From 2d61d5554dd097d601bf6d05d41b5385595e41b8 Mon Sep 17 00:00:00 2001 From: Meta Vens Date: Sat, 2 May 2026 11:54:24 +0100 Subject: [PATCH] Document corpus coverage goals --- docs/architecture/training-plan.md | 62 +++++++++++++++++++++++++ notes/fresh-workspace-handoff.md | 5 ++ notes/ghostlight-implementation-plan.md | 6 +++ state/evidence.jsonl | 1 + state/map.yaml | 1 + 5 files changed, 75 insertions(+) diff --git a/docs/architecture/training-plan.md b/docs/architecture/training-plan.md index 99850cf..4be80e2 100644 --- a/docs/architecture/training-plan.md +++ b/docs/architecture/training-plan.md @@ -47,6 +47,68 @@ Future-branch examples must label: - the post-Rupture concept being exercised - the old social pressure being re-expressed +## Lore Coverage Matrix + +Cold Wake is a fixture, not a corpus. A complete Aetheria training corpus needs +coverage across the setting's institutional ecology before the models can be +trusted to carry tone, conflict, and cultural priors without constant hand +feeding. + +Before calling the corpus complete, generate reviewed story/scenario coverage +for: + +- every major faction +- every minor faction +- every movement and counter-project +- every major historical flashpoint +- every major post-Rupture concept or branch-attractor family used in gameplay +- faction-adjacent institutions that strongly shape daily life, trade, law, + care, warfare, media, faith, logistics, or embodiment + +Major factions need at least two distinct story types: + +- `founding_era_story`: the faction's formative pressure, first stable myth, + early institutional compromise, or founding betrayal. +- `day_in_the_life_story`: ordinary operating texture after the faction has + matured, showing how people actually live inside its systems. + +Minor factions, movements, and flashpoints need at least one strong story or +branching-scene fixture each, plus negative or rejected examples where the first +attempt collapses into stereotype, lore drift, or generic sci-fi wallpaper. + +Every accepted coverage story should involve cultural collision. Do not write +sealed terrariums where one faction monologues at itself and calls that +worldbuilding. The useful data comes from contact: + +- major faction versus major faction +- major faction versus minor faction +- faction versus movement +- movement versus movement +- institution versus care network +- local colony versus imported authority +- species/body affordance versus human-default infrastructure +- technical standard versus ritual, law, class, scarcity, or dignity + +Coverage records should tag: + +- primary faction or movement +- secondary and tertiary factions or movements +- flashpoint or era +- story type: `founding_era_story`, `day_in_the_life_story`, + `flashpoint_story`, `movement_story`, `future_branch_story`, or + `trade_life_story` +- collision axis: law, care, class, logistics, species/body, labor, faith, + media, war, trade, personhood, ecology, memory, technology, or legitimacy +- involved cultures and their mutual misreads +- item/economy hooks discovered, including colony-consumed trade goods +- source refs and open lore gaps +- accepted, rejected, or needs-revision review status + +The first pass can be shallow and uneven. The final corpus cannot be. If one +flashpoint starts producing all the examples, stop and rebalance before the +model learns that Aetheria is mostly one corridor argument wearing different +coats. + Some Elysium events are not globally fixed, but they are not equally optional either. Founding events for Elysium-born factions, technology discoveries, metaphysical practices, and quest hooks should be represented as attractors, diff --git a/notes/fresh-workspace-handoff.md b/notes/fresh-workspace-handoff.md index 52acd84..3ca5092 100644 --- a/notes/fresh-workspace-handoff.md +++ b/notes/fresh-workspace-handoff.md @@ -75,6 +75,11 @@ current gold-data route. only gear assemblies or manufacturing inputs; a Navigator wet-interface cradle is the current example of generated texture that should become a possible demand-profiled trade-good candidate. +- Corpus completeness requires broad Aetheria coverage, not repeated Cold Wake + refinement. Generate stories for every major faction, minor faction, + movement, and major flashpoint. Major factions need both founding-era and + day-in-the-life stories. All coverage stories should involve cultural + collision so faction and movement dynamics appear in the training data. ## Current Next Action diff --git a/notes/ghostlight-implementation-plan.md b/notes/ghostlight-implementation-plan.md index 030c1c4..935ad9c 100644 --- a/notes/ghostlight-implementation-plan.md +++ b/notes/ghostlight-implementation-plan.md @@ -38,6 +38,12 @@ sandboxed training-data pipeline for branching scenes and state consequences. 5. Generalize the loop. - Add more historical grounded fixtures from AetheriaLore. - Add future-branch Elysium fixtures with branch lineage and constraint labels. + - Track coverage across every major faction, minor faction, movement, and + major flashpoint; major factions require both founding-era and + day-in-the-life stories. + - Require cultural collision in coverage stories so training data captures + inter-faction dynamics, mutual misreads, and movement pressure instead of + sealed faction portraits. - Emit technology/item manifest deltas when scenes discover or stress gear, assemblies, supply chains, or faction tech bases. - Keep artifacts database-shaped enough for future game-engine integration. diff --git a/state/evidence.jsonl b/state/evidence.jsonl index b1fbf7f..abf3f5c 100644 --- a/state/evidence.jsonl +++ b/state/evidence.jsonl @@ -16,3 +16,4 @@ {"ts": "2026-05-02T10:41:08+00:00", "type": "design", "status": "accepted", "note": "Research-enabled responder scope is seed-plus-traversal, not a narrow file whitelist: training samples should recover PSC, Jovian territory, resident faction, linked movement, politics, and technology context; parent-visible subagent output does not prove tool calls, so future gold data needs explicit research trace receipts."} {"ts": "2026-05-02T10:44:38+00:00", "type": "experiment", "status": "useful_needs_revision", "note": "Quick research sanity check placed a Navigator route steward in a scene missing embodiment and institutional lore. The no-fork responder recovered Navigator acoustic/translation/fluid-space behavior and PSC/Cold Wake/Ganymede Compact context, but parent-visible subagent output still lacks auditable tool-call receipts; capture saved under experiments/research-sanity."} {"ts": "2026-05-02T10:51:01+00:00", "type": "design", "status": "accepted", "note": "CompoundCommodityData is not limited to gear assemblies or manufacturing inputs; it also covers finished manufactured trade goods consumed by colonies, habitats, ships, clinics, institutions, species communities, and faction enclaves. Generated setting texture like a Navigator wet-interface cradle should be eligible to become a demand-profiled trade-good blueprint candidate."} +{"ts": "2026-05-02T10:54:15+00:00", "type": "design", "status": "accepted", "note": "Corpus completeness requires coverage across every major faction, minor faction, movement, and major flashpoint. Major factions need founding-era and day-in-the-life stories, and all coverage stories should involve cultural collision so training captures faction and movement dynamics rather than isolated faction portraits or Cold Wake overfitting."} diff --git a/state/map.yaml b/state/map.yaml index d2ce74d..8f30c3c 100644 --- a/state/map.yaml +++ b/state/map.yaml @@ -69,6 +69,7 @@ active_subgoals: - Preserve Aetheria grounding through source-backed lore, branch labels, and reviewed assumptions. - Shape every soft judgment as future training data, not permanent prompt folklore. - Keep item/technology exploration database-shaped for Aetheria-Economy instead of prose-only worldbuilding, including colony-consumed manufactured trade goods, not only gear assemblies. + - Track corpus coverage across all major factions, minor factions, movements, and flashpoints so no single fixture becomes the training universe by accident. current_artifacts: canonical_docs: