Hyperlex

Standalone Hermes skill for memetic emergence analysis.

Hermes loads this directory and uses SKILL.md as the behavior contract.
The runtime is the bundled Python package under src/hyperlex/ plus the CLI
at scripts/hyperlex.py. No Abraxas import, no required network for baseline
(mock) mode.

Resolve paths from the installed skill root. Set:

export HERMES_SKILL_DIR="${HERMES_SKILL_DIR:-$HOME/.hermes/skills/hyperlex}"

When to Use

• Detect slang / neologisms and score virality, memetics, hyperstition
• Match slang into historical lineage families with transparent confidence
• Backfill YTD slang packs and backpropagate lineage onto historical receipts (non-mutating)
• Phase 5 simulate cultural transmission, multi-agent memetics, hyperstition risk, phylogeny scaffold
• Risk-schedule advisory LIVE_EMERGENCE_SCAN cadence from risk tiers (never auto-registers cron)
• Emit integrity-hashed receipts for auditable runs
• Extract forecasts from analysis (probabilities only — no fake Brier)
• Settle forecasts as an operator and recompute Brier series from the score log
• Scan betting-sharp, crypto-degen, ai-native, brainrot, kinship, political-status families

When Not to Use

• General multi-platform web research → agent-reach
• Product / opportunity intelligence → neon-genie
• Cinematic continuity / storyboards → kubrick
• Symbolic code architecture mapping → orchestra

Prerequisites

• Python 3.10+
• python3 on PATH
• Optional: requests, jsonschema, crawl4ai for richer ingest / validation
• Optional: network for non-mock sources

Install

bash install.sh --dry-run
bash install.sh
# installs to ~/.hermes/skills/hyperlex by default
python3 "$HOME/.hermes/skills/hyperlex/scripts/hyperlex.py" check
python3 "$HOME/.hermes/skills/hyperlex/scripts/hyperlex.py" smoke

Commands (prefer simplified path)

HLX="python3 $HERMES_SKILL_DIR/scripts/hyperlex.py"

$HLX commands                          # full simplified map (JSON)
$HLX sources                           # sources + routes
# AUTO backend — ingest → full results (receipt, forecasts, phase5 risk)
$HLX pipeline "rizz" --route offline
$HLX ingest "locked in"                # same as pipeline (use --raw-only for signal only)
$HLX pipeline "sigma rizz locked in"   # expands to atoms automatically
$HLX pending                           # open forecasts
$HLX settle --forecast-id <id> --decision TRUE
$HLX score-series --mean-shift --verify-chain
$HLX scan --route offline --receipt --forecasts --append-log
$HLX risk-schedule --tier MODERATE --schedule-out /tmp/hlx-cron
$HLX doctor && $HLX smoke

Ingest routing: prefer --route offline|live|glossary|social over raw --source.
Aliases: real→glossary, x→x_search, firecrawl→crawl4ai. Offline env: HYPERLEX_OFFLINE=1.

Research / advanced (still available): simulate, vector-*, archive-export,
lineage-backfill, lineage-backprop, relay, signal, diagram, ledger-*.

Docs: docs/operator-loop.md, docs/commands.md.

Operator calibration path

run "<query>" --route offline
  → pending
  → settle --forecast-id … --decision TRUE|FALSE
  → score-series [--mean-shift] [--verify-chain]

• Score log default: ~/.hyperlex/score_log.jsonl
• Override: HYPERLEX_SCORE_LOG, --log, or --repo-log → out/calibration/score_log.jsonl
• Never emit numeric Brier without settlement. Empty series → NOT_COMPUTABLE.

Preferred sequence

1. commands — see simplified map.
2. run --route offline — one-shot analyze + receipt + forecasts + score log.
3. pending → settle → score-series — Brier only after operator settlement.
4. Cron — risk-schedule (advisory) + scan --route offline for multi-query.
5. Live ingest — only when network allowed: --route live (or glossary/social).
6. Label claims — OBSERVED / INFERRED / SPECULATIVE; fail closed on missing outcomes.
7. Research — simulate / archive / vector are optional and SPECULATIVE.

Public API (package)

from hyperlex import (
    ingest_signal, fetch_ingest, detect_memetic_patterns,
    match_lineage, emit_receipt, extract_forecasts,
    settle_and_log, recompute_series, score_pair, score_series,
    NOT_COMPUTABLE,
)

result = detect_memetic_patterns(query="rizz", ingest_source="mock")
forecasts = extract_forecasts(result)
# later, after operator review:
# settle_and_log(forecast, outcome_value=1.0, settlement_decision="TRUE")
# recompute_series()

Ensure src/ is on PYTHONPATH when importing outside the CLI
(CLI inserts src/ automatically).

Authority boundaries

Hyperlex may:

• ingest and analyze signals
• match lineages with transparent score breakdowns
• extract forecasts and write receipts / score-log events
• compute Brier only from settled pairs
• export Abraxas-compatible ledger shapes (no Abraxas import)

Hyperlex may not:

• invent numeric Brier on open analysis (provenance.brier stays null)
• auto-settle without authority marker
• promote speculative hyperstition stages as hard truth
• rewrite historical receipt integrity during lineage backprop (report only)
• invent Brier from Phase 5 simulation (always brier: null, SPECULATIVE)
• mutate Abraxas or other systems (export is optional and offline)

Pitfalls

• scripts/hyperlex.py must not shadow the package: always run via the skill path so src/ is first on sys.path.
• Non-mock sources need network and may degrade gracefully — check ingest metadata.
• Lineage confidence is INFERRED; do not treat it as observed ground truth.
• Mean-shift from score-series --mean-shift is advisory for future forecasts only.
• Score log is append-only; series is recomputed from the log, not stored as sole truth.

Verification

python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" check
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" doctor
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" ledger-stats
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" signal --input <result.json>
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" feedback --signal-key hyperstition.stage
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" diagram --from-golden --out-dir out/diagrams
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" scan --config "$HERMES_SKILL_DIR/examples/cron/scan-queries.json" --source mock --receipt --forecasts
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" smoke

Successful packaging:

• ~/.hermes/skills/hyperlex/SKILL.md exists
• check returns "ok": true
• smoke writes a receipt under out/smoke/
• Open analysis has "brier": null

Design references

• DESIGN.md — principles (incl. 11 lineage, 12 Brier requires settlement)
• docs/brier-calibration.md — forecast → settlement → score
• docs/slang-lineages.md — family methodology
• docs/phase5.md / docs/modules/simulation.md — Phase 5 research simulation
• schemas/ — ingest, result, receipt, forecast, settlement, brier_series, lineage
• examples/slang-families/ — Mermaid + HTML family diagrams
• data/backfill/2026/ — YTD slang packs
• references/hermes-runtime-contract.md — path / authority policy

Security

Local stdlib-first CLI. Baseline (mock) needs no network. Real ingest may call public web APIs. Score log and receipts are local files under ~/.hyperlex/ or skill out/.