Everything this platform runs on
Seven systems of record that have never shared a key, four connector classes, one canonical graph, nine agents, and two human gates. Six of the nine capability verdicts are not build — that is the argument, not a hedge.
Embedding-based blocking, then a scorer.
Wells, tracts, leases, agreements, parties.
Source system, resolution method, confidence, timestamp.
Anything under threshold is a human decision, not a guess.
lib/platform/types.ts + the Identity Graph surfaceThis is the layer that is not purchasable. The industry subsurface data standard has no land domain — no tract, lease, mineral ownership, unit or pooling entity — and that is structural rather than a pending release.
Vol 412 / Pg 88·Document feed·Daily sweepCoverage is tiered and uneven. Images existing is not the same as records being geographically indexed and machine-readable, and the gap is precisely where title projects fail.
42-389-31174-00-00·Batch extract·NightlyThe only identifier in the estate that is genuinely canonical. Everything else in the mesh is anchored to it.
BC-14H·REST API·HourlyShort names mutate through the well lifecycle and are reused across pads. Resolution has to be time-varying, not a one-off crosswalk.
Brazos Creek Unit #14H·REST API·HourlyA system of record, not a system of work. Lease-centric and tract-centric models are not interchangeable, and net acreage and continuation logic depend on which was chosen.
CC-80297·Batch extract·DailyCost centers are financial constructs that split and merge on reorganizations, so the mapping to a physical well is many-to-many over time.
48389-311-0014·Spatial service·ContinuousSpatial context sits outside the land system by default, which is why both major land vendors sell a spatial integration rather than owning the geometry.
Assigned on submit·Portal (human)·On submissionRead-only to the platform. The interface is a web form, external accounts must link to the federal identity provider, and its Rules of Use prohibit automated access including form submissions.
No agency accepts a machine as signatory. Every external account links to Login.gov, whose Rules of Use prohibit automated access including form submissions. A named person presses submit.
No agent renders a title opinion. The mesh assembles the chain and the curative list with a citation behind every assertion; a licensed attorney signs.
Any extraction below 0.75 confidence routes to a human reviewer instead of into the graph, and an obligation blocked on unresolved identity cannot be cleared.
Lease-centric and tract-or-formation-centric models are not interchangeable. Net acreage and continuation logic both depend on which one the source system chose, so a field-level mapping silently changes the answer.
ResponseResolve to a canonical subject rather than mapping field to field, and carry the source model on the edge.
Geometry lives in the spatial platform and the land system holds a text description of it. That is why both major land vendors sell a spatial integration rather than owning the geometry themselves.
ResponseTreat geometry as a first-class attribute of the canonical tract, sourced from the spatial platform and reconciled against the parsed description.
Both major vendors in this category have changed private-equity ownership recently. The standard sponsor playbook monetizes data egress and API tiers rather than liberalizing them.
ResponseNegotiate egress rights and API tiering explicitly at renewal, before the identity graph creates a dependency that cannot be unwound.
Not one vendor in this market publishes an accuracy figure for extraction on oil and gas land instruments. The category being sold is an unmeasured capability.
ResponseMake a held-out accuracy score on pre-1950 handwritten assignments in a named county a condition of purchase.
Bounded scope, commodity supply, and a plausible sub-six-month payback.
Smallest data requirement and the fastest evidence, because the operator already owns the training set in its own filing history. Build this first.
A data-governance program wearing an AI badge. Fund it from the deficiency win and defend it as infrastructure.
Highest value and longest horizon. Start with the regulatory deadline corpus — smaller, public, stable — not the contract corpus.
Rides on the same graph. Do not open a second front before the graph is real.
A program that builds everything arrives late and over budget. A program that builds nothing has bought a license and called it a transformation.
parsed instruments
legal descriptions
indexed spans
resolved entities
tract geometry
joined domains
canonical well
allocation keys
grounded rules
agency criteria
volume evidence
priority signal
clean package
submission-ready