \text{ with bit } j \text{ flipped})]$, since $\tilde{X}_j$ flips qubit $j$ in each computational basis component.
+
+**Derived quantities:**
+
+$$V_{jk} = a(j,k) - a(j)\,a(k), \qquad m_j = (-2\theta)\Bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\Bigr)$$
+
+With $\theta = -1$, this simplifies to $m_j = 2\bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\bigr)$.
+
+**Effective Hamiltonians** (diagonal in the computational basis):
+
+$$H' = u\sum_j a(j)\,Z_j, \qquad H_m = u^2\sum_j m_j\,Z_j + u^2\sum_{j\neq k} V_{jk}\,Z_jZ_k$$
+
+**Measurement probability:**
+
+$$p(\tilde{s}) \propto \langle\psi_c|e^{-H_m}|\psi_c\rangle \cdot |\langle\tilde{s}|\psi_a\rangle|^2$$
+
+normalised so that $\sum_{\tilde{s}} p(\tilde{s}) = 1$.
+
+### 3.5 Reduced Density Matrix
+
+The one-site RDM at site $j = 1$ is $\rho_1 = \text{Tr}_{\overline{1}}(|\psi_{\tilde{s}}\rangle\langle\psi_{\tilde{s}}|)$, a $2\times 2$ Hermitian matrix, computed for every measurement outcome.
+
+---
+
+## 4. Input Files
+
+All input files are in the `input/` subdirectory.
+
+| File | Description |
+|---|---|
+| `input/config.json` | Parameters: `N=14`, `u=0.1`, `theta=-1`, `dt=0.0002`, `K=50000`, `ancilla_mode="critical"`, `site_j=1`. |
+
+Since `ancilla_mode` is `"critical"`, no ancilla file is provided; the agent uses $|\psi_a\rangle = |\psi_c\rangle$.
+
+---
+
+## 5. Output Files (Agent Produces)
+
+| File | Shape, dtype | Description |
+|---|---|---|
+| `critical_state.npy` | $(16384,)$, float64 | Ground state $\psi_c$. |
+| `post_probs.npy` | $(16384,)$, float64 | Measurement-outcome probabilities. Entry $i$ is $p(\tilde{s})$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `rdm_site1.npy` | $(16384, 2, 2)$, complex128 | One-site RDM at site 1. Entry $[i, :, :]$ is $\rho_1$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+
+---
+
+## 6. Tools / Software
+
+- **OS**: Ubuntu 22.04 LTS
+- **Python**: 3.10.12
+- **NumPy**: 2.2.6
+- **SciPy**: 1.15.3
+- **QuTiP**: 5.2.3 (optional; may use for exact diagonalisation)
+- Single machine. Peak memory ≈ 8 GB. Hilbert-space dimension $2^{14} = 16384$. Use an efficient implementation that avoids materializing full dense effective-Hamiltonian matrices unless your runtime can support it.
+
+---
+
+## 7. Evaluation Criteria
+
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
+
+| Quantity | Metric | Tolerance |
+|---|---|---|
+| Critical state $\psi_c$ | State infidelity $1 - F$, where $F = \lvert\langle\psi_c^{\text{ref}} \mid \psi_c^{\text{agent}}\rangle\rvert^2$ | $10^{-4}$ |
+| Probabilities $p(\tilde{s})$ | Max absolute error over all 16384 outcomes | $10^{-2}$ |
+| RDM $\rho_1$ | Max element-wise absolute error over all 16384 outcomes | $10^{-2}$ |
+
+**Pass condition:** all three criteria are satisfied simultaneously.
+
+---
+
+## 8. Notes on Reproducibility
+
+- The computation is fully deterministic; no random seed is needed.
+- All $2^{14} = 16384$ measurement outcomes are computed in parallel (no stochastic sampling).
+- The golden reference uses imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$. Agents using ITE with the recommended parameters or exact diagonalisation should meet the tolerances.

tasks/computing_math/ising_post_measurement_1/n16_critical_u01_no_correlators/input/task_specification.md

@@ -0,0 +1,151 @@
+# Post-Measurement State Preparation for a Critical 1D Quantum Ising Chain
+
+## 1. Task Goal
+
+Given a quantum system consisting of a critical 1D Ising chain and an ancilla chain (each with $N = 16$ qubits), execute the following protocol:
+
+1. Compute the critical ground state $|\psi_c\rangle$ of the transverse-field Ising model at criticality.
+2. Set the ancilla state $|\psi_a\rangle = |\psi_c\rangle$ (critical ancilla).
+3. Apply the combined inter-chain unitary and projective measurement operator to produce post-measurement states and their probabilities for all $2^N = 65536$ measurement outcomes.
+4. Compute the one-site reduced density matrix (RDM) at site $j = 1$.
+
+---
+
+## 2. Conventions
+
+The following conventions are used throughout this specification and in all input/output files.
+
+**Pauli matrices.**
+
+$$Z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, \quad X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \quad I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$
+
+**Single-qubit states.**
+
+$$|0\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |1\rangle = \begin{pmatrix} 0 \\ 1 \end{pmatrix}, \quad |{+}\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$$
+
+Note: $Z|0\rangle = +|0\rangle$ and $Z|1\rangle = -|1\rangle$.
+
+**Computational basis ordering (big-endian).** An $N$-qubit computational basis state $|s_1 s_2 \ldots s_N\rangle$ (with $s_i \in \{0, 1\}$) is stored at array index
+
+$$\text{index} = s_1 \cdot 2^{N-1} + s_2 \cdot 2^{N-2} + \cdots + s_N \cdot 2^0$$
+
+Qubit 1 is the most significant bit. For example, with $N = 3$: $|000\rangle \to 0$, $|001\rangle \to 1$, $|010\rangle \to 2$, $|100\rangle \to 4$, $|111\rangle \to 7$.
+
+**Site indexing.** Sites are labeled $1, 2, \ldots, N$ (1-indexed). In the tensor-product Hilbert space $(\mathbb{C}^2)^{\otimes N}$, site $i$ corresponds to the $i$-th tensor factor (from the left).
+
+---
+
+## 3. Physics Protocol
+
+### 3.1 Hamiltonian
+
+The critical 1D transverse-field Ising Hamiltonian with periodic boundary conditions is:
+
+$$H = -\sum_{i=1}^{N} Z_i Z_{i+1} - \sum_{i=1}^{N} X_i, \qquad Z_{N+1} \equiv Z_1 \text{ (PBC)}$$
+
+with $N = 16$.
+
+### 3.2 Critical Ground State
+
+$|\psi_c\rangle$ is the ground state (lowest-energy eigenstate) of $H$. The agent may obtain it by any correct method (e.g., exact diagonalisation, imaginary time evolution, DMRG). For ITE, the recommended Trotterised scheme is
+
+$$|\psi_c\rangle \approx \prod_{k=1}^{K} \Bigl[e^{\Delta t \sum_i Z_i Z_{i+1}} \;\cdot\; \prod_i e^{\Delta t\, X_i}\Bigr]\;|{+}\rangle^{\otimes N}$$
+
+with $\Delta t = 0.0002$, $K = 50000$ (total imaginary time $T = K\Delta t = 10$), maintaining unit norm throughout. The Trotter ordering (whether the $Z$-part or the $X$-part is applied first) does not affect the converged result at the recommended parameters.
+
+### 3.3 Ancilla State
+
+This variant uses the **critical ancilla**: $|\psi_a\rangle = |\psi_c\rangle$ (an identical copy of the ground state). No ancilla file is provided.
+
+### 3.4 Inter-Chain Unitary and Measurement
+
+The site-$j$ unitary coupling the critical chain to the ancilla chain is
+
+$$U_j = \exp\!\bigl(i\,u\,(Z_j - \theta\,I)\otimes \tilde{X}_j\bigr)$$
+
+with coupling strength $u = 0.1$ and constant $\theta = -1$. The tilde denotes operators on the ancilla chain. After applying $\prod_j U_j$ to the initial state $|\psi_c\rangle \otimes |\psi_a\rangle$ and projectively measuring the ancilla in the $Z$-basis, each measurement outcome $|\tilde{s}\rangle = |\tilde{s}_1 \tilde{s}_2 \ldots \tilde{s}_N\rangle$ (with $\tilde{s}_j \in \{|0\rangle, |1\rangle\}$) produces a post-measurement state of the critical chain:
+
+$$|\psi_{\tilde{s}}\rangle = \frac{e^{iH'}\; e^{-H_m/2}\;|\psi_c\rangle}{\bigl\|e^{-H_m/2}\;|\psi_c\rangle\bigr\|}$$
+
+The effective Hamiltonians $H'$ and $H_m$ are both diagonal in the computational basis and **depend on the measurement outcome** $|\tilde{s}\rangle$: each of the $2^N = 65536$ outcomes yields a different pair $(H', H_m)$.
+
+**Ancilla coefficients.** For each outcome $|\tilde{s}\rangle$, define:
+
+$$a(j) = \frac{\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}, \qquad a(j,k) = \frac{\langle\tilde{s}|\tilde{X}_j\tilde{X}_k|\psi_a\rangle}{\langle\tilde{s}|\psi_a\rangle}$$
+
+Since $|\tilde{s}\rangle$ is a computational basis state, $\langle\tilde{s}|\psi_a\rangle = \psi_a[\text{index}(\tilde{s})]$, the amplitude of the ancilla state at array index $\text{index}(\tilde{s})$. Similarly, $\langle\tilde{s}|\tilde{X}_j|\psi_a\rangle = \psi_a[\text{index}(\tilde{s} \text{ with bit } j \text{ flipped})]$, since $\tilde{X}_j$ flips qubit $j$ in each computational basis component.
+
+**Derived quantities:**
+
+$$V_{jk} = a(j,k) - a(j)\,a(k), \qquad m_j = (-2\theta)\Bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\Bigr)$$
+
+With $\theta = -1$, this simplifies to $m_j = 2\bigl(1 - a(j)^2 + \sum_{k \neq j} V_{jk}\bigr)$.
+
+**Effective Hamiltonians** (diagonal in the computational basis):
+
+$$H' = u\sum_j a(j)\,Z_j, \qquad H_m = u^2\sum_j m_j\,Z_j + u^2\sum_{j\neq k} V_{jk}\,Z_jZ_k$$
+
+**Measurement probability:**
+
+$$p(\tilde{s}) \propto \langle\psi_c|e^{-H_m}|\psi_c\rangle \cdot |\langle\tilde{s}|\psi_a\rangle|^2$$
+
+normalised so that $\sum_{\tilde{s}} p(\tilde{s}) = 1$.
+
+### 3.5 Reduced Density Matrix
+
+The one-site RDM at site $j = 1$ is $\rho_1 = \text{Tr}_{\overline{1}}(|\psi_{\tilde{s}}\rangle\langle\psi_{\tilde{s}}|)$, a $2\times 2$ Hermitian matrix, computed for every measurement outcome.
+
+---
+
+## 4. Input Files
+
+All input files are in the `input/` subdirectory.
+
+| File | Description |
+|---|---|
+| `input/config.json` | Parameters: `N=16`, `u=0.1`, `theta=-1`, `dt=0.0002`, `K=50000`, `ancilla_mode="critical"`, `site_j=1`. |
+
+Since `ancilla_mode` is `"critical"`, no ancilla file is provided; the agent uses $|\psi_a\rangle = |\psi_c\rangle$.
+
+---
+
+## 5. Output Files (Agent Produces)
+
+| File | Shape, dtype | Description |
+|---|---|---|
+| `critical_state.npy` | $(65536,)$, float64 | Ground state $\psi_c$. |
+| `post_probs.npy` | $(65536,)$, float64 | Measurement-outcome probabilities. Entry $i$ is $p(\tilde{s})$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+| `rdm_site1.npy` | $(65536, 2, 2)$, complex128 | One-site RDM at site 1. Entry $[i, :, :]$ is $\rho_1$ for the outcome with $\text{index}(\tilde{s}) = i$. |
+
+---
+
+## 6. Tools / Software
+
+- **OS**: Ubuntu 22.04 LTS
+- **Python**: 3.10.12
+- **NumPy**: 2.2.6
+- **SciPy**: 1.15.3
+- **QuTiP**: 5.2.3 (optional; may use for exact diagonalisation)
+- Single machine. Peak memory ≈ 64 GB. Hilbert-space dimension $2^{16} = 65536$. Use an efficient implementation that avoids materializing full dense effective-Hamiltonian matrices unless your runtime can support it.
+
+---
+
+## 7. Evaluation Criteria
+
+Use imaginary time evolution with $\Delta t = 0.0002$ and $K = 50000$.
+
+| Quantity | Metric | Tolerance |
+|---|---|---|
+| Critical state $\psi_c$ | State infidelity $1 - F$, where $F = \lvert\langle\psi_c^{\text{ref}} \mid \psi_c^{\text{agent}}\rangle\rvert^2$ | $10^{-4}$ |
+| Probabilities $p(\tilde{s})$ | Max absolute error over all 65536 outcomes | $10^{-2}$ |
+| RDM $\rho_1$ | Max element-wise absolute error over all 65536 outcomes | $10^{-2}$ |
+
+**Pass condition:** all three criteria are satisfied simultaneously.
+
+---
+
+## 8. Notes on Reproducibility
+
+- The computation is fully deterministic; no random seed is needed.
+- All $2^{16} = 65536$ measurement outcomes are computed in parallel (no stochastic sampling).
+- The golden reference uses imaginary time evolution with $\Delta t = 0.0002$, $K = 50000$. Agents using ITE with the recommended parameters or exact diagonalisation should meet the tolerances.

tasks/computing_math/mp_checkpoint_consolidation_v2/base/input/task_instructions.md

@@ -25,6 +25,5 @@ Your job:
 
 Rules:
 - Treat `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/base/input` as read-only.
-- Do not rely on evaluator-only files under `reference/`, `output_test_pos/`, or `output_test_neg/`.
 - You may self-check by loading the staged reference model and comparing logits on `input/reference_output/input_ids.pt`.
 - If you need a task-local Python environment, you may run `uv sync --frozen --project /media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/base/input/runtime_env`.

tasks/computing_math/mp_checkpoint_consolidation_v2/variant_2/input/task_instructions.md

@@ -0,0 +1,29 @@
+You are working on a Linux checkpoint-consolidation task.
+
+Task directory:
+- `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2`
+
+Visible inputs:
+- Raw checkpoint shards: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/checkpoints`
+- Framework source: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/framework`
+- Reference model config: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/reference_model/config.json`
+- Reference model code: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/reference_model/model.py`
+- Self-check inputs: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/reference_output/input_ids.pt`
+- Self-check target logits: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/reference_output/logits.pt`
+- Expected checkpoint key names: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/reference_output/expected_keys.json`
+- Optional task-local runtime manifest: `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/runtime_env/pyproject.toml` and `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/runtime_env/uv.lock`
+
+Variant facts:
+- Variant name: `variant_2`
+- Visible shard count: `32`
+- Parallel layout described by the submission: `4-way Tensor Parallel x 4-way Pipeline Parallel x 2-way Expert Parallel`
+
+Your job:
+1. Inspect the staged framework code and shard filenames to infer the sharding and serialization rules.
+2. Reconstruct a single-device checkpoint compatible with the staged reference model.
+3. Save exactly one file at `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/output/model.safetensors`.
+
+Rules:
+- Treat `/media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input` as read-only.
+- You may self-check by loading the staged reference model and comparing logits on `input/reference_output/input_ids.pt`.
+- If you need a task-local Python environment, you may run `uv sync --frozen --project /media/user/data/agenthle/machine_learning/mp_checkpoint_consolidation_v2/variant_2/input/runtime_env`.

tasks/education_info/marc_remediation_folio_overlay/base/input/TASK_PROMPT.md

@@ -8,13 +8,7 @@ Implement `scripts/remediate_catalog.py` so it can run on any case directory wit
 python scripts/remediate_catalog.py --case-dir input/public_case --output-dir outputs/public_case
 ```
 
-The evaluator uses the exact same interface for hidden cases:
-
-```bash
-python scripts/remediate_catalog.py --case-dir <hidden_case_dir> --output-dir <temporary_output_dir>
-```
-
-Each hidden case directory has the same file layout and column schema as `input/public_case/`: `legacy_records.xml`, `legacy_records.mrk`, `folio_instances.csv`, `authority_map.csv`, `location_map.csv`, and `cataloging_policy.md`. Your implementation must derive all behavior from those files and from the rules below, not from public record identifiers or public output values. Hidden cases may change the number of records, the create/update/suppress mix, the print-vs-online carrier mix, and the authority substitutions, but they do not introduce new input file types or undocumented rule families. The visible public case is deliberately larger than the hidden cases so you can test generalization across many examples before the evaluator runs the smaller hidden anti-hardcoding checks.
+Implement all behavior from the supplied case files and documented cataloging rules; do not hard-code record IDs or output values.
 
 The workflow should reflect a professional MarcEdit + FOLIO Data Import remediation pass:
 
@@ -34,6 +28,3 @@ Required outputs:
 - `qa_report.json`: machine-readable counts for records received, active records, duplicate suppressions, create/update actions, RDA conversions, authority replacements, and holdings/items emitted.
 - `remediation_summary.md`: short cataloging handoff memo listing decisions, unresolved assumptions, and FOLIO profile names.
 
-Evaluation is deterministic. The grader checks your visible outputs against reference files and reruns the same CLI on hidden synthetic batches bundled only with the evaluator. Do not hard-code public record IDs or output values; hidden cases use the same schema with different records and match conditions. Hidden cases are a small 10-point anti-hardcoding check, not a hard conjunctive pass/fail gate.
-
-Hidden-case fairness contract: every rule needed for hidden cases is present in the visible materials. In particular, `cataloging_policy.md` defines the duplicate survivor rule, FOLIO match precedence, print-vs-online RDA 336/337/338 values, authority-map behavior, and 949 holdings/item mapping. If a hidden record is online, use the same online carrier logic documented for the visible case; if a hidden record is a duplicate, use the same completeness ranking documented for the visible case.

tasks/education_info/marc_remediation_folio_overlay/base/input/starter_project/BUNDLE_MANIFEST.md

@@ -26,13 +26,13 @@ The visible development invocation is:
 python scripts/remediate_catalog.py --case-dir input/public_case --output-dir outputs/public_case
 ```
 
-The evaluator invokes the same script shape on hidden case directories:
+Use the documented script interface for every supplied case directory:
 
 ```bash
-python scripts/remediate_catalog.py --case-dir <hidden_case_dir> --output-dir <temporary_output_dir>
+python scripts/remediate_catalog.py --case-dir <case_dir> --output-dir <output_dir>
 ```
 
-Hidden case directories use the same schema as `input/public_case/` and contain the same required filenames. They may change record count, match status, duplicate presence, carrier type, and authority substitutions, but they do not introduce new file types or undocumented rule families. The visible public case provides 64 incoming records across many create/update/suppress examples; implement against the file schemas and `cataloging_policy.md`, not against public record IDs.
+Derive behavior from the supplied file schemas and `cataloging_policy.md` rather than record-specific values.
 
 ## Dependency Contract
 The grading evaluator is Python 3.12 stdlib-only and network-free. Your CLI may use Python stdlib or common open-source MARC helpers if available, but no external service, live FOLIO tenant, MarcEdit binary, or MarcEdit GUI is required during grading. MarcEdit 7.7.x is freeware workflow context, not an executed runtime dependency. FOLIO Sunflower R1 2025 is free/open-source workflow context and expected professional terminology.

tasks/education_info/marc_remediation_folio_overlay/base/input/starter_project/README.md

@@ -1,3 +1,3 @@
 # Starter Project
 
-Implement `scripts/remediate_catalog.py`. Use the visible inputs in `input/public_case/` and write outputs to `outputs/public_case/`. Hidden evaluator cases use the same schema.
+Implement `scripts/remediate_catalog.py`. Use the visible inputs in `input/public_case/` and write outputs to `outputs/public_case/`. Implement against the documented schema.

tasks/education_info/marc_remediation_folio_overlay/base/input/starter_project/TASK_PROMPT.md

@@ -8,13 +8,7 @@ Implement `scripts/remediate_catalog.py` so it can run on any case directory wit
 python scripts/remediate_catalog.py --case-dir input/public_case --output-dir outputs/public_case
 ```
 
-The evaluator uses the exact same interface for hidden cases:
-
-```bash
-python scripts/remediate_catalog.py --case-dir <hidden_case_dir> --output-dir <temporary_output_dir>
-```
-
-Each hidden case directory has the same file layout and column schema as `input/public_case/`: `legacy_records.xml`, `legacy_records.mrk`, `folio_instances.csv`, `authority_map.csv`, `location_map.csv`, and `cataloging_policy.md`. Your implementation must derive all behavior from those files and from the rules below, not from public record identifiers or public output values. Hidden cases may change the number of records, the create/update/suppress mix, the print-vs-online carrier mix, and the authority substitutions, but they do not introduce new input file types or undocumented rule families. The visible public case is deliberately larger than the hidden cases so you can test generalization across many examples before the evaluator runs the smaller hidden anti-hardcoding checks.
+Implement all behavior from the supplied case files and documented cataloging rules; do not hard-code record IDs or output values.
 
 The workflow should reflect a professional MarcEdit + FOLIO Data Import remediation pass:
 
@@ -34,6 +28,3 @@ Required outputs:
 - `qa_report.json`: machine-readable counts for records received, active records, duplicate suppressions, create/update actions, RDA conversions, authority replacements, and holdings/items emitted.
 - `remediation_summary.md`: short cataloging handoff memo listing decisions, unresolved assumptions, and FOLIO profile names.
 
-Evaluation is deterministic. The grader checks your visible outputs against reference files and reruns the same CLI on hidden synthetic batches bundled only with the evaluator. Do not hard-code public record IDs or output values; hidden cases use the same schema with different records and match conditions. Hidden cases are a small 10-point anti-hardcoding check, not a hard conjunctive pass/fail gate.
-
-Hidden-case fairness contract: every rule needed for hidden cases is present in the visible materials. In particular, `cataloging_policy.md` defines the duplicate survivor rule, FOLIO match precedence, print-vs-online RDA 336/337/338 values, authority-map behavior, and 949 holdings/item mapping. If a hidden record is online, use the same online carrier logic documented for the visible case; if a hidden record is a duplicate, use the same completeness ranking documented for the visible case.

tasks/education_info/marc_remediation_folio_overlay/base/input/starter_project/input/public_case/cataloging_policy.md

@@ -11,4 +11,4 @@ Apply these rules to every visible and hidden batch.
 
 ## Hidden-case rule coverage
 
-The evaluator's hidden cases use these same rules and file schemas. They may contain a different number of records, different OCLC/LCCN/ISBN matches, different authority-map rows, different print-vs-online records, and different duplicate groups. Do not special-case public record IDs. Apply the documented match precedence, duplicate survivor ranking, RDA carrier mapping, authority replacement, and 949 holdings/item mapping uniformly to every case directory passed through `--case-dir`.
+Do not special-case record IDs. Apply the documented match precedence, duplicate survivor ranking, RDA carrier mapping, authority replacement, and 949 holdings/item mapping uniformly to every case directory passed through `--case-dir`.

tasks/education_info/marc_remediation_folio_overlay/base/input/starter_project/scripts/remediate_catalog.py

@@ -12,7 +12,7 @@ def parse_args():
         help=(
             "Directory containing legacy_records.xml, legacy_records.mrk, "
             "folio_instances.csv, authority_map.csv, location_map.csv, and "
-            "cataloging_policy.md. The evaluator passes hidden case directories "
+            "cataloging_policy.md. Process each supplied case directory "
             "with the same schema through this argument."
         ),
     )

tasks/education_info/marc_remediation_folio_overlay/base/input/task_spec.md

@@ -35,5 +35,4 @@ Leave the completed project at:
 output/submission/
 ```
 
-The evaluator checks visible outputs and reruns the same CLI on evaluator-owned hidden cases.
-Hidden cases and reference outputs are not available during solve time.
+Implement the CLI from the supplied case files and documented cataloging rules.

tasks/education_info/moodle_gradebook_closeout_reconciliation/base/input/TASK_PROMPT.md

@@ -66,8 +66,7 @@ Submit these files:
 
 - The backup is a zip-format `.mbz` file. You may inspect it directly with the provided helper script or any standard archive tooling.
 - Do not alter the immutable raw student, item, or submission data inside the backup. Only the benchmark-designated editable files should change.
-- Use `starter_project/tools/rebuild_exports.py` after you repair the backup. The evaluator will compare your rebuilt exports against hidden references.
-- `visible_cases.csv` shows only a subset of the edge cases. Hidden cases stay within the same rule families.
+- Use `starter_project/tools/rebuild_exports.py` after you repair the backup.
 - All student names, SIS identifiers, and records in the bundle are synthetic benchmark artifacts, not real student data.
 - The expected `OneRoster 1.2` output files are exactly:
   - `users.csv`

tasks/education_info/moodle_gradebook_closeout_reconciliation/base/input/bundle_lib.py

@@ -20,7 +20,7 @@ COURSE_TITLE = "Applied Statistics for Public Policy"
 COURSE_SHORTNAME = "MATHSTAT203-SP26"
 SECTIONS = ["SEC-A", "SEC-B", "SEC-C"]
 RNG_SEED = 20260418
-# Scoring pass thresholds are evaluator-only and intentionally omitted from the solve-time helper.
+# This helper implements the documented export contract.
 
 FIRST_NAMES = [
     "Alex", "Avery", "Bailey", "Blair", "Cameron", "Casey", "Devon", "Drew",

tasks/engineering/sumo_urban_am_peak_calibration/base/input/starter_project/README.md

@@ -36,6 +36,6 @@ See `../task_prompt.md` section "What you submit".
 ## Synthetic data disclaimer
 
 All "observed" inputs (detector counts, Bluetooth travel times, ATSPM
-signal timing, LEHD-seed OD) are SYNTHESIZED from a hidden gold SUMO
+signal timing, LEHD-seed OD) are synthesized from a fixed SUMO
 simulation. No real-world proprietary data is redistributed. See
 `inputs/PROVENANCE.md`.

tasks/engineering/sumo_urban_am_peak_calibration/base/input/task_prompt.md

@@ -27,10 +27,7 @@ Linux VM, no GPU**.
 
 ## Inputs you receive (`starter_project/`)
 
-All "observed" data in this task is **synthesized from a hidden gold
-simulation** (documented in `inputs/PROVENANCE.md`); treat the
-observations as ground truth for calibration. No real-world proprietary
-data is used.
+All "observed" data in this task is synthetic.
 
 ```
 starter_project/
@@ -162,7 +159,6 @@ The evaluator is deterministic. It:
    `additionals/tlsLogic.add.xml` referenced by `<additional-files>`),
    NOT the declarative report JSON
 3. Re-runs your `sumocfg` headless on the declared `agent_seeds` from
-   `calibration_report.json` plus **3 evaluator-hidden seeds**
 4. Computes all performance metrics from the evaluator's own outputs,
    not from any files you submit
 
@@ -175,10 +171,7 @@ The evaluator is deterministic. It:
 | 3 | TLS per-phase splits (from **loaded** file) | all within ±3 s each phase |
 | 4 | TLS phase-state sequences (from **loaded** file) | cyclic-equivalent to ATSPM |
 | 5 | AM-peak simulated trip total vs control | ±5% |
-| 6 | Aggregate flow RMSE on 3 **hidden holdout** detectors | ≤ 12% of mean observed |
 | 7 | Corridor TT RMSE on both corridors (from evaluator rerun) | ≤ 15% each |
-| 8 | GEH < 5 on hidden holdout detectors | ≥ 85% |
-| 9 | Evaluator rerun completes without simulation errors across the declared `agent_seeds` plus 3 hidden seeds | 100% |
 | 10 | Public-detector flow means in `calibration_report.json` reproduce from evaluator reruns over the declared `agent_seeds` | per-detector ±1% |
 | 11 | Corridor TT in `calibration_report.json` reproducible from evaluator rerun | ±2% |
 | 12 | `calibration_report.json` schema conformance: cardinalities plus runtime shape checks on sensitivity responses | all |
@@ -204,7 +197,6 @@ Empirical gold-vs-weakest gap: **7 gates**.
 ## Runtime expectations
 
 - Single evaluator rerun (scoring one submission across the declared
-  agent seeds plus 3 hidden seeds, gate aggregation): **~8 s wall on 16 vCPU**.
 - Full 4-baseline discrimination rebuild
   (`generate_data.py` + `make_baselines.py` +
   `evaluate.py × 4` + ZIP packaging via `rebuild_all.py`):

tasks/health_medicine/causal_ihdp_ite_estimation_6a_v1/base/input/feature_schema.json

@@ -76,7 +76,7 @@
   ],
   "notes": [
     "Hidden official test never enters the agent filesystem.",
-    "dummy_test/test.csv matches the hidden test schema, row count, and (replication, unit_id) order, but not the hidden feature values.",
+    "dummy_test/test.csv demonstrates the test schema, row count, and (replication, unit_id) order without supplying target feature values.",
     "ite_hat must equal mu1_hat - mu0_hat for every row.",
     "The GANITE paper is the task context; reproducing GANITE itself is not required."
   ],

tasks/health_medicine/flusight_offline_hosp_forecast_2024_12_14/base/input/TASK_INSTRUCTIONS.md

@@ -49,5 +49,4 @@ Secondary tie-breakers:
 ## Notes
 
 - This is a simplified Agent-HLE adaptation of FluSight: **no full quantile ladder is required**.
-- The hidden evaluator uses finalized future values for the four target weeks.
 - The historical snapshot intentionally reflects the data available **as of** the reference date, not later backfilled revisions.

tasks/health_medicine/simglucose_safe_basal_control_instance_1/base/input/public/submission_format.md

@@ -33,4 +33,4 @@ The evaluator creates a **fresh controller instance per episode**.
 - `report.md`
 - `training_code/*`
 
-The hidden evaluator does **not** install extra packages beyond the pinned runtime. Any extra code needed by `controller.py` should be vendored into the submission directory or otherwise available in the pinned environment.
+No extra packages are available beyond the pinned runtime. Any extra code needed by `controller.py` should be vendored into the submission directory or otherwise available in the pinned environment.

tasks/health_medicine/simglucose_safe_basal_control_instance_1/base/input/public/task.md

@@ -1,6 +1,6 @@
 # Task
 
-Implement `controller.py` so that the hidden evaluator can call `build_controller()` and obtain a controller that safely regulates glucose in SimGlucose.
+Implement `controller.py` with `build_controller()` returning a controller that safely regulates glucose in SimGlucose.
 
 This task is meant to represent **preclinical in silico controller prototyping and validation** for closed-loop diabetes-control R&D.
 

tasks/health_medicine/simglucose_safe_basal_control_instance_1/base/input/public/wrapper_spec.md

@@ -25,7 +25,3 @@ The controller returns a single scalar basal command in **U/min**.
 - no meal announcement is exposed
 - no patient identity is exposed
 - no raw BG is exposed to the controller
-
-## Evaluator-only information
-
-The evaluator may still access native runtime signals such as `bg`, `time`, `meal`, or `patient_name` for scoring and auditing. Those fields are not available to the controller.

tasks/life_sciences/spatial_transcriptomics_spatial_domain_identification/base/input/README.md

@@ -8,4 +8,4 @@ Visible directory structure:
 - `runtime_env/` — agent-facing `uv` manifest for Scanpy / Squidpy tooling
 - `task_prompt.md` — operator-facing requirements
 
-Do not look for manual cortical-layer annotations in `input/`; they are evaluator-only.
+Use only the supplied matrices and metadata.

tasks/physical_sciences/adapt_vqe_molecular_energy/base/input/problem_spec.md

@@ -123,7 +123,7 @@ Write `results.json` with this structure:
 }
 ```
 
-The evaluator computes hidden reference errors and pass / fail outcomes; do not add extra evaluator-only fields.
+Keep the result schema limited to the documented fields.
 
 ## Constraints
 

tasks/physical_sciences/climate_prediction/base/input/task_prompt.md

@@ -25,4 +25,4 @@ Build an end-to-end climate-emulation pipeline from the staged CMIP6-derived Zar
 
 ## Important Note
 
-The staged visible `data.zarr` intentionally masks the held-out `ssp245` test-window target labels (`tas`, `pr`) as `NaN`. Use the training split to learn a predictor; do not expect hidden test labels to be directly readable from the visible input.
+The staged visible `data.zarr` intentionally masks the held-out `ssp245` test-window target labels (`tas`, `pr`) as `NaN`. Use the training split to learn a predictor; the test-window target labels are not supplied.

tasks/transport_safety/abm_hangzhou_metro/base/input/task_prompt.md

@@ -45,7 +45,7 @@ All time columns (`start_time`, `end_time_simulate`, `end_time_real`, `duration_
 must be **integers** (minutes). Round to the nearest minute if your simulation produces fractional values.
 
 The benchmark expects broad coverage of the visible AFC records. A solver that simply copies the observed `end_time` from
-`afc_hangzhou.csv` into `end_time_simulate` will fail hidden evaluation even if the visible correlation looks perfect.
+`afc_hangzhou.csv` into `end_time_simulate` is invalid even if the visible correlation looks perfect.
 
 `validation_report.txt` should contain these summary lines, one per line:
 

tasks/visual_media/skeletal_animation_reproduction/skeletal_animation_reproduction_singing_anime_character/input/task_brief.md

@@ -13,7 +13,6 @@ Required submission:
 
 Constraints:
 - Focus on body motion only. Face and lip sync are not evaluated in v1.
-- Evaluation will compare your submitted `preview.mp4` against a hidden clean reference video.
-- Evaluation will also replay your `final.blend` in a canonical setup and compare that replay to your submitted preview.
+- The submitted `preview.mp4` and the animation replayed from `final.blend` must both reproduce the supplied body motion.
 - Your Blender file must contain the actual rig and animation that produce the submitted preview.
-- The visible reference video is watermarked. Your submission is still judged against the clean hidden reference motion.
+- The watermark in the supplied reference video is not part of the target body motion.

tasks/visual_media/skeletal_animation_reproduction/skeletal_animation_reproduction_white_cyborg_idle/input/task_brief.md

@@ -0,0 +1,7 @@
+Reproduce the body motion from the provided reference video using the supplied white cyborg mesh.
+
+Required submission:
+- Save your Blender scene as output/submission/final.blend
+- Render a matching preview as output/submission/preview.mp4
+
+Ensure both the rendered preview and the replayed Blender scene reproduce the supplied reference motion.

tasks/visual_media/skeletal_animation_reproduction/skeletal_animation_reproduction_white_cyborg_run/input/task_brief.md

@@ -0,0 +1,7 @@
+Reproduce the body motion from the provided reference video using the supplied white cyborg mesh.
+
+Required submission:
+- Save your Blender scene as output/submission/final.blend
+- Render a matching preview as output/submission/preview.mp4
+
+Ensure both the rendered preview and the replayed Blender scene reproduce the supplied reference motion.

tasks/visual_media/skeletal_animation_reproduction/skeletal_animation_reproduction_white_cyborg_walk/input/task_brief.md

@@ -0,0 +1,7 @@
+Reproduce the body motion from the provided reference video using the supplied white cyborg mesh.
+
+Required submission:
+- Save your Blender scene as output/submission/final.blend
+- Render a matching preview as output/submission/preview.mp4
+
+Ensure both the rendered preview and the replayed Blender scene reproduce the supplied reference motion.

tasks/visual_media/uv_reproduction/uv_reproduction_samurai/input/task_brief.md

@@ -0,0 +1,24 @@
+# Samurai UV Reproduction Task
+
+Use Blender to reproduce the target UV configuration for the provided whole-object Samurai mesh.
+
+Official input:
+- Read the source model from `/input/samurai.obj`
+- Use `/input/samurai.mtl` if the OBJ import expects it
+- Use the reference bundle under `/input/reference_images` to understand the target UV direction
+
+Required submission:
+- Export a mesh with UVs as `/output/submission/samurai.obj`
+
+Optional submission:
+- `/output/submission/final.blend`
+
+Submission rules:
+- Do not submit screenshots, UV layout images, checker images, or copied reference pictures as your answer
+- The submitted mesh is the authoritative deliverable; UV evidence may be regenerated in Blender
+- Your submission must contain a UV layer on the mesh
+
+Notes:
+- The task is whole-object, not per-part
+- The input mesh is the provided lowpoly whole-object Samurai asset, not the original highpoly source
+- `/software` is reserved for provenance assets and may be empty in this package version