V_s Foundation Lit-Pull: Schema-Region Granularity, Multi-Map Coexistence, Per-Stream Verisimilitude

Created: 2026-04-22 Origin: Foundation work for MECH-269 base anchor selection + MECH-284 staleness + MECH-287 trigger. Three architectural choices in the implementation plan need biology validation before substrate edits. Earlier pulls covered: replay content/start-state (targeted_review_connectome_mech_269/), ephaptic substrate (mech_270/), routing (mech_271/), and waking V_s invalidation signal-types (waking_v_s_invalidation/). Gap: nothing yet validates (a) the unit over which V_s/staleness is indexed, (b) whether dual-trace maps onto coexisting ensembles for the same place, (c) per-stream vs per-ensemble verisimilitude computation.

Prompt

/lit-pull Hippocampal foundation for MECH-269/284/287 substrate. Three converging architectural questions: (1) schema-region granularity – what is the biologically natural unit over which a hippocampal verisimilitude signal indexes (place-cell field, event/chunk boundary, action-object, learned schema)? (2) multi-map coexistence at the ensemble level – does the literature support multiple distinct place-cell representations for the same physical location coexisting (rate remapping, global remapping, splitter cells, multi-trace), as the substrate for MECH-269’s dual-trace anchor preservation? (3) per-stream vs per-ensemble V_s – does hippocampal verisimilitude/match-mismatch computation operate per modality/stream (allothetic vs idiothetic, exteroceptive vs interoceptive) or per integrated ensemble, or both? The architectural commitment to test: regions are environmentally-grounded chunks (grid cell or action-object), V_s is per-stream (z_world, z_self, z_harm_s, z_harm_a, z_goal), staleness is per-region indexed by the active anchor’s location, dual-trace is preserved by keeping inactive anchors in the anchor set with MECH-272 routing deciding which is operative.

Specific neurobiological systems to cover

  1. Place-cell remapping types and multi-map coexistence
    • Colgin, Leutgeb & Moser 2010 (review of rate vs global remapping)
    • Leutgeb et al 2007 (rate remapping = same ensemble, different rates; global = different ensemble entirely; CA1 vs CA3 differ)
    • Wills et al 2005 (gradual transformation between maps under cue morphing)
    • Whether multiple maps for the same place can coexist as the substrate for a dual-trace inactive-but-retrievable anchor
  2. Splitter cells and context-dependent firing
    • Wood et al 2000 (CA1 splitter cells: same place, different firing depending on past or future trajectory context)
    • Frank, Brown & Wilson 2000 (prospective coding)
    • Whether splitter cells are evidence that anchor-state is composed of (place, stream-mixture) rather than place alone
  3. Schema cells and event boundaries
    • Tse, Langston, Kakeyama, Bethus, Spooner, Wood, Witter & Morris 2007 (hippocampal schemas; rapid one-trial encoding into existing schema)
    • DuBrow & Davachi 2014 (event boundaries segment continuous experience)
    • Brunec & Momennejad 2022 (hippocampal events as schema chunks)
    • Whether the natural granularity of hippocampal “chunks” matches grid-cell-level or action-object-level discretization, or operates at a higher event-segment scale
  4. Multi-trace theory at the population level
    • Nadel & Moscovitch 1997 (multiple-trace theory; each retrieval creates a new trace, old traces persist)
    • Yonelinas, Ranganath, Ekstrom & Wiltgen 2019 (CMS / multi-trace synthesis)
    • Whether dual-trace at the anchor level (active + inactive coexist, routed by state) matches the population-level evidence on multi-trace persistence
  5. Per-modality vs integrated hippocampal coding
    • Eichenbaum 2017 (memory space: time cells, place cells, event cells)
    • O’Mara, Sanchez-Vives, Brotons-Mas & O’Hare 2009 (subiculum integration of allothetic and idiothetic streams)
    • Whether match/mismatch detection in hippocampus is per-stream (allothetic vs idiothetic, exteroceptive vs interoceptive) or per integrated representation
  6. Comparator function in hippocampus
    • Vinogradova 2001 (hippocampus as comparator; CA1 as match/mismatch detector between CA3 retrieval and EC input)
    • Lisman & Grace 2005 (novelty detection via VTA loop; CA1 mismatch upstream)
    • Whether the comparator is per-stream or per-population, and whether it directly drives the broadcast trigger (substrate candidate for MECH-287 anchor-side)

Architectural questions the lit-pull should help answer

  1. Region granularity. Are hippocampal “schema regions” naturally place-cell-field sized (grid cell), action-object sized (sequence chunk), or event-segment sized (multiple actions bound by a context boundary)? The architectural commitment is tractable at any of these scales but the biology should constrain the default.

  2. Multi-map coexistence as dual-trace substrate. Does the literature on rate vs global remapping support multiple distinct ensembles for the same place coexisting in a way that could route by state? If yes, MECH-269’s dual-trace preservation has direct biological backing. If only rate remapping (same ensemble, different rates), the implementation should reflect that — anchors aren’t fully separable, just differentially weighted.

  3. Per-stream vs integrated V_s. The substrate plan computes V_s per stream (z_world, z_self, etc.). Is this consistent with hippocampal comparator biology, which appears to operate at the CA1 ensemble level on integrated CA3 retrieval vs EC input? If V_s is fundamentally integrated, per-stream V_s is an in-silico computational convenience that needs to be flagged as architecture-deviates-from- biology rather than architecture-grounded-in-biology.

  4. Splitter-cell evidence for (place, context) coding. If splitter cells are robust, anchors should be (place, stream-mixture) rather than just place — and the stream-mixture is the context that disambiguates which anchor is active under MECH-272 routing. If splitter cells are a small minority phenomenon, the architectural lift is harder to justify.

  5. Trigger substrate from comparator function. Vinogradova / Lisman & Grace suggest CA1 mismatch detection feeds VTA novelty signals. Is this a candidate anchor-side substrate for MECH-287 distinct from the broadcast LC/DA trigger (which is amygdala/cortex-side)? If yes, MECH-287 may need to register both an anchor-side comparator-fed component AND a broadcast LC/DA component, with crosstalk.

Output structure

Standard targeted_review_*/ format. Suggested directory: evidence/literature/targeted_review_v_s_foundation/

Per-paper records as usual (record.json + summary.md). After the pull, write a short SYNTHESIS.md flagging:

  • Recommended region granularity (place-cell-field / action-object / event-segment) for the in-silico foundation, with one-line justification per option considered.
  • Whether dual-trace at the anchor level has direct biological support or is an in-silico extension requiring an architectural-deviation note.
  • Whether per-stream V_s is biologically supported or an in-silico convenience.
  • Whether the splitter-cell literature supports (place, stream-mixture) anchor encoding.
  • Whether MECH-287 needs an additional CA1-comparator anchor-side component alongside the broadcast LC/DA trigger.

Estimated scope: ~10 papers, single session.

Notes for the agent doing the pull

  • The user is a consultant psychiatrist; clinical mappings welcome (e.g., remapping failure as substrate for context-binding deficits in PTSD; comparator hypoactivity as substrate for novelty-blindness in schizophrenia).
  • Pfeiffer & Foster 2013, Dragoi & Tonegawa 2011/2013, Tang 2017, Foster 2017, Olafsdottir 2018, Anastassiou 2011, Buzsaki 2015, Girardeau 2017, Jadhav 2016 are already pulled in targeted_review_connectome_mech_{269,270,271}/ – do NOT re-pull; cite where relevant.
  • Schultz 1997, Matsumoto 2007, Bromberg-Martin 2011, Wilson 2014, Stalnaker 2015, Gardner 2018, Yassa-Stark 2011, Reagh 2018, Gershman 2010/2017, Bouton 2004, Sara-Bouret 2012, Aston-Jones-Cohen 2005 already in targeted_review_waking_v_s_invalidation/ – cite where relevant.
  • The exemplar that motivates this pull is the substrate foundation work for MECH-269/284/287 implementation; design doc at docs/architecture/v_s_invalidation_runtime.md.
  • Connect to existing claim cluster: MECH-269 (anchor selection + reset), MECH-272 (state-gated routing – load-bearing for dual-trace mixture), MECH-094 (hypothesis tag interaction with anchor-reset events), MECH-284/285/287 (V_s bidirectional cluster + trigger).
  • Be alert to evidence that the architecture should be revised. If multi-map coexistence is NOT supported (e.g., if remapping is exclusively winner-take-all), the dual-trace mechanism may need to be implemented as soft re-weighting rather than hard active/inactive flags. If V_s is exclusively integrated (no per-stream computation in biology), the in-silico per-stream V_s is a deviation that needs a rationale in the architecture doc.

REE is developed by Daniel Golden (Latent Fields). Apache 2.0.