Hyperlex Architecture¶

Version: 0.2.x · Mode: Hermes skill (Python package repo)

Hyperlex is a pure-function-first memetic engine packaged as a Hermes skill. Optional append-only side effects: receipts, score log, receipt ledger. It does not import Abraxas. Relevant Abraxas wire shapes live under hyperlex.compat.abraxas.

Data flow¶

query + source
    → intake (cache, rate limit, fingerprints)
    → analysis (neologism, lineage, virality, memetics, hyperstition)
    → optional: extract_forecasts (no Brier)
    → optional: emit_receipt + receipt ledger
    → optional: relay RUNE.HLX.* envelopes
    → time / operator settlement
    → score_pair / score_series (Brier only if settled)
    → optional: market_signal / forecast_pipeline connectors
    → optional: hyperstition feedback → future mapping advice

Package layout¶

src/hyperlex/
  intake/           # ingest adapters + cache + glossaries + x_search
  analysis/         # memetic core + lineage matcher
  synthesis/        # external signal stub
  receipt/          # emit + hash-chained receipt ledger
  calibration/      # forecast, settle, score, score_log, recalibrate
  relay/            # RUNE.HLX.* envelopes
  connectors/       # market-signal + forecast pipeline packets (generic)
  diagrams/         # Mermaid from receipts / ledger
  compat/abraxas/   # BrierLedger/Score/operator review/claims/runes (no Abraxas import)
  provenance.py     # source fingerprints
  cli.py            # python -m hyperlex
  schemas/          # package-local JSON schemas

Layers¶

1. Intake¶

Source	Role
`mock`	Deterministic, query-aware offline
`glossary` / `real` / `web`	Action Network
`glossary_expanded`	Multi-glossary pack
`reddit`, `urban`, `wikipedia`	Live public APIs
`x_search`	Bearer API → xurl → stub
`crawl4ai` / `firecrawl`	Web crawl
`combined`	Ordered multi-source merge

Disk cache: ~/.hyperlex/cache/ · Rate limits: ~/.hyperlex/rate_limit.json

2. Analysis¶

Neologisms, semantic variation, virality hybrid, memetics protocol, hyperstition stage
Lineage match + confidence (≥ 0.42) with transparent score_breakdown
Open results: provenance.brier = null always

3. Calibration¶

Signal→f maps (lineage, virality, hyperstition stage)
Settlement → atomic/series Brier, Murphy, Ferro–Fricker, Yates, Vieira, Δf
Score log: ~/.hyperlex/score_log.jsonl
Hyperstition feedback: advisory stage-map updates from settled series (future forecasts only)

4. Receipts & ledgers¶

Receipt files: ~/.hyperlex/receipts/
Receipt ledger: ~/.hyperlex/receipt_ledger.jsonl (hash-chained index)

5. Relay & connectors¶

relay → RUNE.HLX.LIVE_EMERGENCE_SCAN, COMMUNICATION_RELAY, CALIBRATION_*
connectors.market_signal → generic market/narrative signal packet
connectors.forecast_pipeline → handoff pack for external forecast systems

6. Host compatibility (optional)¶

Hosts import from Hyperlex:

from hyperlex.compat.abraxas import (
    to_brier_ledger_entry,
    to_brier_score_packet,
    to_operator_brier_review,
    list_hlx_runes,
)

Constraints¶

No fabricated Brier without settlement
Fail-closed NOT_COMPUTABLE
OBSERVED / INFERRED / SPECULATIVE discipline
Python ≥ 3.10, stdlib-first baseline

See SPEC.md, docs/api-v1.md, docs/hermes-skill.md.