SD-033d: Premotor/SMA-analog (sequence-execution substrate)
Claim ID: SD-033d (subdivision of the SD-033 PFC cluster) Subject: pfc.premotor_sma_analog_sequence_execution Status: candidate, implementation_phase: v3, v3_pending: false Registered: 2026-04-19 Design doc written: 2026-06-09 (commitment_closure:GAP-9 — the registration carried design_doc: null until this pass) Instantiates: SD-033 (PFC subdivision cluster) Depends on: SD-033, SD-033a, MECH-094, MECH-261, MECH-090, ARC-028
What this document is
SD-033d is a registration-only graph-completeness subdivision. The premotor / SMA sequence-execution function it names is already implemented inside ree-v3’s E3 trajectory machinery and the MECH-090 commitment loop. This document does the GAP-9 task: it maps the existing machinery onto the premotor/SMA biology so the SD-033 claim graph mirrors the primate frontal-lobe parcellation rather than implying a premotor-less architecture. No new substrate is proposed and none is required. If you are looking for a module called premotor_sma_analog.py, there is deliberately none — the point of this doc is that the function is distributed across modules that already exist.
The companion parent doc is sd_033_pfc_subdivision_architecture.md; this doc expands the one-paragraph SD-033d subdivision entry there into the full machinery mapping.
The biology being mapped
Tanji & Hoshi 2008 (Physiol Rev, “Role of the lateral prefrontal cortex in executive behavioral control”) describe a graded executive-control axis running caudally from lateral PFC:
| Region | Function (Tanji & Hoshi 2008) |
|---|---|
| Lateral PFC | Holds the abstract rule / goal that the sequence serves |
| pre-SMA | Sequence planning, set-switching, ordering of forthcoming actions |
| SMA proper | Sequence execution — unrolling a committed motor program step by step |
| Dorsal premotor (PMd) | Stimulus→action binding (selecting the action a state affords) |
The rostro-caudal gradient is: abstract rule (lateral PFC) → ordered plan (pre-SMA) → executed program (SMA) → stimulus-bound action (PMd). REE’s SD-033a → SD-033d relation captures the rule → sequence-execution segment of this gradient: SD-033a (lateral-PFC-analog) holds the rule/goal; SD-033d is everything downstream that turns that context into an executed action sequence.
The mapping: existing ree-v3 machinery → premotor/SMA biology
Every row below is already in the codebase. SD-033d is the claim-graph name for their collective premotor/SMA role.
| Biological role | ree-v3 component | Location |
|---|---|---|
| PMd — stimulus→action binding (which action does this state afford?) | HippocampalModule.propose_trajectories() generates candidate action sequences from the current state; E2FastPredictor.action_object() is the affordance map the proposer navigates | ree_core/hippocampal/module.py:857, ree_core/predictors/e2_fast.py |
| pre-SMA — sequence planning / ordering / set-switching | E3TrajectorySelector.select() scores the multi-step candidate trajectories and picks the one to commit; the diversity / authority machinery (ARC-065 SP-CEM, MECH-341) keeps the candidate plan-set from collapsing to a single ordering | ree_core/predictors/e3_selector.py:73 |
| SMA proper — sequence execution (unroll a committed program step by step, not re-decide each tick) | MECH-090 Layer-1 committed-trajectory stepping: REEAgent.select_action advances _committed_step_idx through committed_trajectory.actions[:, idx, :] while the bistable BetaGate holds | ree_core/agent.py:1746 / :3796 (the _committed_step_idx counter) |
| Commitment latch — “execute this program, do not re-evaluate every cycle” | MECH-090 bistable BetaGate (beta_gate_bistable): elevates on commit-entry, holds through execution | ree_core/heartbeat/beta_gate.py |
| Sequence-end / program-complete signal → release | ARC-028 + MECH-105: HippocampalModule.compute_completion_signal() → BetaGate.receive_hippocampal_completion() releases the latch at sequence end (Lisman & Grace subiculum→NAc→VTA loop) | ree_core/hippocampal/module.py:1655, ree_core/heartbeat/beta_gate.py:164 |
| Urgency abort of the running program | MECH-091 urgency interrupt: z_harm_a.norm() > threshold releases the latch mid-sequence and resets _committed_step_idx | ree_core/agent.py select_action MECH-091 block |
The two halves Tanji & Hoshi separate — planning the order (pre-SMA) and executing the committed order without re-deciding (SMA) — map cleanly onto REE’s propose+score stage versus the MECH-090 committed-step stepping stage. The bistable BetaGate is exactly the “do not re-evaluate the program each striatal cycle” property SMA-proper execution requires.
MECH-261 write-gating profile (the SD-032 interface)
SD-033d inherits MECH-094 hypothesis-tag gating as the specific case of the MECH-261 mode-conditioned registry (MECH-094 is the sequence-commit instance of the general gate). Per the SD-033 cluster gating table:
- external_task / internal_planning — sequence-level writes licensed (plan and commit real action sequences).
- internal_replay — suppressed unless the hypothesis tag is explicitly set (imagined sequences must not silently overwrite the committed program; this is why MECH-094’s
hypothesis_tagrides on simulated trajectories). - offline_consolidation — consolidative (slow updating of learned state→sequence mappings).
SD-033d is therefore the subdivision where MECH-094 is most visibly load-bearing: the hypothesis tag is the gate that keeps replayed/imagined action sequences from being executed as if real.
Why register it rather than build it
REE registers the PFC as a five-subdivision cluster up front (the SD-010/SD-011 late-split lesson: implementing a monolith and re-splitting later is expensive). The premotor/SMA function is genuinely present in E3 + MECH-090, so SD-033d’s correct status is named-but-not-rebuilt. Building a separate premotor_sma_analog module would duplicate the trajectory machinery and create two competing sequence-execution paths — the opposite of what the cluster registration is for.
v3_pending: false is correct: there is no deferred V3 implementation work. The function exists; only the claim-graph name and this mapping doc were missing.
Falsification / split trigger
SD-033d is registration-only, so it is not falsifiable at the substrate level in the way SD-033a/b are. The load-bearing falsifications for its claimed function live on the components it maps to (E3 trajectory experiments; MECH-090 commitment experiments). What SD-033d does carry is a split trigger:
If future experimental evidence shows that E3’s trajectory-proposal/selection + MECH-090 stepping machinery inadequately covers the pre-SMA / SMA / PMd distinctions — e.g. a sequence-planning vs sequence-execution dissociation that the single E3-commit pathway cannot reproduce — then SD-033d should be split into
SD-033d-i(pre-SMA planning),SD-033d-ii(SMA execution),SD-033d-iii(PMd stimulus-action binding) along the Tanji & Hoshi subdivisions, and the corresponding machinery factored apart.
Until such evidence appears, the single-subdivision registration is the parsimonious reading.
Related claims
- SD-033, SD-033a–e — the PFC subdivision cluster (parent + siblings)
- SD-033a — lateral-PFC-analog; the upstream rule/goal source SD-033d executes
- MECH-090 — bistable commitment latch + committed-trajectory stepping (SMA execution)
- ARC-028 / MECH-105 — hippocampal-completion → BetaGate release (sequence-end signal)
- MECH-091 — urgency interrupt (mid-sequence abort)
- MECH-261 — mode-conditioned write gating (SD-032a interface)
- MECH-094 — hypothesis tag (the specific gate on sequence commits)
- ARC-018 — waking trajectory-proposal loop (the proposer SD-033d executes)
References
- Tanji, J. & Hoshi, E. 2008. Role of the lateral prefrontal cortex in executive behavioral control. Physiol Rev 88(1):37–57. — the lateral-PFC → pre-SMA → SMA → dorsal-premotor executive-control gradient this doc maps onto.
- Lisman, J.E. & Grace, A.A. 2005. The hippocampal-VTA loop. Neuron 46(5):703–713. — the completion → release loop (ARC-028 / MECH-105).
- Primary cluster lit-pull:
evidence/literature/targeted_review_pfc_subdivision_architecture/synthesis.md.