Catalog npm
PACKAGE EXPLORERView in registry

@0xdivyanshh/tab-params

npm

Frozen credit parameter sets for Tab. A published ceiling stays recomputable.

Versions collected2
Direct dependencies2From release 0.1.1
Last observed20 Oct, 23:05 UTC

README

Current package documentation

@0xdivyanshh/tab-params

Tier 0 · versioned constants · no I/O

Every tunable number in the credit product, behind a pinned MODEL_VERSION.

Tier multiples, base APRs, ramp steps and clamps, per-call and per-window caps, the 40% concentration cap, the 0.6 unattested discount, AGE_FULL_DAYS, hop limits for funding ancestry, starter tab parameters, window length, hold TTL, and the rounding direction for interest and ramp.

Why this is a package and not a config file

verify-ceiling recomputes a published ceiling and compares the hash. That only means something if the recomputation uses byte-identical parameters to the ones in force when the ceiling was published. So parameters must be versioned, importable, and published to HCS alongside every ceiling — not read from an environment variable that has since changed.

Every ceiling message carries the MODEL_VERSION it was computed under. Changing any number here is a version bump, and old versions stay in the package so historical ceilings remain verifiable.

Contents

FileHolds
src/versions/v1.tsthe v1 parameter set, frozen
src/index.tsMODEL_VERSION, the current set, and a lookup by version
src/schema.tszod schema for a parameter set, so a new version cannot omit a field

Invariants

  • Parameter sets are frozen once published. Add v2, never edit v1. A ceiling published under v1 must remain recomputable forever.
  • No parameter is read from the environment. The only exception is the demo-mode window length, and that is a documented override that appears in the config dump.
  • AGE_FULL_DAYS is tuned down for testnet and that is stated out loud. On a testnet every account is young, so a naive age factor rejects everyone. The tuned value goes in the config dump and is named in the demo video. The README is right that hiding a tuning knob reads worse than explaining it.
  • Rounding direction for interest accrual and the ramp factor is pinned here. Both compound. Two call sites rounding differently will diverge, and the divergence will be small enough to survive review and large enough to break verify-tab.

Definition of done

A snapshot test that fails if any v1 number changes — that is the whole point of the package. Plus a config-dump function the CLI can print, since the demo depends on being able to show every parameter in force.

Downloads & popularity

Awaiting observations

No metric observations yet

Download counts and popularity will appear here as the collector receives them.

Versions 2

Collected releases
VersionPublishedDirect dependenciesOpen version
0.1.1Registry latest13 Sept 20262
0.1.013 Sept 20262

Package advisories

0
No matching advisories collected

This describes our collected data; it is not a security assessment.

Repository

PENDING
0xnerd100/tab

Shared repository data · GitHub

License detected in repository
Not collected

Metadata history

Meaningful changes
Current collected state Current

Observed since 20 Oct, 23:05 UTC

No superseded metadata states collected. Unchanged observations do not create new revisions.

Record identity

package:6a3512ee50b78db19ad2d1d7c29859630ea1f32dd785535fdd128b9a02f3e0e6