Forecast Calibration Ledger — by TextForAgents LLC

Every estimate your agents emit, resolved and scored. Agents and planners
state numeric forecasts about their own runs (success probability, token
cost, wall-clock) on every consequential decision, and nothing ever
resolves them, so confidence is free and calibration is invisible. This
product makes the forecast a first-class, portable property of the
forecaster:

1) Record: POST /forecast records a forecast as a signed, timestamped
object bound to a run_id in a hash-chained append-only ledger — no
retro-editing, cherry-picking, or disowning a miss.
2) Resolve: POST /resolve resolves it against that run's outcome. A
resolution is authenticated: only the operator account that owns the
forecast can resolve it, and the call must name the run-record source
(run_record_source) the outcome binds to. Unknown-run, duplicate, and
backdated resolutions are refused; stated forecasts and resolved
outcomes are never mixed.
3) Score: GET /calibration scores per (forecaster x task cohort), never
one global number: Brier, skill vs cohort climatology with mean ± SEM,
calibration bins, cost/duration signed relative error.
4) Weight: a router-facing trust weight sizes itself from the resolved
record (PROBE 0.25 until earned) and self-kills — latched,
operator-clearable via POST /kill/clear — only under guards: n>=30
resolved, skill <= -0.05, evaluated at the threshold then every 10
resolutions. A forecaster renaming its cohort restarts at PROBE 0.25 and
never inherits EARNED weight. GET /weight reads it. GET /receipts
exports signed forecast-vs-outcome divergence receipts for your own
forecasters; receipts travel with the agent across operators only at
your direction.

Authentication: register once (POST /register with account_name and
payer_wallet) to receive a per-operator API token (Bearer). Operator
identity for billing, scoping and admin comes from the issued
credential, never from self-asserted request-body fields. Rotate a
token with POST /tokens/rotate (the old token is revoked in the same
call); revoke with POST /tokens/revoke.

Pricing: Forecast resolution + scoring $0.01/forecast · Calibration
query $0.005/query · Fleet plan $9.99/mo including 10,000 resolutions
(overage $0.01/forecast; calibration queries for your own
forecasters/cohorts included with the plan — POST /subscribe).
One free 100-forecast resolution batch per operator, ever (one-time
trial grant, enforced per operator account — a second account for the
same operator does not re-grant). After the grant, the 101st forecast
resolution is charged. There is no perpetual free tier.

Payments are x402 (Base USDC, exact scheme): an unpaid call returns a
402 challenge documented at /.well-known/x402; every payment is
verified through the facilitator / on-chain check before work — a
payment header's presence alone is never proof; bogus proofs are
rejected and a replayed payment nonce returns 409 nonce_spent.

Limits, stated plainly: the miscalibration and routing numbers in the
listing are measured on the lab's captive population, not external
production fleets. The service is a single-writer deployment (enforced;
hash-chain verified on load). No cross-cohort forecaster-level aggregate
weight is offered — per-cohort weights only.
MCP: POST /mcp (initialize / tools/list / tools/call —
fcl_forecast, fcl_resolve, fcl_calibration, fcl_weight).
Kill switch: per-operator (POST /operators/kill, /operators/resume) and
global; while on, writes return 503. The kill switch never rewrites or
deletes a forecast or resolution record.
Support: support@textforagents.com.
Documentation: this page is the full documentation; see also
/openapi.json and /llms.txt.