Hub (hub.tcl): syntax-agnostic Tcl coordinator that manages sim
lifecycle, tick-advancement at 90:1, and query routing via
stdin/stdout JSON to sub-processes in their native runtimes.
BoundedPrediction enforces L2 (bounds on every output) and
confidence scale 0.00-10.00.
M3a statistical/main.R: GBM Monte Carlo, Heston stochastic
volatility (Euler-Maruyama with full truncation), Merton
jump-diffusion, HMM 3-state regime detection (forward filter),
GARCH(1,1) volatility forecasting. All produce BoundedPredictions.
M3b sociological/main.pl: Replicator dynamics over 5 behavioral
archetypes, Hegselmann-Krause bounded confidence opinion dynamics,
Nash equilibrium search via constraint satisfaction. Prolog handles
game logic natively; numerical ODE/PDE delegates to R/Octave.
Hub spec updated to reflect Tcl as coordinator language.
MCP create_pull_request tool auto-appends session URLs to PR bodies.
This skill documents the two-step workaround: create then immediately
update to overwrite the injected content.
Ranked languages per spec by fit; took only the top pick:
- M3a (statistical): R — native stats ecosystem
- M3b (sociological): Prolog — equilibria as constraint satisfaction
- M3c (AMM): Solidity — on-chain-equivalent precision
- M3d (MEV): Fortran — dense PDE/knapsack numerics, no GC
- M3e (tokenomics): Fortran — SDE/VAR matrix loops, same toolchain as M3d
- M3f (consensus): Prolog — Markov/Nash as declarative search
- M3g (microstructure): Zig — tick-level latency, deterministic memory
- Hub: updated to reflect per-spec assignments
Julia removed project-wide: JIT startup cost and large toolchain
not justified when the project isn't going all-in on a single runtime.
M0: economy organ stores local memory ledger; M6 supervises M5
directly, M7 is independent antivirus/guarddog alerting M6 via
Ichor; Ada dependency reframed to economy scope; build sequence
changed to subcomponent-first with ablative tests.
M1: submit_action returns {succeeded|failed}, diagnostics internal
to Conductor; stubs now print "if finished, would respond with..."
for debugging; law script changes require operator + Homunculus
signatures; law script format added as open item.
M2: removed Python/Rust from language options; confidence scale
changed to [0.0, 10.0] per position.
M3 hub: normalized all time horizons to ~40s wall time windows;
confidence scale 0.00-10.00 with "X.XX/10.00" print format; gain
rates as "low - mid - high / 10.00"; removed Python from language
list across all sub-specs (M3a-M3g).
M3a: Julia/R/Fortran/Octave replaces Python; fBM citation added
(Hosking 1984, Wood & Chan 1994); confidence/correctness/certainty
distinguished as 3 separate metrics; models span multiple horizons.
M3b: models span multiple horizons note added; Mesa/Python removed.
M3c: Solidity for on-chain precision; Julia/Octave for analytics.
M3d-M3g: Python removed; confidence values updated to 10.0 scale.
- endocrine/AGENTS.md, ichor/AGENTS.md: fix devCorrectionLog relative
path (../../ → ../../../) so it resolves to repo root .claude/
- bus-topology.md: update stale M2/M6 references to current specs
(M2=Data Feeds, M6=Conductor)
- M3a: combine tick/hourly to match hub's six-horizon convention
- .claude/devCorrectionLog.md: interaction pattern corrections for
future instances (additive clarifications, no idle subagents, etc.)
- M3 hub L3: horizons run in parallel, not sequential
- Mention correction log in CLAUDE.md and all subdirectory AGENTS.md
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
90:1 (1s wall = 90s sim) is the finest tick resolution. Longer
horizons use coarser time steps (1min, 10min, 1hr, 6hr, 1day)
so they cover their full window in seconds, not hours/days.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Grounded ratio based on ABIDES/ECMWF benchmarks. Single global clock
shared by all horizons. Wall time for full window: hourly ~40s,
daily ~16min, weekly ~1.9hr, monthly ~8.3hr, annual ~4.2d, 5yr ~20.8d.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Hourly–weekly: full window in ≤1 wall-second (3600:1 to 604800:1).
Monthly: daily ticks, 30s wall for full window.
Annual: weekly ticks, ~52s wall.
5-year: monthly ticks, ~60s wall.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Hourly 3600:1, daily 86400:1, weekly 604800:1, monthly 2.6M:1,
annual 31.5M:1, 5-year 158M:1. Each horizon tick-advances at the
speed needed to cover its full prediction window in 1 wall-second.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All sims advance every tick continuously. The 360:1 floor means the
slowest system produces 1 simulated hour per 1 wall-clock second.
Updated hub cadence table and all sub-spec references.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The economy organ was the least-documented named component in the
architecture — one line in bus-topology.md, a Pony primitive, and a
smoke-test wire. This adds 8 plan specs following the 10-section template:
M0 hub, M1 ingestion, M2 digestion core (small-model),
M3 cost ledger, M4 budget governor, M5 context yield,
M6 provenance chain (S1/S2 compliance), M7 outer-bus exchange.
Also updates the plans README index and integration DAG, and resolves
bus-topology.md's open item about the organ's placement.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The folder note said "rename this folder — its a bad name." Renamed
payloads → messages to better describe what crosses the Ada border.
Co-Authored-By: Claude <noreply@anthropic.com>
