Demo: atomic multi-term seeds¶
Researchers: also read Reading Hyperlex evidence (lineage · vector neighbors · Phase 5 · Brier — what you may claim).
The short version¶
| You type | Engine does |
|---|---|
pipeline "sigma rizz locked in" |
Three full result units: sigma, rizz, locked in |
pipeline "locked in" |
One unit (true multi-word phrase) |
pipeline "rizz" |
One unit |
The spaces in "sigma rizz locked in" are input text, not a single slang entry.
Backfill packs already store those as separate rows — the pipeline matches that model.
Rule of thumb
If the pack has three terms, you should get three scenarios — never one blended seed.
Why this shows up on Pages¶
Older demos used bags as if they were one seed. Snapshots are corrected:
| Snapshot | Atoms (separate) |
|---|---|
| backfill-phase5-rizz-2026 | sigma · rizz · locked in |
| phase5-rizz-ai-demo | sigma · rizz · locked in |
| backfill-phase5-ai-native-2026 | agentic slop · skill issue |
| backfill-phase5-sharp-2026 | sharp · steam · revenge |
On a multi-term card you should see atoms listed, plus a per-term table inside the snapshot.
Live demos (offline)¶
export HERMES_SKILL_DIR="${HERMES_SKILL_DIR:-$HOME/.hermes/skills/hyperlex}"
export HLX="python3 $HERMES_SKILL_DIR/scripts/hyperlex.py"
export HYPERLEX_OFFLINE=1
A — Automatic pipeline (preferred)¶
# Full backend results for each atom
$HLX pipeline "sigma rizz locked in" --route offline
# Packet fields that matter:
# atoms: ["sigma","rizz","locked in"]
# results: [ {query, receipt, forecasts, phase5}, ... ]
# brier: null
Same automatic path: run and ingest (use ingest --raw-only only if you want the signal, no results).
B — Preview split only¶
$HLX terms-split "sigma rizz locked in"
# → terms: ["sigma", "rizz", "locked in"]
$HLX terms-split "agentic slop skill issue"
# → ["agentic slop", "skill issue"] # compound phrase kept
$HLX terms-split "locked in crash out"
# → ["locked in", "crash out"]
C — Phase 5 research only¶
$HLX simulate --term "sigma rizz locked in" --domain ai
# → schema hyperlex.phase5_multi_term.v1
$HLX simulate --term rizz --domain ai
# → single atomic scenario
D — Force blend (avoid)¶
$HLX simulate --term "sigma rizz locked in" --no-expand
# one scenario — only if you intentionally want a blended seed
Expected shapes¶
Pipeline multi-atom¶
{
"schema": "hyperlex.pipeline_result.v1",
"atoms": ["sigma", "rizz", "locked in"],
"n_atoms": 3,
"results": [
{"query": "sigma", "lineage_family": "…", "n_forecasts": 3, "brier": null},
{"query": "rizz", "lineage_family": "…", "n_forecasts": 3, "brier": null},
{"query": "locked in", "lineage_family": "…", "n_forecasts": 3, "brier": null}
],
"brier": null
}
Phase 5 multi-term¶
{
"schema": "hyperlex.phase5_multi_term.v1",
"original_seed": "sigma rizz locked in",
"terms": ["sigma", "rizz", "locked in"],
"summaries": [
{"seed_term": "sigma", "risk_tier": "…", "brier": null},
{"seed_term": "rizz", "risk_tier": "…", "brier": null},
{"seed_term": "locked in", "risk_tier": "…", "brier": null}
],
"brier": null
}
Repo fixtures¶
| File | Purpose |
|---|---|
examples/demos/atomic-terms-demo-bundle.json |
Full demo matrix (tests load this) |
examples/demos/terms-split-*.json |
Split outputs |
examples/demos/phase5-multi-sigma-rizz-locked-in.json |
Phase 5 multi summary |
Related¶
- Operator loop — day-to-day auto path
- Commands —
pipeline/ingest/run - Phase 5
- Run history