LIVE CLOUD EVIDENCEgemini-structured-extraction · FirestoreRepository · productioncommonsgate-api-00007-wm2 · corr_ae6885de5d314196af8bc6ebbf05f909
CommonsGateRun the proof
Neutral infrastructure for agent-mediated access

Fair access,
even when agents compete.

CommonsGate stops faster, pricier, more persistent AI agents from taking every scarce appointment. Policy decides. Agent power does not.

See 200 agents compete Inspect the protocol
Deterministic core Public replay proof Human review
Manual1×
Free agent1×
Standard3×
Premium8×
ONE PERSONONE CHANCE
ALLOCATION SEALED20 appointmentsPolicy v1.0 · replay verified
Interactive evidence

Don't take our word for it.

Same 200 people. Same policy facts. The only variable is the power of the agent representing them.

LIVE API · SYNTHETIC · 200 PEOPLE · 20 SLOTS
PROOF V1
WHO GETS THE 20 SLOTS?selected / 50 people
ManualOne careful submission
6/50
Free agentBasic automation
6/50
StandardFast monitoring · 3 attempts
4/50
PremiumAlways-on · 8 attempts
4/50
650 incoming attempts450 retries neutralized200 unique people20 policy-selected offers
Capacity never exceededProperty test passed
One person, one chanceProperty test passed
Retries cannot improve oddsProperty test passed
Agent price and speed ignoredProperty test passed
Language excluded from decisionsProperty test passed
Same inputs reproduce outcomeProperty test passed
Agent-swap certificate lab

Measure the risk before changing the queue.

Shadow mode compares the current speed-sensitive process with an agent-neutral charter. The threat lab then executes the controls judges and providers should challenge.

SHADOW MODE · SYNTHETIC

Existing queue comparison

FIFO overall spread0.40
CommonsGate seed mean0.04
Agent-switch change0%
Across 10 committed seed variants
P10 0.04 · P90 0.10

450 retry attempts collapse into one opportunity per person. Small cohorts are suppressed before export.

Run to generate a source-labelled report hashDownload the live agent-swap certificate
ADVERSARIAL EVALUATION

Attack the invariant

Retry floodPrincipal canonicalization

450 retries neutralized

Premium-agent switchAgent-blind manifest

0% outcome change

Capacity overrunHard allocator invariant

20 / 20

Seed substitutionCommit–reveal

Commitment matched

Outcome tamperingIndependent replay

Hash reproduced

Language priority leakFact allowlist

Field absent
Run to execute against the configured APINO CASE DATA
From request to proof

AI helps. It never decides.

Gemini and ADK handle messy human communication at the edge. A small deterministic boundary controls scarce resources.

01

Understand any request

Natural-language intake becomes a source-linked Fair Access Envelope. Low confidence pauses for review.

GEMINI · SCHEMA CONSTRAINED
02

Remove agent advantage

Delegations resolve to one pseudonymous person. Retry volume, speed, price, and language are dropped.

DETERMINISTIC · AGENT BLIND
03

Publish replayable proof

The frozen manifest, precommitted seed, policy version, and outcome hash make the round reproducible.

HASH CHAIN · PUBLIC ARTIFACT
Human review that leaves a trail

Uncertainty cannot become a silent denial.

A reviewer corrects facts—not outcomes. The original extraction is preserved, edits use optimistic versioning, and the deterministic policy reevaluates the request.

Original v1 preserved Reviewer note hashed Policy rerun automatically
CASE · SYNTHETIC-042Needs review
MODEL EXTRACTED · V1Court deadline: unclear
HUMAN VERIFIED · V2Court deadline: Aug 25
POLICY RE-EVALUATEDQualified for round · Priority tier 1
Taskmaster workflow

One steward. A complete background job.

Repeated steward ticks are safe and idempotent. The agent advances time-bound work; policy tools enforce every consequential transition.

  1. 01
    Open intakeProvider-approved charter and seed commitment
    automatic
  2. 02
    Normalize and mergeAny language · one principal · one request
    continuous
  3. 03
    Pause uncertaintyPrompt injection or fact conflicts await review
    human gate
  4. 04
    Freeze and publishCommitted seed · deterministic allocation · public proof
    automatic
  5. 05
    Expire and promoteDeclined offers advance the committed waitlist
    automatic
  6. 06
    Resolve appealsVersioned review · holdback or next-round remedy
    audited
Language is access—not a decision factor

One protocol.
Any language.

Use any BCP 47 language tag supported by the configured Gemini model. The authoritative reason code stays fixed while only approved resident copy is translated.

  • Any script or locale tag at intake
  • Translation has zero allocation authority
  • Failures visibly fall back to English
Resident explanation preview
REASON LOCKED
INCLUDED_IN_ROUND

Included in the allocation round

Your request meets the published intake rules and has one entry in this round. The kind or speed of agent you used does not change your chance.

NextWait for the intake window to close. Applying again will not improve your chance.
Requested hi-Deva-INDelivered ennot run
Public evidence artifact

Verify the round without seeing the people.

No names, requests, principal tokens, or agent identifiers are exposed—only aggregate counts and cryptographic commitments.

SCENARIO · agent-access-200x20-v1

Allocation proof valid

All six safety invariants passed · 450 retries neutralized

PUBLIC · SYNTHETIC
Candidate manifest4fdc947998c2…915b1baf verified
Seed commitmente72f1d53e04f…3b805d12 verified
Outcomeb784864f2805…57449828 verified