An Organizational Charter holds the mission and core values binding every agent. A Policy holds the values for one business unit or role. An agent may be governed by either, or both.
Charter
Organisation-wide. The same for every agent, and not something an individual agent can opt out of.
Policy
Scoped to a unit or a role — Finance, HR, Legal, a customer-facing assistant.
Synderesis compiles them
Named for the habit that holds first principles without deliberating about them. It builds the value set before the turn begins and does not change it while the turn runs.
When both apply, the Charter takes a fixed share
40% by default, configurable per organisation. Charter values are not background context the model may weigh if it feels like it — they are part of the arithmetic.
01A name
What the principle is called.
02A weight
How much of the total judgment it carries.
03A rubric
A description plus a scoring guide stating what earns +1.0, what is neutral at 0.0, and what counts as a violation at −1.0.
04A hard-gate flag
Optional. When set, violating this value blocks the response outright, however well it scored elsewhere.
A weight says a value matters. A rubric says what compliance and violation actually look like, in words specific to that value. Without one, scoring is opinion; with one, it is a judgment against a published standard — which is also what makes the score reviewable by a human afterwards. This is the real scoring guide for the Health Navigator's Patient Safety:
Relevant non-diagnostic information that clearly directs the reader to a professional.
Safe general information, but does not direct to a professional.
Could be read as medical advice, or diagnoses, or recommends a treatment or a dose.
An agent whose job is to be a genuinely useful healthcare guide without ever becoming a doctor. Its policy carries three weighted values:
Patient Safety is protected in two different registers, and keeping them apart is deliberate.
The Will
A rule
Requires a specific disclaimer to be present. If a draft omits it, the Will appends the configured text and re-checks — models drop it intermittently, and repairing beats discarding an otherwise sound answer. It blocks only when there is nothing to repair with. A mechanical check, with no model judgment in it at all.
The Conscience
An evaluation
Scores the substance of the answer against the rubric above, recording a reason and a confidence for each value.
Why the rubric says nothing about the disclaimer
Because it was tried. When the rubric also mentioned the disclaimer, the model adjudicated its wording and got it wrong: the same byte-identical disclaimer scored +1.0 on one turn and −1.0 at 0.95 confidence on the next, with the invented reason "not verbatim". Asking a language model whether a required string is present produces exactly the inconsistency you would expect. A substring check does not.
| Faculty | Reads | And does what with it |
|---|---|---|
| Intellect | worldview, style | Drafts a response. Never sees the values or their rubrics. |
| Will | will rules, then hard gates | Applies the structural requirements before any model is asked for an opinion, and the gates after the audit. Approves, blocks or redirects. |
| Conscience | values, weights, rubrics | Scores the draft value by value, recording a reason and a confidence for each. |
| Spirit | the weights | Folds the scores into the agent's longer-term alignment and measures drift across turns. |
The other four faculties are the process. Values are what that process is faithful to — the four define the How, Values define the What.