Publish hemibrain-heading-fn-2026-09-15 (Fruit Fly Brain, 2026-09-15)

artifacts/hemibrain-heading-fn-2026-09-15/HEMIBRAIN_HEADING_FUNCTIONAL.md

@@ -0,0 +1,74 @@
+# Hemibrain functional velocity-to-heading integrator
+
+**Substrate:** Janelia hemibrain `hemibrain:v1.2.1` (CC-BY), 1015 nodes / 75011 edges. **Functional probe; no biological claim.**
+
+**Task:** `hemibrain-heading-velocity-integrator-v1` (circular-heading-recall, higher better).
+**Outcome:** `topology-advantage` (one of topology-advantage, weight-advantage, null-result).
+
+## Preregistered functional target
+
+- **goal:** A bounded recurrent circuit on the measured EPG compass ring must integrate a signed angular-velocity sequence into a persistent heading bump and report the integrated heading at the end of the rollout.
+- **rationale:** Angular-velocity integration and heading maintenance are the native function of the central-complex compass ring, so the measured EPG wiring is genuinely expected to help here (unlike the wave-5 descriptive scalar readout, which injected an isotropic scalar and scored 0.239 R^2 against 0.390 for an Erdos-Renyi graph).
+- **input:** A signed angular velocity omega_t in {-2,-1,0,1,2} at each of the 12 steps, delivered as the change of a declared 16-column von Mises heading bump reference; positive and negative omega therefore carry opposite signed increments.
+- **dynamics:** Bounded h_{t+1} = tanh(gain * P h_t + velocity_sign * (bump(phi_t) - bump(phi_{t-1}))) on the measured signed, row-normalised EPG operator P restricted to the compass ring, rescaled per graph to unit spectral radius.
+- **readout:** Fixed population-vector decoder over the declared 16-column heading bump basis; the decoded phase is compared with the true integrated heading by a cosine on held-out episodes.
+- **circuit:** Measured EPG compass ring; EPG columns are parsed from neuPrint instance-name suffixes _L1.._L8 and _R1.._R8 (L1..L8 -> columns 0..7, R1..R8 -> columns 8..15). Delta7, PEN_a, PEN_b, PEG and the FB/IB/ExR helpers are excluded.
+
+## Declared circuit
+
+- EPG compass ring: **46** cells over **16** declared columns (L1..L8 = 0..7, R1..R8 = 8..15).
+- Bounded tanh recurrence, `steps = 12`, `omega_max = 2`, `gain = 1.00`, `kappa = 3.00`.
+- Nodes per column: `{'0': 2, '1': 2, '2': 3, '3': 3, '4': 3, '5': 3, '6': 3, '7': 4, '8': 2, '9': 2, '10': 3, '11': 3, '12': 3, '13': 3, '14': 3, '15': 4}`.
+
+## Control ladder (seeds = 10, draws = 5)
+
+| rung | mean | 95% CI |
+|---|---:|---|
+| measured-graph | 0.5042 | [0.4910, 0.5180] |
+| shared-init-random | 0.1971 | [0.1847, 0.2089] |
+| erdos-renyi | 0.1971 | [0.1847, 0.2089] |
+| configuration-model | 0.2089 | [0.1966, 0.2207] |
+| degree-preserving | 0.2066 | [0.1920, 0.2198] |
+| weight-permuted | 0.3767 | [0.3637, 0.3939] |
+| topology-randomized | 0.1891 | [0.1779, 0.1997] |
+| function-matched-positive-control | 0.5042 | [0.4910, 0.5180] |
+
+### Paired contrasts (measured minus null)
+
+| contrast | effect | 95% CI | sign-flip p | Holm reject | MDE (80%%) | power@obs |
+|---|---:|---|---:|:---:|---:|---:|
+| measured-graph minus shared-init-random | 0.3071 | [0.2933, 0.3215] | 0.001953 | yes | 0.0212 | 1.0000 |
+| measured-graph minus erdos-renyi | 0.3071 | [0.2933, 0.3215] | 0.001953 | yes | 0.0212 | 1.0000 |
+| measured-graph minus configuration-model | 0.2953 | [0.2791, 0.3131] | 0.001953 | yes | 0.0256 | 1.0000 |
+| measured-graph minus degree-preserving | 0.2976 | [0.2846, 0.3136] | 0.001953 | yes | 0.0221 | 1.0000 |
+| measured-graph minus weight-permuted | 0.1275 | [0.1155, 0.1421] | 0.001953 | yes | 0.0201 | 1.0000 |
+| measured-graph minus topology-randomized | 0.3150 | [0.3006, 0.3300] | 0.001953 | yes | 0.0224 | 1.0000 |
+| function-matched-positive-control versus its mismatched counterpart | 0.8661 | [0.8479, 0.8870] | 0.001953 | yes | 0.0296 | 1.0000 |
+
+## Sign-flip control (intact velocity sign vs flipped input)
+
+| quantity | value |
+|---|---:|
+| intact mean | 0.5042 |
+| sign-flipped mean | -0.3619 |
+| all six matched nulls positive | True |
+| sign-flip positive control | True |
+
+## Limitations
+
+- The recurrent substrate is the measured EPG compass ring (EPG cells only); Delta7, PEN_a, PEN_b, PEG and the FB/IB/ExR columnar helpers are excluded from the functional circuit, so the reported score is a ring-specific, not whole-circuit, statement.
+- The EPG ring coordinate is parsed from instance-name suffixes L1..L8/R1..R8; all recovered EPG instances carry the PB08 glomerulus label, so the declared 16-column order is an annotation-derived prior, not an electrophysiological measurement.
+- The readout is a fixed population-vector decoder over the declared bump basis; it is identical across every rung, so the contrast isolates the recurrent operator rather than a fitted readout.
+- The dynamics are a bounded (tanh) linearisation of the measured signed operator with a per-graph spectral-radius rescaling; they are a declared model of angular-velocity integration, not a biophysical simulation.
+- Degree-preserving rewiring uses the full 5 x |E| directed double-edge-swap budget and verifies the in/out degree sequences and self-loop count exactly.
+- The functional target was chosen and piloted on the same measured subgraph before the frozen run; it is a preregistered evaluation of a hypothesis about this circuit, not an out-of-sample discovery.
+- The 10 paired seeds give descriptive intervals; the exact sign-flip is a secondary check and the normal-theory MDE/power is optimistic at small n.
+
+## Non-claims
+
+- This functional probe claims no biological advantage, general memory, AGI or programming expertise.
+- The EPG ring coordinate is parsed from neuPrint instance names (L1..L8 / R1..R8); it is the declared heading basis, not a measured physiological map.
+- An advantage over a matched null is necessary but never sufficient for a biological claim; negatives are retained and never reinterpreted.
+
+Runtime: 97.72s; seeds [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]; draws 5; protocol `ffb-hemibrain-heading-fn-v1` sha256 `9ad3b76b98027850`.
+

artifacts/hemibrain-heading-fn-2026-09-15/PROVENANCE.md

@@ -0,0 +1,15 @@
+# HEMIBRAIN functional velocity-to-heading integrator provenance
+
+- Dataset: `hemibrain:v1.2.1` (Janelia hemibrain, CC-BY), accessed through the neuPrint HTTP API at https://neuprint.janelia.org.
+- Recurrent substrate: measured EPG compass ring (46 cells).
+- Preregistration: `ffb-hemibrain-heading-fn-v1` (`hf_space/assets/hemibrain_heading_protocol.json`), sha256 `9ad3b76b98027850b7d3652eaed2d2e72e775ed7f6635405dde6d3e418117aee`.
+- Ladder contract: `fw2k-control-ladder-v1` (`hf_space/assets/control_ladder_protocol.json`).
+- Seeds: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]; draws: 5; bootstrap resamples: 100000.
+- Outcome: `topology-advantage`.
+
+## Non-claims
+
+- This functional probe claims no biological advantage, general memory, AGI or programming expertise.
+- The EPG ring coordinate is parsed from neuPrint instance names (L1..L8 / R1..R8); it is the declared heading basis, not a measured physiological map.
+- An advantage over a matched null is necessary but never sufficient for a biological claim; negatives are retained and never reinterpreted.
+

artifacts/hemibrain-heading-fn-2026-09-15/manifest.json

@@ -0,0 +1,54 @@
+{
+  "dataset": "Janelia hemibrain",
+  "dataset_version": "hemibrain:v1.2.1",
+  "license": "CC-BY (Janelia hemibrain)",
+  "metric_name": "circular-heading-recall",
+  "n_edges": 75011,
+  "n_nodes": 1015,
+  "n_ring_nodes": 46,
+  "outcome": "topology-advantage",
+  "outputs": [
+    {
+      "bytes": 14978,
+      "path": "report.json",
+      "sha256": "86f5398fbb164f9e7244d98d3207fd2b6d6d1b77b5ccc54632569c98eac102e7"
+    },
+    {
+      "bytes": 5866,
+      "path": "HEMIBRAIN_HEADING_FUNCTIONAL.md",
+      "sha256": "943c500b30512327fd5f769188d7f84bc2878f67108b97856ecdc43d0216afb7"
+    },
+    {
+      "bytes": 1053,
+      "path": "PROVENANCE.md",
+      "sha256": "8dad512d2dcedc96100fb8960ad9b41a52bb352a56616347b06c77339c0ed780"
+    }
+  ],
+  "protocol_id": "ffb-hemibrain-heading-fn-v1",
+  "protocol_sha256": "9ad3b76b98027850b7d3652eaed2d2e72e775ed7f6635405dde6d3e418117aee",
+  "provenance": {
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 707,
+    "degree_swap_multiplier": 5,
+    "draws": 5,
+    "protocol_id": "ffb-hemibrain-heading-fn-v1",
+    "seeds": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9
+    ],
+    "subgraph_path": "/home/user/dev/hf/hf-stage/publish/hemibrain-heading-2026-09-14/subgraph.npz",
+    "subgraph_sha256": "2a3047748a3f2cc92ce33e3f107653e086651e1335b92a3a48ff1d43113ec482",
+    "task_name": "hemibrain-heading-velocity-integrator-v1"
+  },
+  "runtime_seconds": 97.72,
+  "schema": "ffb-hemibrain-heading-functional-manifest-v1",
+  "task_name": "hemibrain-heading-velocity-integrator-v1"
+}

artifacts/hemibrain-heading-fn-2026-09-15/report.json

@@ -0,0 +1,420 @@
+{
+  "allowed_outcomes": [
+    "topology-advantage",
+    "weight-advantage",
+    "null-result"
+  ],
+  "controlled_signals": {
+    "all_six_matched_nulls_positive": true,
+    "intact_mean": 0.5041537244890605,
+    "intact_per_seed": [
+      0.5009363067257319,
+      0.5101880382152005,
+      0.4958363141332346,
+      0.49029154210683257,
+      0.5440061055552787,
+      0.4967840513293734,
+      0.46404791462874745,
+      0.4989087940589315,
+      0.5385947234054722,
+      0.5019434547318019
+    ],
+    "sign_flip_positive_control": true,
+    "sign_flipped_mean": -0.361909537365803,
+    "sign_flipped_per_seed": [
+      -0.34114424733449233,
+      -0.37783989041157917,
+      -0.34474249952276115,
+      -0.3728582018219357,
+      -0.39454232222415564,
+      -0.34922854585538066,
+      -0.36081596916071645,
+      -0.37804866799819675,
+      -0.3522300192659332,
+      -0.3476450100628791
+    ]
+  },
+  "dataset": {
+    "license": "CC-BY (Janelia hemibrain)",
+    "n_edges": 75011,
+    "n_nodes": 1015,
+    "name": "Janelia hemibrain",
+    "retrieved_utc": "2026-09-15T16:36:56+00:00",
+    "source": "https://neuprint.janelia.org",
+    "subgraph_path": "/home/user/dev/hf/hf-stage/publish/hemibrain-heading-2026-09-14/subgraph.npz",
+    "subgraph_sha256": "2a3047748a3f2cc92ce33e3f107653e086651e1335b92a3a48ff1d43113ec482",
+    "version": "hemibrain:v1.2.1"
+  },
+  "functional_target": {
+    "circuit": "Measured EPG compass ring; EPG columns are parsed from neuPrint instance-name suffixes _L1.._L8 and _R1.._R8 (L1..L8 -> columns 0..7, R1..R8 -> columns 8..15). Delta7, PEN_a, PEN_b, PEG and the FB/IB/ExR helpers are excluded.",
+    "dynamics": "Bounded h_{t+1} = tanh(gain * P h_t + velocity_sign * (bump(phi_t) - bump(phi_{t-1}))) on the measured signed, row-normalised EPG operator P restricted to the compass ring, rescaled per graph to unit spectral radius.",
+    "goal": "A bounded recurrent circuit on the measured EPG compass ring must integrate a signed angular-velocity sequence into a persistent heading bump and report the integrated heading at the end of the rollout.",
+    "input": "A signed angular velocity omega_t in {-2,-1,0,1,2} at each of the 12 steps, delivered as the change of a declared 16-column von Mises heading bump reference; positive and negative omega therefore carry opposite signed increments.",
+    "rationale": "Angular-velocity integration and heading maintenance are the native function of the central-complex compass ring, so the measured EPG wiring is genuinely expected to help here (unlike the wave-5 descriptive scalar readout, which injected an isotropic scalar and scored 0.239 R^2 against 0.390 for an Erdos-Renyi graph).",
+    "readout": "Fixed population-vector decoder over the declared 16-column heading bump basis; the decoded phase is compared with the true integrated heading by a cosine on held-out episodes."
+  },
+  "ladder": {
+    "contrasts": {
+      "function-matched-positive-control versus its mismatched counterpart": {
+        "available": true,
+        "ci_high": 0.8870005982692515,
+        "ci_low": 0.8479411395106446,
+        "effect": 0.8660632618548636,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      },
+      "measured-graph minus configuration-model": {
+        "available": true,
+        "ci_high": 0.31306091725008356,
+        "ci_low": 0.2791124508679759,
+        "effect": 0.2952527329934235,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      },
+      "measured-graph minus degree-preserving": {
+        "available": true,
+        "ci_high": 0.31359415514520395,
+        "ci_low": 0.28459780221632175,
+        "effect": 0.29757936233109833,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      },
+      "measured-graph minus erdos-renyi": {
+        "available": true,
+        "ci_high": 0.32149242660691146,
+        "ci_low": 0.2933282262021414,
+        "effect": 0.3070656802758186,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      },
+      "measured-graph minus shared-init-random": {
+        "available": true,
+        "ci_high": 0.32149242660691146,
+        "ci_low": 0.2933282262021414,
+        "effect": 0.3070656802758186,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      },
+      "measured-graph minus topology-randomized": {
+        "available": true,
+        "ci_high": 0.330008772517274,
+        "ci_low": 0.30063391311300486,
+        "effect": 0.3150199027993953,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      },
+      "measured-graph minus weight-permuted": {
+        "available": true,
+        "ci_high": 0.14206261631295158,
+        "ci_low": 0.11549838895619689,
+        "effect": 0.12745077861155032,
+        "holm_reject": true,
+        "p_value": 0.001953125
+      }
+    },
+    "draws": 5,
+    "n_seeds": 10,
+    "outcome": "topology-advantage",
+    "primary_metric": "higher_better",
+    "rungs": {
+      "configuration-model": {
+        "ci_high": 0.22065959224452839,
+        "ci_low": 0.19656216850894373,
+        "mean": 0.208900991495637,
+        "per_seed": [
+          0.21096412089236533,
+          0.2137758123900277,
+          0.24025855688351871,
+          0.21571005847766642,
+          0.1907091965368425,
+          0.17020735758046673,
+          0.19829538656707582,
+          0.21547476785292224,
+          0.23452213187156085,
+          0.19909252590392387
+        ],
+        "rung": "configuration-model"
+      },
+      "degree-preserving": {
+        "ci_high": 0.21980862842076504,
+        "ci_low": 0.19201570781020647,
+        "mean": 0.20657436215796215,
+        "per_seed": [
+          0.22459142507721414,
+          0.21792102171702282,
+          0.2121283974462552,
+          0.16097698304141833,
+          0.1917976700783426,
+          0.19616555081914827,
+          0.1819909611605042,
+          0.21397817792063284,
+          0.2387302031467767,
+          0.22746323117230652
+        ],
+        "rung": "degree-preserving"
+      },
+      "erdos-renyi": {
+        "ci_high": 0.20894375481720068,
+        "ci_low": 0.184742271621585,
+        "mean": 0.19708804421324186,
+        "per_seed": [
+          0.20530354759373295,
+          0.18806145941613553,
+          0.2265718730328004,
+          0.19131905338931807,
+          0.19990768757567984,
+          0.19323109109379336,
+          0.17578031249866657,
+          0.2072719916072446,
+          0.22458437757647412,
+          0.1588490483485731
+        ],
+        "rung": "erdos-renyi"
+      },
+      "function-matched-positive-control": {
+        "ci_high": 0.5179820889687388,
+        "ci_low": 0.49097581043289446,
+        "mean": 0.5041537244890605,
+        "per_seed": [
+          0.5009363067257319,
+          0.5101880382152005,
+          0.4958363141332346,
+          0.49029154210683257,
+          0.5440061055552787,
+          0.4967840513293734,
+          0.46404791462874745,
+          0.4989087940589315,
+          0.5385947234054722,
+          0.5019434547318019
+        ],
+        "rung": "function-matched-positive-control"
+      },
+      "measured-graph": {
+        "ci_high": 0.5179820889687388,
+        "ci_low": 0.49097581043289446,
+        "mean": 0.5041537244890605,
+        "per_seed": [
+          0.5009363067257319,
+          0.5101880382152005,
+          0.4958363141332346,
+          0.49029154210683257,
+          0.5440061055552787,
+          0.4967840513293734,
+          0.46404791462874745,
+          0.4989087940589315,
+          0.5385947234054722,
+          0.5019434547318019
+        ],
+        "rung": "measured-graph"
+      },
+      "shared-init-random": {
+        "ci_high": 0.20894375481720068,
+        "ci_low": 0.184742271621585,
+        "mean": 0.19708804421324186,
+        "per_seed": [
+          0.20530354759373295,
+          0.18806145941613553,
+          0.2265718730328004,
+          0.19131905338931807,
+          0.19990768757567984,
+          0.19323109109379336,
+          0.17578031249866657,
+          0.2072719916072446,
+          0.22458437757647412,
+          0.1588490483485731
+        ],
+        "rung": "shared-init-random"
+      },
+      "topology-randomized": {
+        "ci_high": 0.19968330518592112,
+        "ci_low": 0.17791095074864044,
+        "mean": 0.1891338216896652,
+        "per_seed": [
+          0.19335802268669552,
+          0.1916669088421989,
+          0.18079101283587973,
+          0.2056341564668373,
+          0.20416135567095992,
+          0.2166699844071905,
+          0.16374223036710278,
+          0.19562332183641412,
+          0.18326052274437046,
+          0.15643070103900278
+        ],
+        "rung": "topology-randomized"
+      },
+      "weight-permuted": {
+        "ci_high": 0.39386931624211163,
+        "ci_low": 0.36374028249409185,
+        "mean": 0.3767029458775102,
+        "per_seed": [
+          0.37773568987706024,
+          0.37669940272095176,
+          0.36489261431246456,
+          0.38822531441176533,
+          0.4426741813330598,
+          0.3579754909792966,
+          0.3492049312409202,
+          0.3723776189008648,
+          0.3570263319145902,
+          0.3802178830841281
+        ],
+        "rung": "weight-permuted"
+      }
+    },
+    "task_name": "hemibrain-heading-velocity-integrator-v1"
+  },
+  "limitations": [
+    "The recurrent substrate is the measured EPG compass ring (EPG cells only); Delta7, PEN_a, PEN_b, PEG and the FB/IB/ExR columnar helpers are excluded from the functional circuit, so the reported score is a ring-specific, not whole-circuit, statement.",
+    "The EPG ring coordinate is parsed from instance-name suffixes L1..L8/R1..R8; all recovered EPG instances carry the PB08 glomerulus label, so the declared 16-column order is an annotation-derived prior, not an electrophysiological measurement.",
+    "The readout is a fixed population-vector decoder over the declared bump basis; it is identical across every rung, so the contrast isolates the recurrent operator rather than a fitted readout.",
+    "The dynamics are a bounded (tanh) linearisation of the measured signed operator with a per-graph spectral-radius rescaling; they are a declared model of angular-velocity integration, not a biophysical simulation.",
+    "Degree-preserving rewiring uses the full 5 x |E| directed double-edge-swap budget and verifies the in/out degree sequences and self-loop count exactly.",
+    "The functional target was chosen and piloted on the same measured subgraph before the frozen run; it is a preregistered evaluation of a hypothesis about this circuit, not an out-of-sample discovery.",
+    "The 10 paired seeds give descriptive intervals; the exact sign-flip is a secondary check and the normal-theory MDE/power is optimistic at small n."
+  ],
+  "metric_name": "circular-heading-recall",
+  "non_claims": [
+    "This functional probe claims no biological advantage, general memory, AGI or programming expertise.",
+    "The EPG ring coordinate is parsed from neuPrint instance names (L1..L8 / R1..R8); it is the declared heading basis, not a measured physiological map.",
+    "An advantage over a matched null is necessary but never sufficient for a biological claim; negatives are retained and never reinterpreted."
+  ],
+  "outcome": "topology-advantage",
+  "power": {
+    "function-matched-positive-control versus its mismatched counterpart": {
+      "ci_high": 0.8870005982692515,
+      "ci_low": 0.8479411395106446,
+      "effect": 0.8660632618548636,
+      "includes_zero": false,
+      "mde": 0.02963788396959951,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.03345363823516143
+    },
+    "measured-graph minus configuration-model": {
+      "ci_high": 0.31306091725008356,
+      "ci_low": 0.2791124508679759,
+      "effect": 0.2952527329934235,
+      "includes_zero": false,
+      "mde": 0.02561985670034397,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.028918306670243882
+    },
+    "measured-graph minus degree-preserving": {
+      "ci_high": 0.31359415514520395,
+      "ci_low": 0.28459780221632175,
+      "effect": 0.29757936233109833,
+      "includes_zero": false,
+      "mde": 0.022122796341439557,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.024971014338128962
+    },
+    "measured-graph minus erdos-renyi": {
+      "ci_high": 0.32149242660691146,
+      "ci_low": 0.2933282262021414,
+      "effect": 0.3070656802758186,
+      "includes_zero": false,
+      "mde": 0.021240898545118413,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.02397557586928363
+    },
+    "measured-graph minus shared-init-random": {
+      "ci_high": 0.32149242660691146,
+      "ci_low": 0.2933282262021414,
+      "effect": 0.3070656802758186,
+      "includes_zero": false,
+      "mde": 0.021240898545118413,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.02397557586928363
+    },
+    "measured-graph minus topology-randomized": {
+      "ci_high": 0.330008772517274,
+      "ci_low": 0.30063391311300486,
+      "effect": 0.3150199027993953,
+      "includes_zero": false,
+      "mde": 0.022361009660561048,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.025239896631104448
+    },
+    "measured-graph minus weight-permuted": {
+      "ci_high": 0.14206261631295158,
+      "ci_low": 0.11549838895619689,
+      "effect": 0.12745077861155032,
+      "includes_zero": false,
+      "mde": 0.020123768105315798,
+      "mde_target_power": 0.8,
+      "n": 10,
+      "power_at_observed": 1.0,
+      "sd": 0.02271461953269168
+    }
+  },
+  "primary_metric": "higher_better",
+  "protocol_id": "ffb-hemibrain-heading-fn-v1",
+  "protocol_sha256": "9ad3b76b98027850b7d3652eaed2d2e72e775ed7f6635405dde6d3e418117aee",
+  "provenance": {
+    "bootstrap_resamples": 100000,
+    "bootstrap_seed": 707,
+    "degree_swap_multiplier": 5,
+    "draws": 5,
+    "protocol_id": "ffb-hemibrain-heading-fn-v1",
+    "seeds": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9
+    ],
+    "subgraph_path": "/home/user/dev/hf/hf-stage/publish/hemibrain-heading-2026-09-14/subgraph.npz",
+    "subgraph_sha256": "2a3047748a3f2cc92ce33e3f107653e086651e1335b92a3a48ff1d43113ec482",
+    "task_name": "hemibrain-heading-velocity-integrator-v1"
+  },
+  "ring": {
+    "cell_type": "EPG",
+    "column_labels": "L1..L8 = 0..7, R1..R8 = 8..15",
+    "gain": 1.0,
+    "kappa": 3.0,
+    "n_ring_nodes": 46,
+    "n_test": 96,
+    "n_train": 192,
+    "nodes_per_column": {
+      "0": 2,
+      "1": 2,
+      "10": 3,
+      "11": 3,
+      "12": 3,
+      "13": 3,
+      "14": 3,
+      "15": 4,
+      "2": 3,
+      "3": 3,
+      "4": 3,
+      "5": 3,
+      "6": 3,
+      "7": 4,
+      "8": 2,
+      "9": 2
+    },
+    "omega_max": 2,
+    "ring_columns": 16,
+    "steps": 12
+  },
+  "runtime_seconds": 97.72,
+  "schema": "ffb-hemibrain-heading-functional-v1",
+  "task_name": "hemibrain-heading-velocity-integrator-v1"
+}