Why Map the Brain

Instructor script: "This unit is deliberately not about microscopes. It is about the reasoning step that happens before anyone buys a microscope — deciding what claim the data will be asked to support. Most failed connectomics projects skip it."

Read the three outcomes aloud, then name the one that matters most: sorting a claim by what structure can support. The unit page says it plainly — it is the outcome most often failed. Everything in this session is practice for the non-claim in the third bullet.

Instructor script: "This is a motivating question, asked before any instrument is chosen. Before we go further, turn it into something fundable: what is the measurable structural endpoint, what is the null model, and what is the non-claim?"
Collect two or three answers. Each one should imply a different measurement — point that out. This is a framing device, not evidence.

Instructor script: "What counts as brain data? Hold this against the resolution table from the unit: light microscopy tops out around 200–250 nm. An unmyelinated axon in neuropil is 80–300 nm, a spine neck 50–200 nm, a synaptic cleft about 20 nm."
The point to land: at 250 nm, two membranes 20 nm apart are one blur. Light microscopy can show that two arbors overlap in space; it cannot show that they are connected. That is the whole reason the field tolerates the cost of EM.

Instructor script: "A wiring diagram is a constraint on the space of possible dynamics, not a simulation. Any 'this circuit computes X' reading of this analogy is a Bin C claim wearing Bin A clothing."
If the room has not met the bins yet, preview them: A = structure alone suffices; B = structure plus one declared assumption; C = structure cannot establish this. The next two slides fill them in.

These are Bin A claims — each is a census or a comparison. Give the unit's concrete examples: "Neuron X makes 42 synapses onto neuron Y"; "type A targets proximal dendrites of type B"; "reciprocal pairs occur 4x more often than a degree-preserving null predicts"; "this projection is absent, at a stated detection sensitivity."
Then the Bin B middle ground: "this synapse is excitatory" needs the assumption that asymmetric morphology predicts glutamatergic transmission. Rule: name the assumption in the same sentence. "Putatively excitatory (asymmetric morphology)" costs four words.

Bin C. "This circuit computes X during behavior Y" needs physiology. "This synapse is potentiated" needs a functional or molecular measure. "This wiring causes the decision" needs perturbation. Neuromodulatory state is largely invisible in a standard EM volume.
Instructor script: "The most common failure of connectomics writing is a Bin C sentence in the abstract supported by a Bin A measurement in the results."

Walk the unit's worked repair. Version 0: "We will use connectomics to understand how the cortex implements predictive coding" — Bin C, no endpoint, unfalsifiable.
Move 1, structural signature: feedback axons from higher visual areas target a different dendritic compartment of L2/3 pyramidal cells than feedforward inputs.
Move 2, measurement with units: input synapses per source class per compartment per cell, binned by path distance.
Move 3, null: sources distributed in proportion to available postsynaptic surface area — distal synapses are the expected result, not evidence.
Move 4, non-claim: the data constrain where feedback arrives; they do not show it carries prediction error, and they do not establish sign.

Flag the trap this example carries, from unit §4: merge errors do not add noise symmetrically to motif counts. A single merge fuses two neurons' partner lists and manufactures triangles and reciprocal pairs far above the error rate itself. Motif analysis on unproofread segmentation is biased in a predictable direction — toward the interesting answer. Unit 09 returns to this.

Point out that the endpoint is a ratio. Ratios between comparably reconstructed populations quietly control for a great deal of reconstruction bias; absolute counts are sensitive to completeness. Experienced people reach for the ratio automatically — say so, because nobody else will.

Match the claim to the reconstruction state (unit §4). "Cell X synapses onto Y" needs both partners proofread through the synapse — otherwise a merge invents the connection. "Cell X has n inputs" needs the full dendritic arbor closed — split errors make n an undercount of unknown size. "Type A prefers B over C" needs both populations proofread to comparable completeness. "This projection is absent" needs a stated detection sensitivity.

Recovery moves from the unit page: if the analysis is "figured out once we have the data", write the figure caption for the main result now, with fake numbers — if you cannot write it, the experiment is not specified. Whole-graph summaries (mean path length, global clustering) are nearly useless for a local microcircuit claim. Neurons cut by the volume edge have truncated arbors. Populations proofread differently are confounded with effort.

Seed overclaims from the unit's check-yourself: "This interneuron inhibits pyramidal cell firing" (Bin B verging on C — a GABAergic synapse can be depolarizing; write "putatively inhibitory"). "Kenyon cell input is random with respect to glomerular identity" (Bin A only if "random" is defined against a stated null and restricted to the reconstructed population).

Instructor script: "Produce it in this order. Do not skip to the method." Circulate and check step 5 first. The single best predictor of a weak brief is that the non-claim is empty or vague: if you cannot name a sentence you refuse to write, you have not found the boundary of your evidence.

The full lab rubric on the unit page scores falsifiability, measurement, null model, scope discipline and feasibility. For "strong" on the null: it preserves the nuisance structure that matters — degree, distance, cell-type composition — not just "compare to random".

These are the four anchors from unit §5, so "connectomics is useful" is a claim learners can defend with examples rather than enthusiasm. Note what the useful results have in common: each is a census or a comparison, not an assertion about computation. C. elegans's lasting lesson is negative and useful — the complete 302-neuron wiring diagram did not immediately yield an understanding of behavior.