TA-14 · ADMISSIBLE COMPUTATION ARCHITECTURE · PUBLIC PROOF
Before compute becomes demand.
TA-14 has demonstrated an operating governance boundary between externally generated candidates and delivery. The point is not that every determination is perfect. The point is that capability no longer has to equal authority.
WHAT HAS BEEN DEMONSTRATED
A governed commitment boundary can operate in front of delivery.
The live Admissible Search Gateway requests a bounded candidate window from an external search provider, freezes exactly what the provider establishes, preserves original rank, evaluates each candidate independently, issues ALLOW / HOLD / DENY / ESCALATE, controls what crosses the delivery boundary, and preserves admitted and non-admitted evidence.
Discovery ≠ delivery authority. Capability ≠ admissibility.
The proof chain
A bounded request establishes the active context.
The provider discovers candidates. TA-14 does not rewrite what was returned.
The established candidate set and original provider rank become preserved evidence.
Each candidate receives an independent bounded determination.
Only candidates with delivery standing cross the boundary.
Both admitted and non-admitted outcomes remain inspectable.
WHY THE FAILURES MATTER
The governor itself is governable.
Public runs have exposed false DENYs, false ALLOWs, lexical limits, identity-binding defects, structured-identifier defects and evaluative-language defects. TA-14 did not erase those records. The historical execution remains intact; the rule can be corrected, versioned and tested again. That is governance-of-the-governor rather than retrospective repair.
THE DATA-CENTER QUESTION
What if the same boundary moved upstream—before computation is committed?
Today's infrastructure work largely asks how to supply, schedule, cool and optimize computation efficiently. TA-14 Admissible Computation Architecture asks an earlier question: what must be true before a computational proposal has sufficient standing to consume resources and create consequence?
If a provider-side pilot can determine that some proposed computation lacks standing before commitment, then downstream computation may be avoided. The measurable research question is not a promised percentage. It is: downstream compute avoided minus the compute cost of the gate.
What this does—and does not—claim
Demonstrated
An external candidate set can be frozen, independently governed before delivery, selectively prevented from crossing a commitment boundary, and preserved with its determination and evidence lineage.
Still to be measured
TA-14 has not established that the Search Gateway reduces a search provider's internal compute, electricity, cooling or water use. That requires a provider-side admissible-computation pilot upstream of computational commitment.
CALL FOR A BOUNDED PILOT
Test admissibility before computational commitment.
TA-14 is seeking data-center operators, AI infrastructure teams, cloud providers and research partners willing to test one falsifiable proposition: can an admissibility boundary prevent unnecessary downstream computation while preserving evidence, authority, changed-context revalidation and measurable consequence?
Primary metric: downstream compute avoided − governance-gate compute.
One page. One proposition. One URL to send.
The Search Gateway is the public reference implementation. Admissible Computation is the larger research proposition.
TEST THE OPERATING GATE →READ THE ACA →