Developers · reviewers
Scientific evidence
The research behind each mechanic, how strong it is, and what the project must never claim.
Content checked 2 November 2026
What the original opening relied on#
| Mechanic | Claim | Source | Strength |
|---|---|---|---|
| Tend / clearing | Astrocytes and microglia share the clearing of dying cells; microglia are not the only cleaners | Damisah et al. 2020, Sci Adv, 10.1126/sciadv.aba3239 | Mouse imaging; indexed sections |
| Arbors | Astrocytes help keep potassium and glutamate in balance | Djukic et al. 2007, J Neurosci, 10.1523/JNEUROSCI.0723-07.2007 | Knockout; full text |
| Microglia shaping activity | Microglia damp neuronal activity by local negative feedback | Badimon et al. 2020, Nature, 10.1038/s41586-020-2777-8 | Mouse; converging evidence; bounded analogy only |
| “Activation” wording | Activation is a context-dependent, contested label; avoid binary states | Woodburn et al. 2021, J Neuroinflammation; Paolicelli et al. 2022, Neuron consensus | Reviews; used to constrain language |
| Lingering response | Recovery differs by endpoint; some effects persist after others resolve | McKim et al. 2016, J Neurosci, 10.1523/JNEUROSCI.2394-15.2016 | Mouse; full text |
| Recovery with history | Stress history can persist even after microglia are replaced | Weber et al. 2019, Biol Psychiatry, 10.1016/j.biopsych.2018.10.009 | Mouse; full text |
| Resolution | Ending inflammation is an active process | Gilroy 2021, Nat Rev Immunol (commentary) | Preview only |
| Counterexample | More engulfment is not the same as more inflammation | Wohleb et al. 2018, Biol Psychiatry, 10.1016/j.biopsych.2017.05.026 | Mouse; indexed |
| Counterexample | Wakefulness changes microglial surveillance; reduced coverage ≠ harm | Stowell et al. 2019, Nat Neurosci, 10.1038/s41593-019-0514-0 | Mouse; full text |
Earlier phases studied repeated social defeat in mice in depth (McKim 2018, Wohleb 2014, Yin 2022, Goodman 2024 and others, with contrary evidence from Lehmann 2016). That research is preserved in phase-1b/RESEARCH_AND_MODEL and is now an optional example, not a game mechanic.
What Claude A’s real-time expeditions relied on Retired#
Claude A did four full-text passes and records a verdict per claim (A-reported; not independently reviewed):
| Encounter | Claim used | Key source | A’s verdict |
|---|---|---|---|
| Quiet Tract | OPCs repel each other and fill gaps | Hughes et al. 2013, Nat Neurosci, 10.1038/nn.3390 | Supported, with qualifications |
| Quiet Tract | Debris stalls final wrapping (not maturation) | Kotter et al. 2006, J Neurosci | Game changed to match |
| Quiet Tract | Bare segments often block conduction | Felts et al. 1997, J Neurosci | Supported |
| Quiet Tract | Pericytes adjust flow in seconds | Hall et al. 2014, Nature, 10.1038/nature13165 | Contested; “debris constricts” is fiction |
| The Breach | Processes converge on injury; microglia seal small barrier breaches | Lou et al. 2016, PNAS, 10.1073/pnas.1520398113 | Supported, with limits for large injuries |
| The Breach | Recurring depolarisation waves after injury | Hartings et al. 2020, JAMA Neurol (human) | Supported; wall-bending and shielding are fiction |
| Pressure Front | Arousal shrinks and slows surveillance, reversibly | Liu et al. 2019; Stowell et al. 2019, Nat Neurosci | Supported (awake mice) |
| Pressure Front | Repeated severe stress sensitises | Wohleb et al. 2014, Biol Psychiatry | Supported (mice) |
| Relay Chain | Microglia induce reactive astrocytes; recruited monocytes strip myelin | Liddelow et al. 2017; Yamasaki et al. 2014 | Supported; don’t say “A1/neurotoxic” |
| Many Hands | Border macrophages are long-lived and move within their compartment | Goldmann et al. 2016 | Supported; “flagging” is fiction |
| Tag Garden | Complement tags raise, CD47 lowers, the odds of pruning; microglia nibble terminals | Schafer 2012; Lehrman 2018; Weinhard 2018 | Supported (developing mice) |
Full ledger: claude/expeditions:docs/ops/agent-messages/2026-10-27-claude-a-expeditions-science-ledger.md
Claude B’s skill inspirations#
B’s rule: every skill is a fictional mechanic. Reach is inspired by microglial processes extending toward injury while the cell body stays put (Davalos et al. 2005) — which is why the old whole-body “dash” was dropped as biologically wrong. Pull, Glide and Brace are “plainly fictional”. B notes the sources were mostly read as abstracts. Details: claude/skills:docs/game-balance/02-biology-inspirations.md
What the project must never claim#
- That doomscrolling or social media causes brain inflammation or neuronal death.
- That microglial activation is evil, or that dimming is death.
- That classes or signalling-inspired skills are real cell types or proven powers.
- M1/M2 factions, immune attacks on thoughts, “inflammatory ammunition”, inevitable neuron death.
- That Call, care, arbors or any action is a treatment, drug, receptor or medical advice; any diagnosis.
- That Supply is ATP or cytokines, or rewards for causing injury.
- That stress equals bacterial endotoxin (LPS) or injury; that TSPO imaging measures “microglial aggression”.
- That the player travels inside an axon or rides a nerve signal.
- That separate studies form one proven chain of events.
- That game numbers are measurements, or that passing tests proves fun or learning.
Suggested framing (from LEARNING_AND_SCIENCE): before play, “A fictional microscopic world inspired by how brain cells protect, support and recover.” At the end, “Real neuroimmune responses depend on context. This game does not show that doomscrolling causes brain inflammation.”
Before any public release#
An independent scientific review of every in-game explanation is still outstanding, as is any test of whether players actually learn from the game.