Baker auditory neuron correspondence

Anatomical identity audit, 2026-10-17. No physiological response files were opened, no benchmark was scored, and no manifest or model default changed. The M4 v2 proposal remains a proposal.

Result

The cited route is now complete for 12 of 28 Baker classes, up from 2: 11 with unresolved sex compatibility, plus AVLP_pr36, which remains excluded because its MaleCNS matches include dimorphic annotations. Fourteen classes remain partial and two are absent from Table S3. No class is yet eligible for physiological benchmarking.

Here, complete means that every Table S3 reconstruction of the class has an anchored FlyWire v783 root, a published cell-type label and at least one named MaleCNS type linked by an exact flywireType token. It does not mean that an original Baker class is equivalent to one modern type. All one-to-many mappings and individual body IDs remain in the evidence. In particular, B2 splits into CB0090/CB0758, and AVLP_pr23 into AVLP598/WED193; do not pool their responses as though they were single modern types.

The nine newly complete classes other than the dimorphic AVLP_pr36 are A2, B2, AVLP_pr04, AVLP_pr05, AVLP_pr12, AVLP_pr23, IPS/WED_pr01, SAD_pr01 and WED_pr01. AVLP_pr35 and IPS_pr02 retain their original complete routes.

Across 404 source reconstructions, 332 have named MaleCNS routes. Of the remaining 72, two have no FlyWire cell_type label, 63 have a label absent from MaleCNS flywireType tokens, and seven link only to MaleCNS bodies without a type name. These categories are mutually exclusive. AVLP_pr02, AVLP_pr31 and AVLP_pr32 have no named MaleCNS route in this audit. AVLP_pr18 and AVLP_pr24 are absent from S3; CAVE cannot supply identities that the source never lists.

Method and independent checks

Sources:

• Baker et al. (2022), paper, Table S3, publisher workbook. Only the identity/coordinate table is used. Its two root-ID columns refer to synapse tables v11 and v274; those numbers are not the target annotation release.
• FlyWire annotation tag v3.2.0, supplemental_files/Supplemental_file1_neuron_annotations.tsv, associated with materialization 783. Cite Schlegel et al. (2024) and Berg et al. (2026).
• MaleCNS v1.0 data/neurons.parquet, with type, flywireType, dimorphism and fruDsx independently checked against the original downloaded body annotation Feather: zero field differences. Input hashes are in the evidence.

The target is explicitly flywire_fafb_public, version 783, timestamp 2023-09-30T05:10:01.441140 UTC. No query selects the latest release.

1. Read Table S3 identity cells directly from the workbook XML. Root IDs remain exact integer strings; they never pass through floating-point spreadsheet conversion. Select the 28 classes already in the proposal. B1-1 through B1-5 and B1-u belong to the original B1 class; WED-VLP-1/2 belong to WED-VLP.
2. Sample each published FlyWire coordinate at its [4, 4, 40] nm voxel resolution from the static watershed segmentation, with agglomeration disabled. The downloaded volume advertises [16, 16, 40] nm resolution, so CloudVolume performs the coordinate conversion explicitly. Keep the resulting supervoxel.
3. Map each supervoxel to its root at the exact version-783 timestamp. Separately, query CAVE lineage for all 556 distinct historical root IDs in the two source columns. Remove nodes after the target timestamp and require the anchor's root to be reachable from both source IDs. All 404 anchors agree. No split is resolved by selecting the first or largest descendant. Lineage without a coordinate anchor can include unrelated branches after old merges.
4. Join each anchored root to its v3.2.0 cell_type, then to exact, case-sensitive MaleCNS flywireType tokens, split only on comma or semicolon. Retain all male body IDs, types and dimorphism annotations. No synonym or fuzzy-name fallback.
5. Independently ask CAVE whether every anchored root is valid at the target timestamp: 404/404. Map the 404 annotation-table supervoxels, obtained independently of S3 coordinates, to roots at that timestamp: zero mismatches. The two previously complete classes reproduce their original mappings.

The transformation changes 243 v11 IDs and 156 v274 IDs. For 156/404 source reconstructions, the v783 root equals neither original root ID. These counts verify that the audit actually updated identities, rather than merely joining unchanged source IDs.

The first raw binary root request returned an empty body because its content type was unsuitable. A response-length assertion rejected it. The successful request specifies application/octet-stream; the retained successful response contains exactly 404 uint64 roots (3,232 bytes).

Evidence and reproduction

• Summary and input SHA-256 hashes.
• Captured API responses, row-level identity results and offline replay.

The archive contains identity metadata and analysis code, no response traces, credentials, source workbooks, full annotation tables or segmentation volumes. Download the original S3 and v3.2.0 annotations separately, and use the hashed MaleCNS parquet. replay.py verifies input hashes, reconstructs the source rows and both historical lineages, recomputes the joins, checks the independent CAVE oracles and reconciles the counts. See the archive README for the exact command.

Sex annotation audit (2026-10-18)

For the 11 complete classes other than AVLPpr36, every one of the 126 matched MaleCNS bodies has an empty dimorphism, and every one of the 82 anchored materialization-783 roots is isomorphic in flywire_annotations v3.2.0. That column was added in v3.0.0 from the MaleCNS comparison (Berg et al., bioRxiv 2025, now Cell 2026), whose isomorphic category means no consistent _morphological sex difference. Neither paper's retrieved text names these types. audit.py reproduces the counts from hashed inputs into results.json.

The female label does not clear the sex gate. It is the residual value of the column, not a per-type comparison result: v3.2.0 has no empty dimorphism, and 1,528 isomorphic neurons have no cell_type at all, so they cannot have been compared at type level (5,012 more typed ones match no MaleCNS flywireType token). MaleCNS has no isomorphic value; its empty field means unflagged. Both sides therefore say only "not flagged dimorphic", and morphology does not establish physiological equivalence of female recordings and male neurons. The 11 classes keep unresolved sex compatibility. An explicit per-type statement from the MaleCNS authors, or male recordings of the same types, would resolve it.

Remaining limits

Empty sex annotations do not establish isomorphism. The completed routes other than AVLP_pr36 have unannotated MaleCNS sex compatibility, so the female-recording versus male-connectome gate stays open. AVLP_pr36 remains excluded.

Complete identity coverage also does not establish calcium-to-model readout compatibility, the measured protocol-6 JO input, independence from model tuning, or a prospective held-out partition. Those gates must be met and the manifest locked before opening responses. Partial classes need their missing published annotations or cross-connectome identities resolved; no arbitrary renaming is introduced here.