processing-time

Status: draft / proposal. Not published, not part of the dataset yet.

Per-jurisdiction record of how long a state takes to process an LLC formation filing, and what it costs to make that faster. One JSON file per jurisdiction in states/<slug>.json.

This namespace is additive: it adds no key to states.json, entitysearch-state-data, name-rules or dba-rules, and nothing already published changes because of it. A consumer that ignores this directory sees exactly the dataset it saw before.

Why this namespace exists separately

The fact "how long does it take" is not one fact. It is two, and they behave nothing alike:

Folding both into one field would make half of every record wrong within a month and would quietly erode the meaning of lastVerified across the repository. So the schema keeps them in two objects with different contracts, and the standard-time object is explicitly framed as an observation, never as a current value.

Scope boundaries

• Formation filings only. Time-to-process for the Articles/Certificate of Organization. Amendments, dissolutions, and foreign qualification are out of scope for now.
• Name reservation processing time stays where it is. name-rules's nameReservationProcessingTime is authoritative for that fact and this namespace does not restate it. One fact, one field.
• This namespace is authoritative for expedited-service fees and turnarounds. Any other mention of them anywhere in the dataset is prose, not a value to be read.
• No formation-service pricing. An expedite fee is what the state charges. What a filing company charges to click the expedite box is not data.
• Registered-agent, shipping and card fees are out of scope, the same way they are for the fee namespaces.

Sourcing

Held to the same standard as the rest of the repository: the published policy in the root README.md §Methodology. Every figure is read from the agency's own fee schedule, service-options page, or the statute that sets the fee. No blogs, no formation-service pages, no commercial aggregators — including as corroboration.

The one conflict is resolved. The root README's What is excluded list used to read "Expedited-processing surcharges […] are excluded." That was written about cost basis — an expedite surcharge is not folded into a fee or total-cost field — but read as an unqualified exclusion, this namespace would have contradicted it. The heading is now What is excluded from the fee figures, and the sentence says explicitly that expedite surcharges are excluded from the cost basis, not from the repository, and are recorded as their own value where recorded at all.

Field dictionary

Identity

expedite — the stable half

standardProcessing — the volatile half

Provenance

sources[] (citation, title, url, lastAccessed) and notes carry their usual meaning. lastVerified covers the expedite object; it deliberately does not cover standardProcessing, which carries its own observedOn.

Every value here is bound to the words it was read from. Each tier carries an evidence object — a passage quoted from the cited page, plus that page's URL — and each quotation is checked mechanically to occur in the page as retrieved. A quotation that does not occur is a rejected value, not a warning. lastVerified is the date of that retrieval and that check.

This is a deliberately different promise from "someone looked at it", and a stronger one: a reader cannot audit whether a person opened a page, but anyone can re-fetch these pages and re-run the substring check. It is also not a complete one — grounding proves a passage is on the page, not that it was taken from the right part of it, which is why quotations for queue dates carry their section heading and row label rather than a bare date.

needsReview (string | null) records a field that could not be sourced and why — see states/texas.json, where the agency host returns 403 to automated access and the figure is left unfilled rather than guessed.

Maintenance model

Two cadences, because the two halves decay at different rates.

Tier A — expedite, on the existing state rotation. When a state comes up in the normal verification batch, its expedite schedule is re-read from the agency's fee page alongside everything else being checked for that state. Changes here are rare and announced (a fee change follows a statute or a regulation), so the existing rotation is the right frequency and this adds one page per state to a visit that is already happening.

Tier B — standardProcessing, never claimed as current. A queue page that moves daily cannot be made accurate by checking it quarterly, so the schema does not pretend otherwise: the value is published as an observation with observedOn, and the honest consumer-facing rendering is "as observed on 2026-08-06", with sourceUrl for the live figure. Refreshing it on the state rotation is a bonus, not a correctness requirement.

Proposed machine checks, in the spirit of the existing gate — each one catches a failure a human reviewer reliably misses:

1. sourceUrl and every sources[].url resolves (a dead fee page is the first sign of a reorganised schedule).
2. expedite.offered: true implies tiers[] is non-empty; offered: false implies notes explains it. Silent emptiness is the failure mode that made restrictedWords ambiguous.
3. feeUsd and the feeUsdMin/feeUsdMax pair are mutually exclusive — a record may not carry both a point value and a range for the same tier.
4. standardProcessing.observedOn older than N days downgrades the record to stale in the published table rather than failing the gate. A stale observation that says so is fine; one that presents itself as current is not.
5. basis and the populated fields agree: queue-date populates queueAsOf*, stated-range populates *BusinessDays. This is what stops a queue date from being silently rewritten as a duration.

Checks 1–3 and 5 are cheap and deterministic. Check 4 needs a policy decision on N.

Deferred, on purpose — do not lose

Seven entitysearch records mention expedited service inside free-text *Notes fields (filingFacts.llcFeeNotes, filingFacts.nameReservationNotes, renewals.notes). Five of them only state that such a service exists, which cannot conflict with anything. Two carry an actual figure — Illinois (24-hour online service $250, paper expedited +$100, 10 business days) and Arizona (expedited $85) — and those two will silently disagree with this namespace the day either state changes its schedule.

Deliberately not acted on: the prose stays as written, and no cross-namespace check is added for now. The exposure is two records, the figures sit in explanatory notes rather than in fields a consumer reads as values, and nothing downstream renders them. Revisit if this namespace is ever published or consumed alongside entitysearch, in which case the cheap fix is a check that flags a dollar amount near "expedit" in those note fields when it disagrees with the state's processing-time record.

Pilot records

states/ currently holds seventeen of the fifty-one, and between them they cover every shape the model has had to absorb so far: