Skip to content

sadpig70/endowfront

Repository files navigation

EndowFront infographic

EndowFront

CI License: MIT

EndowFront says pay once, frontier forever. Per-query billing cannot survive abundance — as capability gets cheaper, per-token revenue erodes. So stop billing per query: a consortium of capability-dependent institutions places a one-time permanent endowment whose returns fund frontier serving forever, in exchange for irrevocable open access. For a batch of pledges it answers:

Does the corpus fund the frontier across the horizon, does it preserve itself, what is the year-0 coverage, and is irrevocable open access granted?

It aggregates the corpus, simulates it year by year, judges perpetual sustainability, and emits a posture verdict: endowed, underfunded, or depleted.

The Insight

A two-step transform on the L7_Tension that per-query billing cannot survive abundance:

  • L1 direction reversal — reverse who funds: a consortium of institutions funds the lab, not per-query users.
  • L13 frequency shift — move the billing cadence from continuous per-query to a single permanent endowment whose returns fund frontier serving forever. The lab's only KPI becomes keeping the frontier moving; abundance becomes the deliverable.

This is the EVX consensus + innovation winner (12 of 12 personas).

What Makes It Different (vs per-output certification)

GenCert shares the same lens stack (L1 → L13) but is a trust mechanism (certify a generator once, inherit verification). EndowFront is a financing mechanism:

  1. Endow once — the corpus is funded a single time, not per query.
  2. Simulate the whole horizon — perpetuity is tested by a multi-year corpus simulation, not a single year-0 ratio. A return at or below cost growth eventually depletes the corpus and is surfaced honestly.
  3. Grant in exchange — irrevocable open access flips on only when the endowment is perpetual; per-token economics are abolished, not patched.

The regulated audit surface (R2) is first-class: every pledge emits an evidence line and the result carries a full year-by-year corpus projection.

Quickstart

python -m endowfront sample   --output examples/batch_sample.json
python -m endowfront simulate --input examples/batch_sample.json
python -m endowfront report   --input examples/batch_sample.json --output examples/endowfront_report.md

Batch Spec

{
  "policy": {
    "real_return_rate": 0.06, "frontier_cost_year0": 30.0, "cost_growth_rate": 0.02,
    "horizon_years": 50, "min_coverage_ratio": 1.0, "require_open_access": true
  },
  "pledges": [
    { "pledge_id": "p-national-lab", "name": "National laboratory", "amount": 300.0 }
  ]
}

(Amounts are in any consistent unit — the sample uses USD millions.)

How It Works

corpus = Σ pledge.amount                                        # L1: institutions fund the lab

simulate each year (one-time corpus, not per-query — L13):
  payout   = corpus * real_return_rate
  closing  = corpus + payout - cost
  cost    *= (1 + cost_growth_rate)
  stop if closing < 0  (depleted)

sustainability:
  years_funded     = consecutive years with closing >= 0
  preserved        = final_corpus >= corpus0      (real preservation -> perpetuity proxy)
  coverage_year0   = (corpus0 * real_return_rate) / frontier_cost_year0

verdict:
  not funded all horizon                              -> depleted (exit 1)
  funded AND preserved AND coverage >= min_coverage   -> endowed   (open access granted)
  otherwise                                           -> underfunded

An unknown field, a negative magnitude, a nonpositive cost or horizon, or a zero corpus is a blocking issue with a nonzero exit, never a stack trace.

MVP Scope

EndowFront produces a sustainability appraisal and an open-access eligibility verdict under a declared policy, not investment advice, a binding endowment agreement, a guarantee of returns, or a grant of access. It aggregates, simulates, and flags; it does not hold funds or serve a model.

Provenance

EndowFront is the concrete beachhead of IDEA-024 (Permanent Capability Endowment: Pay Once, Frontier Forever — EVX-20260607-010 winner, 12 votes), reached via the L1_DirectionReversal → L13_FrequencyShift lens stack (CIX-20260607-001 ← IDX-20260607-001 ← TCX-20260606-001). It is a sibling of WasteStack, FailureFutures, PowerRoam, CoverGate, ColdMkh, GenCert, BuyBloc, RoboTrace, InferMesh, CertMesh, SettleMesh, SpendMesh, ReleaseMesh, ClimateMesh, SovMesh, FlowMesh, and SignalMesh.

License

MIT — see LICENSE.

Releases

No releases published

Packages

 
 
 

Contributors

Languages