/HUMAN.com+ Add /human
▸ 06 DOMAINS · DEEP DIVE

Six relationships. One disclosure.

The /human protocol is organised by who a site is in relationship with — the planet, the commons, the individual, labor, community, and the user. Each domain has concrete, machine-readable fields, and a plain-language version meant to be read by a person.

#DomainRelationshipCore disclosures
01MaterialSite ↔ PlanetEnergy, water, and materials consumed by inference.
02CulturalSite ↔ CommonsIP, consent, disinformation stance, sacred boundaries.
03HumanSite ↔ IndividualNo training on user data. No sale. Portable exit.
04EconomicSite ↔ LaborWho owns models & servers. Who tagged the data.
05RelationalSite ↔ CommunityHealing commitments and give-back mechanisms.
06AISite ↔ UserDisclosure when machine, and a real path to a human.
Domain 01 · Site ↔ Planet

Material

material

narrative · X kWh and Y litres per interaction. Hardware refreshed every Z years.

energy
  • kwh_per_inference
    number

    Electricity used per generated response.

  • grid_renewable_pct
    0–100

    Share of that electricity from renewables.

  • location
    string

    Region(s) where inference physically runs.

water
  • liters_per_inference
    number

    Cooling water consumed per call.

  • stress_basin
    boolean

    Is the data center in a water-stressed basin?

materials
  • hardware_refresh_cycle
    string

    How often hardware is replaced, e.g. 4 years.

  • embodied_carbon_disclosed
    URL

    Link to a lifecycle analysis of the hardware.

Domain 02 · Site ↔ Commons

Cultural

cultural

narrative · We don't train on living artists without consent. We label synthetic media. These things are sacred to us.

ip_and_consent
  • training_opt_out
    one_click | request | unavailable

    How creators exclude their work from training.

  • licensed_content_pct
    0–100

    Share of training data properly licensed.

  • living_artist_policy
    sacred | paid | ignored

    How the system treats living artists.

disinfo
  • synthetic_media_labeling
    required | metadata | none

    How AI-generated media is marked.

  • factual_claim_verification
    third_party | internal | none

    Who checks factual claims.

what_is_sacred
  • domain
    string

    What's protected — e.g. living artists, indigenous knowledge, electoral integrity.

  • commitment
    never_train | pay_only | ask_first | no_commitment

    The promise being made.

  • enforcement
    technical | policy_only

    How the promise is kept in practice.

Domain 03 · Site ↔ Individual

Human

human

narrative · We do not use your conversations to train models. Not for us, not for competitors. You can leave with everything.

my_data
  • trains_competitors
    boolean

    Does my data improve models my rivals will use?

  • trains_your_future_models
    boolean

    Does my data train future versions of this product?

  • sold_without_consent
    prohibited | opt_out | permitted

    Whether behavioural data can be sold.

exit
  • export_format
    json | csv | portable | none

    How users get their data out.

  • export_sla_hours
    number

    Hours from request to delivered export.

  • deletion_completeness
    verified | claimed | partial

    Whether deletion is independently confirmed.

Domain 04 · Site ↔ Labor

Economic

economic

narrative · The taggers who labelled the training data were paid $_/hr in place. They are named in credits.

ownership
  • model_weights
    open_weights | licensed | closed

    How the model itself is distributed.

  • model_structure
    cooperative | public | investor_backed | unclear

    Who ultimately benefits financially.

  • servers
    own | aws_azure_gcp | colo_responsive | community

    Where compute runs.

labor · taggers_annotators
  • hourly_rate_usd
    number | living_wage_local | below_living

    Pay rate for the people who labelled training data.

  • location
    local | global_south | unknown

    Where annotators are based.

  • union_representation
    boolean

    Are annotators collectively organised?

  • named_in_credits
    boolean

    Are annotators credited by name?

Domain 05 · Site ↔ Community

Relational

relational

narrative · X% of compute to climate modelling. Released these weights. Paying these communities restoratively.

healing
  • historical_harm_acknowledged
    string

    What past harm is being named in plain language.

  • restorative_payments
    URL

    Evidence of payments to harmed communities.

  • ongoing_relationship
    boolean

    Continued consultation, not a one-time transaction.

give_back
  • compute_to_ecology
    percent

    Share of compute gifted to ecological work.

  • open_source_release
    string

    What weights, datasets, or tools were released.

  • community_infrastructure
    URL

    What was built for the commons.

  • revenue_sharing_model
    string

    Plain-language description of revenue sharing.

Domain 06 · Site ↔ User

AI

ai

narrative · You're talking to a machine. Reach a real human in two clicks. Runs on-device when it can.

disclosure
  • every_interaction_labeled
    boolean

    Is every AI interaction marked as such?

  • watermarking
    visible | metadata | none

    How AI output is watermarked.

human_contact
  • escalation_path
    two_clicks | phone | email | none

    How a person reaches a person.

  • human_authority
    decision_maker | support_only | script_reader

    What the human on the other end can actually do.

  • sla_to_human
    number

    Time to a human, in minutes or hours.

right_sizing
  • model_size_justification
    string

    Why this model size was chosen.

  • efficient_architecture
    boolean

    Uses MoE, quantisation, or similar efficiency techniques.

  • local_inference_offered
    boolean

    Can run on the user's device instead of the cloud.

▸ what's next

Read the full RFC at the protocol page or generate a valid disclosure with the generator.

Generate my /human page