Task History
The story of the tracer team’s work — every project since the Omni days, what each one was, and how long it ran. Follow the “Related documentation” links to the protocols that governed the work, and read the verbal history below for how it all fits together.
The sheet-level index that used to live on this page has been retired; the underlying records are kept offline and are available on request.
Jump to era: Pinky, Omni & Semantic Ground Truth · CREMI & Fly Ground Truth · Minnie65 / MICrONS & V1DD · FlyWire / FAFB & Optic Lobe · BANC · Hippocampal CA3 · MEC & Hippocampus · Cross-Project · Verbal History
Pinky, Omni & Semantic Ground Truth 2016–2019
Where it started. The earliest work was done in Omni, hunting errors chunk by chunk through mouse cortex segmentations, with VAST alongside for hand annotation. Out of that grew the first real ground-truth effort: semantic segmentation volumes labeled by hand, then Pinky10 and Pinky100 proofreading as the segmentations got good enough to correct rather than redo, and the first synapse-detection ground truth. This era set the template every later project reused — humans make small volumes of truth, the AI trains on them, and the team’s job shifts from painting to checking.
Related documentation: How to Omni (archived doc) · Omni archive · VAST archive
CREMI & Fly Ground Truth 2019–2020
A short, dense chapter and the team’s pivot from mouse to fly. On the CREMI B volume the team produced dense fly ground truth: membrane review, pre- and post-synapse assignment, twigs and backbones. Pixel-level, patient work — the kind of small perfect volume that teaches a network what a fly synapse looks like.
Related documentation: How To Fly Synapse (archived doc)
Minnie65 / MICrONS & V1DD 2019–2024
Mouse visual cortex at scale: proofreading core cells and building ground truth in Minnie65, the MICrONS dataset — and somewhere in this era’s records is the sheet where the mouse project got its name. Alongside it ran the Allen V1DD cutout synapse assignments. Where Pinky proved the pipeline on a small volume, Minnie ran it on a cubic millimeter.
Related documentation: MINNIE dataset dashboard · Proofreading core cells in Minnie (archived doc)
FlyWire / FAFB & Optic Lobe 2019–2024
The full adult fly brain. FlyWire proofreading across FAFB, synapse ground-truth cutouts, misalignment hunting, and then the long optic-lobe push: cell typing, per-type population studies, and the cell-model illustration work that still fills the gallery’s reference sections. (Plus a zebrafish synapse-tagging cameo.) This is the era that produced most of the site’s visual reference material.
Related documentation: FAFB dataset dashboard · Fly synapse figures
BANC — Brain And Nerve Cord 2023–2026
The Brain And Nerve Cord of the fly — the first dataset spanning both. This was production proofreading at full maturity: queues sorted by synapse count, unproofread-fragment sweeps, orphan backbones, soma checks, sexually dimorphic neurons, NBLAST-matched groups, the optic-lobe pushes, and the can’t-fix review rounds under the three-round triage protocol. If the earlier eras built the machine, BANC is what it looked like running at full speed.
Related documentation: BANC dataset dashboard · Proofreading SOP · Three-Round Triage protocol
Hippocampal CA3 2024–2026
A focused return to the mouse: CA3 proofreading — pyramidal cells, mossy fibers, and inhibitory cells — a small, precise project alongside the larger MEC effort.
Related documentation: CA3 dataset dashboard
MEC & Hippocampus 2024–2026, ongoing
The current chapter: medial entorhinal cortex and hippocampus work with Zetta on the Zheng volumes — semantic labeling in WebKnossos, synapse annotation, and segmentation logs. Full circle in a sense: like the Pinky days, the job is making ground truth in mouse tissue again, with a decade’s better tools.
Related documentation: MEC Synapse Annotation SOP · Voxel Painting SOP
Cross-Project Logs, Templates & Tooling 2016–2025
The connective tissue running under every era: task logs spanning every project since 2016 (one of them literally titled “Every Task Done At Work Since March 12, 2019”), reusable sheet templates, and infrastructure registries.
Related documentation: tracertools Catalog
A Verbal History of the Work
The work began in Omni. In the mid-2010s the team’s days were spent moving chunk by chunk through machine segmentations of mouse cortex, hunting errors by eye — and in VAST, coloring neurons in by hand when the machine couldn’t be trusted at all. It was slow, and that slowness was the point: the only way to make a computer see a neuron was to first show it thousands of examples of humans seeing one.
So the team became ground-truth makers. Semantic segmentation volumes, labeled voxel by voxel. The first synapse-detection truth sets. And as those labels fed back into training, the segmentations improved until the job quietly inverted — by the Pinky10 and Pinky100 era, tracers were no longer redoing the machine’s work but correcting it. Proofreading was born, and with it the basic rhythm every project since has followed: the AI does the volume, the humans do the judgment.
In 2019 the work split into two threads that would run side by side for half a decade. In the mouse thread, the pipeline proven on Pinky was pointed at a cubic millimeter of visual cortex — Minnie65, the MICrONS dataset — with core-cell proofreading and ground truth, plus the Allen V1DD synapse assignments. In the fly thread, the team first ground out dense truth on the little CREMI B volume — membranes, twigs, backbones, every pre and post assigned — and then stepped up to the whole thing: FlyWire, proofreading the full adult fly brain, hunting misalignments, and settling into the long optic-lobe push of cell typing and illustration that still fills the gallery today.
BANC was the culmination. From 2023 to 2026 the team worked the fly’s Brain And Nerve Cord — the first dataset to span both — and it was proofreading as a mature production system: queues ranked by synapse count, fragment sweeps, orphan backbones, soma checks, dimorphic neurons, triage rounds for the cells nobody could fix. By task volume it was bigger than everything that came before it combined, and 2025 was its peak year.
Now the work has come back around to the mouse hippocampus — CA3 proofreading on one side, and the MEC volumes with Zetta on the other, where the team paints semantic labels and validates AI-proposed synapses in WebKnossos. It is, in structure, the same job as the Pinky days: humans making truth in mouse tissue so machines can learn from it. What’s changed is everything around that job — a decade of tools, protocols, and hard-won judgment, which is what this site exists to hold onto.